You are on page 1of 789

Metaphor and metonymy: Making their connections more slippery


Abstract This paper continues the debate about how to distinguish metaphor from metonymy, and whether this can be done. It examines some of the dierences that have been alleged to exist, and augments the already existing doubt about them. The main dierences addressed are the similarity/ contiguity distinction and the issue of whether source-target links are part of the message in metonymy or metaphor. In particular, the paper argues that metaphorical links can always be used metonymically and regarded as contiguities, and conversely that two particular, central types of metonymic contiguity essentially involve similarity. The paper also touches briey on how metaphor and metonymy interact with domains, frames, etc. and on the role of imaginary identication/categorization of target as/under source items. With the possible exception of this last issue, the paper suggests that no combination of the alleged dierences addressed can serve cleanly to categorize source/target associations into metaphorical ones and metonymic ones. It also suggests that it can be more protable to analyse utterances at the level of the dimensions involved in the dierences than at the higher level of metaphor and metonymy as such.

* Correspondence address: School of Computer Science, The University of Birmingham, Birmingham, B15 2TT, UK. Tel: (44) (0)121 414-3816; Fax: (44) (0)121 414-4281. E-mail: Acknowledgments: This research was supported in part by grant EP/C538943/1 from the UKs Engineering and Physical Sciences Research Council (EPSRC), and grant RES-328-25-0009 from the UKs Economic and Social Research Council and EPSRC under the People at the Centre of Communications and Information Technologies programme. I am grateful to colleagues Rodrigo Agerri, Sheila Glasbey, Mark Lee and Alan Wallingon, and to the journal editors and anonymous reviewers, for important suggestions. Cognitive Linguistics 211 (2010), 134 DOI 10.1515/COGL.2010.001 09365907/10/00210001 6 Walter de Gruyter

J. A. Barnden metaphor; metonymy; metaphor/metonymy distinction; contiguity; similarity; part/whole metonymy; representational metonymy; resemblance metaphors; image metaphors.




Specifying the nature of metaphor and metonymy has long been a dicult problem. It has been particularly dicult to specify convincing grounds for dierentiating the two gures from each other (Barcelona 2000a; Cameron 1999a,b; Dirven and Porings 2002; Fass 1997; Haser 2005; etc.). The literature exhibits a wide variety of opinion. In this paper we look mainly at two important alleged grounds for dierentiation, namely (i) that metaphor involves similarity whereas metonymy involves contiguity or related notions of semantic/pragmatic connection (see, e.g., Dirven 2002; Jakobson 2002 [1956]; Lodge 1977; Norrick 1981; Nunberg 1978; Riemer 2001; and many others), and (ii) that metonymy preserves links to the source domain item as part of the message whereas metaphor does not (e.g., Dirven 2002; Haser 2005; Warren 2002). We briey consider several other issues, including the question of whether metaphor and metonymy interact dierently with some postulated structure of domains, frames, idealized cognitive models, or other compartmentalizations of conceptual information. The paper will conclude that these various possible grounds for dierentiation do not, as currently conceived at any rate, provide a rm distinction between metaphor and metonymy. This failure holds even if the putative grounds are combined rather than considered in isolation. The conclusion supports a similar one by Haser (2005: 15), but the paper adds qualitatively new evidence and critique. It leaves it open whether more careful accounts of the alleged dierences could lead to a crisp metaphor/metonymy distinction, or whether additional dierences could help. Our considerations will also not stop metaphor and metonymy having some tendency to dier in particular ways; with, for instance, metaphor tending to involve a rich form of similarity but metonymy tending not to. There have been various notions, in the literature, of a cline or spectrum of phenomena incorporating metaphor and metonymy. Radden (2002) makes one such proposal. Dirven (2002) discusses a phenomenon of post-metonymy, intermediate between metaphor and metonymy, although Riemer (2002) argues that (his own version of ) post-metonymy need not lead in the direction of metaphor. Croft and Cruse (2004: 220) give examples that suggest intermediate possibilities between metaphor

Metaphor and metonymy

and metonymy, while also warning that what may appear to be intermediacy may be the result of combining distinctly dierent processes. The present article regards the idea of a spectrum as broadly being on the right track, but pushes the idea further. It doubts that we should think of just a one-dimensional space of possibilities: rather, there are several and perhaps numerous dierent ways in which metaphor and metonymy can vary in their essenceincluding variance as regards amounts and types of similarity and contiguity. The positioning of metaphor and metonymy within the space created by these dimensions and others may be very complex. We will come back to the question of whether there may be intermediate possibilities between metaphor and metonymy in Section 4. Intermediacy is the question whether there are phenomena that have some of the qualities of both metaphor and metonymy but do not qualify as either. For now we focus briey on the contrasting notion of overlap of metaphor and metonymy. Overlap is the question of whether there is a phenomenon that at one and the same time qualies as being both metaphor and metonymy. We will see that there is evidence of a type of overlap that is distinct from a type that has already been much discussed in the literature. The latter type consists of the sorts of mixing of metaphor and metonymy that have been discussed under the headings of metaphor within metonymy, metonymy within metaphor, chaining of metaphor and metonymy, and so forth (Fass 1997; Goossens 1990; Kovecses 2002; Lan glotz 2006; Ruiz de Mendoza Ibanez and Dez Velasco 2002; Warren 2006). Those types of mixing involve some conceptual item A being linked to some item X, and X to B, where either the AX link is metaphorical and the XB link is metonymic, or vice versa. Rather, the type of mixing or overlap addressed in this paper is where an item A is linked to an item B in such a way that qualies simultaneously as both metaphor and metonymy, but where the situation is not analysable as a chaining of an AX link and an XB link. This simultaneity is also to be distinguished from alternativity: the possibility of alternative interpretations of an utterance taking the AB link to be just metaphorical or just metonymic. Ritchie (2006: 156) says, in referring mainly to metaphor, gurative use of language may itself constitute a eld of meaning, with dimensions such as conceptuality, opaqueness, literalness, triteness, formality, folkishness among others, and Cameron (1999b) has provided nine dimensions on which metaphor can vary. Many dimensions have been important elsewhere in the literature, such as aptness, vividness, memorability, imageability, evaluativeness, persuasiveness, literariness, social divisiveness/cohesiveness, entrenchedness and cultural specicity. But

J. A. Barnden

with the probable exception of literalness, such dimensions arguably do not aect decisions as to whether metaphor or metonymy is involved. The present article is more concerned with dimensions which could be said to be genuinely constitutive of metaphor and metonymy. Metaphoricity and metonymicity are, arguably, language-user-relative in a deep way. They are aected by such things as the particular lexicon, encyclopaedic knowledge, and interconceptual relationships held by a particular language user (whether utterer or understander). Thus, in principle, an expression should not be said to be metaphorical or metonymic in any absolute sense, but only for a particular user. Of course, in practice, many expressions may be metaphorical or metonymic for the vast majority of native users of a language, and the way in which expressions are metaphorical or metonymic may also be the same or similar across such users (e.g., involve the same conceptual metaphor such as LOVE AS JOURNEY, or the same metonymical schema such as CONTAINER FOR CONTENTS). Relativity has been pointed out by various other authors (Cameron 1999b; Dirven 2002; Geeraerts 2002; Norrick 1981; Pragglejaz Group 2007; Radden 2002; Radman 1997; Riemer 2002; Ritchie 2006; Ruiz de Mendoza Ibanez 1999). We will not be exploring it in this article, but we need to distinguish it from the issue of the dierentiation of metaphor and metonymy. User-relativity does not necessarily imply that metaphor and metonymy cannot in general be cleanly dierentiated or that, for a given user, particular cases of metaphor and metonymy cannot be cleanly dierentiated; and, conversely, a lack of a clean dierentiation between the notions of metaphor and metonymy does not necessarily imply user-relativity. Because of its aims, this paper does not rest upon any particular denition of metaphor or metonymy, but instead on other authors claims for metaphoricity or metonymicity of examples used, or on the present authors judgments of how particular examples would be classied in the eld. The paper does nevertheless embody a cognitive assumption, in viewing metaphor and metonymy as being largely to do with cognitive representation and processing issues as opposed to the surface form of utterances. (In the case of lexicalized metaphor or metonymy the representation and processing may have occurred in the past, and thus merely be part of etymological motivation). To the extent that metaphor and metonymy are matters of processing, the issues in this paper amount to questions such as the extent to which the processing creates or traverses similarity and/or contiguity links between conceptual items. However, the word link will be meant in a very general and theoryneutral way. Links certainly include those between target aspects and source aspects that are proposed in an explicitly mapping-based account

Metaphor and metonymy

such as Conceptual Metaphor Theory, for instance when the participants in a love relationship are linked to occupants of a travelling car. But the notion is broader: for example, even though the sentence My job is a jail is analysed in the class-inclusion approach (Glucksberg 2001; Glucksberg and Keysar 1990) in terms of a category of entities that includes both the job and physical jails, we will still describe it as implicitly involving a link between the job and either a hypothetical physical jail or the general category or concept of physical jails, even if no direct link is established in the understanders mind or proposed in the theory. This way of talking is just in the service of having a uniform way of describing the fact that, in metaphor, at least one target item is explicitly or implicitly, and directly or highly indirectly, associated with at least one source item. Although the class-inclusion theorys attention is on the idea of class inclusion as such, our attention is on the nature of the implied link between target item and source item, e.g., between the job and the (hypothetical) physical jail, or between the job and the physical-jail category. This link could, for instance, be regarded as a similarity link based on possession of whatever property or set of properties are held to dene the superordinate category under which both target item and source item are placed. In the job/jail example, the similarity might consist of both the job and a physical jail being constraining and dicult to escape. (Care is needed here to postulate suitably abstract notions of constraining and escape, in light of the circularity objections raised by Ritchie (2006) and others). Also, a link can take a collapsed, degenerate form: namely, that of an imaginary identication. For example, if in a blend space (Fauconnier and Turner 1998; Turner and Fauconnier 1995, 2000, 2002) one entity is identied with another, that identication is itself a link in our sense. And if the entities come from input mental spaces, the corresponding entities in those spaces can also be said to be linked indirectly with each other via the identication link in the blend space. Finally, to the extent that metaphor and metonymy are in part processing issues, we need to keep in mind the possibility that links might only be classiable as metonymic or metaphorical by taking into account the way the links are used in processing rather than or as well as any other aspect of their nature. The structure of the rest of the paper is as follows. In Section 2, we mainly consider whether a distinction between metaphor and metonymy can be found in a distinction between similarity and contiguity. In Section 3 we look at the extent to which source/target links are themselves kept as part of the messages conveyed by metaphorical or metonymic utterances. Section 4 briey examines two further possible grounds for dierentiation, namely the interaction with conceptual compartments such as domains and frames, and the role of imaginary identication/

J. A. Barnden

categorization of target items and/or source items. Section 4 also comments further on intermediacy and overlap of metaphor and metonymy, and briey considers claims that (some) metaphor can be viewed as double metonymy. Section 5 concludes, and explains how it can be protable to analyse utterances at the level of the dimensions of variation discussed in the article (similarity, etc.), as opposed to the higher level of metaphor and metonymy as such.


Similarity versus contiguity?

A traditional view has been that metaphor is a matter of similarity between source and target items, and metonymy a matter of contiguity between them (Dirven 2002; Feyaerts 2000; Jakobson 2002 [1956]; Lodge 1977; Norrick 1981; Ullman 1962). Haser (2005) provides a review. There is an enduring intuition that in metonymy the source and target item are related in some salient and easily accessed way, making the metaphor of spatial contiguity reasonable. In this sense, a composer is contiguous to his/her music, a time period to an important event occupying it, etc. Contiguity also includes, of course, more physical cases, as of a bottle being contiguous to its contents. Contiguity is meant to have no whi of likeness. But many authors have noted the slipperiness of the notions of similarity and contiguity (e.g., Chiappe 1998; Cooper 1986; Dirven 2002; Haser 2005; Riemer 2002). Norrick (1981: 27) says that the line between principles of similarity and of contiguity is at times fuzzy (although he does not go on to explore this fuzziness). The slipperiness of the notions compromises their ability to dierentiate metaphor from metonymy. In addition, metaphor can impose similarity rather than resting entirely on already noticed similarity (Black 1993 [1979]; Haser 2005; Indurkhya 1992). Consider an utterance that metaphorically casts a particular cloud as a camel, such as The camel was high up in the sky. For speaker or hearer, the cloud might initially only have a slight visual similarity to a camel; but the act of using the metaphor causes speaker and/or hearer to view the cloud much more as a camel. The structure of a camel is imposed on the cloud, giving the latter a structure that would not, or could not as easily, have been discerned otherwise, and may be partly articial. Eects can include the division of a part of the cloud into subparts in a non-obvious way, or conversely the agglomeration of two distinct parts of the cloud into one undierentiated part from the point of view of camel shape. Analogously, viewing a marriage as a business may cause one to add a structure to marriage that one had not previously perceived.

Metaphor and metonymy

But we will see that similarity and contiguity are not as distinct as is assumed even by previous critics of them as a basis for dierentiation. The argument is in two parts. First we argue (in Section 2.1) that there appears to be no bar to viewing metaphorical linkage between source and target in metaphor as a special type of contiguity. Thus, if or when metaphorical linkage amounts to similarity, similarity is a special type of contiguity. Secondly, we argue a partial converse to this (in Sections 2.2 to 2.4): that certain familiar forms of contiguity involve similarity in an essential way, where moreover this similarity can sometimes be akin or even identical to the similarity underlying some metaphor. The discussion takes contiguity to include not just relations that have specically been labelled as such but also other relations such as pragmatic functions (Barcelona 2002, 2004; Nunberg 1978). The sets of relations discussed under the heading of contiguity and those under the heading of pragmatic function are very similar. In eect the discussion regards any relationship that has been held to be metonymic as a possible type of contiguity. Similarity can at one extreme be a matter of sharing some simple features and at the other a matter of a complex structural analogy. We also need to keep in mind a broad distinction in the ways in which things can be similar. A road can be similar to a snake in virtue of the shape of each, and one plan of action can be similar to another in virtue of their structure, similarity of individual steps, etc. These are examples of two things being intrinsically similarsimilar because of their own natures, independently of relationships to other entities outside themselves. On the other hand, in the context of metaphorically casting an organization as a solar system, the term planets could describe major employees even though there is no (relevant) intrinsic similarity between a planet and an employee. Relative to the overall similarity between the organization and a solar system we can say that employees and planets are extrinsically similarsimilar because of their relationships with other things taking part in an overall structural analogy. Of course, two things may be similar because of some mix of intrinsic and extrinsic similarity. We will not assume that all metaphor is necessarily based on similarity. In particular, some metaphors have been held to be based on experiential correlations between source and target (Barcelona 2000b; Grady 1997; Kovecses 1990; Radden 2002). For example, an experienced correlation between seeing and knowing (in that seeing can lead to knowledge) may be at the root of a metaphorical view of KNOWING AS SEEING. Now, whether this metaphorical view, once created, is a matter of similarity is a contentious matter. For instance, in one such case (SADNESS IS DOWN), Barcelona (2000b) claims that the metaphor involves similarity,

J. A. Barnden

but Haser (2005: 44) throws doubt on this. Fortunately, we do not need to adjudicate the matter in this paper. The argument of Section 2.1 actually applies to correlation-based metaphorical links as well as to similarity links; and the arguments of Sections 2.2 to 2.4 are not aected in their nature or signicance by the presence or otherwise of non-similaritybased metaphor. Hence, the remainder of the paper will refer both to similarity-based and to correlation-based metaphor, but will leave open the possibility that some or all of the latter is also within the former. One move that might be attempted is simply to disallow similarity from the class of associations called contiguity (Feyaerts 2000, following Ullman 1962). Then, of course, one would get a crisp distinction between (similarity-based) metaphor and metonymy. However, such a move seems unprincipled and to be made just to save the distinction, as opposed to examining the phenomena to see what the useful distinctions are, if any. So, we will assume that the notion of contiguity does not in itself contain a stipulation against similarity. 2.1. Metaphorical linkage as contiguity

Basically, we ask why metaphorical links shouldnt themselves be regarded as contiguities. The fact that (at least) similarity links in metaphor are not normally regarded as contiguity links, or the fact that authors uncritically portray similarity and contiguity as being dierent types of relationship, is hardly a valid answer to the question, as there is no accepted denition of how broad contiguity is. The term contiguity is in itself highly metaphorical and susceptible to a wide range of interpretations, as has often been observed. It is thus perhaps surprising that the question of metaphorical linkage counting as contiguity has not been raised more often. Consider the widespread phenomenon of referential metaphor. Typically, referential metaphor is said to occur when a denite noun phrase is used metaphorically to refer to some target item, as in (1), from Gibbs (1990): (1) The creampu didnt even show up.

A boxer in the context is being metaphorically viewed as a creampu and is being referred to by the phrase The creampu. Another, more mundane and conventional, example would be They have reached the third milestone on the project using the phrase the milestone to refer metaphorically to an important, planned event in the project. Thus, assuming that underlying (1) there is some postulated similarity link between the boxer in question and a hypothetical creampu (in the literal sense), we can use

Metaphor and metonymy

this link to achieve indirect reference to the boxer (target item) via direct reference to the creampu (source item), just as we can use an alleged contiguity link in a metonymy to achieve indirect reference to a target item via a direct reference to a source item. Referential metaphor can also use correlation-based metaphorical schemata. Consider the passage: Susan sank into a pit of sadness. She stayed at the bottom for many months. Let us assume for the sake of argument that the passage is to be analysed using the SADNESS IS DOWN metaphor and that this is correlation based. Then in the second sentence the phrase the bottom is a referential metaphor for the worst phase of her sadness state. So, is there anything about metaphorical links that should prevent us from regarding them as a special case of contiguity links, at least when they are being used in referential metaphor? One sharpened version of this question could be: If contiguity links in general are salient semantic or pragmatic associations or salient applications of pragmatic functions, is there anything about similarity or correlation-based links in referential metaphor that should prevent them from qualifying as contiguities along with other types of salient association/function? Before going on we should dispose of one alternative to an assumption made a moment ago: the assumption that the phrase The creampu in (1) refers to a hypothetical literal creampu. One might argue instead that while creampu in the noun phrase does refer to the category of literal creampus, there is no act of postulating a member of that category: rather, the noun phrase acts much as if it had been The person who is, metaphorically speaking, a creampu using a creampu purely predicatively. (This would be consistent with a class-inclusion account of metaphor). However, we can still say that there is an (alleged) similarity between the boxer and (literal) creampus in general, or a similarity relationship between our concept of the particular boxer and the general concept of (literal) creampus. Our question would then become: Is there anything about this similarity that disqualies it from being a type of contiguity? For simplicity, in the following we will stick to the assumption that the phrase The creampu in (1) does refer to a hypothetical literal creampu, on the understanding that the discussion could be adjusted to t category-based accounts. Another distinction to note before going on is that the issue of whether correlation-based metaphorical links can be regarded as contiguities is dierent from the issue of whether the original correlations themselves are so regarded. If Susans state of sadness does not in fact cause any physical downness (e.g., drooping body), but is metaphorically cast as an imaginary physical downness, then the fact that some sadness states


J. A. Barnden

can cause, and therefore be contiguous to, downness does not force us to consider Susans sadness and the imaginary downness to be contiguous. However, it is certainly possible to allow the notion of contiguity to encompass potential as opposed to actual causation, in which case the metaphorical link would be one of contiguity, whatever else it might be. However, we do not rely on this argument in the following. Some grounds could be imagined for trying to maintain that metaphorical links, whether similarity-based or correlation-based, should not qualify as contiguity links. We will treat these in turn and argue against them. Naturally, there may be grounds beyond those considered here. First, it might be claimed that metaphorical links are more a matter of (possibly culture-wide) mental imposition upon the world than are the contiguity links in generally recognized forms of metonymy. Or, to paraphrase, perhaps metaphorical links are much more in the mind whereas metonymic links are much more a case of reecting what is objectively in the world. Thinking of a love relationship as a physical container is arguably more an imposed, mental view than regarding a physical container as being related to its physical contents.1 However, it is dicult to sustain a rigid contrast in general. As Dirven (2002) says, contiguity is itself to some extent partly in the eye of the beholder, and Norrick (1981) takes a similar view. It is partly a mentally, socially and culturally constituted matter that, for instance, a particular group of people is the football team representing Finland, and yet the word Finland can metonymically refer to the group, as in Finland lost the match. The situation is similar for many other types of metonymy where the source item plays some sort of social or political role with respect to the target item or vice versa, as in Bush attacked Iraq with Bush as source item and the USA or the US military as actual attacker. The Representational metonymies to be discussed in Section 2.2where, say, a pictorial image in a painting is used to refer to the depicted object, or vice versainvolve a mentally imposed representational link; also, Goodman (1968) argues that there are conventional and stipulative aspects to the way that paintings, etc. do their representing. Finally, in the celebrated example of using the phrase The ham sandwich to refer to the restaurant customer who ordered the sandwich (Nun-


Of course, we might claim that the structuring of the world into, say, containers and contents is itself a mental imposition, and not part of the objective, real world. If that is factored in, then the distinction being drawn is between the real, objective world as naively conceived to exist in common-sense as opposed to what even common-sense would concede is imposed on the world by minds.

Metaphor and metonymy


berg 1995), the act of ordering something in a restaurant only makes sense given a suitably constituted socioeconomic culture, one where certain discourse acts are regarded as constituting ordering. Furthermore, we can argue that many metaphorical similarity relationships do exist in the world, or arise objectively from it, just as much as many metonymic relationships do. This is clearest when the metaphor rests on a complex structural analogy such as that between an army and a society of army ants (example from Goatly (1997: 163)) or that between a commercial company and a solar system. There is a sense in which the partial isomorphism of structure really exists. It is a mathematical aspect of the world that exists just as much as a simple, familiar mathematical object such as the number 9 does. And, the partial isomorphisms exist just as much as the link between, say, the date 11th September 2001 and certain terrorist events does. So, given that dates and events are used metonymically for each othere.g., the (abbreviated, US-style) date 9/11 for some terrorist eventsand given that their relationship is a contiguity, it seems articial not to regard the abovementioned analogical links as contiguities. Secondly, perhaps contiguity and metaphorical linkage could be distinguished on grounds of structural correspondence (Barcelona (2004) provisionally suggests this). Perhaps similarity in metaphor involves a correspondence of some structure between source and target whereas contiguity does not: contiguity just relates two wholes that it leaves unanalysedso that their intrinsic similarity or otherwise is not an issueand whose relationships to other things are irrelevantso that extrinsic similarity is not an issue either. Now, similarity-based metaphor can indeed be seen to involve at least some small amount of structural correspondence. Even when the similarity consists only of the source and target items having one corresponding feature, such as perhaps some sort of weakness in the boxer/creampu case, we can see a structural correspondence: the boxer corresponds to the creampu, the weakness on the target side corresponds to the weakness on the source side, and the boxer having the former weakness property corresponds to the creampu having the latter weakness property. But, the trouble is that some metonymy involves structural correspondence as well, as we will see in Sections 2.2 and 2.4, where the structural correspondence is at least as rich as the minimal sort found in the boxer/creampu example of metaphor. As for correlation-based metaphor, if this does not (always) involve structural correspondence, then, given that metonymy does not usually involve structural correspondence, structural correspondence cannot be used as a metaphor/metonymy dierentiator. If on the other hand (some) correlation-based metaphor does involve some structural correspondence,


J. A. Barnden

as claimed by Barcelona (2000b), then we are back to the point that so does some metonymy. Thirdly, perhaps contiguity links should be restricted to associations that are conventional or rmly established (cf. discussion in Haser (2005: 22)). But even if this is correct it will not work to provide a distinction with similarity in metaphor, because most similarity in metaphor is highly conventional or rmly established, and the metaphorical links of correlation-based metaphor are also rmly established. Finally, it might be claimed that in metaphorical similarity there is no real source-side entity corresponding to the target-side entities, whereas in metonymy there is. For example, metaphorically casting a person Richard as a lion does not involve a particular, real lion, whereas metonymically referring to some real artworks via an artist does involve the artist being real as well. However, some core types of metonymy are open to having merely hypothetical source items. In a conversation in a library about the location of books about certain topics, we can say Santa Claus is on the top shelf meaning that books about Santa Claus are there, just as we can say Alexander the Great is on the top shelf or Car engines are on the top shelf with analogous intent. Equally, we can say Santa Claus is in the left-hand part of the picture, with Santa Claus being a case a Representational metonymy (Section 2.2). We can also similarly say Lions are on the top shelf and Theres a lion in the left-hand part of the picture without assuming that any particular real lion is discussed or depicted. Conversely, a metaphor source item can be real, as in Singapore is the Britain of the Far East (example quoted by Wee (2006) and following a common pattern of using a well-known existing entity as a metaphor for another entity). The conclusion so far is that there is nothing to stop us regarding the metaphorical links traversed in (at least) referential metaphor as special cases of contiguity. However, our claim is not restricted to referential metaphor as normally conceived, i.e., as being about metaphoricallyused denite noun phrases, any more than metonymy is conned to denite noun phrases. Rather, whenever a metaphorical link is used for accessing something in the target via something in the source, irrespective of the surface linguistic forms involved, we can claim the link is being used as a type of contiguity just as much as we can in standard referential metaphor examples such as (1). For one thing, our discussion would not be essentially changed if (1) had used an indenite noun phrase, as in Some creampus didnt even show up. More distantly, consider the common use of the phrasal verb eat up to refer to commercial taking-over, as in a sentence of form Company A tried to eat up Company B. We can say that there is a hypothetical act of physical eating-up that is conceptually

Metaphor and metonymy


contiguous to the real taking-over, just as much as we can say that in the case of (1) there is a hypothetical creampu that is conceptually contiguous to the real boxer. 2.2. Contiguity involving similarity, 1: Representational metonymy

In this subsection and the next (2.3) we will be arguing that two (salient) types of contiguity can be viewed as involving similarity. This point is a partial converse to that of the previous subsection (2.1). Representations (things that represent) and their representatees (the things they represent) are often used to stand for each other in metonymy. Thus we have REPRESENTATEE FOR REPRESENTATION and REPRESENTATION FOR REPRESENTATEE. These are both covered by Warren (2006), but with dierent names from ours. We will use the term Representational metonymy to cover both directions of the metonymy. Warrens examples of REPRESENTATEE FOR REPRESENTATION include Ari painted a tanker (quoted by Warren from Fass (1997); a tanker is the metonymic source phrase). Let us assume that in context the sentence means that Ari painted a picture of a tanker, or a picture of various things including a tanker. Here the source REPRESENTATEE is the (possibly imaginary) tanker and the target REPRESENTATION is either the picture as a whole or the image of the tanker in the picture. Other examples would be Theres a tanker in the left hand side of the picture and Tony Blair is on the left hand side of the photo, both of which make it more explicit that a physical representation is being indirectly referred to (by the phrases a tanker and Tony Blair respectively). But the REPRESENTATION need not be a visual representation of the REPRESENTATEE. For instance, it can be an acoustic representation, as in . . . Knechts symphony begins amid beautiful countryside, but, rather more rapidly than in [Beethovens Pastoral Symphony] a storm approaches, breaks and fades away . . . (emphasis added).2 It is clear from the surrounding text that what is under discussion is an acoustic image of a represented storm. (This is worth noting because the above quotation, taken by itself, bears a possible contrasting interpretation in which storm is a metaphorical description of non-representational musical events in the symphony). And the REPRESENTATION in a Representational metonymy need not be any type of perceptual representation, as it could be an idea in someones mind and could concern something

2. From notes on the Pastoral symphony by Jan Smaczny for City of Birmingham Symphony Orchestra, England, April 2000.


J. A. Barnden

non-perceptual, as in Sallys disappointment was at the back of Johns mind all day long. Here the phrase Sallys disappointment metonymically refers to some IDEA OF Sallys disappointment (which is itself being metaphorically viewed as a physical object in Johns mind, which is metaphorically viewed as a container). An example of the reverse metonymic pattern, REPRESENTATION FOR REPRESENTATEE, is: In Goldnger Sean Connery saves the world from a nuclear disaster (Warren 2006: 48), with actor Sean Connery himself or the moving images of him in the lm as the REPRESENTATION and character James Bond as the REPRESENTATEE. It is of course the (ctional) person James Bond who saves the (ctional) world. Now, it is certainly true that even when the REPRESENTATION is a visual item or physical object (e.g., an image or an actor) and the REPRESENTATEE is a physical object, the REPRESENTATION need not bear any signicant visual resemblance to the representatee. In the sentence The town is on the left hand side of the map the visual representation of the town could be just a small black dot, and thus have little resemblance to the actual appearance of the town. However, we are concerned henceforth with the prominent special case of Representational metonymies where the representation is indeed based, at least in part, on some substantial sort of intrinsic perceptual similarity, and will take the visual-similarity subcase of Representational metonymy as particularly salient. In situations involving photos, pictures and the like the representation relationship is normally based at least in part on visual similarity, though other factors such as convention and stipulation can also be present (cf. Goodman 1968). Naturally, the visual similarity often involves a high degree of structural correspondence. Clearly, if we are to claim that metonymic links in general are contiguities, then we must admit that in similarity-based Representational metonymies the contiguity between source and target happens to take the form (at least in part) of similarity. Moreover, the similarity is central to the metonymy, not some incidental feature of it. These obvious points have not been given due weight in discussions of the metaphor-metonymy distinction. Perhaps this is because examples of Representational metonymy are rare in the metonymy literature, according to Warren (2006). This investigative rarity, however, does not do justice to the prevalence and ordinariness of the phenomenon. Warren (2006: 49) herself seeks to distinguish the type of similarity involved in similarity-based Representational metonymies from the type of resemblance involved in metaphor. She says that the former simply involves matching the REPRESENTATION with the REPRESENTATEE whereas in (many) metaphors the resemblance involves prop-

Metaphor and metonymy


erty selection and adaptation. These claims of Warrens are disputable. First, she does not explain what this matching in the metonymic case amounts to. Secondly, the visual similarity in the metonymic case will almost always involve signicant property selection (except in the most photographic of depictions) and often some measure of adaptation (e.g., colours may be intensied, shapes simplied). Thirdly, she only says many metaphors. Fourthly, other authors have regarded the resemblance relation that visual depictions have to what they depict as usable in metaphor: for example, such resemblance is an example of one of Norricks (1981) iconic semiotic principles, and this general semiotic principle induces a metaphoric principle in the special case of language.3 Finally, it is dicult to see any fundamental dierence between the types of visual resemblance possible in the metonymies and those possible in imagebased metaphors (Lako 1993; Lako and Turner 1989). In image-based metaphorsalso sometimes called resemblance metaphors, though this is too vague a termtwo physical objects or images are put into a metaphorical relationship on the basis of their visual appearance (which can include motion), examples being the road snaked through the desert, where the word snaked uses an image metaphor to describe the shape of the road, and The rock that saved him was lathered and fringed with leaping strings of foam (quoted by (Goatly 1997: 271)) where the word lathered uses an image metaphor to describe the foamy rock and the words leaping and strings set up further image metaphors to describe the foam itself. It could be argued (and perhaps this is Warrens point) that in image metaphor one needs to understand just how the two things are similar in order to understand the utterance, whereas in Representational metonymy one can simply take it on trust that there is a similarity or some other sort of representational connection. The claim would be that to understand, say, what it is for a road to snake we need to think about what a moving snake looks like; whereas to understand Theres a snake in the left hand side of the picture we only need to know that there is some subimage or other in the picture that is intended to depict a snake. However, this dierence is at most a matter of degree and particular circumstances, and does not support any sharp distinction between Representational metonymy and image-based metaphor. In the case of Tony Blair is on the left hand side of the photo, could we be said to have properly understood the utterance without understanding something about how a representation in a photo could be similar to Tony Blairs actual appearance,

3. However, Norrick (1981) does not discuss Representational metonymy.


J. A. Barnden

so that the understanding involves more than just realizing that some similarity exists? Conversely, in the case of The road snaked through the desert we do not, for most purposes, need to have more than a vague idea of the bendiness of the road. It is not clear that the visual similarity considered by the understander is more detailed than the visual similarity the understander needs to consider in the Blair photo case. In response one might claim that in the Blair photo case, the act of metonymy as such is a very bare one, consisting simply in assuming that there is some visual representation of Blair in the photo, and does not require knowledge of what Blair looks like; but understanders perform an additional, pragmatic inference that that representation is probably a shape that looks like Blair, and if they happen to know what Blair looks like they can esh this looks-like relationship out. However, we could counter this with the parallel claim that in a sentence like The road snaked across the desert, the act of metaphorical understanding is a very bare one, consisting simply in assuming that there is some visual similarity between the road and a snake, and does not require knowledge of what a snake looks like; and there is an additional pragmatic inference that that similarity is probably one of physical shape, and if understanders happen to know what a snake and its movements look like they can enrich the similarity. Ultimately, the point is that in both Representational metonymy and image-based metaphor we have a relationship of similarity; the degree to which that similarity is apprehended by the understander is a matter of how rich an understanding the understander comes to, irrespective of whether we have a case of metonymy or a case of metaphor. Also, note that even if a town is represented by merely a small dot on a map, so that there is little or no intrinsic similarity, the spatial relationships within the map itself between the dot and other representations will be structurally analogous to the spatial relationships of the town to the representatees of those other representations. Thus, just as in metaphor, it is often the external associations of source and target items, rather than their internal structure, that are important in source/target similarity, so that the similarity is largely extrinsic to any given source/target link. In Representational metonymy the external associations are often an implicit part of the representational relationship between a particular REPRESENTATION and a particular REPRESENTATEE. Analysed in this way the map could be said to be a metaphorical representation of the geographical region in question, and the metaphorical links between town-denoting dots on the map to real towns are similarity links used as contiguities, on the lines of Section 2.1. It is just that we are now arguing that something we started o assuming was a contiguity link is also a similarity link like those used in a type of metaphor, rather than arguing as

Metaphor and metonymy


we did in Section 2.1 that a similarity link in general can also be used as a contiguity link. It is not just that it is dicult to put a wedge between the general nature of similarity in Representational metonymy and that of similarity in image-based metaphor. Rather, we can even argue that the just the same particular similarity can operate in both metonymy and metaphor. On the metonymy side, consider the following, where the label 2my is short for metonymy case of 2 (2my) Theres a snake on the left-hand side of the drawing.

referring to a wavy line in the drawing that is intended to depict a snake. On the metaphor side, consider the following, where 2or is short for metaphor case of 2 (2or) Theres a snake on the left-hand side of the drawing.

as a way simply of describing a line in a drawing, where the line is not intended to depict a snake (it might depict something unconnected with snakes). (2my) and (2or) are deliberately the same sentence, but involved in dierent utterance events and we use dierent labels for ease of reference. Suppose the snake lines in the two drawings are identical. Then both examples involve exactly the same type and degree of visual snake/line similarity, and it is likely that there is no other type of similarity in play (such as would arise if, say, in one of the examples the line were drawn in ink containing snake bile!). This raises the possibility that we should classify at least some Representational metonymy that relies on similarity as also simultaneously being metaphor. But we should also wonder whether the contiguity in Representational metonymy is more than just the similarity. We might argue that in Representational metonymy we also have the extra feature that the similarity is being used in a particular way, namely in an act of representation, whereas metaphorical source items do not represent corresponding target items. But there is a case for suggesting that metaphor itself does involve a representation relationship: that source-domain items in metaphor do represent corresponding target-domain items, in the mind of someone using the metaphor. For example, in (2or) an imagined snake could be said to represent the line. Or, the persons idea of a snake could be said to represent the line (in the special source-to-target sense envisaged), as well as representing a snake in the ordinary way. It all depends on one what one means by the word representation. Whatever the relative merits of these arguments as to whether Representational metonymy has a feature beyond similarity that metaphor does


J. A. Barnden

not have, the point holds that a dierence between metaphor and metonymy cannot solely be based on metaphorical and metonymic links always diering as to whether similarity is involved or even on what forms of similarity they involve. 2.3. Contiguity involving similarity, 2: Partitive metonymy

We now turn to a very dierent type of contiguity, namely the contiguity involved in WHOLE FOR PART and PART FOR WHOLE metonymy. We will use the term Partitive contiguity or metonymy to cover both directions. We will see another important way in which contiguity can involve similarity. Similarity will be highly relevant to the way that some Partitive metonymy works, and in these cases we will say that the PART and WHOLE are relevantly Partitively similar or that the PART is relevantly whole-similar for short. Relevant Partitive similarity arises when the WHOLE and the PART to some extent share particular features that are important in the motivation for the metonymy. This applies, for example, in the traditional metonymic use of hand to mean a sailor. Important (traditional) functions of sailors, such as grasping a rope, are performed partly by their hands. So, there is a sense in which a sailor and his/her hands are functionally similar to some degree (of course, what the hands do in grasping is only part of what the whole body does in grasping). It is precisely this partial function sharinghence, partial functional similaritythat motivates the metonymy. But we can go further: to a degree, the whole person has the function in question because of having a part that has that function or an approximation to it. The parthood is central to the similarity, and the similarity is central to the signicance (in context) of the parthood. We are not concluding that, because some sort of similarity is involved, metaphor is therefore involved. This would prejudge the question of whether metaphor restricts the type of similarity, or involves more than just similarity. However, it is worth mentioning here the like test for distinguishing whether two items A and B have a metonymic or metaphorical relationship (Gibbs 1999). The core intent of the test is presumably to test for (metaphor-supporting) similarity. According to the test, if saying A is like B is appropriate in the context then we have a case of metaphor. But it might be thought odd to say that Sailors are like their hands or The hands are like the sailor, even in a context where sailors use of their hands is salient. However, if anything this shows the dubious validity of the test for our current purposes. The conditions under which we might judge any given form of words like Sailors are like their hands to be appropriate will be aected by many factors. Also, the test is unfair

Metaphor and metonymy


in using an impoverished likeness sentence: we should really be assessing a sentence like Sailors are like their hands in that they have functions such as that of grasping rope in common. Surely this sentence is appropriate and true in the relevant context. In short, the suspicion is that the test picks up on, if anything, just default preconceptions of likeness rather than the more specialized, less obvious but nevertheless relevant and technically important forms arising in specic contexts. We will look at further examples of relevant Partitive similarity. The shared features in the examples happen to be ones of function or appearance, but the phenomenon could apply much more widely. The central observation about our examples will be that matters of appearance or function will be what the utterance is getting at, and the PART will be similar to the WHOLE precisely in appearance or function respectively. We will also continue to see in the examples that the WHOLE has the appearance or function it has partly because the PART has it, so the similarity arises in part from the parthood; and (in the PART FOR WHOLE direction) the particular part is chosen in the metonymy because of the similarity. It is common for one important contribution, or even the main contribution, to the appearance of something to come from a certain type of part of the thing, notably the outer surface of the whole thing, or the outer surface of an especially salient part of the thing. Warren (2006: 42) mentions the metonymic PART-FOR-WHOLE use of the word palefaces by Native Americans (at one time) to refer to white people. So consider a sentence such as We run away when we see palefaces, uttered perhaps in a 1950s cowboy lm. The persons face is relevantly similar, in appearance, to the person as a whole (or more precisely to their skin as a whole): you can normally tell someone is a white person by looking just at their face, and the fact that someone is white is highly relevant to the understanding of the sentence. Also, the very motivation for the particular type of metonymy in question is precisely the similarity of appearance of face to whole skin and then to the whole person. Consider now the following example (mentioned by Warren (2006: 43), but of course an example of a common way of speaking): (3) Everyone who wants a roof should have one.

Although the phrase the roof could be referring literally just to roofs, it is more likely to be metonymically referring to roofed dwellings. Part of the function of a dwelling is to shelter the occupants, and an important aspect of that function is provided by the roof. Assuming the sheltering function is relevant to the understanding of the sentence in context, we see that roofs and dwellings are relevantly Partitively similar.


J. A. Barnden

The above examples are about parthood within physical objects, but relevant Partitive similarity is not conned to these. A more abstract example is The meal was enjoyable when the phrase the meal refers not just to the food served and the eating of it but the whole occasion, including conversation, etc. (This metonymy is similar in style to the metonymy that Norrick (1981: 9394) discusses of [to] cook referring to the whole meal-preparation process). The meal in the narrow sense is a PART of the whole occasion. They share the function of providing food to the eaters; the occasion has that function because (in part) the narrow meal does so; and the narrow meal is by default an important aspect of the enjoyability. The hand (for sailor), palefaces, roof, and meal examples are all PARTFOR-WHOLE, but the point works for WHOLE-FOR-PART as well. Consider the sentence She has a good head (Warren 2006: 44). Warren mentions an interpretation in which the head links metonymically to the persons intelligence viewed as a PART of the head. Clearly, the heads function of engaging in intelligent thought comes directly from the PART in this analysis. Some researchers might object that what is happening in WHOLE FOR PART cases is merely zone activation rather than metonymysee, e.g., Croft and Cruse (2004), following Paradis (2004). There is no room to argue against this stance here, but we do not need to rely on the WHOLE FOR PART direction anyway: the above consideration of PART FOR WHOLE is enough for our purposes. The similarity in Partitive metonymy can be like that used in some metaphor. It is dicult to drive a wedge between appearance-based relevant Partitive similarity and appearance-based similarity in image-based metaphor. As regards function-based relevant Partitive similarity, note that functional similarity is key in some metaphors, as in brain-as-computer or vice versa, insofar as both entities are viewed as having problemsolving as one function. Finally, we return to the issue addressed in Section 2.1 of whether metaphor and metonymy can be clearly divided as regards the degree of subjectivity and mental imposition. While it is plausible that metaphor generally has these qualities to a higher degree, it is by no means clear that the dierence across the board is enough for a clear dierentiation. In particular, parthood can be subjective and mentally imposed, especially when the entities are at least somewhat abstract. For instance, in the meal/ occasion case, the dividing line between the whole occasion and the meal as a sub-occasion of eating is fuzzy and subjective. Are drinks included within the (narrow) meal? Are snacks, coee etc. away from the main table included within it? Also, food might be available buet-style on a table for guests to get and take to other positions whenever they want,

Metaphor and metonymy


so that there are no spatial or temporal boundaries to the narrow meal at all, and yet it is still appropriate to use the word meal to refer metonymically to the whole occasion (and the whole occasion is still more than just a narrow meal aspect). In such cases the very idea that the occasion has a part constituting a narrow meal is a mental imposition of structure, quite aside from the question of what the exact boundaries of that part are.


Contiguity involving similarity, 3: Other cases

Other types of metonymy may involve similarity in an essential way. Panther (2006) discusses phrases like a Pearl Harbour where a proper name is used as if it were a common noun. He analyses this particular example as involving, rst, a metonymic step to an event, and then a metonymic step from that specic event to other events of the same kind. Panther (2006: 181 at note 24) says that this second metonymic step goes from the specic event to events that are, as he says, like the original one in significant respects. Thus, the essential nature of the metonymic step is to traverse a similarity link. There is even some appreciable similarity in the case of a sports team or a single athlete representing a country. Such representation only makes sense in the context of more than one country (or other social, political or geographical unit) being represented. So, there is a one-to-one correspondence between some countries and their athletes/teams. This can by itself be viewed as a minimal type of structural correspondence (hence similarity)minimal in the sense that correspondence of relationships is not yet involved. But we can enrich the correspondence by taking competitive relationships between teams/athletes to correspond to competitive relationships between countries. The structural correspondence is extrinsic to any one country and its team/athlete, but this is not a problem as the similarity in metaphor generally can also be extrinsic relative to individual items involved (cf. the solar-system/organization similarity in the preface of Section 2). In addition, in the case of a team as opposed to a single athlete, there is additional intrinsic similarity that conceivably contributes to the metonymy. The team is similar to the country in being composed of people who work cooperatively for the good of the team (and hence also for the country), just as the people in a country work cooperatively in some measure for the good of the country. Part of the very reason the team is viewed as representing the country may be that it has the features just mentioned, though this deserves further discussion.

22 3.

J. A. Barnden Source/target links as part of the message (link survival)

Warren (1999, 2002, 2006) asserted versions of the claim that, in a metonymy, the source/target link itself is kept as part of the message of the utterance. Similar or strongly related views have also been expressed by Croft (2006), Dirven (2002), Haser (2005), Panther (2006) and Radden and Kovecses (1999). As an example of the phenomenon, when people understand the sentence Finland lost the [football] match they surely construct semantic mental representations in which the football team in question is indeed identied as being the team associated with the country Finland; the mental identication of the team is not (or at least not solely) done by some other means. Thus, the metonymic link between Finland and the team is preserved as part of the representation of sentence meaning: the role that the target item plays in relation to the source item is an important part of the message, not just a processing route to determining the message. The Finland-to-team link is not ( just) used to determine a mental list of player names, say. On the other hand, inclusion of links within the message supposedly does not happen in metaphor. Dirven (2002) claims this, in eect; and Warren (2006: 15) even says that metaphor involves the annihilation of the source by the target. Consider the sentence They have reached the third milestone on the project. Even if understanding of this is obtained on-line by considering a hypothetical physical milestone or the physicalmilestone category, it is reasonable to suppose that there is no need to have, within the resulting representation for the gurative meaning of the sentence, a record of the fact that the plan component in question is linked to that hypothetical milestone or milestone category. The claim that, in metonymy, the source-target link (at least often) survives into sentence meaning is appealing. Indeed, in many cases of metonymy there is no way of specifying the target item other than by reference to the source. Someone can utter or understand Finland lost the match without having any knowledge of the players names, or any way of referring mentally to the team other than by some analogue of the description the football team of Finland. And even when there is some other readily available way of mentally referring to the target, part of the point of the sentence would often be lost if the explicit link to the source were thrown away. However, it remains to be seen how universally the source-target link needs to survive. For instance, consider the sentence John has brains where brains are being used metonymically to refer to intelligence (cf. Haser 2005: 46). As Haser indicates, it is perfectly adequate for the mental representation of the meaning to state simply that John has intelligence, with no reference to brains.

Metaphor and metonymy


And, on the other hand, we now argue that many uses of metaphor do appear to keep links to source items. (See also Croft (2006), and the notion of knowledge by metaphorical character in Stern (2000)). Part of the point of a poetic or other literary work is often the way in which information is expressed, and in particular the metaphors used. For example, in Shakespeares play As You Like It the character Jaques says All the worlds a stage . . . and then lengthily elaborates this idea. Part of the message is the comparison of world to theatre (especially in view of the dramatic irony that Jaquess real world is already theatre for us), not just the information about the world in itself that we may get as a result of comparing world to theatre, and then forgetting about the link to the theatre. More mundanely, many names for things have a metaphorical quality, and it is plausible that use of the names involves remembering the links to source items. An example is the name army ant. The reason army ants are so-called is a rich behavioral similarity to soldiers and other army units (see the popular science exposition quoted by Goatly 1997: 163), and at least in case of someone rst learning about them it is dicult to believe that reference back to the behavior of real army units is not active in the persons mind as an important part of the conception of the ants. Again, consider someone using the phrase the camel to refer to a cloud that looks like a camel, and saying The camel has broken its neck to describe the cloud coming apart at the place of its so-called neck. Now, Indurkhya (1992) uses the case of a similarity between a camel and a cloud as an example of how structure from a metaphor source (the camel) can be imposed on a target (the cloud), rather than residing intrinsically in the target. (See comments above, in the preface of Section 2). Thus, the identication of some part of the cloud as corresponding to the neck may be ineliminably dependent on the comparison to the camel, and indeed the head and the rest of the body may only be vaguely related to the subshapes in the cloud, so that we cannot necessarily just regard the neck part as the place where the head part joins the rest-of-body part, because those parts may not themselves be clearly specied. Under these conditions, consider what the understanders representation of the sequence of events has been. How is the place of the breakage to be internally represented? We have to assume either that (a) he/she has kept a detailed, entirely spatial representation of the original cloud, and, after somehow picking out a subregion N of it as the referent of its neck by considering the shape of camels, remembers only the spatial characterization of N as the internal representation of where the breakage was, or that (b) he/she refers mentally (whether consciously or not) to the supposed neck part


J. A. Barnden

by some representation that could be glossed as the cloud part, whatever it is, corresponding to the camels neck or the cloud part between the head part and the body part (perhaps together with a representation of roughly where that part is). Consider also: Theres a camel in the sky, with the word Theres interpreted existentially (as opposed to deictically), and in a context where it is clear that a cloud is being mentioned. Especially if the understander is not looking at the sky, there is now no specic cloud to compare to a camel, so unless the understander arbitrarily imagines a particular cloud shape, the most natural suggestion is that he/she mentally describes the cloud as having a shape similar to that of a camel. This point transfers directly to metaphor example (2or) in Section 2.2 in a context where the understander has not seen the drawing and so has no specic line shape to describe unless he/she mentally invents one. Thus, the link is likely to survive into the message of (2or) under certain conditions. Equally, the mental representation of the shape alluded to by metonymy example (2my) is presumably something like a shape representing a snake. So, for both (2my) and (2or), having links as part of the message is potentially equally important, depending on discourse context. Although the cloud/camel example and line/snake example (2or) are cases of image-based metaphor, the point they make is not specic to such metaphor. In particular, in any metaphor where source structure is imposed on the target, aspects of the target may best be mentally identied via corresponding aspects of the source. By contrast, there is no particular reason to insist that, with unreective use of unremarkable metaphorical phraseology such as milestones in everyday discourse, we should take source/target links (even if used during understanding) to be part of the message.

4. 4.1.

Additional discussion Two other possible dierences between metaphor and metonymy

We consider rst the idea that a metonymic step stays within a single conceptual compartment of some sort (a domain, Idealized Cognitive Model, frame, etc.) whereas metaphor crosses between dierent compartments. We concentrate here on compartmentalizations that are static in the sense of not relying on decisions made about the interpretation of the very utterances under consideration, so that when A and B are alluded to in an utterance, the question of whether A and B are in the same compartment or not cannot be varied by the utterance itself.

Metaphor and metonymy


The idea of using a static compartmentalization to distinguish metonymy and metaphor has major problems. For instance, a country can not only be used metonymically to stand for its team, but a country and a football team (even its own actual team) can be put into metaphorical relationship to each other, either way round. The captain can be likened to a national leader, other roles within the team can be likened to roles within society, manipulations of the ball can be likened to interactions within society, etc. Either a country and its team are in the same compartment or they are not, and, in either case, either metaphor or metonymy can link the country and team in an utterance. The particular utterance is not allowed to help dene whether the country and team are in the same compartment or not. Utterances (2my) and (2or) supply another example, especially if we assume that in each situation the same line is present, and in each situation the snake is the same imagined, prototypical snake. Utterance (2my) is metonymic and (2or) is metaphorical, but they involve exactly the same interaction with any static compartmentalization. Even if dierent lines and imagined snakes were allowed in the two situations it would be dicult to nd a principled way in which the line and snake could be in the same compartment for (2my) and dierent compartments for (2or). A variety of other authors have found problems with compartmentbased dierentiations between metaphor and metonymy, including Barcelona (2002), Cameron (1999a), Croft (2002), Feyaerts (2000), Haser (2005), Kittay (1989), Moore (2006), Panther (2006) and Peirsman and Geeraerts (2006). However, some authors criticize one proposed compartment-based distinction only to introduce a proposal that itself has aws. As just one example, Moore (2006) claims that metonymy operates within a frame whereas metaphor crosses between frames. He says that the days of the week form a frame. But days can be used metaphorically for each other, as in the following part of a blog posting, created on a Monday: Sunday felt like Monday, so in honor of it [i.e., today] being honorary Tuesday, I am doing two minis [mini-blogs] today . . .4 Of course, if one were allowed to propose a compartmentalization that depended on decisions about metonymy and metaphoricity of utterancese.g., one put a snake and a line in the same compartment because of a metonymic utterance like (2my)then the compartmentalization would not be of much use in dening the dierence between metaphor and metonymy. But one general problem even about static compartmentalizations is that rarely if ever is any clear constraint placed in what

4. http:/ / (accessed 3rd July 2008).


J. A. Barnden

one is allowed to propose as being packaged into a compartment, so that for any particular set of examples one can often suggest a static compartmentalization that makes the postulated metaphorical links cross between compartments and the postulated metonymic links not do so. Haser (2005 Ch. 2) makes similar points. We now turn to another possible ground for dierentiating metaphor and metonymy, namely imaginary identication or categorization. As a feature of metaphor this has appeared in various forms and under a variety of names, for example in blending theory (Fauconnier and Turner 1998; Turner and Fauconnier 1995, 2000, 2002)in that corresponding target and source items become identied as a single item in a metaphorical blendand somewhat similarly in the ATT-Meta approach (Barnden 2001, 2006; Barnden et al. 2004). It has also been suggested by Warren (2002). The idea is that in the course of understanding a metaphorical utterance the understander imagines that the target item and source item are the same thing (e.g., imagines Richard to be a specic hypothetical lion) or imagines the target item to be in the source item when this is construed as a category (e.g., imagines Richard to be in the physical lion category). Not only does imaginary identication or categorization fail to be a generally accepted tenet about metaphor, but also it is possible to construe metonymy as involving it. Turner and Fauconnier (2000, 2002) and Fauconnier (2009) apply imaginary identication to some metonymically linked items, as well as to metaphorically linked ones: for example, in Turner and Fauconnier (2000, 2002) a printing press and a newspaper company that are metonymically related to each other become one item in a blend. Fauconnier (2009) rearms the point from Turner and Fauconnier (2002) that in their blending treatment of anger as heat, the heat, the anger and the bodily reactions correlated with angerand thereby metonymically related to angerbecome identied as one element in the blend. Thus, even if imaginary identication were established as essential in metaphor it would not uncontroversially distinguish it from metonymy. 4.2. Overlap, intermediacy, and combinations of dierences

Our arguments do not prevent some combination of the discussed dimensions, rather than some single dimension, from serving to distinguish metaphor and metonymy. But it is going to be very dicult to come up with such a combination. The snake/line examples (2my) and (2or) from Section 2.2 are similarly if not identically situated on the dimensions of (a) compartmentalization, (b) similarity, (c) contiguity, (d) structural corre-

Metaphor and metonymy


spondence, (e) link survival, and (f ) source-item hypotheticality. As for (a): the two examples interact identically with any static compartmentalization, under the assumptions of Section 4.1. As for (b) and (c), in metaphorical (2or) the link is not only a similarity link but can be regarded as a metonymically-used contiguity link on the lines of Section 2.1. The contiguity, while dierent from that in (2my), is similar to it in many respects, e.g., the extent to which it is mentally imposed. For (d): both examples involve the same type and degree of visual similarity, and therefore of structural correspondence. For (e): link survival into the message may need to happen just as much for metaphorical (2or) as for metonymic (2my), as we saw at the end of Section 3. For (f ): example (2or) involves a hypothetical snake and (2my) may well do so. It is possible also that the question of imaginary identication/categorization would not distinguish (2my) and (2or), given the discussion in Section 4.1. Even if there is exact co-positioning of two dierent utterances in the spaceensuring an overlap of the metaphor and metonymy regions in the spaceit does not necessarily follow that there is overlap in the stronger sense dened in the Introduction: namely that there is a phenomenon that at one and the same time qualies fully as being both metaphor and metonymy (in a sense dierent from chaining or metaphtonymy, as mentioned in the Introduction). This sort of overlap might fail to occur because there might be constitutive dimensions that we have not considered. However, the arguments in Section 2.1 do suggest that we should take seriously the idea that referential metaphor is always also a type of metonymy, and the arguments in Section 2.2 suggest that similarity-based Representational metonymy is (often or perhaps always) also a type of metaphor. Of course, the reason for the snake/line link in (2my) being metonymic can be construed as being dierent from the reason for its being metaphorical in (2or). The metonymicity could reside just in the fact that the link is crossed for the purpose of achieving the appropriate mental reference, rather than in the intrinsic nature of the link; and the metaphoricity could reside in the use of the link as part of likening one thing to another and/or imaginatively identifying one thing with another. This would make metonymy and metaphor be, by denition, dierent properties of the use of links, no matter how much these properties coincide in application and no matter what the links are like intrinsically. In this view, the overlap of metaphor and metonymy in the use of a given link is a mixing of two dierent phenomena, but it is a dierent sort of mixing from the chaining of two links, one metonymic and the other metaphorical. The situation as regards intermediacy is dierent. Recall from the Introduction that intermediacy is the issue of whether there are phenomena


J. A. Barnden

that have some of the qualities of both metaphor and metonymy but are not classied as either. Notice rst that intermediateness and overlap do not imply each other, and are strongly contrasting concepts. Two types of thing can overlap without there being anything outside both types that is nevertheless close enough to both to qualify as intermediate between them; and two non-overlapping types of thing can have things between them. Prime numbers and even numbers overlap on the number 2, but this does not imply there is anything intermediate between evenness and primeness; and a 30-year old is intermediate between being a teenager and a middle-aged person whereas of course teenagers and middle-aged people have no overlap. However, the question of intermediacy versus overlap can be more complicated. The fringes of two categories might overlap and we could say that items within this overlap are intermediate between more central parts of the categories. Now, although each individual dimension that we have discussed has some similarity to a spectrum or continuum as proposed by other researchers, and the word spectrum or continuum suggests possibilities intermediate between metaphor and metonymy, our arguments do not necessarily imply intermediate phenomena. This article leaves the issue to future investigation. If we could show that on at least one constitutive dimension there were phenomena intermediate between metaphor and metonymy then the phenomena would necessarily be intermediate in the whole space. The intermediacy on the one dimension would be enough to stop the phenomena qualifying fully as either metaphor or metonymy. But in principle at least there could be phenomena that are intermediate in the whole space but that are not intermediate between them on any one dimension. Indeed, metaphor and metonymy might use largely the same portions of every dimension separately and yet still be completely separate in the whole space and allow intermediate possibilities there, because of complex positioning within the whole space.5 The way in which the dimensions enrich and sharpen the analysis of overlap and intermediacy between metaphor and metonymy is one way in which it is fruitful to consider the dimensions as opposed to conning discussion to the broader concepts of metaphor and metonymy.


This could happen much as with shapes in geometric space. Consider an upright square in 2D space. Split the square on a diagonal and separate the two resulting triangles a little. The points within both triangles involve roughly the same range of horizontal positions and the same range of vertical positions. Yet the triangles do not overlap at all and there are intermediate points.

Metaphor and metonymy 4.3. (Some) metaphor as double metonymy


One type of theoretical move that considerably muddies the gurative water is to claim that metaphor in general, or some type of metaphor, is actually composed of metonymy in some way. For instance, the analyses of Riemer (2001, 2002) and Barcelona (2000b) indicate that any metaphor whose meaning is construed as transferring features identically from source to target can be viewed as metonymy, because those features can be viewed as metonymically linked to both the source item and the target item. (See also Haser 2005: 25). In this vein Riemer mentions the view of Group m (1981) that metaphor is double synecdoche (therefore double metonymy). That iswith some deviation from Group ms own terminologythere is a metonymy from the source concept to a feature also possessed by the target concept, and from that feature to the target concept. Alternatively, this can be viewed as a metonymy from the source concept up to a shared superordinate category (the category of things that possess the feature in question) and then down to the target concept. A theoretician is free to regard metaphor as composed of metonymy in some such way, but there is a sense in which it does not really matterit is a labelling move that leaves undisturbed the important point, namely that certain types of language involve certain types of conceptual link, or collections of links, where individual links may have qualities of similarity, contiguity or both, or may have other qualities; where the assembly of links used may be some strange mixture of types; and where dierent instances of language can structurally and procedurally arrange the links involved in a large number of dierent ways. Notice also that casting metaphor as metonymy does not of itself mean that the importance of, say, similarity in metaphor is downgraded. For example, in a double-metonymy view we still have the issue of what particular shared attribute (or set of attributes) or covering abstraction is being used, i.e., what the similarity is. It is just that this similarity is being theoretically structured as being via metonymies involving shared properties or via a covering abstraction. And of course we cannot couple just any two metonymies together to get a metaphor.



We have seen that several ways in which metaphor and metonymy have been thought to dier do not work, because the alleged dierences are fuzzy or are at best only general tendencies. The fuzziness and slipperiness of the dierences is even greater than some authors have already considered them to be. In particular, we have argued that distinctions between


J. A. Barnden

metaphor and metonymy based on the following issues fail: contiguity versus similarity; source/target links surviving as part of the message; interaction with conceptual compartments; and (in passing) structural correspondence and hypotheticality of source items. While the authors own ATT-Meta theory of metaphor involves imaginary identication, as does blending theory, it is possible that metonymy involves it as well. Not only do the mentioned dimensions not serve individually to distinguish metaphor from metonymy, we have seen evidence that no combination of them does, because clear cases of metaphor and metonymy can be at the same position in the multidimensional space; and even that sometimes the use of a source/target link can be simultaneously metonymic and metaphorical. Some of our arguments echo points made by other authors, but we have added qualitatively new evidence and critique. Of course, the arguments do not bear upon the distinguishing power of other possible dimensions. Another question is whether our arguments bear against the possibility that there are (proto)typical forms of metaphoricity and metonymicity that can be cleanly distinguished. Haser (2005) argues against even this being possible on the dierentiating grounds that have been put forward in Cognitive Linguistics, but the present article leaves the possibility open, partly because it is by no means clear what counts as (proto)typical. In that metaphor and metonymy involve fuzzily dened ranges of complex combinations of contiguity, similarity, link survival, etc., it is helpful in the interests of more precise, richer, deeper and more liberated analysis to disentangle these properties from each other, even though the individual notions of contiguity, similarity, link survival, etc. are themselves fuzzily dened. We have argued that metaphor and metonymy can each involve types of contiguity and similarity, thus violating tacit, simplistic assumptions that these properties are opposed to each other. By arguing that some metonymy involves similarity in an essential way we encourage attention on investigating just what are the forms and extents of similarity that appear in metonymy and metaphor. Again, instead of seeking a way of rmly dierentiating metaphor and metonymy through their interaction with conceptual compartments such as domains and frames, we can concentrate on neutrally examining the ways in which metonymy and metaphor stay within or cross between compartments in particular regimes of compartments. We have also seen that the dimensions enrich the analysis of overlap and intermediacy between metaphor and metonymy. In particular, we advance beyond single-spectrum views and bring to light new possibilities for intermediacy. The question of whether more distant regions of the multidimensional space are interesting also arises.

Metaphor and metonymy


Thus, a major conclusion from the discussion in this article is that instead of worrying about whether some utterance is metaphorical or metonymic, or even about how far along a literal/metonymic/metaphorical continuum it is, we should often be asking instead: What degree and type of similarity does it involve, if any? What sort of contiguity does it involve, if any? Does it involve link survival? Is the source item hypothetical, and in what way? Is there any imaginary identication? And so forth. Considering the dimensions in themselves helps to free us from a mindset that seeks clear-cut dierences between metaphor and metonymy when these may not exist. The most radical form this conclusion might take is the eliminativist possibility that the words metaphor and metonymy are just pragmatically useful labels in approximate discussions, not legitimate foci for detailed technical attention. Ritchie (2006: 11) says that Metaphor, and gurative language generally, is but a convenient way of identifying and discussing a widely-recognized but fuzzily dened subset of [certain interpretive connections he discusses]. Fauconnier (2009) says that metaphor, metonymy, etc. elude rigorous denition and that these categories do not provide deep insight; that insight comes from looking at the detailed underlying cognitive operations involved, such as blending, and the way they are combined. However, irrespective of whether such eliminativist suggestions are correct, the points made above about the usefulness of the dimensional analysis hold good. Received 8 July 2008 Revision received 14 May 2009 University of Birmingham

Barcelona, Antonio (ed.). 2000a. Metaphor and metonymy at the crossroads: A cognitive perspective. Berlin & New York: Mouton de Gruyter. Barcelona, Antonio. 2000b. On the plausibility of claiming a metonymic motivation for conceptual metaphor. In Antonio Barcelona (ed.), Metaphor and metonymy at the crossroads: A cognitive perspective, 3158. Berlin & New York: Mouton de Gruyter. Barcelona, Antonio. 2002. Clarifying and applying the notions of metaphor and metonymy within cognitive linguistics: An update. In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and contrast, 207277. Berlin & New York: Mouton de Gruyter. Barcelona, Antonio. 2004. Metonymy behind grammar: The motivation of the seemingly irregular grammatical behavior of English paragon names. In Gunter Radden & Klaus-Uwe Panther (eds.), Studies in linguistic motivation, 357374. Berlin & New York: Mouton de Gruyter. Barnden, John A. 2001. Uncertainty and conict handling in the ATT-Meta contextbased system for metaphorical reasoning. In Varol Akman, Paolo Bouquet, Richmond


J. A. Barnden

Thomason & Roger A. Young (eds.), Modeling and using context: Third International and Interdisciplinary Conference (Lecture Notes in Articial Intelligence 2116), 1529. Berlin: Springer. Barnden, John A. 2006. Articial intelligence, gurative language and cognitive linguistics. In Gitte Kristiansen, Michel Achard, Rene Dirven & Francisco J. Ruiz de Mendoza Ibanez (eds.), Cognitive linguistics: Current applications and future perspectives, 431459. Ber lin & New York: Mouton de Gruyter. Barnden, John A., Sheila R. Glasbey, Mark G. Lee & Alan M. Wallington. 2004. Varieties and directions of inter-domain inuence in metaphor. Metaphor and Symbol 19(1). 130. Black, Max. 1993 [1979]. More about metaphor. In Andrew Ortony (ed.), Metaphor and thought, 2nd edn., 1941. Cambridge & New York: Cambridge University Press. Cameron, Lynne. 1999a. Operationalising metaphor for applied linguistic research. In Lynne Cameron & Graham Low (eds.), Researching and applying metaphor, 128. Cambridge & New York: Cambridge University Press. Cameron, Lynne. 1999b. Identifying and describing metaphor in spoken discourse data. In Lynne Cameron & Graham Low (eds.), Researching and applying metaphor, 105132. Cambridge & New York: Cambridge University Press. Chiappe, Dan L. 1998. Similarity, relevance, and the comparison process. Metaphor and Symbol 13(1). 1730. Cooper, David E. 1986. Metaphor. Oxford: Blackwell. Croft, William. 2002. The role of domains in the interpretation of metaphors and metony mies. In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and contrast, 161205. Berlin & New York: Mouton de Gruyter. Croft, William. 2006. On explaining metonymy: Comments on Peirsman and Geeraerts, Metonymy as a prototypical category. Cognitive Linguistics 17(3). 317326. Croft, William & D. Alan Cruse. 2004. Cognitive linguistics. Cambridge & New York: Cambridge University Press. Dirven, Rene. 2002. Metonymy and metaphor: Dierent mental strategies of conceptualiza tion. In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and contrast, 75111. Berlin & New York: Mouton de Gruyter. Dirven, Rene & Ralf Porings (eds.). 2002. Metaphor and metonymy in comparison and con trast. Berlin & New York: Mouton de Gruyter. Fass, Dan. 1997. Processing metaphor and metonymy. Greenwich, CN: Ablex. Fauconnier, Gilles. 2009. Generalized integration networks. In Vyvyan Evans & Stephanie Pourcel (eds.), New directions in cognitive linguistics, 147160. Amsterdam: John Benjamins. Fauconnier, Gilles & Mark Turner. 1998. Conceptual integration networks. Cognitive Science 22(2). 133187. Feyaerts, Kurt. 2000. Rening the Inheritance Hypothesis: Interaction between metaphoric and metonymic hierarchies. In Antonio Barcelona (ed.), Metaphor and metonymy at the crossroads: A cognitive perspective, 5978. Berlin & New York: Mouton de Gruyter. Geeraerts, Dirk. 2002. The interaction of metaphor and metonymy in composite expres sions. In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and contrast, 435465. Berlin & New York: Mouton de Gruyter. Gibbs, Ray W., Jr. 1990. Comprehending gurative referential descriptions. Journal of Experimental Psychology: Learning, Memory and Cognition 16(1). 5666. Gibbs, Ray W., Jr. 1999. Researching metaphor. In Lynne Cameron & Graham Low (eds.), Researching and applying metaphor, 2947. Cambridge & New York: Cambridge University Press. Goatly, Andrew. 1997. The language of metaphors. London & New York: Routledge.

Metaphor and metonymy


Goodman, Nelson. 1968. Languages of art: An approach to a theory of symbols. Indianapolis: Bobbs-Merrill. Goossens, Louis. 1990. Metaphtonymy: The interaction of metaphor and metonymy in expressions for linguistic action. Cognitive Linguistics 1. 323340. Glucksberg, S. 2001. Understanding gurative language. Oxford & New York: Oxford University Press. Glucksberg, Sam & Boaz Keysar. 1990. Understanding metaphorical comparisons: Beyond similarity. Psychological Review 97(1). 318. Grady, Joseph E. 1997. THEORIES ARE BUILDINGS revisited. Cognitive Linguistics 8(4). 267290. Group m. 1981. A general rhetoric. Baltimore & London: The John Hopkins University Press. Haser, Verena. 2005. Metaphor, metonymy and experientialist philosophy: Challenging cognitive semantics. Berlin & New York: Mouton de Gruyter. Indurkhya, Bipin. 1992. Metaphor and cognition: An interactionist approach. Dordrecht: Kluwer. Jakobson, Roman. 2002 [1956]. The metaphoric and metonymic poles. In Rene Dirven & Ralf Po rings (eds.), Metaphor and metonymy in comparison and contrast, 4147. Berlin & New York: Mouton de Gruyter. Reprinted with added abstract by Rene Dirven from Roman Jakobson & M. Halle (eds.) 1971 [1956], Fundamentals of Language 2, 9096. The Hague & Paris: Mouton.) Kittay, Eva F. 1989. Metaphor: Its cognitive force and linguistic structure. Oxford: Clarendon Press. Kovecses, Zoltan. 1990. Emotion concepts. Berlin & New York: Springer. Kovecses, Zoltan. 2002. Metaphor: A practical introduction. Oxford & New York: Oxford University Press. Lako, George. 1993. The contemporary theory of metaphor. In Andrew Ortony (ed.), Metaphor and Thought, 2nd edn., 202251. Cambridge & New York: Cambridge University Press. Lako, George & Mark Turner. 1989. More than cool reason: A eld guide to poetic metaphor. Chicago: University of Chicago Press. Langlotz, Andreas. 2006. Idiom creativity: A cognitive-linguistic model of idiomrepresentation and idiom-variation in English. Amsterdam & Philadelphia: John Benjamins. Lodge, David. 1977. The modes of modern writing: Metaphor, metonymy and the typology of modern literature. London: Edward Arnold; Ithaca, NY: Cornell University Press. Moore, Kevin E. 2006. Space-to-time mappings and temporal concepts. Cognitive Linguistics 17(2). 199244. Norrick, Neal R. 1981. Semiotic principles in semantic theory. Amsterdam: John Benjamins. Nunberg, Georey. 1978. The pragmatics of reference. Bloomington, IN: Indiana University Linguistics Club. Nunberg, Georey. 1995. Transfers of meaning. Journal of Semantics 12(2). 109132. Panther, Klaus-Uwe. 2006. Metonymy as a usage event. In Gitte Kristiansen, Michel Achard, Rene Dirven & Francisco J. Ruiz de Mendoza Ibanez (eds.), Cognitive linguistics: Current applications and future perspectives, 147185. Berlin & New York: Mouton de Gruyter. Paradis, Carita. 2004. Where does metonymy stop? Senses, facets, and active zones. Metaphor and Symbol 19(4). 245264. Peirsman, Yves & Dirk Geeraerts. 2006. Metonymy as a prototypical category. Cognitive Linguistics 17(3). 269316.


J. A. Barnden

Pragglejaz Group 2007. MIP: A method for identifying metaphorically used words in discourse. Metaphor and Symbol 22(1). 139. Radden, Gu nter. 2002. How metonymic are metaphors? In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and contrast, 407434. Berlin & New York: Mouton de Gruyter. Radden, Gunter & Zoltan Kovecses. 1999. Towards a theory of metonymy. In Klaus-Uwe Panther & Gunter Radden, Metonymy in language and thought, 1759. Amsterdam & Philadelphia: John Benjamins. Radman, Zdravko. 1997. Diculties with diagnosing the death of a metaphor. Metaphor and Symbol 12(2). 149157. Riemer, Nick. 2001. Remetonomyzing metaphor: Hypercategories in semantic extension. Cognitive Linguistics 12(4). 379401. Riemer, Nick. 2002. When is a metonymy no longer a metonymy. In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and contrast, 379406. Berlin & New York: Mouton de Gruyter. Ritchie, L. David. 2006. Context and connection in metaphor. Basingstoke & New York: Palgrave Macmillan. ez, Ruiz de Mendoza Iban Francisco J. 1999. From semantic underdetermination via metaphor and metonymy to conceptual interaction. Paper No. 492, LAUD Linguistic Agency, SERIES A: General and Theoretical Papers. Ruiz de Mendoza Ibanez, Francisco J. & Olga I. Dez Velasco. 2002. Patterns of conceptual interaction. In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and contrast, 489532. Berlin & New York: Mouton de Gruyter. Stern, Josef. 2000. Metaphor in context. Cambridge, MA & London, UK: Bradford Books, MIT Press. Turner, Mark & Gilles Fauconnier. 1995. Conceptual integration and formal expression. Metaphor and Symbolic Activity 10(3). 183204. Turner, Mark & Gilles Fauconnier. 2000. Metaphor, metonymy, and binding. In Antonio Barcelona (ed.), Metaphor and metonymy at the crossroads: A cognitive perspective, 133 145. Berlin & New York: Mouton de Gruyter. Turner, Mark & Gilles Fauconnier. 2002. Metaphor, metonymy, and binding. In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and contrast, 469 487. Berlin & New York: Mouton de Gruyter. Ullmann, Stephen. 1962. Semantics: An introduction to the science of meaning. Oxford: Blackwell. Warren, Beatrice. 1999. Aspects of referential metonymy. In Klaus-Uwe Panther & Gunter Radden, Metonymy in language and thought, 121135. Amsterdam & Philadelphia: John Benjamins. Warren, Beatrice. 2002. An alternative account of the interpretation of referential metonymy and metaphor. In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in compar ison and contrast, 113130. Berlin & New York: Mouton de Gruyter. Warren, Beatrice. 2006. Referential metonymy. Scripta Minora of the Royal Society of Letters at Lund, 20032004: 1. Stockholm: Almqvist and Wiksell International. Wee, Lionel. 2006. Proper names and the theory of metaphor. Journal of Linguistics 42. 355371.

Grammatical weight and relative clause extraposition in English


Abstract In relative clause extraposition (RCE) in English, a noun is modied by a non-adjacent RC, resulting in a discontinuous dependency, as in: Three people arrived here yesterday who were from Chicago. Although discourse focus is known to inuence the choice of RCE over truth-conditionally equivalent sentences with canonical structure (Rochemont and Culicover 1990; Takami 1999), Hawkins (2004) and Wasow (2002) have proposed in addition that RCE should be preferred when the relative clause is long (or heavy) relative to the VP because such structures are processed more eciently in comprehension and production. The current study tested this hypothesis based on Hawkins (2004) domain minimization principles. In an acceptability judgment task, canonical sentences were rated signicantly higher than extraposition sentences when the RC was light, but this dierence disappeared when the RC was heavy. In a self-paced reading task, extraposition sentences were read signicantly faster than canonical sentences when the RC was heavy, but there was no dierence when the RC was light. In an analysis of RCE in the ICE-GB corpus, extraposed RCs were signicantly longer than the VP on average, whereas canonical RCs were signicantly shorter, and the proportion of sentences with extraposition decreased as the ratio of VP-to-RC length increased. These ndings support Hawkins

* Correspondence address: Department of English and Linguistics Program, 500 Oval Drive, Purdue University, West Lafayette, IN 47907, USA. E-mail: 3ejfranci@purdue. edu4. Acknowledgements: I am grateful to research assistants Yanhong Zhang and Najeong Kim for all their help with data collection and processing. I would also like to thank Ewa Da browska, Stephen Matthews, and three anonymous CL referees for their detailed and insightful comments on earlier versions of the paper, as well as Bill Croft, Pat Deevy, Alex Francis, Jack Hawkins, and Tom Wasow for their helpful discussions of the data. This research was funded by the Department of English and the College of Liberal Arts at Purdue University. Cognitive Linguistics 211 (2010), 3574 DOI 10.1515/COGL.2010.002 09365907/10/00210035 6 Walter de Gruyter


E. J. Francis

(2004) domain minimization principles and help explain why a discontinuous dependency is allowed and sometimes preferred even in a language with relatively xed word order. Keywords: grammatical weight; relative clause extraposition; sentence processing; English syntax.



In relative clause extraposition from subject NP, a relative clause modies a noun within the subject NP but occurs in a position following the VP. In sentence (1a) from the International Corpus of English Great Britain (henceforth, ICE-GB), for example, the VP soon appeared intervenes between the noun sets and the relative clause that were able to receive all the TV channels. The extraposition sentence in (1a) can be used to express the same proposition as the canonical sentence in (1b), in which the relative clause directly follows its head noun. (1) a. b. New sets soon appeared that were able to receive all the TV channels. (ICE-GB) New sets that were able to receive all the TV channels soon appeared.

Relative clause extraposition (henceforth RCE) in (1a) violates the prescriptive rule banning misplaced modiers and would likely be corrected to (1b) if found in a students essay (see Trenga 2006: 5156). Nevertheless, this type of extraposition occurs naturally in formal speech and formal writing as well as informal styles, as illustrated in the additional examples in (2) from the ICE-GB corpus: (2) a. The indications are that Europe will once again become the priority because slowly and awkwardly, a treaty is taking shape which will most likely emerge by the end of this year as a new European union. (formal speech) However, a close look at the nature of these countries development experiences will reveal that certain conditions existed which cannot be applied to all other countries at all times. (formal writing) I think it would take uh oh three-quarters of an hour to an hour for somebody to start who didnt have any any any experience at all. (informal speech)



Linguists agree that the misplaced modier in RCE involves some kind of discontinuous dependency between the subject NP and the extra-

Grammatical weight


posed relative clause.1 However, the precise nature of this dependency has been a matter of debate within formal approaches to syntax. Since the relative clause (henceforth, RC) occurs in a position completely outside the subject NP, this dependency apparently violates the requirement of X-bar structure that modiers should occur as either complements to the head or as adjuncts to a phrasal projection of the head. As a way of reconciling RCE with X-bar theory, the earliest generative accounts assumed rightward movement of the RC from its canonical position (e.g., Baltin 1981; Ross 1967). However, as Rochemont and Culicover (1990) point out, this solution was problematic because RCE is both more and less constrained than typical movements like wh-movement. For example, RCE is subject to a locality condition that prevents the extraposed RC from occurring outside the clause containing its antecedent (originally formulated as the Right Roof Constraint, Ross 1967), while at the same time freely allowing certain island violations, for example extraction of the extraposed element out of the subject NP, as in the examples in (1a) and (2ac) above. Because of its special properties, RCE and related extraposition constructions have been problematic for syntactic theories and have resulted in a number of dierent analyses, all of which involve special formal or interpretive mechanisms. In addition to rightward movement (Ross 1967, Baltin 1981), other accounts have included discontinuous constituency of the NP (McCawley 1987, 1998) and stranding of the extraposed phrase due to leftward movement of non-extraposed elements (Kayne 1994). Another more surface-oriented approach favors base-generated adjunction of the extraposed constituent to IP or VP (a normal case of adjunction as far as constituent structure is concerned), with the addition of a special interpretive rule to state the discontinuous dependency and the relevant locality constraints (Rochemont and Culicover 1990). Along the same lines, Culicover and Jackendo (2005) characterize RCE as a case of simple adjunction in the syntax, but as involving a radical mismatch between syntax and semantics. In his review of the literature on extraposition, Baltin (2006) argues that each of these approaches has its merits, but that none of them provides a fully satisfactory explanation of all the relevant data (see also Bianchi 2002). Regardless of which formal approach one takes, however, it is clear that RCE is unexpected from a purely formal perspective. Why should English allow a discontinuous dependency that requires special syntactic

1. I remain neutral as to whether nominals with determiners are best analyzed as NP or DP since this issue is not important for the research questions addressed in the current study.


E. J. Francis

or interpretive mechanisms when a canonical structure can be used to express the same proposition? Why are RCE structures allowed and even preferred in some contexts in a language with relatively xed word order and constituent structure? The answer that has been proposed most often is that RCE is used to express a particular type of discourse information. Although the details of various accounts dier, the general consensus is that RCE tends to occur when the RC expresses new, contrastive or important information and the VP expresses old or backgrounded information (cf. Huck and Na 1990, 1992; Kuno and Takami 2004; Rochemont and Culicover 1990; Takami 1999). In (2b) above, for example, the RC expresses the most important information of the sentence while the VP existed merely asserts the existence of the subject certain conditions, helping to introduce the subject. More generally, RCE adheres to the tendency of focused constituents to occur later in a sentence than old or backgrounded information. Thus, the marked word order of RCE may be exploited by speakers to highlight the information status of the RC. Another factor that has been discussed in relation to this type of extraposition is predicate type. Rochemont and Culicover (1990: 65) observe that extraposition from subject typically occurs with unaccusative predicates such as predicates of existence and appearance. However, they show that other predicate types such as unergative and transitive predicates are also permissible as long as the predicate can be understood as c-construable (i.e., old or backgrounded) information. This is in contrast to other constructions such as Presentational there Insertion (PTI), which appear to be more strictly limited to occur with unaccusative predicates, as illustrated by the contrast in (3ab). (3) a. b. A man phoned her up who she didnt know. (RCE) *There phoned her up a man who she didnt know. (PTI)

Thus, Rochemont and Culicover (1990: 66) argue that the preference for unaccusative predicates derives from their typical use in discourse rather than from any lexically specied constraints on extraposition. While acknowledging the importance of discourse information structure, the current study investigates the possible role of a dierent factor grammatical weightin licensing RCE in English. Exact denitions vary, but the term grammatical weight usually refers to the length and/or complexity of a phrase in relation to other phrases in the same sentence: heavier constituents are longer or structurally more complex than lighter constituents. Grammatical weight has been implicated in a number of phenomena involving non-canonical word order in several languages (Arnold et al 2000, 2004; Cheung 2006; Hawkins 1994, 2004; Konieczny 2000; Lohse et al 2004; Matthews and Yeung 2001; Siewierska 1993; Stal-

Grammatical weight


lings et al 1998; Uszkoreit et al 1998; Wasow 1997; Yamashita and Chang 2001) and I will propose that it plays an important role in English RCE as well. Following Quirk et al (1972), Wasow (2002: 3) presents the following descriptive generalization: (4) Principle of End Weight (PEW): Phrases are presented in order of increasing weight.2

For example, in a phenomenon known as Heavy NP Shift (henceforth, HNPS), the direct object of a transitive verb occurs at the end of the sentence following an oblique argument or adjunct (usually a PP), as in (5b), rather than occurring in its canonical position adjacent to the verb, as in (5a). In accordance with the PEW, corpus studies have shown that HNPS normally occurs when the object NP is heavier than the PP, as in (5b) (Arnold et al 2000; Wasow 1997). Sentences with a light object NP in a shifted position, as in (5c), do sometimes occur, but are uncommon and are typically judged as less acceptable in the absence of special intonation or discourse conditions. (5) a. b. The waiter brought the wine we had ordered to the table. (Canonical) The waiter brought to the table the wine we had ordered. (HNPS) (Arnold et al. 2000: 28) ?The waiter brought to the table the wine.


Why should longer, more complex constituents tend to occur later in the sentence? Wasow (1997) and Arnold et al (2000) have argued that placing heavy constituents later facilitates production and planning of utterances by giving speakers extra time to formulate the heavier constituent (see also Arnold et al 2004, Stallings et al 1998, Yamashita 2002, and Yamashita and Chang 2001). In addition, Hawkins (1994) has proposed that placing heavy constituents at the end allows listeners and readers to process sentences more eciently by allowing faster recognition of the major constituents of the sentence (see also Matthews and Yeung 2001).3 Similarly, Gibson (1998) has proposed that moving heavy constituents to

2. Wasow (1997: 82) points out that this idea has been around for a long time and was, for example, noted by Otto Behaghel (1909). 3. Note that in Hawkins theory, this is only true for head-initial languages. For head-nal languages such as Japanese and Turkish, it is predicted that heavy constituents should be shifted frontward to facilitate constituent recognition (2004: 108109).


E. J. Francis

the end can reduce integration costs in the processing of non-local dependencies. To the extent that these explanations can be distinguished from each other empirically, experimental and corpus studies of production and comprehension of alternative word orders have provided evidence in favor of both production and parsing-based explanations. As a result of such ndings, Hawkins (2004) has extended his approach to apply more globally to both production and comprehension through a general principle of domain minimization (see also Cheung 2006, Lohse et al 2004). Culicover and Jackendo (2005: 167) and Wasow (2002: 67) have suggested that RCE in English should be sensitive to grammatical weight in a similar manner to HNPS. As Wasow states, Both the nal position of the (usually heavy) extraposed element and the lightening of the NP serve to increase the probability of satisfying the PEW (Wasow 2002: 7). Although Wasows remarks are suggestive, there have been no previous empirical studies testing this prediction for English. The current study seeks to ll this gap by investigating the role of grammatical weight in the processing, acceptability, and usage of RCE in English. One challenge for investigating the role of grammatical weight is that it is correlated with discourse status: lighter constituents tend express old information while heavier constituents tend to express new information (e.g., Givon 1983). Furthermore, only discourse status has been investigated in previous studies of RCE, and even this factor has not been investigated using quantitative data. However, previous studies of other syntactic alternations have shown that discourse status and grammatical weight can have independent eects. For example, Arnold et al (2000) found that each factor had an independent eect on the likelihood of people producing a particular variant for HNPS and Dative sentences. While discourse status is not directly investigated in the current study, it is held constant in the experiments so that the eect of weight can be isolated, and it is investigated indirectly in the corpus study by means of an analysis of predicate types. In the following sections, I report on the results of two psycholinguistic experiments and a corpus study of weight eects in RCE. The results of Experiment 1, a self-paced reading experiment, showed a signicant reading time advantage for RCE when the RC was heavy and no dierence between RCE and canonical structure when the RC was light. The results of Experiment 2, an acceptability judgment task, showed that canonical sentences were judged as signicantly more acceptable than RCE sentences when the RC was light, but that this dierence disappeared when the RC was heavy. Finally, a corpus study found that on average, extraposed RCs were longer than the VP while non-extraposed RCs were

Grammatical weight


shorter, and that the proportion of sentences with RCE decreased as the ratio of VP length to RC length increased. I argue that these results support Hawkins (2004) theory and help explain why RCE is preferred in some contexts despite the discontinuous dependency. The paper is structured as follows. Section 2 discusses previous research on grammatical weight and RCE in German and sets out the predictions of Hawkins (2004) theory for RCE in English. Sections 34 discuss two psycholinguistic experiments that tested the eects of grammatical weight on sentence processing: a reading time task (Experiment 1) and an acceptability judgment task (Experiment 2). Section 5 discusses a corpus study of naturally-occurring examples of RCE to show how grammatical weight aects speakers and writers structural choices. Section 6 concludes the paper and discusses some of the implications of the results reported here. 2. Relative clause extraposition, grammatical weight, and processing

As Wasow (2002: 7) points out, the discontinuous dependency in RCE not only complicates the syntax but also presumably increases processing complexity. However, studies of grammatical weight have shown that moving heavy constituents to the end, as typically happens with RCE, can also facilitate processing. To see how these two opposing constraints might interact in the case of RCE, it is necessary to consider some examples within a particular theoretical framework. Two major theories have dealt specically with weight eects in sentence processing: Gibsons (1998) Syntactic Prediction Locality Theory (SPLT) and Hawkins (1994, 2004) performance-based theory of constituent order. Both theories are locality based in that both predict that there should be a greater cost to working memory when listeners or readers must integrate linguistic information across a distance. Thus, in the case of RCE, integrating the RC with its head noun across an intervening VP constituent should incur some cost to working memory. However, the same theories predict that in cases where the RC is heavy, the cost of integrating the subject NP with the verb will be greater, and processing eciency may be maximized by placing the heavier constituent at the end. The two theories dier in a number of details, most importantly in the way that the distance between constituents is measured and in how the processing domains are dened. However, they make similar predictions with respect to end weight eects, and I will not attempt to distinguish between them empirically in this study. Rather, I will couch the study in terms of Hawkins (2004) theory because this theory spells out the predictions for extraposition phenomena in a more detailed manner. In


E. J. Francis

addition, Hawkins theory is more comprehensive in making predictions for choice of structure in production as well as for processing ease in comprehension, and for dealing with diverse linguistic phenomena in a variety of typologically distinct languages. Hawkins (2004) theory of eciency and complexity in grammars sets out very specic predictions for weight eects in the production and comprehension of language. The main thrust of this theory is the principle of Minimize Domains, as dened in (6): (6) Minimize Domains: The human processor prefers to minimize the connected sequences of linguistic forms and their conventionally associated syntactic and semantic properties in which relations of combination and/or dependency are processed. The degree of this preference is proportional to the number of relations whose domains can be minimized in competing sequences or structures, and to the extent of the minimization dierence in each domain. (Hawkins 2004: 104)

One thing this principle predicts is that speakers should prefer to rearrange heavy constituents to minimize the domains in which relations between linguistic elements are processed. Several syntactic and semantic domains are relevant for processing, including syntactic dependencies in phrase structure, lexically specied dependencies between head words and their complements (e.g., subcategorization, theta roles, collocations), semantic dependencies between modiers and modied phrases, and various kinds of co-indexation (e.g., pronominal coreference, ller-gap dependencies). The most important domain for our purposes is the Phrasal Combination Domain (called Constituent Recognition Domain in earlier versions of the theory) as dened in (7): (7) Phrasal Combination Domain (PCD): the smallest string of elements required to construct a mother node and its immediate constituents (Hawkins 2004: 107).

The assumption is that constituents can be constructed as soon as the head word (or other constructing category) is encountered. In English Heavy NP-Shift, for example, the PCD for the VP includes the verb and head word of each of its complements (i.e., the verb, the preposition head of PP, and the determiner or noun introducing the NP). When the NP is heavy, the PCD for VP can be made smaller by moving the NP to a position following the PP, as shown in (7ab). (8) a. PCD for canonical VP The waiter brought the wine we had ordered to the table.

Grammatical weight b. PCD for VP with HNPS


The waiter brought to the table the wine we had ordered. A simple way of calculating PCDs is in terms of IC-to-word ratios (Hawkins 1994). In (8ab) the VP has three immediate constituents: the verb, the object NP, and the PP. Thus, the IC-to-word ratio for the VP in (8a) is 3/7 (43%), while the IC-to-word ratio for the VP in (8b) is 3/5 (60%). Thus, the shifted structure in (8b) is predicted to be more ecient than the canonical structure in (8a). Although no previous studies of grammatical weight have investigated RCE in English, Hawkins theory has been tested in two studies of a similar type of extraposition in German (Uszkoreit et al 1998; Konieczny 2000). Both studies looked at the eect of grammatical weight on preferences for constituent order in sentences similar to those in (9ab) below from Hawkins (2004: 137). Note that in German, the main verb comes at the end of the verb phrase (when there is an auxiliary verb), and the object NP comes before the verb. In this type of extraposition, the RC is moved from the canonical object position as in (9a) to a position following the verb as in (9b). I will refer to this type of extraposition as RCE from Object NP. (9) a. PCDs for canonical sentence
Er hat gestern das Buch das der alte Professor verloren hatte gefunden. he has yesterday the book that the old professor lost had found.

Object NP VP He found the book yesterday that the old professor had lost. IC-to-word ratios: Object NP 3/3 (100%), VP 3/10 (30%) b. PCDs for extraposition sentence
Er hat gestern das Buch gefunden das der alte Professor verloren hatte. he has yesterday the book found that the old professor lost had.

Object NP VP He found the book yesterday that the old professor had lost. IC-to-word ratios: Object NP 3/4 (75%), VP 3/4 (75%) As shown above in (9ab), the PCD for the object NP (i.e., the distance for integrating the head noun with its RC) is minimized in the canonical sentence, with an IC-to-word ratio of 100%. However, the PCD for VP (i.e., the distance for integrating the verb with its object) is minimized in the extraposition sentence, with an IC-to-word ratio of 75%. To determine


E. J. Francis

which structure is most ecient for a given sentence, what matters is the overall minimization of all the relevant domains. For example, when the RC is heavy and there is only a one-word intervening verb, as in (9b), the most ecient structure should be extraposition because the combined PCDs for Object NP and VP are minimized. This is shown in both the higher average IC-to-Word ratio for extraposition (75% vs 65%) and in the lower number of words needed to construct both the NP and VP domains in the extraposition sentence (8 words vs. 13 words) (Hawkins 2004: 138). More generally, extraposition should be more ecient than the corresponding canonical structure to the extent that RC length exceeds extraposition distance (i.e., length of the main verb and any modiers preceding it). This is because making the RC longer increases the VP domain in the canonical sentence, while making the extraposition distance shorter decreases the NP domain in the extraposition sentence. Both Uszkoreit et al (1998) and Konieczny (2000) found some evidence in support of Hawkins predictions for German RCE. In a corpus study that looked at sentences similar to those in (8ab) above, Uszkoreit et al (1998) found that the rate of extraposition was highest when extraposition occurred over a short distance (a one-word verb) and when the RC itself was long. For example, extraposition occurred about 95% of the time for sentences with a one-word main verb (gefunden in 8 above) but only about 10% of the time when the main verb was preceded by a four-word modifying phrase (Uszkoreit et al 1998: 7). The length of the RC also affected the rate of extraposition. For example, when a one-word phrase preceded the verb and the RC was short (two or three words), extraposition only occurred in about 33% of the relevant sentences. However, this increased to 82% when the RC was long (1015 words) (Uszkoreit et al 1998: 9). Uszkoreit et al also conducted an acceptability judgment task for which they systematically varied the extraposition distance and the length of the RC. As expected, extraposition sentences were rated higher as RC length increased and as extraposition distance decreased. Contrary to expectation, however, canonical sentences were rated as more acceptable than extraposition sentences in almost all cases. Extraposition sentences were rated as highly as canonical sentences only when the extraposition distance was short (one word) and the RC was long (1998: 12). Konieczny (2000) conducted an acceptability judgment experiment and a self-paced reading experiment to investigate weight eects in the processing of German RCE. Results of the acceptability judgment task were very similar to the results of the acceptability study by Uszkoreit et al (1998). Acceptability of extraposition was highest when the extraposition distance was short (one word) and when the RC was medium or long (Konieczny 2000: 638639). Also similar to Uszkoreit et als (1998) re-

Grammatical weight


sults, acceptability of canonical sentences was overall higher than acceptability of extraposition sentences in all conditions. Konieczny (2000) also conducted a self-paced reading experiment in which it was found that, as expected, the relative pronoun was read slower when the RC was extraposed, indicating some processing cost for integrating the noun with a nonadjacent RC. However, contrary to locality-based predictions, reading time at the main verb was signicantly slower in the extraposition sentences than in the canonical sentences even though the verb was closer to its object in the extraposition sentences. Konieczny (2000: 643) explains this eect (sometimes called antilocality since there is a processing advantage for non-local structures) as possibly the result of the RC in the canonical sentence helping readers anticipate the phrase-nal verb by providing additional information about one of its arguments and by allowing extra time for readers to deduce information about the verb. Also contrary to locality-based predictions, longer RCs did not signicantly aect reading time of the verb, nor did longer extraposition distance have any signicant eect on reading time of the relative pronoun. Given the very dierent results of the acceptability task (which patterned with Uszkoreit et als acceptability data) and the self-paced reading task, Konieczny (2000: 644) suggests that locality-based predictions are clearest for production, whereas online comprehension is subject to other eects such as anticipatory processing. However, he does not explain why locality eects in comprehension have been shown in other studies.4 RCE from Object NP in German is not quite the same as RCE from Subject NP in English, but the predictions of Hawkins theory are quite similar. As illustrated in (9ab), the PCD for the subject NP is minimized in the canonical sentence, while the PCD for the matrix clause is minimized in the extraposition sentence. Similar to the case of German RCE, extraposition is preferred to the extent that the length of the RC exceeds the length of the matrix VP. When the RC is heavier than the VP, as in (10ab), there is predicted to be an overall advantage for extraposition because the combined minimization, as measured in total number of words needed to construct both domains (Hawkins 2004: 138), is smaller for extraposition. In (10ab), for example, the total is 14 for the canonical sentence and 6 for the extraposition sentence. The same advantage for extraposition is also shown in the IC-to-Word ratios, where the average

4. See Vasishth and Lewis (2006) for additional evidence of anti-locality eects with headnal structures in Hindi. The authors attempt to explain both locality and anti-locality eects using a theory of activation decay.


E. J. Francis

ratio is higher for extraposition than for canonical structure (87.5% vs 59%). (10) a. PCDs for canonical sentence
New sets that were able to receive all the TV channels appeared.

Subject NP Matrix S IC-to-word ratios: Subject NP 3/3 (100%), Matrix S 2/11 (18%) b. PCDs for extraposition sentence
New sets appeared that were able to receive all the TV channels.

Subject NP Matrix S IC-to-word ratios: Subject NP 3/4 (75%), Matrix S 2/2 (100%) The experiments reported here are based on the predictions of Hawkins theory as outlined above.5 Experiment 1 uses a self-paced reading task to test the eects of grammatical weight on the processing of canonical vs. extraposition structures in English. Similarly, Experiment 2 investigates the eects of grammatical weight on acceptability judgments for the same kind of sentences as in Experiment 1. With the weight of the VP held constant (longer than the RC in the light RC condition and shorter than the RC in the heavy RC condition), predictions for these experiments (based on PCDs alone) are as follows:6



It should be noted that Hawkins (2004) theory takes account of several domains in addition to PCD. Within the Subject NP, there is a semantic dependency between the head noun and the RC, since the RC restricts the meaning of the noun, and there is co-indexing between the head noun and the relative pronoun. Similarly, within the Matrix S, there are additional dependencies between the predicate and the subject (e.g., selectional restrictions, theta role assignment). However, since there are additional dependencies within both the NP and VP domains, the overall predictions turn out to be very similar to the predictions for PCD alone. Therefore, following John Hawkins suggestion (p.c. 2007) the hypotheses for the current study are based on the simpler metric of PCDs alone. As an anonymous reviewer points out, additional factors other than weight or domain minimization could invalidate the rst two hypotheses. However, since weight is the only factor that was manipulated, with other known factors held constant, and since sentence materials were designed to be fully felicitous with both extraposed and canonical structures (apart from any weight eects), I will assume as a starting point that the rst two hypotheses should hold.

Grammatical weight


When the RC is shorter than the VP, canonical sentences should be read faster and judged as more acceptable than extraposition sentences. When the RC is longer than the VP, extraposition sentences should be read faster and judged as more acceptable than canonical sentences. As grammatical weight of the RC increases, acceptability ratings for canonical sentences should decrease and reading times for canonical sentences (mean RT per word) should slow down because of the longer distance for integrating the subject NP with the main verb.

However, note that if reading time is subject to anti-locality eects as found in Koniecznys (2000) study of German RCE, canonical sentences should be read faster than extraposition sentences regardless of RC weight because the RC should facilitate processing of the verb. An analysis of RCE in the ICE-GB corpus investigates the related issue of how grammatical weight aects speakers choice of structure in language use. Since the principle of Minimize Domains applies to production as well as perception, the theoretical predictions for the corpus analysis are similar to those of the experiments. All else being equal, extraposition should be preferred in language use in cases where RC length exceeds VP length, and this preference should be stronger as the dierence in length becomes greater. When all relevant examples of sentences with RCs modifying the subject are included in the analysis, we predict the following: The proportion of sentences with extraposition should be highest when the ratio of VP length to RC length is lowest (i.e., when the RC is much longer than the VP) and should decrease as this ratio increases. For extraposition sentences, the RC should be longer on average than the VP, while the converse should be true of canonical sentences. These weight eects are distinct from the eects of predicate type and should hold when predicate type is held constant.

Sections 35 report the results of Experiments 12 and the corpus analysis. We will see that all of the results showed signicant eects of grammatical weight in the expected direction, and that the predictions were borne out most clearly in the corpus analysis.


Experiment 1: Reading time

The goal of Experiment 1 was to test whether grammatical weight aected processing eciency of sentences with an extraposed or non-extraposed RC modifying the subject NP. A self-paced reading task, for which whole


E. J. Francis

sentence reading time was measured, was used to assess processing eciency. As is standard for self-paced reading tasks, it was assumed that faster reading times indicate faster, more ecient processing, at least for sentences that are understood correctly. 3.1. Methods

3.1.1. Participants. Forty participants were recruited from the student body at Purdue University. All were native speakers of American English between the ages of 18 and 39 (average age 22). There were 17 men and 23 women. Participants gave informed consent and were each paid $6 for completing a 3545 minute session. 3.1.2. Materials. Experimental stimuli consisted of ve sets of nine sentences each in a 3 3 repeated measures design. Three levels of RC weight (4, 8, and 15 words) and three levels of structure (canonical RC, extraposed RC, adjunct clause) were manipulated, and VP length was held constant at ve words. Lexical content of the sentences was chosen to be maximally felicitous in both RCE and canonical congurations. To satisfy semantic and pragmatic conditions on RCE, only indenite, quantied subject NPs were used, and only intransitive, unaccusative VPs were used (see Rochemont and Culicover 1990: 6068). In addition, sentences were constructed so that the RC could readily be construed as new information. A sample stimulus set is shown in Table 1. Grammatically acceptable ller sentences of varying lengths were also included in the experiment. 3.1.3. Procedure. Following a brief background questionnaire, participants were presented with a series of sentences on a computer screen. Each sentence was presented in its entirety, and participants were instructed to press a button as soon as they had read and understood the sentence. Following each sentence, participants were presented with a true-or-false question about the content of the sentence (see Appendix A for exact instructions). To ensure validity of the results, only sentences with accurate responses to the comprehension question were included in the analysis of reading time. An E-Prime program was used to present the stimuli and record the whole-sentence reading times and true-false responses. Sentences were presented in 4 blocks of 33 sentences each. Each block consisted of 11 test sentences and 22 llers, except for block 4, which consisted of 12 test sentences and 23 llers. For each participant, sentences were ordered randomly within each block, and ordering of blocks was

Grammatical weight


also random. Participants were given the opportunity to take a break at the end of each block. 3.1.4. Hypotheses. Following Hawkins (2004), it was predicted that that there should be a reading time advantage for extraposition over canonical structure when the RC is heavier than the VP due to the longer distance for integrating the verb with its subject. Conversely, there should be a reading time advantage for canonical structure over extraposition when the RC is lighter than the VP because integration of the subject noun with the RC is easier when they are adjacent. Finally, reading time for canonical sentences should get slower as RC weight increases due to the increased distance for integrating the subject with the main verb. These predictions are reected in the total number of words for combined PCDs, as shown in Table 1. Extraposed relatives stay the same in all three weight conditions, at 10 words. Canonical relatives start at 9 words in the light condition, where they should be read faster than extraposed relatives, but increase to 13 and 20 words in the medium and heavy conditions, where they should be read slower than extraposed relatives. A third structureadjunct clausewas included as a control condition. This type consisted of a main clause identical to the main clause of the other two structures and a nite subordinate clause adjoined to the main clause following the VP (see examples in Table 1). Similar to extraposed RCs, adjunct clauses occur at the end of the sentence, but adjunct clauses are not involved in any discontinuous dependency with the subject NP. Therefore, reading times for sentences with adjunct clauses are predicted to be faster than reading times for extraposition sentences when the clause is light, similar to canonical sentences. However, since adjunct clauses do not intervene between the subject and the verb (unlike canonical RCs), reading times for adjunct sentences are not expected to slow down in the medium and heavy conditions. In terms of total number of words for combined PCDs, sentences with an adjunct clause stay at 9 words in all three conditions. 3.2. Results

For purposes of comparison across light, medium, and heavy conditions, mean reading time per word rather than whole sentence reading time was used in the analysis. Mean reading time per word (henceforth RTW) was calculated by dividing each whole sentence reading time by the number of words in the sentence. All test sentences within each category (light, medium, heavy) were of the same length in words and the three length conditions diered only in the length of the relative clause. Sentences within


E. J. Francis

Table 1. Sample stimulus set consisting of three clause weights and three structures Condition Light 4 words Canonical RC Sample stimulus sentence Three people who were from Chicago arrived here early yesterday morning. Three people arrived here early yesterday morning who were from Chicago. Three people arrived here early yesterday morning after they left Chicago. Three people who were from a northern suburb of Chicago arrived here early yesterday morning. Three people arrived here early yesterday morning who were from a northern suburb of Chicago. Three people arrived here early yesterday morning after they left a northern suburb of Chicago. Three people who were originally from a far northern suburb of Chicago which is called Lake Forest arrived here early yesterday morning. Three people arrived here early yesterday morning who were originally from a far northern suburb of Chicago which is called Lake Forest. Three people arrived here early yesterday morning after they secretly left a far northern suburb of Chicago which is called Lake Forest. IC-to-word ratios Subject NP: 3/3 (100%) Matrix S: 2/6 (33%) Total words: 9 Subject NP: 3/8 (37.5%) Matrix S: 2/2 (100%) Total words: 10 Subject NP: 2/2 (100%) Matrix S: 3/7 (43%) Total words: 9 Subject NP: 3/3 (100%) Matrix S: 2/10 (20%) Total words: 13 Subject NP: 3/8 (37.5%) Matrix S: 2/2 (100%) Total words: 10 Subject NP: 2/2 (100%) Matrix S: 3/7 (43%) Total words: 9 Subject NP: 3/3 (100%) Matrix S: 2/17 (12%) Total words: 20

Extraposed RC

Adjunct clause

Medium 8 words

Canonical RC

Extraposed RC

Adjunct clause

Heavy 15 words

Canonical RC

Extraposed RC

Subject NP: 3/8 (37.5%) Matrix S: 2/2 (100%) Total words: 10

Adjunct clause

Subject NP: 2/2 (100%) Matrix S: 3/7 (43%) Total words: 9

each condition of each stimulus set consisted of the same words, with the exception of the adjunct clause condition where the rst three words of the adjunct clause had to be changed to accommodate the dierent clause type (see Table 1). As shown in Figure 1 and Table 2, RTW for all three structures decreased (got faster) as clause weight increased. This trend was strongest

Grammatical weight


Figure 1. Mean reading time per word by clause weight

Table 2. Mean reading time per word by clause weight Light Canonical mean std. error n mean std. error n mean std. error n 356.44 15.38 40 381.88 23.84 40 361.93 15.32 40 Medium 322.07 16.90 40 346.40 21.66 40 341.43 16.43 40 Heavy 330.53 21.22 40 292.94 14.83 40 346.62 20.99 40



for extraposition sentences, for which mean RTW decreased from 382 ms in the light condition to 293 ms in the heavy conditiona dierence of 89 ms. In contrast, RTW for canonical and adjunct sentences decreased by only 25 ms and 15 ms, respectively. A two-way repeated measures ANOVA showed a signicant main effect of clause weight by participants, but the eect did not reach signicance in the by items analysis: F12; 38 6:56, p < 0:01; F22; 3 4:07, p 0:14. There was also a signicant interaction between clause weight and structure in the participant analysis, but the interaction did not reach signicance in the item analysis: F14; 36 3:18, p 0:02; F24; 1 11:15, p 0:22. No main eect for structure was found:


E. J. Francis

F12; 38 0:66, p 0:52; F22; 3 0:33, p 0:74. The dierence between participant and item analyses is most likely due to the fact that there were forty participants, but only ve items (sentence sets) were used. Pairwise t-tests conrm that the main eect of weight and the interaction between weight and structure were primarily due to the decrease in RTW for heavy extraposition sentences. While there was a signicant difference in RTW between light and heavy extraposition sentences (t 4:77, p < 0:01), there was no signicant dierence between light and heavy canonical sentences (t 1:33, p 0:19) or between light and heavy adjunct sentences (t 0:91, p 0:37). RTWs for extraposition sentences were signicantly faster than for canonical sentences in the heavy clause condition (t 2:59, p 0:01). In the light clause condition, RTWs for canonical sentences were slightly faster than for extraposition sentences, but this dierence was not signicant (t 1:45, p 0:16). 3.3. Discussion

As predicted by Hawkins (2004) theory, there was a signicant reading time advantage for extraposition over canonical structure in the heavy clause condition. Unlike Koniecznys (2000) ndings for German RCE, in which readings times on the main verb were uniformly faster for canonical sentences than for extraposition, the present study found no antilocality eects. The dierent results for German vs. English are not necessarily at odds with each other, however. Konieczny (2000) measured reading time on the main verb, whereas the present study measured only whole sentence reading time.7 While Koniecznys results show a consistent advantage for canonical structure with respect to reading time on the verb, possibly reecting readers early anticipation of the verb, it is still possible that the reading time for the entire sentence might have shown no such advantage, or even an advantage for extraposition, if such data had been collected. Arguably, the whole sentence reading times collected for the present study are a more direct reection of overall processing time for the sentence than are the localized reading times reported in the German study, and therefore might be expected to align more closely with corpus data, as they in fact do (see Section 5 below).8 Due
7. Another dierence between Koniecznys (2000) study and the present study is that in German RCE from Object NP, a head-nal transitive verb must be integrated with the direct object preceding it, but in English RCE from Subject NP, a head-initial intransitive verb must be integrated with the subject. Since all verbs need a subject in English, and since the subjects in the stimuli were equally plausible with or without the relative clause, there might be less reason for anticipatory eects to occur. I am grateful to an anonymous CL reviewer for pointing this out.


Grammatical weight


to limitations of both studies, however, it is not possible to compare the results directly. Some of the other ndings are not exactly as predicted by Hawkins (2004). Although reading times were slightly faster for canonical and adjunct sentences than for extraposition sentences in the light condition, these dierences were not signicant. Despite the discontinuous dependency in the extraposition sentences, there was no evidence of added processing cost for extraposition. This result may, however, be due to limitations of the experimental design. In the light clause condition, the RC had four words and the VP had ve words, so that only a one-word advantage for canonical structure was predicted. It may be that the dierence between the relevant NP and Matrix S domains was too small to show up as a dierence in reading time. If, instead, the experiment had included a light clause condition in which the VP was twice as long as the RC, as in (11ab) below, we might have found a signicant advantage of canonical structure over extraposition. (11) a. b. Three people who were from Chicago arrived in Albuquerque early yesterday morning at 6 am. (Canonical) Three people arrived in Albuquerque early yesterday morning at 6 am who were from Chicago. (Extraposition)

Experimental evidence for a dierence in reading time between sentences like (11a) and (11b) awaits future research. However, in Section 5 below, we will see evidence from a corpus analysis that canonical structure is in fact strongly preferred in language use when VP length exceeds RC length. Also contrary to prediction, RTW for canonical sentences did not slow down as clause weight increased, but instead got slightly (though not signicantly) faster. In addition, RTW for extraposition sentences got signicantly faster as clause weight increased even though increased clause weight would have had no eect on the relevant PCDs. Because only whole sentence reading times were measured, it is dicult to determine why this happened. However, one possibility is that there was a general eect of sentence length such that longer sentences were read faster per word than shorter sentences. If so, this could help explain both patterns. For canonical sentences, the general speeding up with longer sentences might have counteracted the predicted slowing eect of heavy RCs, resulting in no signicant change in RTW. For extraposition sentences, the speeding up of RTW in the heavy condition could be entirely due to a general eect of sentence length, since heavier RCs would have had no effect on the PCDs for extraposition. However, since I am unaware of any


E. J. Francis

previous studies showing such an eect, this explanation must be considered speculative. 4. Experiment 2: Acceptability judgment task

The goal of Experiment 2 was to test whether grammatical weight affected acceptability judgments of sentences with an extraposed or nonextraposed RC modifying the subject NP. A written survey was used to collect sentence judgments using a 9-point scale. It was expected that acceptability judgments should reect domain minimization preferences in a similar manner to reading time. 4.1. Methods

4.1.1. Participants. Thirty-one participants were recruited from the student body at Purdue University. All were native speakers of American English between the ages of 18 and 54 (average age 24). There were 19 men and 12 women. Participants gave informed consent and were each paid $6 for completing a 3545 minute session. Data from one participant were excluded from the analysis because the participant turned out to be a native speaker of Spanish. Data from thirty participants were included in the analysis. People who had participated in Experiment 1 were excluded from the subject pool. 4.1.2. Materials. Sentence materials were the same as in Experiment 1 (see Table 1), with the exception of the ller sentences. As in the previous experiment, ve sets of nine test sentences were used. Each set included three levels of RC weight (4, 8, and 15 words) and three levels of structure (canonical RC, extraposed RC, adjunct clause). VP length was held constant at ve words. Unlike in Experiment 1, where only grammatically acceptable ller sentences were used, ller sentences covered a wide range of dierent levels of acceptability as well as dierent sentence lengths. Fillers were categorized in advance as good, medium, or bad in acceptability based on judgments of the same or similar sentences given in published sources such as syntax textbooks and research articles. 4.1.3. Procedure. Following a brief background questionnaire, participants were asked to complete a written survey. The survey consisted of a series of sentences for which participants were asked to rate each sentence on 9-point scale, where 9 means completely acceptable and 1 means completely unacceptable (see Appendix B for exact instructions). Rating scores were entered by circling a number from 1 to 9 below each sentence.

Grammatical weight


As in Experiment 1, sentences were presented in 4 blocks of 33 sentences each. Each block consisted of 11 test sentences and 22 llers, except for block 4, which consisted of 12 test sentences and 23 llers. Participants lled out one of four dierent survey scripts with two dierent orderings of sentences within each block and two dierent orderings of blocks. Pseudo-random order was used to arrange sentences within each block to avoid similar sentences on the same page of the survey. Participants were given the opportunity to take a break at the end of each block. Responses were later entered by hand into an Excel spreadsheet. 4.1.4. Hypotheses. Because canonical and extraposition sentences are equally grammatical (in the sense of being permitted by the grammar), and acceptability judgments are predicted to follow domain minimization preferences, weight-based predictions for acceptability are similar to those for reading time in Experiment 1. Following the domain minimization preferences shown in Table 1, it was predicted that extraposition should be rated higher than canonical structure when the RC is heavier than the VP (medium and heavy conditions) due to the longer distance for intergrating the verb with its subject. Conversely, canonical sentences should be rated higher than extraposition sentences when the RC is lighter than the VP (light condition) due to the longer distance between the subject noun and the RC. Acceptability of canonical sentences should decrease as RC weight increases due to the increased distance between the subject and the main verb. Finally, adjunct sentences should be rated similarly to canonical sentences (and higher than extraposition sentences) in the light condition, but unlike canonical sentences, ratings of adjunct sentences should not decrease in the medium and heavy conditions. 4.2. Results

As shown in Figure 2 and Table 3, mean ratings for canonical sentences decreased from 8.05 in the light clause condition to 6.69 in the heavy clause condition. Ratings for adjunct sentences similarly decreased from 7.63 in the light condition to 6.97 in the heavy condition. In contrast, ratings for extraposition sentences started lower, at 6.33 in the light condition, and increased slightly to 6.41 in the heavy condition. A two-way repeated measures ANOVA showed a signicant main effect for weight in both the participant and item analyses: F12; 28 11:06, p < 0:01; F22; 3 19:15, p 0:02. There was also a signicant main eect of structure for both participant and item analyses: F12; 28 13:13, p < 0:01; F22; 3 60:84, p < 0:01. There was a signicant interaction between clause weight and structure in the participant analysis


E. J. Francis

Figure 2. Mean acceptability ratings by clause weight9

Table 3. Mean acceptability ratings by clause weight Light Canonical mean std. error n mean std. error n mean std. error n 8.05 0.16 30 6.33 0.33 30 7.63 0.24 30 Medium 8.09 0.15 30 6.67 0.29 30 7.60 0.18 30 Heavy 6.69 0.25 30 6.41 0.31 30 6.97 0.24 30



but this trend did not reach signicance in the item analysis: F14; 26 8:59, p < 0:01; F24; 1 17:87, p 0:18. Pairwise comparisons suggest that the main eect of structure was due to the lower acceptability of extraposition sentences as compared with canonical and adjunct sentences in the light and medium conditions. Canonical sentences were rated signicantly higher than extraposition sen-


Error bars in all gures represent standard error of the mean and are based on the by participant data.

Grammatical weight


tences in the light condition (t 5:94, p < 0:01), and in the medium condition (t 5:83, p < 0:01), while there was no signicant dierence between canonical and extraposition sentences in the heavy condition (t 1:24, p 0:23). Adjunct sentences patterned similarly to canonical sentences, except that ratings did not decrease as much in the heavy condition. Adjunct sentences were rated signicantly higher than extraposition sentences even in the heavy condition (t 2:60, p 0:01). The main eect of weight and the interaction between weight and structure were apparently due to the decrease in ratings for canonical and adjunct sentences (but not for extraposition sentences) in the heavy condition. Pairwise t-tests conrm that there was a signicant dierence between light and heavy canonical sentences (t 6:34, p < 0:01) and between light and heavy adjunct sentences (t 3:25, p < 0:01), but no signicant dierence between light and heavy extraposition sentences (t 0:45, p 0:66). 4.3. Discussion

Canonical and adjunct sentences started at high acceptability in the light condition and decreased in acceptability in the heavy condition. On the other hand, extraposition sentences stayed at moderate acceptability in all three conditions, converging with canonical sentences in the heavy condition (though adjunct sentences were still rated higher). This pattern of results partially conrms the initial hypotheses. As predicted, acceptability ratings for canonical sentences decreased as clause weight increased. Also as predicted, there was a signicant advantage for canonical sentences and adjunct sentences over extraposition sentences in the light clause condition. Contrary to prediction and unlike in Experiment 1, however, there was no advantage for extraposition in the heavy condition. Rather, the advantage for canonical sentences shown in the light and medium conditions simply disappeared in the heavy condition. Also unexpectedly, ratings for adjunct sentences decreased in the heavy condition, though not as much as ratings for canonical sentences did. While clear eects of grammatical weight were shown in both Experiment 1 and Experiment 2, it is interesting that the results patterned in a dierent way. Reading time results showed no dierence among structures in the light condition, but an advantage for extraposition in the heavy condition. In contrast, acceptability results showed an advantage for canonical structure in the light condition, but no signicant dierence among structures in the heavy condition. It is likely that the dierent results are due to dierences between the two tasks. While reading time is an online measure assumed to reect processing diculty in a relatively


E. J. Francis

direct manner, acceptability judgments are collected oine and may be subject to additional factors such as prescriptive rules and stylistic considerations. Since extraposition from NP occurs naturally in both spoken and written English, we know that this structure is permitted by the grammar of English. Therefore, all the test sentences used in Experiments 1 and 2 should be grammatical in the sense of being part of speakers grammatical competence. However, in prescriptive terms, extraposed RCs are a type of misplaced modier of which highly educated people (in this case, university students) are likely to be aware. The misplaced modier is especially noticeable since the subject-modifying RC occurs completely outside the subject NP in a position normally reserved for VP or sentence modiers. In addition, it is possible that the lower acceptability for RCE reects a frequency-based preference for adjacency between the head noun its modifying relative clause. As shown in a corpus study reported in section 5 below, RCE occurred in only 15% of the relevant cases. Although both canonical order and RCE are licensed by the competence grammar, speakers may have a sense that the canonical order is more frequent and therefore more basic, resulting in lower acceptability ratings for RCE sentences.10 Finally, although the lexical content was chosen to be fully felicitous with extraposition, the isolated sentences in the experiment were not situated in any discourse context, possibly drawing more attention to the extraposition structure than a natural context would have done. It is plausible, therefore, that the discrepancy between reading time and acceptability results could be explained as follows. In the light clause condition, there was no dierence in reading time between canonical and extraposition sentences, suggesting that processing eciency for the two structures is similar. Thus, a prescriptive and/or frequency-based bias in favor of canonical order might help explain the lower acceptability of extraposition in the light condition. In the heavy clause condition, there was a clear reading time advantage for extraposition over canonical structure, suggesting that extraposition sentences are processed more eciently than canonical sentences when the RC is heavy ( just as the theory predicts). The weight-based advantage in processing eciency for extraposition sentences might have counteracted the normative bias in favor of canonical structure, resulting in approximately equal acceptability for canonical and extraposition structures in the heavy condition. However, it is still not clear why adjunct sentences decreased in acceptability in the heavy condition rather than staying the same in all conditions.

10. For example, see Bresnan (2006) for evidence that fully grammatical but infrequent or improbable structures are commonly judged as less acceptable.

Grammatical weight


It is interesting that Uszkoreit et al (1998) and Konieczny (2000) got very similar results to these in their acceptability judgment experiments for RCE sentences in German. Uszkoreit et al (1998: 12) found that canonical sentences were rated higher than extraposition sentences overall, but that this dierence was neutralized when the extraposition distance was short and the RC was long. Similarly, Konieczny (2000: 638639) found that canonical sentences were rated higher in all conditions, but that there was the least dierence between canonical and extraposition sentences when the extraposition distance was short and the RC was long. A frequency-based explanation is less plausible for German than for English, since RCE is more common in German. For example, Uszkoreit et al (1998: 6) found that extraposition occurred in 43% (340 out of 789) of the relevant cases. Therefore, I conjecture that prescriptive rules or other stylistic considerations aected metalinguistic judgments in both English and German, causing extraposition sentences to be rated somewhat lower than the domain minimization preferences would predict, while a frequency-based bias in favor of canonical structure might have been an additional factor for English. We now turn to an additional kind of evidence for grammatical weight eectsa corpus study of naturally-occurring speech and writing.


Corpus analysis

Hawkins (2004) idea of processing eciency through domain minimization applies not just to comprehension but also to production. It is therefore predicted that preferences for choice of one structure over another in language use should reect domain minimization preferences. This prediction has been conrmed in previous corpus studies of Heavy NP Shift, Particle Shift, and Dative Shift in English (Arnold et al. 2000; Lohse et al. 2004; Wasow 1997) as well as RCE in German (Uszkoreit et al 1998). For the current study, it was predicted that RCE from Subject NP should occur most often in naturally occurring speech and writing when the RC is longer than the VP, and that incidence of extraposition should decrease as the ratio of VP to RC length increases. 5.1. Methods

5.1.1. Corpus. The International Corpus of English Great Britain (henceforth ICE-GB) was used for this study (Nelson, Wallis, and Aarts 2002). The corpus includes about one million words of British English in a variety of genres of both speech and writing.


E. J. Francis

5.1.2. Data collection and coding. Finite and non-nite clauses for which the subject NP was modied by a RC were collected by means of category and tree fragment searches in the ICE-GB corpus. Each clause was then coded by hand for the following categories: extraposition status, Subject NP length, VP length, RC length, verb complex, main predicate, predicate type, RC type, head noun of RC, relative pronoun, and discourse type. RCs were classied as extraposed or canonical depending on their position, with canonical RCs occurring within the subject NP and extraposed RCs occurring at the end of the sentence following the VP. Phrase length was counted in words as dened by units of text separated by blank spaces. Repetitions and restarts were excluded from the word count. Verb complex included the main verb and its auxiliaries in the exact form used in the sentence. Main predicate was coded as the lemma form of the main verb (or adjective, in the case of predicate adjectives). The main predicate was then assigned to one of following predicate types: transitive action verb, transitive stative verb, intransitive unergative verb, intransitive unaccusative verb, passive verb, copular verb, raising verb, and predicate adjective. The distinction between unaccusative and unergative was based on the presentational there test: unaccusative verbs are those verbs that t into a sentence such as: There arrived three guests. Passive verbs, which were almost always forms of transitive action verbs, were included as a category separate from both transitive and unaccusative verbs. RC type was coded according to the following categories based on the grammatical function of the relative pronoun within the RC: subject, direct object, object of preposition, possessive, and adjunct. Grammatical function of the head noun was not coded, since the head noun was always the subject. The form of the relative pronoun was also noted (e.g., which, that, who, where, to whom, etc.). Discourse type was coded as either speech or writing depending on whether the sentence came from a spoken or written source. 5.1.3. Hypotheses. Based on the same domain minimization preferences used in Experiments 1 and 2, extraposition should be preferred to the extent that RC length exceeds VP length. This is because extraposition minimizes the combined PCDs for Subject NP and Matrix S in just those cases for which the RC is longer than the VP. (See predictions for reading time in Section 3 above). It was therefore predicted that, in a corpus of sentences with a subject-modifying RC, the RC should be longer on average than the VP when extraposition is used. Conversely, the RC should be shorter on average than the VP when the canonical structure is used. In addition, the proportion of sentences with extraposition should

Grammatical weight


be highest when the ratio of VP length to RC length is lowest and should decrease as this ratio increases. Although spoken English is arguably a more direct reection of online demands in sentence production than written English is, previous studies of grammatical weight have found similar eects in speech and writing (Wasow 1997: 99). Therefore, it was predicted that similar weight eects would show up in both spoken and written sentences in the ICE-GB corpus. The category of predicate type was coded to see whether lexical and semantic properties of the RCE sentences were in line with previous ndings and to test whether grammatical weight eects are independent of predicate type. Gueron (1980), Kuno and Takami (2004), Rochemont and Culicover (1990), and Takami (1999) have proposed that extraposition from subject NP is subject to certain semantic or pragmatic constraints that aect which predicate types occur with RCE. Based on this previous research, intransitive, unaccusative predicates were predicted to occur most frequently in RCE sentences because of the tendency for predicates of this type to represent old or backgrounded information and to serve a presentational function with respect to the information following the verb (see Rochemont and Culicover 1990: 6568). Transitive predicates, which rarely serve this kind of function, were predicted to occur only infrequently with extraposition. However, following Rochemont and Culicover (1990: 6568), it was expected that no predicate types should be completely excluded from occurring with extraposition because the relevant restriction is contextual rather than strictly lexical or semantic in nature. Finally, the eects of grammatical weight were predicted to hold independently of predicate type. 5.2. Results

Of the 391 sentences collected for this study, 332 (85%) had canonical structure, while 59 (15%) had extraposition. As predicted, extraposed RCs were signicantly longer than the VP on average (12.36 vs. 3.44 words; t 8:19, p < 0:01), while nonextraposed RCs were signicantly shorter than the VP (8.37 vs. 12.94 words; t 8:06, p < 0:01), as shown in Figure 3. Also as predicted, the proportion of sentences with extraposition consistently decreased as the ratio of VP length to RC length increased. While 91% of sentences for which the RC was at least ve times longer than the VP (VP-to-RC ratio of 0.2 or less) involved extraposition, only 2% of sentences for which the RC was the same length or shorter than the VP (VP-to-RC ratio of 1.0 or greater) involved extraposition, as shown in


E. J. Francis

Figure 3. Mean length of VP and RC for sentences with extraposed vs. canonical RCs

Figure 4. Percentage of extraposed RCs by ratio of VP length to RC length

Figure 4. No examples of extraposition were found when the VP was more than 1.3 times longer than the RC. Thus, the weight of the RC in relation to the VP appears to be a strong predictor of extraposition. Similar results were found looking at VP length alone: 90% of sentences with a one-word VP involved extraposition while only 32% of sentences with four-word VPs had extraposition, and no examples of extraposition were found when the VP had more than 11 words. Results for RC length alone were less dramatic, but in the predicted direction. Only 12% of three and four word RCs were extraposed, but this increased to 33% for RCs of 15 words or more. Results for discourse type (spoken vs. written) and predicate type were in line with previous ndings. Similar to the results of Wasows (1997: 99) study of weight eects in double object constructions, length of VP in relation to RC showed the same general pattern for speech and writing. As shown in Table 4, canonical RCs were shorter than the VP on aver-

Grammatical weight
Table 4. Mean VP and RC length (words) in spoken vs. written sentences Speech Extraposition Canonical VP length RC length VP length RC length 3.60 12.88 12.95 6.92


Writing 3.32 11.97 12.94 9.31

age while extraposed RCs were longer than the VP on average for both spoken and written sentences. Findings for predicate type were compatible with Rochemont and Culicovers (1990) observations. As shown in Figure 5, the two most common predicate types occurring with extraposition were passive (46%) and unaccusative (24%), together accounting for 70% of all RCE sentences. In contrast, transitive and copular predicates were the two most common predicate types occurring with canonical clauses. It is interesting that passives were more frequent than unaccusatives, since only unaccusatives had been discussed in previous work on RCE. This novel nding underscores an advantage of using quantitative corpus data, since previous studies used mostly constructed examples or individual attested examples (e.g., Huck and Na 1990, 1992; Rochemont and Culicover 1990; Takami 1998). However, these results are still in line with the prediction that unaccusatives should be the most frequent predicate type, provided that passive predicates are classied together with unaccusatives. Although passives and unaccusatives do not have exactly the same behavior, this categorization is justiable based on the similar syntactic, semantic, and aspectual properties of passive and unaccusative predicates in English and other languages (cf. Burzio 1986) as well as their similar discourse functions.11 As expected, there were no strict constraints, since all six predicate types were represented in both data sets (Figure 5). These results are consistent with Rochemont and Culicovers claim that the apparent restriction on predicate type is pragmatic in nature rather than strictly semantic, syntactic, or lexical (1990: 6668). As shown in Table 5, the frequency of extraposition was lower than the frequency of canonical structure overall (15% vs 85%), but it was much

11. For this study, (active) unaccusative predicates were identied based on their ability to occur with presentational there. Note that many passive predicates also allow presentational there: i. There appeared no evidence of any mistakes. (unaccusative) ii. There was found no evidence of any mistakes. (passive)


E. J. Francis

Figure 5. Predicate types for canonical and extraposition sentences

Table 5. Frequency of extraposed and canonical clauses by predicate type Predicate type Unaccusative/passive Other predicate types Total Extraposed 37% (n 41) 6% (n 18) 15% (n 59) Canonical 63% (n 69) 94% (n 263) 85% (n 332) Total 100% (n 110) 100% (n 281) 100% (n 391)

higher with unaccusative/passive predicates (37%) than with other predicate types (6%). Thus, it appears that both predicate type and weight help predict the occurrence of extraposition. Furthermore, it appears that predicate type is independent of weight, since weight eects were apparent even when predicate type was held constant. Figure 6 shows the proportion of clauses with extraposition by weight ratio for all clauses containing passive or unaccusative predicates. Similar to the overall results shown in Figure 4, 100% (17 of 17) showed extraposition when the RC was at least ve times longer than the VP, while only 8% (4 of 51) showed extraposition when the RC was the same length or shorter than the VP. Similar results were found when examining only the clauses containing other predicate types (i.e., all clauses except those with a passive or unaccusative predicate): 67% (4 of 6) showed extraposition when the RC was at least ve times longer than the VP, while only 1% (2 of 190)

Grammatical weight


Figure 6. Percentage of extraposed RCs by ratio of VP length to RC length: Unaccusative and passive verbs only

showed extraposition when the RC was the same length or shorter than the VP. Following the approach of Diessel (2008), a binary logistic regression analysis was conducted to conrm whether grammatical weight and predicate type were really independent factors predicting the occurrence of extraposition vs. canonical structure. The two independent variables included in the model were weight ratio (i.e., ratio of VP to RC length, a continuous variable) and predicate type (i.e., passive/unaccusative vs. all others, a binary categorical variable). Both factors were signicant in predicting extraposition, with weight being the stronger of the two predictors: weight ratio (X 2 1 38:91, p < 0:01), predicate type (X 2 1 10:47, p < 0:01). In addition, there was no signicant interaction between the two factors (X 2 1 0:28, p 0:59), meaning that the eects of weight were independent of the eects of predicate type. Note that these results are consistent with an account in which discourse status is also independent of weight and perhaps a stronger predictor of extraposition than predicate type. However, since the corpus was not coded for discourse status, additional studies would be needed to conrm this possibility. 5.3. Discussion

Results of the corpus analysis showed a strong eect of RC weight in accordance with the domain minimization preferences of Hawkins (2004) theory. Although RCE was relatively infrequent overall, at only about 15% of the sentences with subject-modifying RCs, extraposition was


E. J. Francis

strongly preferred in cases where the VP length (extraposition distance) was one or two words or where the RC was at least four times longer than the VP. In contrast, extraposition happened in only about 2% of cases in which the RC was the same length or shorter than the VP. Furthermore, in sentences where extraposition occurred, the RC was almost always longer than the VP and was more than three times longer on average. Conversely, in sentences with canonical structure, the VP was about 1.5 times longer than the RC on average. Although the trends are clearly in the expected direction, these results are not exactly as the domain minimization preferences predict. In cases where the RC and VP were exactly the same length, for example, extraposition occurred in only 9% of sentences even though the PCDs for Subject NP and Matrix S domains would have been approximately equal. These results suggest that grammatical weight needs to be strongly skewed in favor of extraposition before speakers will use extraposition consistently. But of course grammatical weight is not the only relevant factor. As we have already seen, semantic and pragmatic constraints, as identied in previous work on RCE, limit the occurrence of extraposition. Since canonical structure tends to occur in a wider range of semantic and pragmatic contexts, we would expect canonical structure to occur more frequently in general. Indeed, we saw that the rate of extraposition went up to 37% with passive and unaccusative verbs, as compared with only 6% with other predicate types (Table 5). However, 37% is still a minority of cases, and most of those cases occurred when the RC was at least twice as long as the VP (Figure 6). Thus, even when considering only semantically felicitous unaccusative and passive predicates, the rate of extraposition was still lower than domain minimization principles would predict. An additional reason for the relative infrequency of RCE may be that RCE is in competition with other constructions besides the canonical structure. For example, RCE sentences with unaccusative main verbs can often be paraphrased using the presentational there construction, as illustrated in (11c). (11) a. b. c. New sets that were able to receive all the TV channels soon appeared. (Canonical) New sets soon appeared that were able to receive all the TV channels. (Extraposition) There soon appeared new sets that were able to receive all the TV channels. (Presentational there)

Presentational there serves a discourse function similar to that of extraposition, and, in addition, allows a heavy subject NP to occur at the end

Grammatical weight


of the sentence together with its modiers.12 Because the presentational there construction does not require extraposition and minimizes the NP and Matrix S domains at the same time, it may be preferable to either canonical structure or RCE in sentences with a heavy subject-modifying RC and an unaccusative main verb.13 Further investigations would be needed, however, to conrm this hypothesis. A nal reason why RCE only occurred consistently when the RC was much heavier than the VP might be related to conventionalized preferences for certain phrase structure congurations. Hawkins (1994: 90) observed a similar result for HNPS (Heavy NP Shift) in a corpus study of English: HNPS only applied in about 30% of cases for which the NP was longer than the PP and only occurred consistently when the NP was at least ve words longer than the PP. Hawkins (1994) proposed that the large weight dierence required to induce a preference for Heavy NP Shift (in contrast to the smaller dierence required to show a preference for placing a heavy PP at the end in a V-PP-PP sequence) may be due to a conventionalized preference for verb-object order in English phrase structure. A similar explanation could apply to the present case as well, since adjacency between a head noun and its modifying RC is also a highly conventional property of English phrase structure, with extraposition requiring special syntactic and interpretive rules. It is not clear, however, where such an adjacency preference would come from. One possibility is that the preference reects speakers sensitivity to stylistic considerations, such as the prescriptive rule banning misplaced modiers. Alternatively, an adjacency preference could be simply the result of biases developed through encountering the canonical structure much more frequently in language usea case of infrequent use of RCE perpetuating infrequent use of RCE. This would then be similar to the idea of frequency bias used to help explain the acceptability results in Section 4. Finally, the adjacency preference might be encoded in the grammar itself. Hawkins (1994: 8990) proposed that canonical verb-object order is licensed by the phrase structure grammar whereas HNPS may be licensed only by weight-based principles. I believe that all of these possibilities may be at least partially correct, and I will suggest in Section 6 that such

12. The underlined portion of (11c) is not the surface grammatical subject, but it is the logical subject in that it represents the single argument of a one-place predicate (with there being an expletive subject) and corresponds to the grammatical subject of the canonical sentence in (11a). 13. Presentational there is restricted to occur with unaccusative predicates, and unlike RCE is usually infelicitous with other types of predicates regardless of pragmatic context (Rochemont and Culicover 1990: 66).


E. J. Francis

an adjacency preference might best be stated in the grammar in terms of a default constraint on the syntax-semantics interface, as proposed by Culicover and Jackendo (2005). 6. General discussion and conclusions

RCE from Subject NP involves a discontinuous dependency which deviates from the normal X-bar structure of phrases (at least on the surface), adding complexity to the grammar with little or no eect on the meaning of the sentence. In addition, RCE is subject to a prescriptive rule banning misplaced modiers and is singled out in some style guides as an example of poor writing (e.g., Trenga 2006: 56). From this perspective, it is perhaps surprising that RCE occurs naturally in both formal and informal language use. While it is known from previous research that discourse information structure plays a role in licensing this construction, the current study shows that grammatical weight is also an important factor in determining how and when speakers use RCE. Furthermore, the current study provides evidence that the increased incidence of RCE with heavy RCs is related to processing eciency. Hawkins (2004) theory of domain minimization predicts that the degree to which RCE is be preferred in both comprehension and production should be correlated with the degree to which RC length exceeds VP length. In support of this hypothesis, corpus analyses reported here showed that the incidence of RCE increased as the proportion of VP to RC length decreased and that RCE was strongly preferred over canonical structure when the RC was four or more times longer than the VP. Incidence of RCE was lower overall than was predicted by domain minimization principles alone, but this can plausibly be explained by semantic and pragmatic constraints on RCE, competition from other non-canonical sentence types, and perhaps a conventionalized preference for adjacency between a head noun and its modifying clause. Importantly, a logistic regression analysis showed that weight was a signicant predictor of extraposition and that the eects of weight were independent of predicate type. Also in support of Hawkins proposals, reading time results showed that when the RC was three times heavier than the VP (the heavy condition), RCE was processed signicantly faster than canonical structure. Importantly, lexical content and information structure were controlled in the reading time experiment and only grammatical weight was manipulated, meaning that there was an eect of weight independent of lexical and informational factors. The results of the acceptability judgment task also supported the idea of domain minimization, though perhaps not as clearly as the results for

Grammatical weight


reading time and corpus analysis. Results showed that canonical structure was rated higher than RCE overall, but that this preference disappeared when the RC was heavy. Although RCE was not the preferred structure in the heavy condition, the eect of grammatical weight was still in the predicted direction and was very similar to the eect shown in previous acceptability judgment studies of German RCE (Konieczny 2000; Uszkoreit et al 1998). It may be that the lack of preference for RCE in the heavy condition is related to the nature of the judgment task, since asking for judgments of isolated sentences may serve to highlight the fact that RCE deviates from the canonical (and much more frequent) phrase structure pattern in which the relative clause it adjacent to its head noun and also violates a prescriptive rule. In contrast, the self-paced reading task (which supported Hawkins predictions more clearly) did not ask participants to make any evaluative judgments and may be viewed as a more direct measure of processing eciency. The overall pattern of these results strongly suggests that grammatical weight is a gradient factor related to processing eciency rather than part of a categorical rule of grammar. Nevertheless, these results also have implications for the syntactic analysis of RCE. For example, one nding of the corpus analysis was that RCE never occurred in cases where the VP was more than 1.3 times longer than the RC. Similarly, in the acceptability judgment experiment, RCE sentences in the light and medium conditions were rated signicantly lower than canonical sentences. If one were to ignore the gradient eects of weight that were also shown in these studies, this might give the appearance of a categorical constraint in which weight is somehow specied as a binary syntactic feature. Although weight (or heaviness) is not a feature of any of the major theories of RCE syntax (unlike for some analyses of Heavy NP Shift, for example), syntactic locality conditions such as Subjacency are. Given the controversial nature of locality conditions in the literature on extraposition (cf. Baltin 2006) and the heavy reliance on constructed examples within this body of research, there is a real danger that gradient eects related to processing might be mistaken for categorical syntactic rules. Strunk and Snider (2008) in fact provide systematic evidence that syntactic locality conditions on RCE from Object NP are gradient rather than categorical. In their acceptability experiments on RCE from Object NP in English and German, they found that many Subjacency-violating sentences were judged no worse than non-violating sentences and that the Subjacency effects that did show up were gradient in nature, depending on the depth of embedding of the extraposed RCs antecedent. In addition, Strunk and Snider (2008) report a corpus study of RCE in German showing that Subjacency-violating sentences occurred naturally in discourse and that


E. J. Francis

the eects of syntactic locality on frequency of RCE were gradient, patterning similarly to the eects of grammatical weight in Uszkoreit et als (1998) corpus study of German and the current study of English. Thus, if structurally-dened locality conditions are gradient and reside outside the syntactic component of grammar, the syntax of RCE may be simpler than has been assumed in most generative approaches to extraposition. The current study does not speak to this issue directly, but provides a pattern of data to look for in determining whether a proposed constraint is gradient and performance-based, or whether it is more accurately stated as a categorical syntactic rule.14 An additional implication for the grammatical analysis of RCE comes from the reading time results. These results suggest that the dependency between the RC and its head noun is processed similarly to the dependency between the verb and its subjectboth are subject to weight-based locality eects that can be measured in terms of IC-to-word ratios. Following Hawkins (1994, 2004), the ratios are calculated as though the head noun and its extraposed RC were part of the same NP just as the subject NP and its predicate are part of the same clause. This is reminiscent of McCawleys (1987, 1998) analysis of RCE as involving a single, discontinuous NP constituent. While McCawleys proposal is probably too strong a conclusion on the basis of these results, the similarity of the two dependencies suggests that extraposing the RC is not, in itself, a signicant source of additional complexity. Rather, what contributes to complexity is the cost of integrating information across a distance. Culicover and Jackendo s (2005: 166167) approach is appealing from this perspective because it states the usual correspondence between syntactic constituency and semantic modication as a default pattern (soft constraint) that is violated in RCE rather than trying to reconcile RCE with X-bar structure through some kind of rightward or leftward syntactic movement. Hawkins (1994) idea of a conventionalized grammatical preference for adjacent word order can perhaps best be thought of in terms of such a default constraint. Rather than claiming, as Hawkins (1994: 89) does for HNPS, that the non-canonical word order pattern is not licensed by the grammar at all, I would propose instead that RCE is licensed by

14. Following the general approach of Culicover and Jackendo (2005), my assumption is that a categorical syntax exists, but need not include rules for patterns that are predictable from semantics, pragmatics, processing, or other non-syntactic factors. Another possibility I have not considered here is that of gradient rules within the syntax (cf. Featherston 2005). Since grammatical weight is evidently a performance-based phenomenon in the current study, I will not address this possibility any further here. See Wasow (2002: 115158) for relevant discussion.

Grammatical weight


the grammar, but that it violates a default constraint which species the preferred syntax-semantics correspondence. Such a constraint violation captures the intuition that canonical ordering is the more basic option, as reected in corpus frequency and acceptability judgments, but does so without complicating the syntactic structure (extraposed RCs are simply clausal adjuncts in syntax). Although the current study does not directly argue in favor of a particular syntactic analysis, the reading time results, which showed no preference for canonical structure, lend themselves nicely to such an approach. In sum, the results of all three experiments (self-paced reading, acceptability judgment, and corpus analysis) suggest that grammatical weight plays a signicant role in licensing RCE from Subject NP in English. Grammatical weight is therefore helpful for explaining why RCE is permitted by the grammar and preferred in certain contexts of language use, despite the discontinuous dependency that is incurred. Appendix A: Instructions for Experiment 1 You will be presented with a series of sentences, each followed by a statement. Your rst task is to read the sentence and then press the left button as soon as you have understood the sentence. Following each sentence, you will be presented with a simple statement. Your second task is to decide whether the statement is true or false based on the information in the sentence you just read. To make your response, press the left button for true and the right button for false. In making your decision, use only the information contained in the sentence itself. Avoid making any inferences beyond the actual content of the sentence. Please make your responses as quickly and accurately as possible. After you have made your selection, there will be a brief pause and the next sentence will appear. There will be four sets of sentences. Following each set, the computer will prompt you to take a short break. After you have rested, you may press the space bar to continue with the next set. When the last set is nished, you will be prompted to inform the experimenter. Any questions? Please place your index and middle nger on the response pad to select the left button for true, right button for false. You may press the spacebar with your other hand when you are ready to begin. Appendix B: Instructions for Experiment 2 Please read each of the sentences listed below. For each sentence, we would like you to indicate your reaction to the sentence. Mark your


E. J. Francis

response sheet by circling a number from (1) to (9). Use (9) for sentences that seem fully normal, and understandable to you. Use (1) for sentences that seem very odd, awkward, or dicult for you to understand. If your feelings about the sentence are somewhere between these extremes, use one of the middle responses from (2) to (8). Please try to use the entire scale, not just the end points of the scale. THERE ARE NO RIGHT OR WRONG ANSWERS. Please base your responses solely on your gut reaction, not on rules you may have learned about what is proper or correct English. Please work straight through the survey and DO NOT turn back to a page after you have completed it. You will have the opportunity to take three short rest breaks during the survey. Rest breaks are indicated at the bottom of certain pages. For example, you may encounter sentences like the following in the survey: Worst
We persuaded there to be strike. Phil wrote a poem for his mother in honor of her fortieth birthday. She asked that whether she should come to the party.


1 2 1 2 1 2

3 3 3

4 4 4

5 6 5 6 5 6

7 7 7

8 8 8

9 9 9

Please turn to the following page when you are ready to begin.

Arnold, Jennifer E., Thomas Wasow, Ash Asudeh & Peter Alrenga. 2004. Avoiding attachment ambiguities: The role of constituent ordering. Journal of Memory and Language 51. 5570. Arnold, Jennifer E., Thomas Wasow, Anthony Losongco & Ryan Ginstrom. 2000. Heaviness vs. newness: The eects of structural complexity and discourse status on constituent ordering. Language 76(1). 2855. Baltin, Mark. 1981. Strict bounding. In Carl L. Baker & John McCarthy (eds.), The logical problem of language acquisition, 257295. Cambridge, MA: The MIT Press. Baltin, Mark. 2006. Extraposition. In Martin Everaert & Henk van Riemsdijk (eds.), The Blackwell companion to syntax, vol. 2, 237271. Malden, MA: Blackwell. Behaghel, Otto. 1909. Beziehungen zwischen Umfang und Reihenfolge von Satzgliedern. Insogermanische Forschungen 25. 110142. Bianchi, Valentina. 2002. Headed relative clauses in generative syntax-part I. Glot International 6(7). 197204.

Grammatical weight


Bresnan, Joan W. 2006. Is syntactic knowledge probabilistic? Experiments with the English dative alternation. In Sam Featherston & Wolfgang Sternefeld (eds.), Roots: Linguistics in search of its evidential base, 7596. Berlin: Mouton de Gruyter. Burzio, Luigi. 1986. Italian syntax. Dordrecht: Reidel. Cheung, Ki Shun Antonio. 2006. Processing factors in language comprehension and production: The case of Cantonese dative constructions. Hong Kong: University of Hong Kong MPhil thesis. Culicover Peter. W. & Ray S. Jackendo. 2005. Simpler syntax. Oxford: Oxford University Press. Diessel, Holger. 2008. Iconicity of sequence: A corpus-based analysis of the positioning of temporal adverbial clauses in English. Cognitive Linguistics 19(3). 457482. Featherston, Sam. 2005. That-trace in German. Lingua 115(9). 12771302. Gibson, Edward. 1998. Linguistic complexity: Locality of syntactic dependencies. Cognition 68(1). 176. n, Givo Talmy. 1983. Topic continuity in discourse. Quantitative cross-language studies (Typological Studies in Language 3). Amsterdam: John Benjamins. Gueron, Jacqueline. 1980. On the syntax and semantics of PP-Extraposition. Linguistic Inquiry 11. 637678. Hawkins, John A. 1994. A performance theory of order and constituency. Cambridge, UK: Cambridge University Press. Hawkins, John A. 2001. Why are categories adjacent? Journal of Linguistics 37(1). 134. Hawkins, John A. 2004. Eciency and complexity in grammars. Oxford: Oxford University Press. Huck, Georey J. & Youngnee Na. 1990. Extraposition and focus. Language 66(1). 51 77. Huck, Georey J. & Youngnee Na. 1992. Information and contrast. Studies in Language 16(2). 325334. Kayne, Richard S. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press. Konieczny, Lars. 2000. Locality and parsing complexity. Journal of Psycholinguistic Research 29(6). 627645. Kuno, Susumu & Ken-ichi Takami 2004. Functional constraints in grammar: On the unergative-unaccusative distinction. Amsterdam: John Benjamins. Lohse, Barbara L., John A. Hawkins & Thomas Wasow. 2004. Domain minimization in English verb-particle constructions. Language 80(2). 238261. Matthews, Stephen & Louisa Y. Y. Yeung. 2001. Processing motivations for topicalization in Cantonese. In Kaoru Horie & Shigeru Sato (eds.), Cognitive-functional linguistics in an East Asian context, 81102. Toyko: Kurosi. McCawley, James D. 1987. Some further evidence for discontinuity. In Georey J. Huck & Almerindo E. Ojeda (eds.), Discontinuous constituency (Syntax and Semantics 20), 185 200. New York: Academic Press. McCawley, James D. 1998. The syntactic phenomena of English. Chicago: University of Chicago Press. Nelson, Gerald, Sean Wallis & Bas Aarts. 2002. Exploring natural language: The British component of the International Corpus of English. Amsterdam: John Benjamins. Quirk, Randolph, Sidney Greenbaum, Georey Leech & Jan Svartvik. 1972. A grammar of contemporary English. London: Longman. Rochemont, Micahel S. & Peter W. Culicover. 1990. English focus constructions and the theory of grammar. Cambridge: Cambridge University Press. Ross, John R. 1967. Constraints on variables in syntax. Cambridge, MA: Massachusetts Institute of Technology dissertation.


E. J. Francis

Siewierska, Anna. 1993. Syntactic weight vs. pragmatic factors and word order variation in Polish. Journal of Linguistics 29(2). 233265. Stallings, Lynne M., Maryellen C. MacDonald & Padraig G. OSeaghdha. 1998. Phrasal ordering constraints in sentence production: phrase length and verb disposition in HeavyNP Shift. Journal of Memory and Language 39. 392417. Strunk, Jan & Neal Snider. 2008. Extraposition without subjacency. Paper presented at the 30th Annual Convention of the German Society of Linguistics (DGfS 2008), Bamberg, Germany, 2729 February. Takami, Ken-ichi. 1999. A functional constraint on Extraposition from NP. In Akio Kamio & Ken-ichi Takami (eds.), Function and Structure, 2356. Amsterdam: John Benjamins. Trenga, Bonnie. 2006. The curious case of the misplaced modier: How to solve the mysteries of weak writing. Cincinnati, OH: Writers Digest Books. Uszkoreit, Hans, Thorsten Brants, Denys Duchier, Brigitte Krenn, Lars Konieczny, Stephan Oepen & Wojciech Skut. 1998. Studien zur performanzorientierten Linguistik: Aspekte der Relativsatzextraposition im Deutschen. In CLAUS Report No. 99, 114. Saarbrucken: Universitat des Saarlandes. Vasishth, Shravan & Richard L. Lewis. 2006. Argument-head distance and processing complexity: explaining both locality and antilocality eects. Language 82(4). 767794. Wasow, Thomas. 2002. Postverbal behavior. Stanford: CSLI Publications Wasow, Thomas. 1997. End-weight from the speakers perspective. Journal of Psycholinguistic Research 26(3). 347361. Yamashita, Hiroko. 2002. Scrambled sentences in Japanese: Linguistic properties and motivations for production. Text 22(4). 597633. Yamashita, Hiroko & Franklin Chang. 2001. Long before short preference in the production of a head-nal language. Cognition 81. B45B55.


Abstract We propose a constructionist approach to the polyfunctionality of the Italian focus particle magari (roughly corresponding to maybe, but also I wish). The sheer syntactic versatility of this word leads us to detect its formal regularities at the level of discourse congurations. This level of analysis, identied within the French linguistic tradition, is dened by the maintenance of a predicate-argument-adjunct structure in discourse. The salient feature of discourse congurations is their shape, which can be described by referring to a number of topological patterns: lists of elements in the same syntactic position, repetition of syntactic structures, shifting of elements from a post-verbal to a pre-verbal position and so on. These topological patterns are meaningful to an extent and they are eligible to be regarded as constructions. Magari is shown to be regularly associated with a general topological pattern, namely a list of items that occupy the same syntactic position as the item focused by magari. Each semantic function of magari correlates with one particular kind of list. These associations of a form (the dierent types of lists) and a meaning (the functions of magari) are shown to be related to one another by means of inheritance links.

` * Correspondence address: Dipartimento di Linguistica, Universita Roma Tre, Via Ostiense 236, 00146 Rome, Italy. E-mails: 3fmasini@uniroma3.it4; 3pietrand@uniroma3.it4. Acknowledgements: The research was carried out within a project on the topology of ` grammatical meaning in discourse constructions (Topogram, Universita Roma Tre). We would like to express our graditude to Claire Blanche-Benveniste, Elisabetta Bonvino, Claudio Iacobini, Sylvain Kahane, Alessandro Lenci, Henning Nlke and Raaele Simone, who read previous drafts of this paper and provided invaluably helpful comments and suggestions. A special thank goes to Ewa Dabrowska, Editor-in-Chief at Cognitive Linguistics, who guided us through a very fair and stimulating revision process, and to an anonymous Associate Editor and two anonymous referees, who sensibly improved our paper with their comments. The usual disclaimers apply. Cognitive Linguistics 211 (2010), 75121 DOI 10.1515/COGL.2010.003 09365907/10/00210075 6 Walter de Gruyter


F. Masini and P. Pietrandrea non factuality; focus particle; construction grammar; discourse conguration; topology; lists.



The polyfunctionality of magari

The word magari has attracted considerable attention among Italian linguists because of its especially intriguing polyfunctionality that knows no parallel in its counterparts in other European languages (cf. Arcaini 1997, 2000; Licari and Stame 1989; Schiemann 2008). Firstly, magari can have the function of a general marker of non factuality. In this case, it roughly corresponds to the English adverb maybe. See example (1): (1) ` Magari e a casa Maybe (s)he is at home

Magari can also function as a scalar operator (in the sense of Fillmore et al. 1988 and Kay 1990), triggering a scale of non factuality whose extreme position is occupied by the constituent in the focus of magari. See (2): (2) Bisognerebbe negoziare una tregua, un armistizio, magari la pace It would be necessary to negotiate a ceasere, an armistice and maybe peace

Besides, magari may act as a non factual concessive marker, as in (3), where the speaker concedes that the subject is clever despite thinking that he has not studied enough: (3) ` ` Magari e intelligente, ma non e abbastanza preparato He might be clever, but he has not studied enough

In imperatival contexts, magari weakens the illocutionary force of the order, as in the following example: (4) Magari parlagliene tu! Perhaps you yourself could talk to him about it!

Finally, magari functions as an optative marker. This happens when it occurs in exclamative contexts: (5) Vorrei tanto vedere un lm come quello. Magari ne facessero ancora! I really would like to watch a movie like that. I wish they still made some! A: Vuoi un po di riposo? Would you like to rest a bit?


Magari B: Magari! Id love to! / I wish I could!



The problem

The problem that arises is: how can we account for the polyfunctionality of magari? First of all, one should decide whether the functions of magari are somewhat related to each other or are completely independent, i.e., homonymous. There is a good reason for rejecting the latter hypothesis, namely: the set of functions held by magarinon factual, scalar non factual, non factual concessive, imperative, optativerecalls in most respects the semantic network developed by several irrealis markers of non factuality in various non-European languages (cf., e.g., Elliott 2000; Lazard 1998). The crosslinguistic presence of similar semantic networks makes it fairly unlikely that we are dealing with pure homonymy. We therefore consider the various functions of magari as microsenses of this word, that is distinct sense units [ . . . ] that occur in dierent contexts and whose default construals stand in a relation of mutual incompatibility at the same hierarchical level (Croft and Cruse 2004: 126127). Under this perspective, the word magari has a hyperonymic reading and a cluster of hyponymous readings, whose default construals are sister incompatibles (Croft and Cruse 2004: 127). The question now arises of identifying the contexts that licence the various functions of magari and the nature of the relations holding between these functions. Such a task is made more complicated by the sheer syntactic versatility of magari. Indeed the contexts in which magari occurs can be properly detected only by adopting a wide-ranging notion of context. In Section 3 we describe the practical diculties encountered in the analysis of magari and the theoretical approach and tools adopted for solving them, whereas in Section 4 we provide a qualitative and quantitative description of magari and its various functions. In Section 5, we give a construction grammar account of our ndings.

3. 3.1.

The theoretical approach Construction grammar

A fruitful theoretical approach to the kind of problem outlined in the previous section is to place the analysis of magari in the wide framework of construction grammar. As is well known, construction grammar comprises a number of dierent models (cf., among others, Croft 2001; Fillmore et al. 1988; Goldberg 1995, 2006; Kay and Fillmore 1999), which


F. Masini and P. Pietrandrea

nevertheless share a set of basic tenets. The main tenet regards the very notion of construction, which is dened as a conventionalized association of a form and a meaning and is considered the basic unit of linguistic analysis. This denition virtually captures every meaningful unit of language, ranging from simple words to more complex and abstract sentence-level structures, such as argument structure constructions (Goldberg 1995) or sentence types (Michaelis and Lambrecht 1996). It is thus clear that certain higher-level abstract patterns, those that are commonly considered as context for lower-level lexically specied units, may be treated as full linguistic objects in this framework, provided that they prove to be meaningful to some extent. Another assumption of constructionist approaches that is crucial for our purposes is that they take into account not only syntactic and semantic information, but also lexical and/or pragmatic information (Kay 1990: 61) and that all this information is coded simultaneously in the construction and contributes to characterize the constructions itself. This provides the tools for the detection of the correlations between certain contexts, or rather constructions, and the various functions of magari. This latter theoretical issue has been recently addressed by Fried (2007), who has convincingly argued that the relations between the dierent functions of the same polyfunctional lexical unit are better understood if one takes into account the entire construction in which they occur, rather than the single item under examination. Therefore, the entire construction becomes the true linguistic form to be investigated. Still another aspect of constructionist approachesand in particular of Goldbergs cognitive construction grammar (Goldberg 2006)that will prove useful in our analysis is the use of inheritance links, which account for the relations holding among constructions. The inheritance system works this way: if construction A shares some formal properties with construction B, then construction A also shares some semantic properties with B, and the two constructions are related by an inheritance link. As is well known, according to Goldberg (1995: 75 ), there are four major types of inheritance links: polysemy links (IP ), subpart links (IS ), instance links (II ) and nally metaphorical extension links (IM ).1 Given this


Polysemy links (IP ) capture the nature of the semantic relations between a particular sense of a construction and any extensions from this sense (Goldberg 1995: 75); subpart links (IS ) are posited when one construction is a proper subpart of another construction and exists independently (1995: 78); instance links (II ) are posited when a construction is a more fully specied version of the construction it is linked to (1995: 79); nally, metaphorical extension links (IM ) are posited when two constructions are found to be related by a metaphorical mapping (1995: 81).



framework, all the possible abstract constructions that host the adverb magari can be regarded as constructions that share at least one formal propertynamely the presence of magariand that are eligible to be linked to one another at the representational level. Therefore, once we have identied the form of the constructions in which magari occurs, we will have to determine which kind of inheritance links connect these various constructions to one another. First of all, though, we have to address the formal analysis of the contexts of magari, which is not unproblematic. 3.2. A practical diculty

A practical diculty in the formal analysis of magari concerns the abovementioned syntactic versatility of this word, which makes it particularly hard to dene its contexts of occurrence. Indeed, magari occurs in every illocutionary act: assertions (cf. examples (1) to (3)), orders (4), exclamations (5). It can also occur in questions, as exemplied in (7):2 (7) ` Non potrebbe essere uscito con un amico? Non sara magari con suo fratello? Dont you think he might have gone out with a friend? Couldnt he be with his brother?

What is more, the categorial status of magari is not easy to dene: it can be used either as an interjection (6) or as an adverb (cf. examples (1) through (5) and (7)). Occasionally, and retaining some of the semantic properties of its adverbial function, it can also be understood as a clause connective, as exemplied in (8): (8) ` Magari un po debolina, magari me la sono immaginata, magari e solo un eetto ottico . . . Ma vi giuro che lho vista [Web] Maybe a bit feeble, maybe I dreamt it, maybe its just an optical eect, but I swear I saw it

2. The analysis presented in this article was carried out with the aid of real examples. In particular, we made use of corpora of contemporary written and spoken Italian, respectively la Repubblica corpus [laR] and Lessico di frequenza dellitaliano parlato [LIP] (see Section 4 for further information about these corpora); in addition, we took examples from the Web [Web] and contemporary novels. Examples taken from the mentioned corpora or the Web are marked with the corresponding abbreviation in squared brackets at the end of the example. Texts taken from novels include the full reference of the novel. Intuition-based examples, on the contrary, have no indication.


F. Masini and P. Pietrandrea

As an adverb, magari can have scope on units of dierent size and category. First, it can have both a clausal scope, as in (1), (3), (4) and (5), and a phrasal scope, as in examples (2), (7) and (9) to (11): (9) ` E un piacere venir qui e vedere tutta questa gente che si commuove per me e che magari ha pianto per la mia vittoria [Web] Its nice to come here and see all these people who are moved because of me and who, maybe, cried for my victory Gli aerei piccoli e molto utilizzati, con personale poco pagato e maga` ri stanco, non possono dare la massima tranquillita ne agli utenti ne ai sindacati [Web] Small and thoroughly exploited aircrafts, with a badly paid and possibly tired sta, cannot fully reassure either users or trade unions Si discute magari male, ma sempre molto a lungo We discuss perhaps badly, but always at great length



When magari has scope on a phrase, the latter can be a verb phrase (9), an adjectival phrase (10), an adverbial phrase (11), a prepositional phrase (7) and even a noun phrase as in (2). Besides, magari is endowed with an almost unrestrained syntactic mobility: it can occur in fact at every major phrasal boundary. For example, if we consider the proposition in (12), we may have the patterns in (13): (12) (13) ` Luigi e venuto Luigi has come ` a. Magari Luigi e venuto ` b. Luigi magari e venuto ` c. Luigi e venuto, magari

Some regularities can be easily detected even in such a complex picture. For example, magaris use as an optative is preferably expressed with an interjection in exclamative contexts. The imperative use is associated with orders. Nevertheless, some diculties remain in detecting the relevant context associated with non factual, scalar and concessive uses of magari. The assertive context, for example, appears to be associated with all these functions. The size and category of the unit within the scope of magari is not a relevant factor in determining the function of this word. As a matter of example, we have shown above that magari retains the same scalar function whether it has scope on a verb phrase (9), an adjectival phrase (10) or a nominal phrase (2). Such a function is also compatible with a clausal scope of magari, as shown in (14):

Magari (14)


` ` Forse e venuto ieri, ha passato qui tutto il pomeriggio e magari si e fermato a dormire Perhaps he came yesterday, he spent the whole afternoon here and possibly he stopped for the night

Not even the distribution of magari within the clause seems to be a relevant parameter to detect its correct function. For instance, when it has a sentential scope, magari may have the same concessive function whether it fronts the clause (15), is pre-verbal (16) or, nally, post-verbal (17): (15) (16) (17) Magari Luigi ha sbagliato, ma io non me ne sono accorta Luigi might have made a mistake, but I couldnt spot it Luigi magari ha sbagliato, ma io non me ne sono accorta Luigi ha sbagliato, magari, ma io non me ne sono accorta

This syntactic versatility of magari makes it not trivial to identify the structural constraints that characterize the maximally abstract magari construction (the hyperonymic magari) and all other sub-constructions. As things stand, we could simply propose that, in all its functions, magari can be described as the only lexically specied part of a maximally abstract construction in which the only relevant information regards the internal make-up, i.e., the presence of the unit magari with its phonetic properties (even the information about its categorial status is underspecied, since it may behave both as an adverb and as an interjection). Such a characterization is obviously largely unsatisfactory for our purposes. 3.3. Theoretical tools: Discourse congurations, topological structures, topological patterns

The diculty described in the previous section has led us to look for tools that could help in better dening the trans-categorial and trans-level nature of magari. Within the French linguistic tradition, a level of analysis has been identied that is called conguration de discours (discourse conguration) (cf., among others, Blanche-Benveniste 1993, 1997; BlancheBenveniste et al. 1979, 1990; Gerdes and Kahane in print.). In order to dene discourse congurations, we assume as a primitive what BlancheBenveniste et al. (1979) called construction maximale (maximal construction), i.e., the predicate-argument-adjunct structure. The predicateargument-adjunct structure is hardly ever realized together in a sequence in discourse. More frequently, it is gradually built by means of repetitions, rewordings, additions, and other kinds of insistences on one or


F. Masini and P. Pietrandrea

more of its positions. So, for example, the predicate-argument-adjunct structure in (18) may be realized as in (19) as well as in (20):3 (18) (19) ADJ1 -ARG1 -PRE-ARG2 ` Forse chissa io ho scelto il momento sbagliato Maybe, who knows, I have chosen the wrong moment 1 2 forse maybe ` chissa who knows ADJ1 (20) io I ARG1 ho scelto have chosen PRE il momento sbagliato the wrong moment ARG2

Magari lui rincorre un sogno, unutopia, un ideale qualunque Maybe he pursues a dream, a utopia, an ideal whatsoever 1 2 3 ADJ1 ARG1 PRE magari maybe lui he rincorre pursues un sogno a dream unutopia a utopia un ideale qualunque an ideal whatsoever ARG2

A given predicate-argument-adjunct structure can also be instantiated more than once in discourse. For example, the spoken sequence in (21) features two repetitions of the ADJ1 -ARG1 -PRE-ARG2 structure, beside the multiple instantiations of the ARG1 and ARG2 positions: (21) praticamente per ogni tipo di gioco cera un edicio specico. Per esempio il circo serviva alle corse dei carri, lanteatro alle lotte dei gladiatori, lo stadio per i giochi atletici In practice, for every kind of game there was a specic building. For example the circus was for the chariot races, the amphitheatre for the gladiator ghts, the stadium for athletic games [from Bonvino 2005: 61]


Apart from the ARG, ADJ and PRE abbreviations for argument, adjunct and predicate, respectively, we sometimes use other labels, namely: ASP (aspectual element), CAUSE (causative element) and MOD (modal element). Besides, note that the translations of the examples throughout the paper are deliberately as literal as possible in order to facilitate the grid representation (see below).

Magari 1 praticamente in practice per ogni tipo di gioco for every kind of game il circo the circus cera there was serviva was un edicio specico a specic building alle corse dei carri for the chariot races


per esempio for example

lanteatro the amphitheatre lo stadio the stadium

alle lotte dei gladiatori for the gladiator ghts per i giochi atletici for athletic games PRE ARG2



The chunk made up of the sequence of units that instantiate or repeat a given predicate-argument-adjunct structure is called discourse conguration (Pietrandrea 2008a). As this denition makes clear, discourse congurations are objects dened in purely syntactic terms. Interestingly, though, they may have a semantic investment. Let us consider the following example: (22) Io mangio. Il mondo crolla I eat. The world collapses

In this case there is a discourse conguration dened by one repetition of the syntactic structure ARG-PRE. The two sentences making up the discourse conguration depict totally unrelated situations. Still, the syntactic parallelism between the two sentences forces the addressee to nd a semantic relation between them: in this case a relation of contrast. Interestingly, when the same situations are depicted by two contiguous but distinct structures, which hence do not form a proper discourse conguration, the addressee is not invited to postulate a relation between the two situations: (23)
?? ??

Sono io che mangio. Il mondo crolla It is me who eat. The world collapses


F. Masini and P. Pietrandrea

Discourse congurations can be more or less extended: while the discourse congurations in (19) and (20) are limited to a clause, the discourse conguration in (21) spans an entire text. Therefore, crucially, discourse congurations are dened regardless of the boundary between the clausal and the supra-clausal level. The discourse congurations from (19) through (21) are represented in grids, i.e., through a rewriting procedure elaborated mainly by Blanche-Benveniste and colleagues (1979), Bilger (1982), BlancheBenveniste and colleagues (1990), Bilger and colleagues (1997), and Bonvino (2005) and Pietrandrea (2008a) for Italian. This rewriting procedure consists in a representation of the speech ow on a bi-dimensional plane and is constrained by three rules: (i) the horizontal axis of the plane should feature the sequence of the positions that dene the predicateargument-adjunct structure; (ii) the vertical axis should list all the actual realizations within each position; (iii) a left-to-right and top-down reading of the string contained in the grid should render the linear order of the represented chunk. This representation highlights an important fact, namely that the salient feature of discourse congurations is their shape and not the categories they are made up of. We refer to the shape of a discourse conguration with the term topological structure. Such a topological structure can be described by referring to a number of bi-dimensional topological patterns: lists of elements in the same position, repetitions of syntactic structures, chiasms of elements shifting from a pre-verbal to a post-verbal position (or viceversa) and so on.4 It should be noted that units belonging to dierent levels and categories may enter the same topological structure. For example, the two discourse congurations used as answers in (24) and (25) are made up of clauses and nominal constituents respectively. Nevertheless, they present the very same topological structure, characterized by a list of instantiations of the rightmost position, whose last element is preceded by magari (cf. Section 4.4 for more details on this structure).


For example, the discourse conguration in (21) is characterized by two lists of arguments in the ARG1 and ARG2 positions; one repetition of syntactic structure (line 1 and line 2) and a chiasm between the rst two realizations of the ARG1 and ARG2 positions. The pre-verbal hyperonym in ARG1 -line1 position per ogni tipo di gioco for every kind of game is exemplied by a post-verbal hyponym in ARG2 -line2 alle corse dei carri for the chariot races, whereas the post-verbal hyperonym in ARG2 -line1 position un edicio specico a specic building is exemplied by a pre-verbal hyponym in ARG1 -line2 il circo the circus (Bonvino 2005: 61).

Magari (24) A: B: 1 ` ` Come mai e cos tranquillo? Why is he so calm? ` ` ` Sara rientrato presto, si sara riposato, magari avra dormito He probably came back home early, rested, maybe slept ` sara rientrato presto he probably came back home early ` si sara riposato he probably rested magari maybe ADJ1 (25) A: B: 1 2 3 magari maybe ADJ1 ` avra dormito he probably slept PRE


2 3

` Chi puo essere stato? Who could have done this? Un gatto, un cane, magari una scimmietta A cat, a dog, possibly a small monkey un gatto a cat un cane a dog una scimmietta a small monkey ARG1

The body of research carried out on discourse congurations has shown, albeit incidentally, that certain topological patterns, as well as the various topological structures within which those patterns are unied, may have very abstract meanings. For example, we have seen above in (22) that the repetition of the same syntactic structure may carry a meaning of contrast (see also Blanche Benveniste 1997: 113). Another example of meaningful topological pattern comes from the listing of elements in only one position of the grid. The abstract topological pattern list has the very general meaning of relation among the conjuncts and may assume more specic meanings according to the exact way in which it is instantiated. For instance, it is acknowledged that a list instantiated by conjuncts preceded by one or more additive conjunctions, is interpreted as an additive list (26).

86 (26)

F. Masini and P. Pietrandrea Ha comprato il pane e il latte (S)he bought bread and milk 1 2 PRE ha comprato (s)he bought e and ADJ1 il pane bread il latte milk ARG1

Accordingly, a list instantiated by conjuncts preceded by one or more disjunctive conjunctions is interpreted as an alternative list (27). (27) Torna domani o dopodomani (S)he will come back tomorrow or the day after 1 2 PRE torna (s)he comes back o or ADJ1 domani tomorrow dopodomani the day after ADJ2

Yet other types of lists are possible (cf. Bonvino et al. 2009 for a preliminary study, but also Gerdes and Kahane in print). A list which features the repetition (for two, three or more times) of the same lexical material in the same position conveys a general meaning of intensication that, we may suppose, specialises according to the categorial nature of the repeated constituent. For example, the repetition of the adjective piccola small in (28) acts as a superlative, while the repetition of the verb in (29), as noted by Bertinetto (1991: 50), is a special way to express continuous aspect in Italian. (28) Ho visto una casa piccola piccola I saw a little little house 1 2 PRE ARG1 ho visto I saw una casa a house piccola little piccola little ADJ1

Magari (29)


Leroe cerca cerca cerca ma non trova nulla The hero searches, searches, searches but does not nd anything [from Bertinetto 1991: 50] 1 2 3 4 ma but non not ADJ1 ADJ1 ARG1 PRE leroe the hero cerca searches cerca searches cerca searches trova nds PRE niente anything ARG1 ARG2

A list which features the repetition of semantically related elements, especially co-hyponyms may convey a meaning of lexical approximation (30). (30) Cera un elenco, un sommario, un indice insomma There was a list, a summary, an index lets say 1 2 3 PRE cera there was un elenco a list un sommario a summary un indice an index ARG1 insomma lets say ADJ1

We propose in Figure 1 a tentative representation of the relations observed between the listing phenomena mentioned above. Albeit preliminary, this representation shows that the use of a topological methodology allows to provide a unied account for a number of constructional phenomena usually treated under dierent domains. To sum up, topological patterns can be viewed as indenitely extended, bi-dimensional, syntactic patterns, dened regardless of the boundary


F. Masini and P. Pietrandrea

Figure 1. The constructional network for lists

between the clausal and the supra-clausal level and (at least at the most abstract level) regardless of the categories they are made up of. These formal patterns are meaningful to a certain extent. 3.4. Topological patterns as constructions

The existence of meaningful abstract patterns naturally recalls the notion of construction in construction grammar. We would propose therefore to consider topological patterns as a type of constructions that operate at the level of discourse congurations. Including topological patterns among the array of constructions is in line with some important recent attempts to break the boundary of the clause/sentence (Mithun 2005, 2008) and to extend the notion of con struction to upper-level entities (Fried and Ostman 2004; Fried and Ost man 2005; Ostman 2005). In fact, as made clear by Fried and Ostman



(2005) and Ostman (2005), construction grammar, as a theory, has no built-in limitation with respect to the extension of the notion of construction to larger stretches of discourse. Yet, only a very limited number of works on constructions deal with upper-level entities. Among them, Ostman (2005) should be mentioned, who suggests that constructions can be detected at the textual level, claiming that there exist discourse patterns with a form (i.e., text type) and a meaning or function (i.e., genre). Our paper follows this line of research. However, whereas Ostman (2005) extends the notion of construction to entire texts and claims that this textual setting is essential for interpreting certain sentences correctly, in this paper we hypothesize that there exist constructions insensitive to the boundary of the clause/sentence and dened by their topological structure. Like all constructions, they can be more or less specied and enter inheritance systems. The analysis of magari in what follows will be driven by these theoretical hypotheses. 4. The analysis of magari

Pietrandrea (2007) hasalbeit cursorilyobserved that, although distributional regularities cannot be found at the clause level, magari has a regular topological distribution in discourse congurations. In particular, she noted that 42 out of the 75 tokens of magari occurring in a small corpus of the Roman variety of spoken Italian (about 56 percent), are associated with a specic kind of topological pattern: the focus of magari belongs to a list of items that realize the same syntactic position. In utterance (31), for example, the ARG2 position is realized by four dierent arguments (in a scene, in a forest, in a jungle, in the desert), the rst of which is in the focus of magari. (31) Che ne so poteva comparire una scenograa che che magari li riportava ne in un ambiente, in una foresta piuttosto che in una giungla nel deserto [LIP] I dont know a set could appear that that maybe reconveyed them in in a scene, in a forest rather than in a jungle, in the desert


F. Masini and P. Pietrandrea

1 che ne so poteva una che I dont comparire scenograa that know could a set appear che that magari maybe li riportava ne reconveyed in them in un ambiente in a scene in una foresta in a forest piuttosto che rather than in una giunga in a jungle nel deserto in the desert ARG1 ADJ1 ADJ1 PRE ARG1 ADJ2 PRE ARG2

This strong tendency to occur in lists induced Pietrandrea (2007) to semantically characterize magari, commonly understood as an epistemic adverb, as a generic marker of non factuality, more precisely as a marker of non exclusion of factuality (NEF). In other words, putting forward the constituent in the focus of magari as but one of a set of possible options, the speaker does not fully subscribe to the factuality of the proposition realized through that constituent (that reconveyed them in a scene): (s)he simply does not exclude that that proposition could be factual. We have further extended Pietrandreas analysis with the aid of two large, diatopically balanced corpora of both spoken and written contemporary Italian, namely: the la Repubblica corpus (written, approx. 380 million tokens, cf. Baroni et al. 2004) and the Lessico di Frequenza dellItaliano Parlato (LIP) corpus (spoken, approx. 500.000 tokens, cf. De Mauro et al. 1993).



We randomly selected 600 occurrences of magari (300 in the written corpus and 300 in the spoken corpus). For the sake of consistency, we subtracted from this rst corpus 35 occurrences (32 spoken and 3 written) that could not be easily interpreted, such as for example (32), where the speaker interrupts himself right after uttering magari, thus making a proper classication impossible: (32) ` ` Cioe non vorrei scartare questa possibilita a priori insomma magari I mean I wouldnt want to rule this out from the outset, I mean, magari

We analyzed the remaining 565 occurrences of magari within the context of their discourse congurations; i.e., we took as a relevant unit of analysis the whole chunk made up of the sequence of units that instantiate or repeat the predicate-argument-adjunct structure that each occurrence of magari contributes to dene. A rst thorough analysis of this corpus allowed us to identify the ve main functions of magari mentioned in Section 1. Afterwards, both authors coded separately the entire data set. The more problematic cases were discussed together. This led to exclude, for the sake of simplicity, the 20 occurrences of magari (9 spoken and 11 written) that fullled more than one function. Thus, for example, we excluded occurrences such as (33), where magari functions at the same time as a concessive and as a scalar marker.5 The nal corpus therefore amounts to a total of 545 (286 written and 259 spoken) occurrences of magari. (33) Ce la mettono tutta, magari scrivono anche bei pezzi, ma sono troppo limitati, possono esprimersi solo parzialmente [laR] They try hard, they might even write nice pieces, but they are too limited, they can only partially express themselves

The discussion of the more problematic cases also led to a more precise semantic characterization of the ve functions associated to magari, which can be dened as follows: equipotential non exclusion of factuality (ENEF); scalar non exclusion of factuality; scalar concessive conditional; weakened imperative; optative.

5. These are cases of unications, which are extremely interesting from a theoretical point of view. However, for our current purposes, they would have biased the overall picture.


F. Masini and P. Pietrandrea

Table 1. Functions of magari in the corpus ENEF Written With list Without list Spoken With list Without list Total With list Without list % 30 23 7 87 47 40 117 70 47 21% Scalar 223 145 78 91 63 28 314 208 106 58% Concessive 27 27 0 28 28 0 55 55 0 10% Imperative 3 1 2 47 15 32 50 16 34 9% Optative 3 0 3 6 0 6 9 0 9 2% Total 286 196 90 259 153 106 545 349 196 100% % 100% 69% 31% 100% 60% 40% 100% 64% 36%

The quantitative results of our investigation on the nal sample are given in Table 1. The data in Table 1 show a complex picture. First of all, the ve functions of magari are not equally distributed across the corpus: the scalar and the ENEF functions cover together 79 percent of the occurrences of magari in the corpus (58 percent and 21 percent respectively), whereas concessives and imperatives are more marginal (10 percent and 9 percent respectively) and optatives are very infrequent (2 percent of the occurrences). Secondly, the association of magari with lists is regular: 349 occurrences of magari out of 545 (64 percent) have in their focus a constituent belonging to a list. It should be noted that, far from being a phenomenon typical of spoken language, the written occurrences of magari present an even more regular association with lists (69 percent). It can also be observed that, with the exception of optatives, which are by the way very rare, magari tends to be associated with lists no matter its exact function. See Figure 2. Yet the association of magari with lists is not equally distributed across the various functions (w 2 71;48, df 4, p < 0;001). As is shown in Figure 3, scalar and concessive magari clearly prefer the association with lists, whereas optative, imperative and (to a lesser extent) ENEF magari tend to be associated with lists less than expected. The ve functions of magari are not even equally distributed across modalities: indeed, there is a signicant interaction between the various functions of magari and the spoken vs. written modality (w 2 121;05, df 4, p < 0;001).6 In particular, scalar uses are more frequent in writing
6. We thank the Associate Editor who reviewed the paper for pointing this out.



Figure 2. The association of the ve functions of magari with lists

Figure 3. Interaction between the ve functions of magari and their association with lists

than expected, while all other uses are preferred in spoken data. These regularities are indicated in Figure 4. Another signicant regularity emerges from the observation of the distribution of the ve functions across modalities x [list] vs. [list] constructions (w 2 204;57, p < 0;001).7 As represented in Figure 5, in fact, it
7. See note 6.


F. Masini and P. Pietrandrea

Figure 4. Interaction between the ve functions of magari and their occurrence across modalities

Figure 5. Interaction between the functions of magari and their occurrence across modalities and list constructions

is clear that the ENEF magari retains its preference for spoken modality, regardless of its association with lists, whereas the strong preference of imperatives for spoken modality breaks down with list constructions. The concessive magari prefers list constructions regardless of modality,



whereas the scalar magari prefers written modality regardless of its association with lists. Finally, optatives disprefer the association with list constructions regardless of modality. All in all, the analysis of the distribution of the ve functions of magari underlines the peculiar behaviour of the optative function, which is much less frequent than the others and does not occur in lists regardless of modality. It will be clearer in what follows that, as theoretically hypothesized within the Behavioral Prole approach (Divjak 2006; Divjak and Gries 2006, Gries 2006), the behavioral regularities observed at the distributional level have a relevance at the cognitive and constructional levels as well. 4.1. Magari as a focus particle

The tendency of the elements focused by magari to occur in lists has led us to consider this word as a particular type of focus particle. As shown by Nlke (1983, 2001) and Konig (1991), focus particles, such as the En glish also, even, only or the French meme are particles endowed with a remarkable syntactic mobility, which have scope on a constituent and focus on a part of it, thereby interacting with the focus structure of the sentence in which they occur (Konig 1991: 10). By focusing on a part of the scope, in fact, focus particles relate the value of the focused expression to a set of paradigmatic alternatives. For example, in (34) also has scope on the entire sentences and focuses on Piero, relating the value Piero to a set of paradigmatic alternatives. This entails the presupposition that someone else has left: (34) Piero has also left

This property, that derives from the very notion of focus (Rooth 1992), has been highlighted by Nlke (1983, 2001), who denes the focus particles of French adverbes paradigmatisants (paradigmatizing adverbs), i.e., adverbs presupposing the existence of a paradigm of variables that act as alternatives to the element in their focus. Magari presents all the features that are typical of focus particles: it is characterized by a noticeable syntactic mobility, it has scope on constituents of various type and size and it focuses on a part of them, relating this focused part to a set of alternatives. In (31), for example, magari has scope on the constituent li riportava in un ambiente reconveyed them in a scene and focuses on in un ambiente in a scene, which is therefore related with a set of alternatives (in a forest, in the jungle, in the


F. Masini and P. Pietrandrea

desert). The peculiarity of magari is that, in the vast majority of its occurrences, the set of alternatives is not merely presupposed, but concretely realized by the list of units occupying the same position of the focused element. As we will see, the characterization of magari as a paradigmatizing adverb will have important consequences for our analysis. 4.2. Dening the topological structure associated with magari

Considering magari as a focus particle anchored to both a focus and a scope enables us to provide a more rigorous denition of the relevant portion of topological space associated with this word. This can be dened as the space on the grid delimited on the horizontal axis by the position of magari plus the extension of its scope and on the vertical axis by the extension of the list of elements occupying the same position of the focus of magari. While the focus of magari is easily identiable through classical tests, the extension of its scope is a less apparent matter. Following the rules established by Nlke (2001: 274) for detecting the extension of the scope of French focus particles, we will distinguish two cases. If magari is pronounced with a neutral intonation, as in (35), it scopes over the whole sequence of units to its right, until the intonational phrase ends. If magari is pronounced with a parenthetical intonation, as in (36), not only the sequence of units to its right, but also the immediately preceding phrase is included in its scope: (35) (36) ` Magari TORNA SUBITO, se non e proprio scemo He might come back immediately, if hes not completely stupid ` STARA CANTANDO, magari SOTTO LA DOCCIA, LA SUA CANZONE PREFERITA He might be singing, maybe in the shower, his favourite song

How are sentences with broken scope like (36) to be represent in grids? The rules mentioned in Section 3.3 impose a representation of the sentence in (36) as in (37). In order to account for the fact that in the abstract predicate-argument-adjunct structure magari has scope over the entire clause, we should write it in the lower left position; however, in order to preserve the linear order of the sequence, we should also write it one line below with respect to the rst constituent uttered.

Magari (37) Grid representation of (36) 1 ` stara cantando he will be singing magari maybe sotto la doccia in the shower PRE ADJ2 la sua canzone preferita his favorite song ARG1



The topological structure relevant for our analysis is now univocally dened. Henceforth, it will be visually delimited by a thicker border, as shown in (38) and (39). It should be noted that this is a mere topological unit, which can be instantiated by items of very dierent type and size ranging from the sole magari, as in (38), to an entire text, as in (39). (38) Magari! I wish (it were like this)! 1 magari I wish PRE (39) Magari stava mangiando, o passeggiando, semplicemente, sul ponte ` . . . magari era l che si stava aggiustando i pantaloni Maybe he was eating, or strolling, simply, on the deck . . . maybe he was over there straightening his trousers [from Alessandro Baricco, Novecento, Milan, Feltrinelli, 1994] 1 2 magari maybe o or stava he was mangiando eating passeggiando semplicemente sul ponte simply strolling on the deck ` era l che si stava he was over there ASP ADJ1 PRE aggiustando i pantaloni straightening his trousers PRE

magari maybe


F. Masini and P. Pietrandrea

This structure can be dened in constructional terms as a semi-specied topological structure characterized not only by the presence of a fully lexically specied item (magari), but also, in most cases, by a specic topological pattern. The latter can be described as a list of equivalent items that occupy the very same position as the focus of magari and that can be of dierent type and size. Although this abstract structure recurs, no matter what the exact function of magari is, the exact form of the list changes according to the function of magari. In the following sections, this phenomenon will be examined in detail. 4.3. Equipotential non exclusion of factuality (ENEF)

About 21 percent of the occurrences of magari fulll the function of presenting the focus of magari as an element whose factuality is not excluded on a par with the factuality of other elements. We call this function equipotential non exclusion of factuality (ENEF). The speaker puts the element in the focus of magari and its alternatives on the same level. In doing so, (s)he does not exclude, but (s)he neither subscribes to the focused element, which is considered equally possible with respect to the other options. Examples of ENEF magari are provided in (40) through (42): (40) ` ` Tentero magari la corona Ibf o Wbc, insomma continuero [laR] Maybe I will try (to win) the Ibf or the Wbc title, in any case I will go on 1 2 3 insomma in any case ` continuero I will go on ARG1 ADJ1 (41) PRE ADJ2 ARG1 ADJ1 ` tentero I will try magari maybe o or la corona the title Ibf Ibf Wbc Wbc

Avremo modo di discutere sui nostri capolavori e sui titoli che magari sono stati messi una o poche volte [Web] We will have a chance to talk about our masterpieces and about the titles that maybe have been quoted one or few times

1 avremo modo well have a chance di discutere to talk dei nostri capolavori about our masterpieces


e dei titoli and about the titles

che that

magari maybe

sono stati messi have been quoted

una one

o or

poche volte few times PRE ADJ2






` ` Magari e arrivato lautobus o e passato un suo amico in macchina [Web] Maybe the bus has come or maybe a friend of his in a car has passed by 1 2 magari maybe o or ADJ1 ` e arrivato has come ` e passato has passed by PRE lautobus the bus un suo amico a friend of his ARG1 in macchina in a car ADJ2

When used in this function, the focus of magari regularly occurs (60 percent of the occurrences in the two corpora, 77 percent in the written corpus) at the top of a list of elements that either occupy one and the same position, as in (40) and (41), or instantiate the same syntactic structure, as in (42). Although in all the examples above magari fullls the same semantic function, it should be noted that the constituent in its focus may belong to very dierent categories: it can be a prepositional argument (31), a nominal argument (40), an adjunct (41) or also a clause (42). This fact supports the hypothesis that it is the topological structure of the construction (in particular the position of the focus of magari at the top of a list), rather than other categorial variables, that is relevant for licensing the non factual reading of magari.


F. Masini and P. Pietrandrea

It is worth mentioning that the conjuncts that are listed below the constituent focused by magari can be introduced by disjunctive conjunctionssuch as piuttosto che rather than in (31) or o or, as in (40) through (42)or by a second occurrence of magari, as in (43), which can function as a disjunctive connective according to Mauri (2008a, 2008b). Sometimes, especially when the list is long enough, the items can be listed without explicit conjunction markers. In all cases the list is interpreted as a disjunctive list. (43) ` ` ` ` Magari e l da un attimo magari e l da sempre Maybe hes been there for a second, maybe hes been there forever [from Alessandro Baricco, Oceano Mare, Milan, BUR, 1999]

The regular association of magari with disjunctive lists suggests that the overall eect of equipotential non exclusion of factuality is constructional in nature. As already argued, magari is a general marker of non factuality. This marker happens to be regularly associated with lists. It is precisely this regular association with lists that turns magari into a more specic kind of marker, i.e., a marker of non exclusion of factuality. The fact that the list we are dealing with is disjunctive in nature adds still another feature. Indeed, the disjunctive list can be characterized as the semantic relation which obtains between two (or more) items that are equally possible [ . . . ] and are potential substitutes for each other (Mauri 2008a: 25). Therefore, the fact that the focus of magari belongs to a disjunctive list suggests that it is put forward as an option not to be excluded on a par with the other listed options. This combination of features contributes to produce the interpretation of magari as a marker of equipotential non exclusion of factuality. 4.4. Scalar non exclusion of factuality

As much as 58 percent of the occurrences of magari in the corpus (78 percent in the written corpus) fulll the function of scalar operator of non factuality, in that they trigger a scale of non factuality whose extreme position is occupied by the element in its focus. Examples are in (44) through (47): (44) I lm di oggi saranno stati approvati dallalto tre, quattro, magari cinque volte [laR] Todays movies have been probably approved from on high three, four, maybe ve times

Magari 1 i lm doggi Todays movies saranno stati approvati have been probably approved dallalto from on high tre three


2 3 magari maybe

quattro four cinque volte ve times ADJ2


Vorrei strapparle una parola, una battuta, magari un mezzo sorriso [Web] I would like to get a word, a quip, maybe a faint smile out of her 1 vorrei I would like to strapparle get out of her una parola a word una battuta a quip magari maybe MOD PRE PRE ADJ1 ARG2 un mezzo sorriso a faint smile

2 3


` Li condanna a vivere in una societa che non a torto e non per razzismo li vede con sospetto, li sfugge e magari li respinge [laR] It condemns them to live in a society that not injustly and not for racism views them with suspicion, keeps away from them and maybe rejects them


F. Masini and P. Pietrandrea

li condanna it condemns them a vivere to live in una ` societa in a society che that non a torto not injustly e non per razzismo and not for racism li vede con sospetto views them with suspicion li sfugge keeps away from them e magari and maybe ARG1 PRE ARG1 ADJ1 ADJ2 ADJ1 ADJ2 li respinge rejects them



Alla ne io mi sarei sentita in colpa e magari lui avrebbe nito per detestarmi [Web] In the end I would have felt guilty and maybe he would have ended up hating me 1 alla ne in the end e magari and maybe ADJ1 ADJ2 io I lui he mi sarei sentita in colpa would have felt guilty avrebbe nito per detestarmi would have ended up hating me PRE


The scalar function can be considered a particular instance of the focusing character of magari. It was shown in Section 4.1 that, as a focus particle, magari entails the existence of a certain number of propositions that form a paradigm. In (44), for example, the paradigm is comprised of the following propositions: they have been approved from on high three times, they have been approved from on high four times, they have been approved from on high ve times. In the function under examina-



tion, magari indicates that the constituent in its focus (the proposition they have been approved from on high ve times) realizes the most extreme proposition in the paradigm, i.e., the most non factual one or, rather, the last one for which the speaker would not exclude the factuality. This imposes a directionalityand consequently a scalarityto the paradigm, which turns into a scalar domain of non factuality, in which the proposition realized by the constituent in the focus of magari has the highest degree of non factuality. As examples (44) through (47) make clear, when fullling this function, the focus of magari regularly (66 percent) occurs at the bottom of a list of constituents occupying the same syntactic position or realizing the same syntactic structure. Also in this case, it is clear that the constituents in the list may dier largely from each other in size and category: they can be nominal arguments (45), predicates (46), adjuncts (44) or even entire clauses (47). This suggests that the scalar meaning of magari is licensed by the peculiar topological structure associated with it, i.e., by the occurrence of the focus of magari at the bottom of a list of constituents. Our analysis so far shows that the function of magari as a scalar operator of non exclusion of factuality is constructional in nature. The general non factual meaning of magari combined with a list yields an overall meaning of non exclusion of factuality, as shown in Section 4.3. The fact that magari focuses on the last conjunct of a listas already noted by Fauconnier (1976) and Kay (1990) in their analyses of the French word meme and the English word evenintroduces in the same construction an entire domain (corresponding to the items listed above the one in the focus of magari plus the item in the focus of magari) and, at the same time, the most extreme item of that domain (corresponding to the focus of magari). Representing one of the listed items as the most extreme in the domain induces a ranking. Given the semantic nature of magari, the listed items are ordered for increasing degree of non factuality, more precisely they are ordered from the most likely to the last one for which the speaker would not exclude the factuality. In spite of the well established association of scalar magari with list constructions, 42 percent of the occurrences of magari with this function are not associated with a list. We will account for these exceptions in Section 4.8. 4.5. Scalar concessive conditional

About 10 percent of the occurrences of magari have a particular type of scalar function: they are scalar concessive conditionals. An example was provided in (3), another one is in (48):

104 (48)

F. Masini and P. Pietrandrea ` ` Ciascuna di queste vicende e, magari, piccola; ma la loro somma e un grande dramma [laR] Each of these events is, maybe, small; but their sum is a great tragedy

The name scalar concessive conditional has been proposed by Haspelmath and Konig (1998) to indicate concessive constructions such as that in (49): (49) Even if we do not get any nancial support, we will go ahead with our project

This construction can be regarded as a particular conditional construction in which a set of protases is related to an apodosis (Haspelmath and Konig 1998: 565). In (49) the set of protases is made up of the vari ous conditions evoked by the scalar operator even (if we get great nancial support, if we get some nancial support, if we do not get any nancial support). These conditions are clearly ranked in a scalar domain according to degree of adversity for the situation described in the apodosis, the condition in the focus of even (if we do not get any nancial support) being considered as the most adverse. In scalar concessive conditionals, the set of protases describes non factual conditions, whereas the apodosis is normally factual.8 It is exactly the combined eect of the factuality of the apodosis and the adversity of the circumstances described in the adverbial clause that triggers the concessive interpretation of this conditional construction. Examples such as (48) can be considered as particular cases of scalar concessive conditionals. Magari in fact evokes a set of conditions arranged in a scalar domain: whether each of these events is remarkable, whether each of these events is normal, or whether each of these events is small. These conditions are non factualdue to the presence of magariand they are ranked not only according to degree of adversity but also to degree of non factuality. The condition in the focus of magari ` (ciascuna di queste vicende e piccola each of these events is small) is therefore not only the most unfavorable, but also the most unlikely. The main clause, however, is clearly factual. As shown in the grid representations in (50) and (51), when magari fullls the function of scalar concessive conditional, its focus is always a non factual item that occupies the rst position of a list made up of at least


Konig and Haspelmath (1998: 573) discuss some marginal exceptions to the factuality of the apodosis, but they are not relevant for our purposes.



two conjuncts, the last of which is introduced by an adversative conjunc` tion (ma/pero but, invece whereas): (50) Il comandante Arguelles si aspettava quindi un temporale, magari violento ma facile da superare [laR] Captain Arguelles therefore expected a possibly violent, but easy to overcome storm
il comandante Arguelles Captain Arguelles si aspettava expected quindi therefore un temporale a storm

magari arguably ma but

violento violent facile da superare easy to overcome ARG1 ADJ2



` ` Magari andra per le lunghe, ma non nisce cos Maybe it will go overtime, but it doesnt end like this 1 3 magari maybe ma but non not ADJ1 ADJ1 PRE ` andra it will go nisce it ends PRE per le lunghe overtime ` cos like this ARG1


Also in this case, the conjuncts in the focus of magari may be constituents of dierent category and size, such as adjuncts (50) or clauses (51). Therefore, it is only the kind of topological structure associated with this use of magari that licenses its scalar concessive meaning. On the one hand, the list within the topological structure of magari instantiates a particular type of list: [x1 (x2 , . . .), adversative conjunction, xlast , where x1 (x2 , . . .) 3non factual4, xlast 3factual4]. This abstract scheme is also typically associated with non factual concessive meanings in Italian:

106 (52)

F. Masini and P. Pietrandrea ` ` Puo essere che sia intelligente, pero non lo dimostra It is possible that (s)he is clever, but (s)he doesnt show it

On the other hand, the occurrence of magari in a conditional concessive context licences its scalar reading. Magari, as well as other focus particles (such as the Italian anche/pure also or the German auch also), always acquires a scalar meaning in conditional concessive contexts. As shown by Konig (1991: 64), this reinterpretation depends on the Gricean maxim of Relevance: if a conditional connection between two eventualities is asserted and presupposed, it is invariably the more remarkable case that it is asserted. Thus in (50) it would be trivial to assert that the storm expected by the captain may be mild. This eventuality is presupposed, whereas the more remarkable case the storm may be violent is asserted. This tendency entails that the focus of the conditional concessive magari is usually interpreted as the most extreme item in a scalar domain of conditions not to be excluded, i.e., the conditional concessive magari is always interpreted as scalar. 4.6. Imperative

About 9 percent of the occurrences of magari in the corpus have in their focus an imperative (53) or a related construction (Konig and Siemund 2007), such as a hortative (54) or a deontically modalized sentence (55). (53) Magari diglielo, faglielo comunque capire che ci tieni a lui! Maybe tell him, anyway make him understand that you care about him! Senti questo teniamolo, magari vediamolo alle prime bozze! [LIP] Listen, lets keep this, maybe lets see it at the proofreading stage! Bisogna seguire un certo regime alimentare, bisogna magari mangiare un po meno It is necessary to follow a certain diet, it is necessary, maybe, to eat a little less

(54) (55)

As shown by Elliott (2000: 76) and De Haan (2004), the presence of a marker of non factuality in imperative and related constructions is quite a common phenomenon across languages. This association may be motivated by the fact that commands describe non factual situations, which favors the presence of a non factual marker (see Elliott 2000: 76). Nevertheless, as mentioned in Section 1, magari does not merely harmonically mark the non factuality of the command, butas often happens crosslinguistically (Mithun 1995)it also fullls another function: it serves to weaken the force of the command (or exhortation). For example, the



commands in (53) are considered as less mandatory, and consequently more polite, than their counterpart in (56) where magari is absent: (56) Diglielo, faglielo comunque capire che ci tieni a lui! Tell him, anyway make him understand that you care about him!

Needless to say, the weakened imperative function of magari is marginal in the written corpus, while it is attested in as much as 18 percent of the occurrences in the spoken corpus. As the examples (53) to (55) and the grid representations below show, the imperatives in the focus of magari often occur in lists. This holds for about 32 percent of the imperative magari in our corpus. The imperatives may occur at the top of a disjunctive list, as in (57), or at the bottom, as in (58), in which case they also have a scalar meaning: (57) Grid representation of (53) 1 2 magari maybe diglielo tell him faglielo make him comunque anyway capire understand che ci tieni a lui that you care about him




Prova a calmarti un po [ . . . ] e magari chiedi scusa alla mamma [Web] Try to calm down a bit and possibly apologize to your mother 1 2 e magari and possibly ADJ1 prova try chiedi scusa apologize PRE a calmarti to calm down alla mamma to your mother ARG1 ADJ2 un po a bit

The occurrence of the imperative focused by magari in a list of imperatives makes it clear how magari weakens the illocutionary force of the imperative. When the speaker puts the focus of magari at the top of a


F. Masini and P. Pietrandrea

list of alternatives (s)he invites the listener to take his/her command into account as but one of other options. When (s)he puts the focus of magari at the bottom of a list of commands he puts forward the focused command as to be executed as a last resort. It is clear that the weakened imperative function of magari, is, as well as the other functions, constructional in nature. The imperative (or its related constructions) endows the construction with the illocutionary force of a command. The non exclusion value of magari, combined with the occurrence of its focus in a list, is used to present the focused command as an option which is not to be excluded. 4.7. Optative

Less than 2 percent of the occurrences of magari in the corpus have an optative function such as that represented in the following example: (59) ` Magari fosse cos semplice! I wish it were so simple!

It has been shown by Pietrandrea (2008b) that, when introduced in Italian in the 13th century, the word magarietymologically related to the Greek makarios (blissful)only had an optative meaning. It was usually employed as a predicative adjective uttered with an exclamative intonation referring to a sentential subject introduced by the complementizer ke that, as in (60): (60) Makare ke mme abberanno uccisa! If only they killed me! [Iacopone da Todi, XIII laude del Laudario Urbinate, 13th century]

When fullling the optative function, magari is always associated with an exclamative intonational prole. As the examples below show, in the scope of magari there can be a past subjunctive (61), an innitive (62), a non verbal element (63) or even a element as in (6), reproduced in (64): (61) (62) (63) (64) Magari venisse! I wish he would come! Magari averne! I wish I had! Magari due! I wish (there would be) two of them A: Vuoi un po di riposo? Do you want to rest a bit? B: Magari! Id love to!



Apparently the focus of the optative magari is never associated with topological structures characterized by lists. Pietrandrea (2008b) suggests that a relation of semantic bleaching exists between the optative and the non exclusion of factuality function of magari. Such a bleaching may have been historically induced by ambiguous contexts and reinforced precisely by the coalescence of magari within constructions characterized by lists. In fact, as pointed out by Pietrandrea (in print), an optative meaning can be conceived of as the indication of a selection among a set of alternative (SoAs). Consequently, the occurrence of magari within list constructions (where more than one alternative option is expressed) has the eect of weakening the meaning of selection and favouring a more general non factual reading. Synchronically speaking, the optative and the non exclusion of factuality uses of magari are nevertheless related to one another in that they both express non factuality. 4.8. Exceptions

The fact that the focus of the optative magari never belongs to a list has been theoretically explained in the previous section. However, we know from the data in Table 1 that the foci of ENEF, scalar and imperative magari are also not necessarily part of a list. This phenomenon characterizes 36 percent of the occurrences in the corpus and, therefore, it represents an exception to be explained. Let us consider the cases in which magari fullls an ENEF function, but its focus does not belong to a list, as in the following example: (65) ` ` Non rischio, non opero scelte che magari potevano suscitare contrasti [laR] (S)he didnt risk, (s)he didnt make choices that maybe could cause disagreements
1 ` non rischio (s)he didnt risk ` non opero (s)he didnt make scelte choices che that magari maybe potevano suscitare could cause PRE contrasti disagreements





F. Masini and P. Pietrandrea

We may hypothesize that magari behaves in this context as a regular focus particle. It merely presupposes, without realizing it, the existence of a paradigm of unspecied alternatives to the focus of magari. For example, the clause in the focus of magari in (65) ( potevano suscitare contrasti could cause disagreements) evokes a paradigm of other unspecied, but semantically related, possible alternatives which are not explicitly mentioned in the text, such as: potevano provocare proteste could cause protests, potevano procurare inimicizie could cause hostility, etc. Therefore, it may be hypothesized that this type of magari has inherited the property of evoking paradigms of elements alternative to its focus precisely from its more frequent association with fully realized lists. A similar line of reasoning might be applied to the occurrences of magari in imperatival contexts. Thus a sentence like (66) evokes, without realizing it, a paradigm of alternative commands (tell him, dont tell him), thereby weakening the illocutionary force of the imperative. (66) Magari diglielo Maybe tell him/her

Another exception to be dealt with regards the scalar function of magari. About 34 percent of the scalar occurrences of magari appear in contexts without a list. These are cases like those represented in the following example: (67) [ . . . ] dovrei parlarvi di vini, magari toscani [laR] [ . . . ] I should talk to you about wines, possibly Tuscan (wines)

In this construction magari and its focus have a parenthetical intonation. This characteristic leads us to put forward a possible explanation for these apparently exceptional cases. Indeed, according to the rules established in Section 3.3 for grid representations, when the constituent immediately preceding magari is parenthetical, it should be included in its scope. As a consequence, the grid representations of (67) would be akin to that provided in (68): (68) Grid representation of (67) 1 dovrei I should parlarvi talk to you magari possibly MOD PRE PRE ADJ1 ARG1 ARG1 di vini about wines toscani Tuscan ADJ2



If we rely on this representation, we can hypothesize that magari has focus on a constituent that lies at the bottom of a partially instantiated list. The constituent in the focus of magari is an adjunct that modies the backgrounded part of the scope. In (67), for example, magari focuses on the adjunct toscani Tuscan, which modies the backgrounded item (di vini about wines). This item is given the rst time without modications in the sequence dovrei parlarvi di vini I should talk to you about wines. It is then modied, without an explicit repetition, in the sequence magari toscani possibly Tuscan. The modication, without reiteration of the backgrounded part of the scope, means that the latter elided. The overall eect is that a sequence such as (67) is interpreted as equivalent to the following: (69) [ . . . ] dovrei parlarvi di vini, magari di vini toscani 9 [ . . . ] I should talk to you about wines, possibly Tuscan wines 1 dovrei I should parlarvi talk to you magari possibly MOD PRE PRE ADJ1 ARG1 di vini about wines di vini about wines ARG1 toscani Tuscan ADJ2

In (69), the modifying adjunct focused by magaritoscani Tuscan lies at the bottom of a list of two adjuncts. The rst of these adjuncts (position ADJ2 -line1) is a element, i.e., the position is empty. Consequently, it conveys a meaning such as non qualied (wines) and the whole sequence is interpretable as:

9. Constructions like (69) are indeed grammatical and attested. See, for example, the following instances: ` ` (i) Poi si vedra, se trovero il tempo per dedicarmi ad un uomo, magari un uomo vero Then we will see, if I will nd the time to dedicate myself to a man, possibly a real man (ii) Cera una volta sarebbe un inizio perfetto per cominciare una storia, magari una storia per bambini Once upon a time would be a perfect beginning to start a story, possibly a story for children

112 (70)

F. Masini and P. Pietrandrea [ . . . ] dovrei parlarvi di vini quali che siano, magari di vini toscani [ . . . ] I should talk to you about whichever wines, possibly Tuscan wines

In conclusion, the presence of the modifying adjunct at the bottom of the partially instantiated list would trigger a scale of non factual propositions. The actualization of a more specic eventsuch as that described by the proposition I talk to you about Tuscan winesis in fact to be conceived as less likely than the actualization of a more general event such as that described by the proposition I talk to you about wines. To sum up, two hypotheses can be put forward in order to explain the cases in which the focus of magari does not belong to a list and to relate these cases to other more frequent cases with lists. The rst hypothesis, which virtually applies to all exceptional cases, is that magari evokes a paradigm of possible alternatives with respect to the element in its focus by virtue of its frequent association with concretely realized lists. The second hypothesis only applies to scalar occurrences. In these cases the constituent in the focus of magari is always parenthetical and can be seen as the second conjunct of a partially instantiated list. In both cases, the presence of a list is posited. This entails that from a cognitive, if not linguistic, point of view the focus of magari always belongs to a list, either fully instantiated, or partially instantiated, or simply evoked. This association of magari with a list would mark its general non exclusion of factuality meaning. 5. The network of magari constructions

As mentioned in the introductory sections of this paper, the main goal of our investigation is to understand which contexts license the various functions of magari and whether there is a relation between these functions (and of course which sort of relation). In order to reach this goal, we set our analysis within the general theoretical framework of construction grammar, which in principle allows to treat contexts as linguistic objects, and then we used a specic working methodology, namely the topological methodology, which allowed us to identify a set of topological structures in which magari regularly occurs. In this section we will give a more rened constructionist account of our ndings. The analysis carried out in Section 4 shows that we can distinguish two main magari constructions: the optative magari construction; the map of non exclusion of factuality (NEF) magari constructions: ENEF;

Magari scalar NEF (with fully and partially instantiated list); scalar concessive conditional; weakened imperative.


As we briey discussed in Section 4.7, the optative magari presents specic distributional properties that distinguish it from other magari constructions: it is very infrequent, it does not occur with lists and it is always associated with an exclamative intonational prole. However, the two constructions are not completely independent from one another. Firstly, Pietrandrea (2008b) showed that there exists a diachronic semantic bleaching from optative to non exclusion of factuality, thus positing a sort of diachronic link between the two constructions. Secondly, from the point of view of our synchronic analysis, the two constructions share the presence of magari and a general non factuality feature. As for the set of NEF magari constructions, we identied a class of topological structures in which magari occurs regularly. All these structures refer to a more general topological structure that can be represented as in Figure 6: the lexically specied adverb magari is followed by its scope, which is made up of a background and a focus; the latter is part of a list, i.e., is one of the listed elements. It is important to note that there is no explicit information about levels, categories, word order or sentence types, so all this information is underspecied, as well as the type of list involved. All the dierent magari constructions analysed in the sections above are more specied instances of the maximally abstract construction outlined in Figure 6. In other words, these constructions have some properties that specify the abstract construction (partially) described in Figure 6. These properties are listed below: ENEF magari construction: list disjunctive focus x1

Figure 6. The topological structure of the Abstract NEF magari construction


F. Masini and P. Pietrandrea scalar NEF magari construction (fully instantiated list): focus xlast scalar concessive conditional magari construction: x1 (x2 , . . .) 3non factual4 vs. xlast 3factual4 focus x1 imperative magari construction: sentence type imperative speech act command, exhortation, etc.

If we accept the hypothesis of the partially instantiated list put forward for the exceptional cases of magari with a scalar function (cf. Section 4.8), then we still have another construction with the following overriding properties: scalar NEF magari construction (partially instantiated list): list [x1 , x2last ], in which x1 focus x2last intonation: parenthetical

Therefore, from a constructionist perspective, the distribution of the non exclusion of factuality magari can be accounted for by positing a hierarchy of closely related topological structures, each of which is regularly associated with one determined function of magari and all of which are linked to a more abstract construction with the general meaning of 3non exclusion of factuality4 and the topological structure described in Figure 6. The network of magari constructions emerging from our results is presented in Figure 7. Before commenting on this gure, it is worth discussing some of the conventions used. First, we made use of the inheritance links proposed by Goldberg (1995) (cf. Section 3.1) to relate the various constructions at issue. Second, regarding the representation of the constructions themselves, we adapted the boxes-within-boxes notation (Fried 2007; Fried and Ostman 2004) to our needs by incorporating the outline of the topological structure associated with magari as the formal part of the construction. Third, some elements are graphically highlighted in order to facilitate the reading of the network: the various magari constructions endowed with a topological structure are enclosed in boxes with thicker borders, whereas the overriding properties of each subconstruction are put in boldface. Constructions with an uncertain status are marked by a question mark near the inheritance link and enclosed in a box framed by a dotted line. Finally, as can be seen, Figure 7 does not contain the imperative magari. The representation of this construction is given in Figure 8, which we will comment on later. Now let us return to Figure 7.



Figure 7. The network of magari constructions


F. Masini and P. Pietrandrea

Overall, the network of constructions proposed in Figure 7 reveals that the network of magari constructions is basically governed by Instance inheritance links (II ). A maximally abstract Non factual magari construction in instantiates by both the Optative magari construction and the Abstract NEF magari construction. The other magari subconstructions are inherited from the Abstract NEF magari construction by means of instance inheritance links. The Abstract NEF magari construction as represented in Figure 6 is also linkedby means of a Subpart inheritance link (IS )to an independent List construction with the maximally abstract meaning of 3relation between the listed items4, whose existence has been proposed in Section 3.2. As already pointed out, it is precisely the presence of a list that somehow turns the general non factual meaning of magari into a 3non exclusion of factuality4 meaning. At the same time, the ENEF magari construction is linkedby means of an IS to the Disjunctive list construction, which is an instance of the general List construction.10 The main property of the Abstract NEF magari construction is that it features a topological list that includes the element which is in the focus of magari. At this level, however, the interaction between the list and the focus is still underspecied, as well as the list itself. This information becomes more specied as we reach the lower levels of the hierarchy. Both the ENEF magari construction and the Scalar NEF magari construction (with a fully specied list) specify which element of the list is in the focus of magari. In the former, the focus is at the top of the list, and the list is disjunctive; in the latter, the focus is at the bottom of the list. According to the hypothesis put forward in Section 4.8, the Scalar NEF magari might be instantiated by another constructionthe Scalar NEF magari construction with a partially instantiated listwhose topological structure is even more constrained. The list cannot contain more than two elements and the rst one is a null element. Finally, the Abstract NEF magari construction is instantiated by the Scalar concessive conditional construction, which includes a contrastive list in which one or more non factual elements are contrasted with a fac-

10. It should be noted that this link is not strictly necessary, since the Disjunctive list may be instantiated by directly within the ENEF magari construction, that inherits the more general List construction from the Abstract NEF magari construction. We however decided to maintain this link for the sake of explicitness, and more precisely to highlight the semantic contribution of the disjunctive list to the whole ENEF construction.



Figure 8. Imperative magari constructions

tual element. The Scalar concessive conditional construction is an instantiation of both the Abstract NEF magari construction and the (abstract) concessive construction, therefore we are dealing with a case of multiple inheritance. In addition, this construction is linked by a purely semantic link (represented here by a dotted line) to the Scalar NEF magari construction, since they share the scalarity feature. As mentioned above, the Imperative magari construction is not present in this network. In fact, this is due to the fact that we interpret the Imperative magari construction as a further instantiation of both the ENEF magari construction and the Scalar NEF magari construction. As mentioned in Section 4.6, and as reproduced in Figure 8, the Imperative magari construction may have the topological structure of both the former and the latter. In both cases, the corresponding meaning of magari is maintained and a general function of weakening of the command/ exhortation is added. Therefore, the Imperative magari construction can be seen as a lower-level construction in which sentence type and speech act information is specied. Also, the two constructions are linked to one another (by a dotted link), since they share the Sentence type and Speech act features. In conclusion, the proposed constructionist analysis allows to connect all magari constructions with one another in an inheritance hierarchy and therefore gives us a better understanding of the speakers knowledge of this piece of grammar.

118 6.

F. Masini and P. Pietrandrea Conclusions

The word magari has a number of grammatical meanings: equipotential non exclusion of factuality, scalarity, concessivity, weakening of the illocutionary force of the imperative and optativity. All these meanings proved to be constructional in nature, i.e., they are determined by the various constructions in which magari occurs. These constructions were eciently identied by looking at topological patterns, i.e., structures that are recognizable at the discourse conguration level. This level of analysis, dened by the maintenance of a given predicate-argumentadjunct structure in discourse, crosses the traditional divide between clausal and supra-clausal level and can only be characterized in terms of its topological structure. The distribution of magari within discourse congurations has revealed interesting regularities. With the exception of the more marginal and more ancient optative function, magari is regularly associated with certain abstract topological patterns that can be characterized as lists containing the element focused by magari. The exact shape of this topological pattern is the only distinctive property that allows to univocally identify each type of magari. Finally, these meaningful topological structures can be reinterpreted as proper constructions, whose peculiarity consists in that they are insensitive to the boundary between clauses and are bi-dimensional in nature. They can also be represented in an inheritance hierarchy, which shows how the dierent magari constructions are inherited from a maximally abstract construction. Received 7 April 2008 Revision received 27 October 2008 Dipartimento di Linguistica ` Universita Roma Tre Rome, Italy

Arcaini, Enrico. 1997. Le connecteur magari dans une perspective comparative. In Gerd Wotjak (ed.), Studien zum romanisch-deutschen und innerromanischen Sprachvergleich, 5976. Frankfurt am Main: Peter Lang. Arcaini, Enrico. 2000. Italiano e francese. Unanalisi comparativa. Turin: Paravia Scriptorium. Baroni, Marco, Silvia Bernardini, Federica Comastri, Lorenzo Piccioni, Alessandra Volpi, Guy Aston & Marco Mazzoleni. 2004. Introducing the La Repubblica corpus: A large, annotated, TEI(XML)-compliant corpus of newspaper Italian. In Proceedings of the 4th International Conference on Language Resources and Evaluation (LREC 2004, Lisbon, 2628 May), 17711774. Paris: ELRA.



Bertinetto, Pier Marco. 1991. Il verbo. In Lorenzo Renzi, Giampaolo Salvi & Anna Cardinaletti (eds.), Grande grammatica italiana di consultazione, vol. II, 13161. Bologna: Il Mulino. ` Bilger, Mireille. 1982. Contribution a lanalyse en grille. Recherches sur le francais parle 4. 195215. ` Bilger Mireille, Mylene Blasco, Paul Cappeau, Frederic Sabio & Marie-Josee Savelli. 1997. Transcription de loral et interpretation: illustration de quelques dicultes. Recherches sur le francais parle 14. 5585. Blanche-Benveniste, Claire. 1993. Repetitions de lexique et glissement vers la gauche. Re cherches sur le francais parle 12. 934. Blanche-Benveniste, Claire. 1997. Approches de la langue parlee en francais. Paris: Ophrys. Blanche-Benveniste, Claire, Bernard Borel, Jose Deulofeu, Jacky Durand, Alain Giacomi, Claude Loufrani, Boudjema Meziane & Nelly Pazery. 1979. Des grilles pour le francais parle. Recherches sur le francais parle 2. 163205. Blanche-Benveniste, Claire, Mireille Bilger, Christine Rouget & Karel Van den Eyende. 1990. Le francais parle. Etudes grammaticales. Paris: Editions du Centre National de la Recherche Scientique. Bonvino, Elisabetta. 2005. Le sujet postverbal. Une etude sur litalien parle. Paris: Ophrys. Bonvino, Elisabetta, Francesca Masini & Paola Pietrandrea. 2009. List Constructions: a semantic network. Paper presented at the Third International Conference of the French Association of Cognitive Linguistics (AFLiCo) Grammars in Construction(s), Paris, Nanterre, 2729 May. Croft, William. 2001. Radical Construction Grammar. Oxford: Oxford University Press. Croft, William & Alan D. Cruse. 2004. Cognitive Linguistics. Oxford: Oxford University Press. de Haan, Ferdinand. 2004. On representing semantic maps. Tucson, AZ: University of Arizona manuscript. URL: http:/ / (accessed 3 March 2008). De Mauro, Tullio, Federico Mancini, Massimo Vedovelli & Miriam Voghera. 1993. Lessico di frequenza dellitaliano parlato. Milan: ETAS libri. Divjak, Dagmar S. 2006. Ways of intending: Delineating and structuring near synonyms. In Stefan Th. Gries & Anatol Stefanowitsch (eds.), Corpora in cognitive linguistics: Corpusbased approaches to syntax and lexis, 1956. Berlin & New York: Mouton de Gruyter. Divjak, Dagmar S. & Stefan Th. Gries. 2006. Ways of trying in Russian: Clustering behavioral proles. Corpus Linguistics and Linguistic Theory 2(1). 2360. Elliott, Jennifer R. 2000. Realis and irrealis: Forms and concepts of the grammaticalization of reality. Linguistic Typology 4. 5590. Fauconnier, Gilles. 1976. Etude de certains aspects logiques et grammaticaux de la quantication et de lanaphore en francais et en anglais. Paris: Champion. Fillmore, Charles J., Paul Kay & Mary Catherine OConnor. 1988. Regularity and idiomaticity in grammatical constructions: The case of let alone. Language 64(3). 501538. Fried, Mirjam. 2007. Constructing grammatical meaning. Isomorphism and polysemy in Czech reexivization. Studies in Language 31(4). 721764. Fried, Mirjam & Jan-Ola Ostman. 2004. Construction Grammar: A thumbnail sketch. In Mirjam Fried & Jan-Ola Ostman (eds.), Construction Grammar in a cross-language perspective, 1186. Amsterdam & Philadelphia: John Benjamins. Fried, Mirjam & Jan-Ola Ostman. 2005. Construction Grammar and spoken language: The case of pragmatic particles. Journal of Pragmatics 37(11). 17521778. Gerdes, Kim & Sylvain Kahane. In print. Speaking in Piles. Paradigmatic Annotation of a Spoken French Corpus. In Proceedings of the Fifth Corpus Linguistics Conference, Liverpool.


F. Masini and P. Pietrandrea

Goldberg, Adele. 1995. Constructions. A Construction Grammar approach to argument structures. Chicago: The University of Chicago Press. Goldberg, Adele. 2006. Constructions at work. Oxford: Oxford University Press. Gries, Stefan Th. 2006. Corpus-based methods and cognitive semantics: The many meanings of to run. In Stefan Th. Gries & Anatol Stefanowitsch (eds.), Corpora in cognitive linguistics: Corpus-based approaches to syntax and lexis, 5799. Berlin & New York: Mouton de Gruyter. Haspelmath, Martin & Ekkehard Konig. 1998. Concessive conditionals in the languages of Europe. In Johan van der Auwera (ed.), Adverbial relations in the languages of Europe, 277334. Berlin & New York: Mouton de Gruyter. Kay, Paul. 1990. Even. Linguistics and Philosophy 13. 59111. Kay, Paul & Charles J. Fillmore. 1999. Grammatical constructions and linguistic generalizations: The whats X doing Y? construction. Language 75(1). 133. Ko nig, Ekkehard. 1991. The meaning of focus particles: A comparative perspective. London: Routledge. Ko nig, Ekkehard & Peter Siemund. 2007. Speech acts distinctions in grammar. In Timothy Shopen (ed.), Language typology and syntactic description, vol. I: Clause structure, 2nd edn, 276304. Cambridge: Cambridge University Press. Lazard Gilbert. 1998. Lexpression de lirreel: essai de typologie. In Leonid Kulikov & Heinz Vater (eds.), Typology of verbal categories: Papers presented to Vladimir Nedjalkov on the occasion of his 70 th birthday, 237248. Tubingen: Max Niemeyer. Licari, Carmen & Stefania Stame. 1989. Pour une analyse contrastive des connecteurs prag matiques italiens et francais: magari/peut-etre, anzi/au contraire. Studi Italiani di Linguis tica Teorica e Applicata 18. 153161. Mauri, Caterina. 2008a. The irreality of alternatives: Towards a typology of disjunction. Studies in Language 32(1). 2255. Mauri, Caterina. 2008b. Coordination relations in the languages of Europe and beyond. Berlin & New York: Mouton de Gruyter. Michaelis, Laura & Knud Lambrecht. 1996. Towards a construction-based theory of language function: The case of nominal extraposition. Language 72(2). 215247. Mithun, Marianne. 1995. On the relativity of irreality. In Joan Bybee & Suzanne Fleischman (eds.), Modality in grammar and discourse, 367388. Amsterdam & Philadelphia: John Benjamins. Mithun, Marianne. 2005. On the assumption of the sentence as the basic unit of syntactic structure. In Zygmunt Frajzyngier, Adam Hodges & David S. Rood (eds.), Linguistic diversity and language theories, 169183. Amsterdam & Philadelphia: John Benjamins. Mithun, Marianne. 2008. The extension of dependency beyond the sentence. Language 84(1). 69119. Nlke, Henning. 1983. Les adverbes paradigmatisants: fonction et analyse. Copenhagen: Akademisk Forlag. Nlke, Henning. 2001. Le regard du locuteur 2: Pour une linguistique des traces enonciatives. Paris: Kime. Ostman, Jan-Ola. 2005. Construction discourse: A prolegomenon. In Jan-Ola Ostman & Mirjam Fried (eds.), Construction grammars. Cognitive grounding and theoretical extensions, 121144. Amsterdam & Philadelphia: John Benjamins. Pietrandrea, Paola. 2007. The grammatical nature of some epistemic-evidential adverbs in Spoken Italian. Italian Journal of Linguistics 1. 3964. Pietrandrea, Paola. 2008a. Certamente and sicuramente: Encoding dynamic and discursive aspects of commitment in Italian. Belgian Journal of Linguistics 22. 221246.



Pietrandrea, Paola. 2008b. Constructionalization, grammaticalization and discourse. The case of magari. Paper presented at the 4th New Reections on Grammaticalization Conference, Leuven, 1619 July. Pietrandrea, Paola. In print. The conceptual structure of irreality. A focus on non-exclusionof-factuality as a conceptual and a linguistic category. Language Sciences. Rooth, Mats. 1992. A theory of focus interpretation. Natural Language Semantics 1. 75 116. Schiemann, Anika. 2008. La polisemia di magari (e forse). Analisi corpus based su CORAL-ROM italiano. In Emanuela Cresti (ed.), Prospettive nello studio del lessico italiano. Atti SILFI 2006, vol. I, 299307. Florence: Firenze University Press.

Reviewing imagery in resemblance and non-resemblance metaphors


Abstract This article analyses the nature of mental imagery in metaphoric thought as envisaged by the contemporary theory of metaphor in Cognitive Linguistics (Lako 1993). Our study of metaphor in the eld of marine biology draws on two crucial aspects of mental imagery, namely dynamicity and pervasiveness. Image metaphors and behaviour-based metaphors have generally been regarded as two dierent types of resemblance metaphor. In our view, the dynamicity of certain mental images highlights inherent similarities between these two types of metaphor, and makes the dierences between them more apparent than real. For this reason, we propose a more rened description of resemblance metaphors in terms of the static or dynamic nature of the mental images underlying them. Our study also underlines the fact that mental images permeate all classes of metaphor, and that the pervasiveness and dynamicity of mental images aords insights into both resemblance metaphors and non-resemblance metaphors. Keywords: mental imagery; metaphor; dynamicity; marine biology; cognitive linguistics.



Conceptual Metaphor Theory and Primary Metaphor Theory establish a sharp distinction between metaphors that arise from physical or behavioural analogy and metaphors motivated by abstract or subjective cognitive processes. In the case of Conceptual Metaphor Theory, Lako (1993)

* Correspondence address: University of Granada, Buensuceso Street 11, Postcode 18002, Granada, Spain. Tel.: (34) 958 240517. E-mails: 3jmurena@ugr.es4; 3pfaber@ugr.es4. Cognitive Linguistics 211 (2010), 123149 DOI 10.1515/COGL.2010.004 09365907/10/00210123 6 Walter de Gruyter


J. M. Urena and P. Faber

and Lako and Turner (1989) distinguish between conceptual-structural/ conventional metaphors and image metaphors. However, Gradys (1997, 1999) Primary Metaphor Theory distinguishes between correlation metaphors and resemblance metaphors. These classes of metaphor arise by virtue of our embodied conceptualisation system. In this article, we use the term resemblance metaphor to refer to image metaphors and behaviour-based metaphors, and non-resemblance metaphor to refer to conceptual/conventional metaphor and correlation metaphor. We prefer the term non-resemblance metaphor for three reasons. Firstly, it encompasses any type of metaphor that does not arise from resemblance. Secondly, the term conceptual metaphor is not felicitous because resemblance metaphors are also conceptual, as underlined in other studies (Kovecses 2002; Alexiev 2005). Thirdly, the term conventional metaphor is not a good choice because resemblance metaphors are also conventional. According to Gradys (1999) characterisation of resemblance metaphors, image metaphors are associated with motionless visual images, whose motivation for metaphorical transfer is based on physical properties (e.g., shape and colour). In contrast, there are other metaphors that result from behavioural comparison, and therefore, are typically linked to motion and dynamicity. This article challenges this classication. Our study of resemblance metaphor in the eld of marine biology indicates that image metaphors and behaviour-based metaphors are closely linked. In fact, what truly differentiates these metaphors is the static or dynamic nature of their underlying images1. Rather than belonging to two dierent categories, they should be regarded as belonging to a graded category in which members dier in terms of the dynamicity of their images. As shall be seen, imagery is also important in non-resemblance metaphors. This means that images and their analysis should go far beyond mere physical or behavioural resemblance. 2. Dening imagery

Imagery has two related senses. First of all, it refers to quasi-perceptual experience, which signicantly resembles perceptual experience, but oc-


In this regard, this research work is on a par with other studies. For instance, Caballero (2006) explores a host of image metaphors that are dynamic, and Pena (2003) makes a distinction between situational and non-situational metaphors that is based on the feature of dynamicity.

Reviewing imagery


curs in the absence of the appropriate perceptual stimuli (Thomas 1999: 208). This denition includes image schemas and mental images, both of which are key ideas in Cognitive Linguistics (cf. Johnson 1987; Lako and Johnson 1999). An image schema can be regarded as an instance of imagery simpliciter or an especially unsaturated form of imagery, produced by simulating only the very earliest and most generally applicable stages of the process of a perceptual exploration (Thomas 2009, A note on schema and image schema, para. 9). Apart from being an unsaturated form of imagery, they are non-intentional because they do not participate in the conscious act of perceiving. In other words, image schemas are emergent properties of unreective bodily experience (Gibbs and Colston 2006: 247). In contrast, a mental image is a more substantiated kind of mental representation. It cross-cuts any sensory mode, and embodies our perceptual and imagistic awareness2. Mental images are intentional, insofar as they involve a conscious mental act of perceiving. In other words, they are the result of more eortful cognitive processes (Gibbs and Colston 2006: 247). Mental images also have content specicity, the complexity of which is constrained by linguistic and environmental situatedness. Image schemas may also be complex, but in the sense that they can combine to give structure to conceptual domains (cf. Cienki 1997: 9; Kimmel 2005). This structure entails conceptual relationships, and accordingly, it has been shown that some image schemas are subsidiary or subordinate to others (Pena 1999). The second sense of imagery is related to a well-entrenched view in cognitive psychology that mental images are a key factor in creative thought (cf. Finke et al. 1992; Weisberg 1986). Consequently, imagery not only refers to true imagination, but also to the production of mental images that arise from our capacity to separate, shue, distort and recombine simpler mental images in the rst sense. This ability has its cognitive uses

2. The notion of mental image is admittedly rather vague. Because of this, a mental image need not refer to a mental picture, but can also refer to sensory images or image simulations in dierent sensory modes. As Pylyshyn (2003: 113) aptly points out, this is due to the fact that neither language nor pictures are sucient to represent the content of thought and that most thought is not available to conscious inspection [ . . . ] If there is something special about the format in which we think when we have the experience of seeing with the minds eye, nobody has satisfactorily articulated what it is.


J. M. Urena and P. Faber

(Finke et al. 1989; Finke et al. 1992), including the generation of metaphoric thought. In our view, both senses of imagery are essential to account for mental images in metaphoric thought.


Image metaphors and behaviour-based metaphors

Grady (1999), Lako (1993), and Lako and Turner (1989) agree that the core feature of image metaphors is the comparison between the images of two entities. Lako (1993: 230) writes that because two images are being compared, these metaphors are called image metaphors:
Metaphoric image-mappings work in just the same way as all other metaphoric mappings: by mapping the structure of one domain onto the structure of another. But here, the domains are conventional mental images.

The prototypical conception of an image metaphor is a metaphor based on resemblance in shape and/or colour. In this way, clear instances of marine biology image metaphors are seahorse (Hippocampus), which refers to a sh with a horse-like head (see picture in Table 1 in the Appendix), and milksh (Chanos chanos) because of the white underside of this sh (see picture in Table 1 in the Appendix). These metaphors are based on visual perception, which is the dominant component of our embodied conceptualisation system (Watt 1991). For this reason, this type of metaphor has the highest degree of iconicity or mental imagery shared by the source and target concepts. These metaphors clearly dier from behaviour-based metaphors, such as hermitcrab (Paguroidea) in which the crab acts like a hermit instead of looking like one. According to Grady (1999), behaviour-based metaphors cannot be called image metaphors because they are based on behavioural rather than physical resemblance. By way of example, Grady (1999: 89) mentions the well-known metaphor Achilles is a lion. Since Achilles courageous actions resemble the aggressive behaviour of lions without any claim about his physical appearance, this metaphor cannot be considered an image metaphor. Strictly speaking, according to Gradys classication, marine biology terms such as sea nettle, archersh or triggersh are not image metaphors either, since they are based on behavioural or functional resemblance. Nevertheless, we argue that these metaphors also evoke mental images, and that mental images are not exclusively associated with metaphors based on physical comparison.

Reviewing imagery 3.1. Images in behaviour-based metaphors


There is a growing body of research in cognitive psychology showing that mental representations of perceptual experience are central to cognition (cf. Damasio 1994; Finke 1989; Paivio 1971, 1986; Thomas 19993). In this section we show that behaviour-based metaphors are grounded in mental images that can be either dynamic or static. 3.1.1. Dynamic images in behaviour-based metaphors. Behaviour-based metaphors, such as Achilles is a lion, emerge from the visual experience of a motor action, which yields a set of images that are eshed out by spatial-dynamic actions. In this metaphor, we evoke images of Achilles bravely confronting his enemies and a lion ercely ghting other animals for survival (also in Ruiz de Mendoza and Pena 2008). The nature of these images is constrained by the image-schematic topology of the target domain, which cannot be violated by the cognitive topology of the source domain, while still remaining consistent with it. This is in consonance with Lako s Invariance Principle (Lako 1990, 1993). In spite of this constraint, the images retrieved by this type of metaphor are disperse since each individual may re-create these actions in dierent mental scenarios. As Deane (2005: 247) points out, the same spatial relation may receive distinct representations in multiple representational modalities. Although most research on imagery in contemporary cognitive psychology focuses on visual perception (and, to a much lesser extent, on audition), there is growing evidence that kinaesthetic, somaesthetic and haptic perception is also pivotal to mental image formation (cf. Gibbs 2006; Gibbs et al. 1994; Gibbs and Colston 2006 [1995]; Popova 2005). This means that mental images need not necessarily be visual in nature, and that visual imagery and kinaesthetic imagery share a common representational, and possibly neuropsychological substrate (Gibbs 2006: 124). According to Paivios (1971, 1986) dual coding approach, cognitive tasks are mediated not only by linguistic processes, but also by a nonverbal imagery model of thought as well. What Paivio calls the image system in our brains refers to both non-verbal objects and events, and arises not only from visual stimuli, but also from auditory, kinesthetic, and other

3. Thomas (1999) dwells on the three major theories of imagery in Conceptual Science, namely Picture Theory, Description or Propositional Theory, and Perceptual Activity Theory. He aligns himself with Perceptual Activity Theory, according to which perceptual learning is not viewed as a matter of storing descriptions (or pictures) of perceived scenes or objects, but as the continual updating and rening of procedures or schemata (Thomas 1999: 218).


J. M. Urena and P. Faber

sensory components of non-verbal information. Accordingly, behaviourbased and function-based metaphors can also be regarded as image metaphors because they are closely linked to conventional mental images representing events, which are not necessarily based on visual stimuli. Moreover, since behaviour and function mostly involve (loco)motion on account of a correlation or cause-eect event, most behaviour-based images (i.e., images that feature the behaviour of a living being) and function-based images (i.e., images that feature the functioning of an instrument, device or machine) are unquestionably dynamic. Sea wasp. In the eld of marine biology, many specialised concepts have basic-level category denominations. This guarantees richly contoured and easily retrievable mental images, since the basic level is the level of rich mental images and rich knowledge structure (Lako 1993: 212). For example, the metaphor sea wasp, which is an alternative scientic name for jellysh Chironex eckeri (see picture in Table 1 in the Appendix), evokes an easily retrievable image that primes kinaesthetic perception. It also gives priority to the more subjective sensory image of actually participating in an event, rather than to the objective and visual pattern of observing it. In this case, the perceptual experience foregrounded is touch, which is a somaesthetic and kinaesthetic sense, and like vision, also a spatial sense (Popova 2005: 402). This metaphor evokes the dynamic event image of our touching a wasp, its stinging us, and our subsequent experience of pain. This image is mapped or superimposed onto the image of a jellysh injecting its stinging capsules or nematocysts under our skin, which causes the pain. This metaphor has a metonymic basis. The close relationship and interaction between metaphor and metonymy has been underlined in recent research (cf. Barcelona 2003; Radden 2002). More precisely, the sea wasp metaphor is based on two conceptual metonymies operating on the two domains or categories connected by the metaphor. In the metonymies, the source is the stinging capacity, which is a shared attribute of the targets wasp and jellyfish. In other words, both wasps and this type of sh have to be metonymically understood from their salient property stinging capacity as metonymic source, which creates the abstract similarity that makes the metaphorical connection between the source (wasp) and the target (jellyfish). Archersh. Another example of a behaviour-based metaphor relying on dynamic images is archersh (Toxotidae). The behaviour of this sh is compared to that of an archer, which includes the function of an archers bow, which shoots arrows at a target. The reason for this

Reviewing imagery


comparison is that archersh have the ability to spit water droplets at aerial insects (either on the wing or resting on surfaces above the water), and thus knock them onto the water to be eaten (see picture in Table 1 in the Appendix). Thus, the dynamic image of an archer shooting an arrow at his target is superimposed onto the image of an archersh spitting water at an insect. This metaphor also has a metonymic basis. The source domain of the metonymies is shooting capacity as instantiated by: (i) the archers use of a bow and arrow; (ii) the archershs projection of water droplets to hit insects. The source domain of the metonymies stands for the targets, archer and archerfish, and is in turn responsible for the abstract similarity that makes the metaphorical connection between the source (archer) and the target (archerfish). All of these metaphors can also be approached from the perspective of Conceptual Blending Theory (Fauconnier and Turner 1998, 2002). They are clear instances of formal blending, more specically, of compounding. For example, archersh involves two input spaces relating to archer and sh, plus the conventional array of meanings linked to these lexical items. However, the projection to the blended space is selective, including only the subset of semantic features associated with the concepts of both archer and sh, along with their forms (i.e., word projection). Thus, both conceptual structure and linguistic structure are projected onto the blend, giving rise to a new emergent structure. Figure 1 illustrates this scenario. This structure is novel both from a linguistic point of view (the creation of a new word) and from a semantic point of view (the creation of a new meaning). 3.1.2. Static images in behaviour-based metaphors. Although behaviour most frequently implies dynamic mental images, curiously enough, we have found behaviour-based metaphors in the eld of marine biology that are based on static images. For instance, the metaphor hawksh (Cirrhitidae) refers to a sh that behaves like a hawk because it rests atop the highest point on the coral reefs, waiting for suitable prey to appear (see Table 1 in the Appendix). The sh then dives down to capture its prey. The initial image of a motionless hawksh awaiting its prey on a high vantage point maps onto that of a motionless hawk on a tree branch or cli, waiting to capture its prey. Still another example is the metaphor garden eel (Heterocongridae). Garden eels receive this name because they live in colonies, keeping the main portion of their bodies buried in the sandy sea bottom while the rest remains upright in the open sea (see Table 2 in the Appendix). This behaviour retrieves a motionless image which resembles that of slim


J. M. Urena and P. Faber

Figure 1. Blended space of archersh

plants in a garden. The garden eel metaphor also entails a physical aspect motivation: the mass-eect shape of the eels allows for the comparison between these animals and a garden. On this basis, we argue that this is another metonymy-based metaphor. The source of the metonymies, i.e., the state of standing still, maps onto the targets, i.e., plants in the garden and eel, and prompts the metaphorical connection between them. 3.2. Dynamic image metaphors

In marine biology, most of the examples refer to either a behavioural/ functional model or a physical-aspect model. For instance, the metaphors sea nettle and sea wasp are based on behaviour. Archersh integrates behavioural and functional motivations, whereas triggersh arises from resemblance in function. Independently, horseshoe crab is a shape-induced metaphor, and sea lettuce emerges as a result of comparison in shape and colour (see Table 1 in the Appendix).

Reviewing imagery


Generally speaking, we tend to think of shape and colour as more static than dynamic attributes. However, it is evident that an entity can change its shape as well as its colour. Accordingly, there are also dynamic metaphors based on physical comparison. This fact supports the claim that people nd it easier to make sense of [ . . . ] moving objects over those that are stationary (Gibbs and Colston 2006: 252). Concerning shape, Deane (2005: 249250) arms that there are multiple representations of shape: one representation depicts static forms; the other depicts dynamic form. Lako (1993: 229) provides the following example when describing the characteristics of image metaphors: the image of the slow, sinuous walk of an Indian woman is mapped onto the image of the slow, sinuous, shimmering ow of a river. Though involving dynamicity, this example features an image metaphor because it is moving shapes or lines that are compared. However, dynamism also entails behavioural or functional patterns. These are processed in our brains, and create interrelated experiencebased concepts that become meaningful because of these regular patterns. This evidently leads to behaviour-based or function-based metaphors. Thereby, the slow and sinuous walk of the Indian woman is part of the way she walks, and thus, of her behaviour. Likewise, the slow and sinuous ow of a river is also part of its behaviour. Thus, this is a resemblance metaphor which integrates physical and behavioural motivations. There are also resemblance metaphors in marine biology that combine both behaviour and physical appearance. Such is the case of the anglersh (Lophius). This sh behaves like, and thus, resembles an angler for two reasons: (i) the shape of the foremost spine of its dorsal n looks like a shing rod with its shing line and eshy bait at its tip (see picture in Table 2 in the Appendix); (ii) this spinal shing rod is used as a lure for attracting prey which stray close enough for the anglersh to swallow. Since catching a prey is an action or event, this can be regarded as a dynamic image. Still another example is the metaphor boxer crab (Lybia tessellata). This crab holds an anemone in each pincer, and uses these anemones for protection (usually against octopuses) in the same way as a boxer uses his sts against his opponent (see picture in Table 2 in the Appendix). These little round-shaped anemones resemble boxing gloves, while the action of attacking predators with the anemones is a type of behaviour that resembles that of a boxer. Regarding colour, an example of a dynamic resemblance metaphor is chameleon sh (Badis badis). This is a freshwater sh that changes its skin colour when hungry, threatened or protecting its eggs, ospring, or territory (see pictures in Table 2 in the Appendix). This change of skin


J. M. Urena and P. Faber

colour occurs within a single static locus (i.e., locomotion is not involved). Yet, this type of eect creates mental video-clips of sequentially unfolding images, which naturally implies change or dynamic structure. This is the reason why we recruit the superimposed dynamic images of a real chameleon and of this sh, altering their skin colour. This physical ability is part of their behaviour. In summary, image metaphors and behaviour-based metaphors are not clearly dierentiated categories, since there is a group of metaphors that possess characteristics of both, and thus reside in a transition zone between the two. 3.3. Fictive dynamicity in resemblance metaphors

Our tendency to think in terms of dynamic patterns has been documented (cf. Talmy 1999 [1996]). Such tendency hinges upon representations that are motionless in nature. These representations emerge from what Talmy (1999 [1996]: 245) calls ception or ctive motion, which involves sensory stimulation, mental imagery, and ongoingly experienced thought and aect. Metaphor is found in ctive motion constructions dealing with spatial description (Talmy 1996). Regarding specialised language, Caballero (2006) identies instances in architectural discourse where metaphor plays a role in ctive motion. Example (1) given by Caballero (2006: 180) includes motion verbs codifying actual static scenes, which are conceptualised as non-veridical dynamic images through metaphorisation: (1) Based on a boomerang shaped plan, the new building steps down from a prow at its south end to embrace a new public space.

In the eld of marine biology, we have also found metaphorically extended motion verbs that evoke visual mental images involving ctive dynamic structure, as shown in the following examples: (2) (3) The Rst Reef, the worlds largest known deep water coral reef, forms a structure that fades away to depths between 300 and 400 m. The seaward edge of a reef is fairly steep and slopes down to deeper water. Since the water is generally clearer, corals may grow to the depths of 50 m depending on light available. In tile Pulmonate tile rudimentary velum, v, is marked by a line of granular ciliated cells, which [ . . . ] bends up towards the dorsal surface, in such a way as to almost encircle the tentacles.


These metaphors are clearly imagistic in nature, and form a part of the experts visual thinking (Caballero 2006: 3). What makes this type of

Reviewing imagery


metaphors interesting is their complex nature. They can be regarded as instantiations of the more general metaphor form is motion (Lako and Turner 1989: 142144). At the same time, these metaphors emerge because the form that they evoke matches the actual shape of the entities, and is based on how they are visually scanned. In other words, despite the fact that they are often classied as non-resemblance metaphors, resemblance is involved here, but of a more sophisticated kind. We can thus conclude that while the two types of resemblance metaphor cannot be regarded as clear-cut categories because dynamic structure and static structure permeate both categories, in some resemblance metaphors it is the boundaries between static structure and dynamic structure that are fuzzy. However, the fuzzy boundaries between both schemas in these resemblance metaphors answer psychological strategies rather than reect the actual state of aairs. In short, the conceptualisation of factive statis or stationariness through this kind of metaphor is biased because it results in images involving ctive change.


Non-resemblance metaphors

Resemblance metaphors emerge from the superimposition of easily retrievable mental images. Yet, non-resemblance metaphors also involve the retrieval of mental images. Precisely, the great bulk of research on gurative mental imagery is currently on non-resemblance metaphors. Consequently, both types of metaphor are more closely linked than previously assumed. The pervasiveness of mental imagery is due to the logic of our embodied conceptual system, which licenses the creation of any type of metaphor on the basis of mental images. Therefore, strictly speaking, mental imagery constitutes the grounding of metaphoric thought. 4.1. Mental images in non-resemblance metaphors

Lako (1993: 229) writes that the rationale of conceptual metaphors, namely understanding abstract concepts through concepts directly grounded in bodily experience, involves mental imagery, which is the mental realisation of such experience:
Abstract reasoning is a special case of imaged-based reasoning. Image-based reasoning is fundamental and abstract reasoning is image-based reasoning under metaphorical projections to abstract domains.

As a general rule, words can designate portions of conventional mental images (Lako and Johnson 1999: 69). Recent research provides evidence


J. M. Urena and P. Faber

that language makes much greater use of the brains mental imagery than previously thought (Rohrer 2005: 166). In keeping with the two-domainof-experience mapping system proposed by Conceptual Metaphor Theory, when both domains are active, imagery associated with sourcedomain entities can be activated, and thereby associated with the targetdomain entities neurally connected to them (Lako and Johnson 1999: 56). As previously discussed, kinaesthetic perception involves motor activity or bodily (loco)motion, which occurs in space. In fact, mental imagery is prominent in the form of spatial-dynamic images, especially when it comes to real or imagined body action. As pointed out by Rohrer (2005: 169), mental imagery can also be kinaesthetic, as in the felt sense of ones own body image. Mediation of the lived body action for mental image formation is called embodied simulation (Gallese 2005). Gibbs and Perlman (2006: 223) arm that processing metaphoric meaning is not just a purely cognitive act, but involves some imaginative understanding of the bodys role in structuring abstract concepts. Examples of embodied simulation can be found in expressions such as chewing on the idea and grasping an idea, which arise from the conceptual metaphor ideas are objects. Gibbs et al. (2006) demonstrated that people imaginatively engage in the act of chewing or grasping something to better understand these metaphorical phrases. Furthermore, it has been shown that the literal re-enactment of gurative verbal cues activates the primary motor and somatosensory cortices in our brains (Rohrer 2005). This underscores the signicance of embodiment or sensorimotor experience for metaphorical concept formation. Since Conceptual Metaphor Theory posits that abstract concepts are ultimately grounded in perceptual or bodily grounded experience (Ko vecses 2005; Lako 1990; Lako and Johnson 1980), mental imagery is thus an integral part of all metaphors. As Caballero (2003a: 152) stresses in the eld of architecture, if a distinction is to be made between images and concepts, such a distinction should not lie in the image component, since all the information organised and processed in our minds is essentially imagistic. In a like way, it can also be argued that all the information organised and processed in our minds, including images, is also conceptual. Mental images are likewise present in an extensive class of metaphorical or imageable idioms (Lako 1987; Lako and Johnson 1999). An imageable idiom comes with a conventional rich mental image and knowledge about that image (Lako and Johnson 1999: 68). According to Lako and Johnson (1999), a signicant portion of the array of linguistic expressions stemming from the conceptual metaphor love is a journey

Reviewing imagery


consists of idioms. They give the expression spinning ones wheels as an example. In marine biology we have also found non-resemblance metaphors based on dynamic mental images. This is the case of recruitment, which refers to the incorporation of new members of one species to the stock of the already existing individuals, particularly those living in communities. This includes shoals of sh and planktonic aggregates. This metaphor can be subsumed by the more general metaphor marine communities are military structures, which gives rise to metaphorical terms, such as intrusion, cohort, sentinel organism, invasive exotic species, evolutionary arms race and line of defence. Recruitment activates the generic dynamic mental image of a group of organisms that increases as new organisms join them. The specic details of this image largely depend on the context in which the metaphor is embedded. For instance, the recruitment of individuals of the species Engraulis encrasicolus, which is a type of anchovy, evokes a dierent mental image from that evoked by the recruitment of individuals of the species Labidocera scotti, a kind of marine planktonic copepod (i.e., a small crustacean). A second factor constraining and modelling the mental image activated is encyclopaedic meaning. According to Lako and Johnson (1999: 69), a metaphorical word is not just a linguistic expression of a metaphorical mapping, but the linguistic expression of an image plus knowledge about the image plus one or more metaphorical mappings. Thus, the mental image of the recruitment of anchovies evoked by an expert in marine biology is certainly richer than that evoked by a layman. However, it is also true that cognitive patterns give priority to the objective aspects of images rather than to their subjective implications (Dewell 2005: 386). Moreover, culture, as a specic type of contextual factor, also has a decisive role in forming conventional rich images, which appear to be pretty much the same from person to person in the same culture (Lako 1987: 450). However, when cultures dier, so do images. Consequently, when a European biologist builds a mental image of the recruitment of anchovies, in all likelihood the image brought to mind is that of an anchovy of the species Engraulis encrasicolus, most frequently found in the Mediterranean Sea. The physical features of this species of anchovy are dierent from those of the species Encrasicholina heterolobus, which inhabits the Indo-Pacic region. A biologist from Australia would probably activate an image of this species of anchovy, when he or she is thinking about recruitment. It can thus be concluded that mental images permeate both resemblance metaphors and non-resemblance metaphors.

136 4.2.

J. M. Urena and P. Faber Similarities between non-resemblance metaphors and resemblance metaphors

The previous section showed that mental images, traditionally associated with resemblance metaphors, have an important role in non-resemblance metaphors as well. In this section we argue that resemblance metaphors also have features that are traditionally considered to pertain exclusively to non-resemblance metaphors. Conceptual Metaphor Theory has primarily focused on conceptual/ conventional metaphors, which emerge from multiple mappings between two content-rich domains of experience. In other words, they have rich knowledge and rich inferential structure (Lako and Turner 1989: 91). Although this work is of undeniable interest, it has also meant that image metaphor has been more or less left out in the cold, and has been regarded as a kind of second-class metaphor. The main reason for this is that image metaphor is regarded by Lako and co-workers as a eeting, ad hoc kind of metaphor with an impoverished inner structure (Lako 1987, 1993; Lako and Turner 1989). Nevertheless, in recent years there has been a renewed interest in resemblance metaphor. Corpus-based research both in general language (Deignan 2007) and specialised discourse (Caballero 2003a, b, 2006 in architecture) shows that resemblance metaphors are well-established, conventional metaphors that arise from enduring and productive patterns of gurative thought. For example, our research in marine biology shows that certain wellentrenched resemblance metaphors can be brought together under productive, encompassing metaphors. Accordingly, terms like elephant seal (Mirounga), seahorse (Hippocampus), sea lion (Otariidae), hawksh (Cirrhitidae), spider crab (Maiidae), boarsh (Capros aper) and sand tiger shark (Carcharias taurus) can be subsumed by the general metaphor sea animals are land animals. Another such metaphor is marine organisms are workers, which stems from the multiple-correspondence process involving metaphorical terms, such as surgeonsh (Acanthuridae), pilot sh (Naucrates ductor), anglersh (Lophius), ddler crab (Uca), harvestsh (Peprilus alepidotus), by-the wind sailor (Velella spirans), nurse shark (Ginglymostoma cirratum), innkeeper worm (Urechis), and rock cook (Centrolabrus exoletus)4. Still another aspect that places resemblance metaphors on the same level as non-resemblance metaphors is that resemblance metaphors
4. As shown in Section 3.3, the resemblance metaphors involving ctive dynamicity can also be subsumed by a more general metaphor (i.e., form is motion).

Reviewing imagery


meet the two generalisation principles proposed by Lako (1993: 209) for non-resemblance metaphors, namely, the polysemy generalisation and the inferential generalisation. According to the polysemy generalisation, certain linguistic expressions of the source domain acquire related senses. For example, terms like thresher, and sponge have two or more senses, one of which refers to marine organisms. The main sense of thresher refers to a man who threshes the grain by beating it with a ail (a long, thin tool). In marine biology, a thresher is a shark of the genus Alopias. The metaphorical motivation is resemblance in both shape and behaviour. Regarding shape, the sharks abnormally long, thin, caudal n looks like a ail, and insofar as behaviour is concerned, the shark uses its aillike n to strike its preys and render them dazed. In the case of sponge, the central meaning of the concept is the marine biology sense. Sponge refers to a marine invertebrate animal of the phylum Porifera, characteristically having a porous skeleton composed of brous material or siliceous or calcareous spicules. The metaphoric sense of sponge, namely, porous plastics, rubber, cellulose, or other material chiey used for washing, bathing, and cleaning, arises on account of physical resemblance because the porous structure of this object looks like the skeleton of the marine organism. By virtue of the inferential generalisation, each mapping denes an open-ended class of potential correspondences across inference patterns (Lako 1993: 10). This is true for metaphors in the domain of marine biology because new species are continually being discovered. Such species are usually given metaphorical names that t into existing metaphorical systems within the domain, and thus increase the number of cross-domain correspondences and mappings that characterise a given resemblance metaphor. This capacity to infer, which emerges from the topological or gestaltic structure of conceptual (as opposed to linguistic) metaphors (Lako and Johnson 1980), follows a robust domain logic to create terms for marine organisms, as well as other terms. Accordingly, surgeonsh have scalpels; anglersh use their baits; harvestsh harvest food for survival; pilotsh usually travel together with sharks (see picture in Table 1 in the Appendix); and a burrow of innkeeper worms is occupied by several commensals. These examples clearly reect the role played by visually biased gurative language as an ecient instrument to organise thought through prolic inferential processes and evaluation (Caballero 2006: 3). More concretely, these examples show the importance of visual thinking in the creation of conceptually rich domain knowledge, which is enhanced by the high number of systematic crossdomain correspondences.


J. M. Urena and P. Faber

In this respect, the dierence between both classes of metaphors lies in the type and nature of mappings involved since both have multiple mappings as well as polysemic and inferential conditions. 4.3. Mental images in primary metaphors and correlation metaphors

Grady (1999: 87) clearly highlights the role of imagery in primary metaphors:
Quantity, desire [ . . . ] may take place at the level of cognition whose operation is not directly accessible to consciousness. In order to manipulate them at the conscious level it may be necessary to tie these elements of mental experience to specic sensory images.

Grady (1997: 100) arms that the direct bodily basis of primary source concepts is processed in our brains in the form of images (image content), which are paired with target concepts (response content) to build primary scenes. Although he states that primary scenes are not necessarily eshed out by rich content, at the same time, he holds that primary scenes are local structures, motivated by particular moments in our experience. He thus argues for the participation of down-to-earth conceptual structure in the source domain that cannot be as abstract as image schemas5. This conceptual structure would therefore have a greater level of specicity than image schemas, and illustrate the need of our cognitive system to resort to more or less specic images when constructing (metaphoric) meaning. Gradys view is likewise endorsed by Lima (2006: 115):
For instance, all cases of containers can be included in the image schema of a container, but each case may involve many primary scenes, such as (a) going into a room or (b) taking something out of a box, which can generate distinct metaphors. Even if we can have a schematic mental representation that is abstract enough to include all cases, the experiences that generate the metaphors do not seem to be the same in all of them. For example, in scene (a) going into a room, we experience going into spaces with certain characteristics and certain limits; in (b) taking something out of a box, we experience interacting with a container and its contents.

It could be argued that high-level primary metaphors such as organisation is physical structure are not based on images of a specic type.


Far from criticizing image schemas, we are underlining the important role of images in metaphoric thought that is not resemblance-based.

Reviewing imagery


However, we should not forget that these are metaphoric formulations, namely, abstractions or generalisations of concrete linguistic and environmental situations. As research shows, language processing draws on location-specic perceptual images of entities and their attributes (Bergen et al. 2007: 734). Therefore, we can only have mental images of more or less specic types of physical structure (e.g., solids, liquids, etc.) if the abstraction is substantiated and situated. Evidently, the more information that we have about an entity, the richer its mental image will be. Moreover, mental images are generated by assembling the parts of the image one part at a time (Gibbs and Colston 2006: 247). In other words, we can speak of a procedural representation, in which the mental image of an entity is not built all at once, but rather sequentially by scanning its parts. It is our claim that mental images rather than solely image schemas are often the grounding for both resemblance metaphor and nonresemblance metaphors. According to Grady (1997), primary metaphors involve experiential correlation, which consists of establishing a strong conceptual link between two distinct events that iteratively co-occur. This phenomenon usually gives rise to correlation metaphors because after repeated cooccurences in our experience of the world, we come to conceive one event in terms of another. This evidently makes correlation metaphors dierent from resemblance metaphors. These basic correlation-induced physical experiences are generally recurring events throughout our life. For example, the experience of ones body moving through space generates the metaphor actions are selfpropelled motions and expressions, such as I am moving right along on the project (Lako and Johnson 1999: 52). This expression is linked to the dynamic image of ourselves moving through space. Since experiential correlation often implies a cause-eect event, it would seem that the retrieval of dynamic mental images for primary metaphors is at the center of this process. However, this is not always the case, because the range of sensorimotor domains activated in primary metaphors also includes static experiences, involving domains such as temperature (affection is warmth), size (important is big), physical conguration (uninteresting is flat) and location (states are locations)6.

6. See Grady (1997) for an account of other primary metaphors, or Lako and Johnson (1999: 5054), who also provide an inventory.

140 4.4.

J. M. Urena and P. Faber Similarities and dierences between correlation metaphors and resemblance metaphors

In the eld of marine biology, we have found resemblance metaphors based on cause-eect relationships, rather than physical similarity. The reliance on the cause-eect schema brings this type of resemblance metaphors close to correlation metaphors. In resemblance metaphors of this kind, the two entities involved share some facet of their behaviour or function (in the case of the source concept). In the case of the sea nettle, this property is a defence mechanism shared by both the plant and the marine organism. The experience involves a cause-eect experience of touching a nettle/sea nettle (cause/stimulus) and the subsequent perception of an itchy sensation (eect/response). The example of the sea nettle is a cause-eect metaphor that highlights active (kinaesthetic) sensorial experience, i.e., direct interaction. However, there are also cases in which cause-eect events are backgrounded because passive (visual or auditory) experience is primed. For example, the term ghost crab, which is based on a resemblance metaphor, only implies the visual experience of a motor action. The ghost crab (Ocypode) receives its name because of its ability to disappear from sight almost instantly by sprinting and scuttling at speeds up to 10 miles per hour while making sharp directional changes. This crustacean behaves this way when it establishes visual contact with a potential predator (correlation or cause-eect event). Consequently, marine biology resemblance metaphors could be classied in terms of the sensorimotor experiences that the terminological designations are based upon. The specication of the motivations for metaphorical transfer is essential for a clear classication of resemblance metaphors. Grady (1999: 98) acknowledges that an in-depth analysis is needed in order to further rene the dierent types of metaphor by carefully considering the motivations for these metaphors. This emphasises the signicance of our embodied conceptualisation system, which licenses the formation of any type of metaphor on the basis of mental images, and involves all manner of sensorimotor experiences. Cause-eect structure can also be at work in resemblance metaphors where behavioural comparison operates together with physical comparison. This is the case of the cookie-cutter shark (Isistius brasiliensis). This shark remains motionless at the sea bottom while its body emits a vivid, greenish phosphoresecent gleam, except for a black band around its throat. The prey of this shark is usually large fast-swimming sh. They are lured by what appears to be the silhouette of a small sh, which is actually the sharks non-luminiscent black collar. The shark behaves like

Reviewing imagery


a cookie-cutter in that once it has locked onto its lured prey (cause), the shark extracts cookie-shaped plugs of esh (physical resemblance) from the victim (eect). It should be noted that correlation in resemblance metaphors, while also entailing a cause-eect schema7, diers from correlation in primary metaphors in that in primary metaphors there is correlation between the source concept and the target concept, i.e., the source has a bearing on the target. Such source-target correlation does not exist in resemblance metaphor. For instance, in the sea nettle metaphor, the correlation between the hand touching the sea nettle (marine organism) or nettle (plant), and the resulting painful sensation is limited to either the source or the target. In other words, there is no co-occurrence of the events of touching the marine and the non-marine nettle, but only a resemblance between the overall cause-eect structure of source and target. What occurs in resemblance metaphors of this kind is that the type of correlation operating in the source concept is mapped onto the target concept. This mapping is sanctioned by the nature of the experience, which is the motivation for the correlation in the target concept. Accordingly, the stinging plant is an appropriate source for the sea nettle because the former does not violate the conceptual topology of the latter (Invariance Principle), and both share their overall cause-eect structure. In this way, the source helps to activate relevant aspects of the target, and even allows the perceiver to infer other properties about it. Still another dierence between resemblance metaphors grounded in a cause-eect schema and correlation metaphors is the metonymic nature of the latter. The prototypical example which illustrates this claim is the more is up metaphor. Radden (2002: 414) writes:
In order to correlate two variables, they have to be conceptually contiguous. The correlation of quantity and verticality provides a perfect example of conceptual contiguity in that both variables originate from the same experiential basis.

Radden argues that the metonymic grounding of correlation metaphors is based on a continuum ranging from literalness via metonymy to metaphor. This process ties in with the notions of conation and deconation,

7. Correlation is not exclusively a matter of cause-eect links. In fact, Radden (2002: 414) argues that there are positive correlations, which tend to evoke a causal interpretation, and negative correlations, which do not invite a causal interpretation. However, in dealing with metaphor, we bind correlation to cause-eect structure because cause-eect correlation (i.e., positive correlation) is the only type of correlation that pertains to metaphor (Radden 2002: 414).


J. M. Urena and P. Faber

as well as the developmental model of primary scenes, which gives rise to primary metaphors (Grady 1997). The metaphor more is up comes into its own in four stages. The rst stage involves up being literally conceptualised. At the second stage, the variable or dimension quantity is conceptualised through partial metonymy (up for up). At the same time, quantity is linked to the dimension verticality by means of the experiential basis of conation (i.e., up more). Conceptual conation takes place in primary scenes, such as seeing the level of uid in a container go up when more uid is poured into it. As Radden (2002: 10) points out, the two manifestations of this highly frequent scene, rise of a level and rise of quantity, occur simultaneously and are intimately correlated. The third stage is deconation, whereby up and more become a full metonymy (i.e., up for more). At the nal stage of the continuum, this full metonymy becomes the primary metaphor more is up. Thus, it can be stated that the immediate basis of primary metaphors is metonymic in nature. Radden (2002: 414) further states that the causal relation between quantity and verticality strengthens the metonymic basis of both these variables and licenses the reversibility principle of metonymic relationships, according to which the ow of causation may be seen in either direction: something is more because its level is higher or the level is higher because its quantity is more . In contrast, the pairs of cause-eect correlations that give rise to resemblance metaphors such as sea nettle occur independently (one pair in the source domain and another pair in the target domain). Therefore, they are not contiguous, which is the essential feature of metonymic mappings.



Within the category of resemblance metaphors there are certain very clear examples of image metaphors and behaviour-based metaphors. Prototypical image metaphors are based on visual stimuli in the form of static images. In contrast, prototypical behaviour-based metaphors are based on motion and dynamicity. However, other resemblance metaphors do not clearly belong to one category or the other. In our opinion, resemblance metaphor is a graded category, in which members are better or worse exemplars, based on the static or dynamic nature of the mental images underlying them. In other words, image metaphors and behaviour-based metaphors can be dierentiated in terms of their degree of dynamicity. The existence of resemblance metaphors which combine behaviour and shape or colour seems to point to the need for a new, more rened classication.

Reviewing imagery


Evidence from specialised elds, such architecture and marine biology, shows that there is a more complex type of resemblance metaphors that are grounded in ctive dynamicity. Such dynamicity answers psychological strategies rather than reects the actual state of aairs. In this case, it is the boundaries between static structure and dynamic structure that are fuzzy. Recent corpus-based research shows that resemblance metaphors and non-resemblance metaphors share features, which up until now have been attributed exclusively to non-resemblance metaphor. The reason for this is the nature of mental imagery, which is an integral part of every stage of metaphor formation. Received 10 December 2008 Revision received 9 September 2009 University of Granada

Appendix Tables 1 and 2 show the marine biology resemblance metaphors referred to in this article. The metaphors in Table 1 represent separate, clear-cut categories, and are either image metaphors or behaviour/function-based metaphors. In contrast, Table 2 contains resemblance metaphors that are based both on behaviour/function and physical appearance. Pictures are provided to help understand the motivation for metaphorical transfer.


J. M. Urena and P. Faber

Table 1. Resemblance metaphors that are either image metaphors or behaviour/functionbased metaphors Metaphorical term Type of metaphorical motivation Shape Picture

Image metaphors (motionless mental images)




Sea lettuce

Shape colour


8. Picture provided at Last access 21 May 2009. 9. Picture provided by Bryan Harry at http:/ / sh_pops/chanidae/milksh01.htm. Last access 21 May 2009. 10. Picture provided by Guy Werner at http:/ / htm. Last access 21 May 2009.

Reviewing imagery
Table 1 (Continued ) Behaviour/ function-based metaphors (dynamic mental images) Sea wasp Behaviour



Pilot sh




Behaviour Function


Behaviourbased metaphor (motionless mental image)




11. Picture provided by Dr. Zoltan Takacs at http:/ / php?idx=18. Last access 21 May 2009. 12. Picture provided by Eric Orchin at http:/ / =5270518. Last acces 21 May 2009. 13. Picture provided by Stefan Anitei at http:/ / Archersh-Tunes-its-Shot-Power-to-the-Prey-Size-2.jpg. Last access 21 May 2009. 14. Picture provided by Mark Pidcoe at http:/ / index.html. Last access 21 May 2009.


J. M. Urena and P. Faber

Table 2. Resemblance metaphors that can be regarded as both image metaphors and behaviour/function-based metaphors Metaphorical term Type of metaphorical motivation Behaviour Colour Picture

Metaphors based on dynamic mental images

Chameleon sh


Boxer crab

Behaviour Shape



Behaviour Function Shape


Metaphor based on a motionless mental image

Garden eel

Behaviour Shape


15. Pictures provided by LA Productions at http:/ /, %20Badis%20badis.htm. Last access 21 May 2009. 16. Pictures provided by Linda Johnston at http:/ / _pic_boxer_crab_by_linda_johnston_g8m38.html. Last access 21 May 2009. 17. Picture provided by Bruce Robison/Corbis at http:/ / animals/sh/anglersh.html. Last access 21 May 2009. 18. Picture provided by the Gull Dive Center at http:/ / aspx/Dive_Shop. Last access 21 May 2009.

Reviewing imagery References


Alexiev, Boyan. 2005. Contrastive aspects of terminological metaphor. Soa: University of Soa PhD dissertation. Barcelona, Antonio. 2003. On the plausibility of claiming a metonymic motivation for conceptual metaphor. In Antonio Barcelona (ed.), Metaphor and metonymy at the crossroads. Cognitive approaches, 3158. Berlin & New York: Mouton de Gruyter. Bergen, Benjamin, Shane Lindsay, Teenie Matlock & Srini Narayanan. 2007. Spatial and linguistic aspects of visual imagery in sentence comprehension. Cognitive Science 31. 733764. Caballero, Mara del Rosario. 2003a. Metaphor and genre: The presence and role of metaphor in the building review. Applied Linguistics 24. 145167. Caballero, Mara del Rosario. 2003b. How to talk shop through metaphor: Bringing metaphor research to the ESP classroom. English for Special Purposes Journal 22(2). 177194. Caballero, Mara del Rosario. 2006. Re-viewing space: Figurative language in architects assessment of BuiltSpace. Berlin & New York: Mouton de Gruyter. Cienki, Alan. 1997. Some properties and groupings of image schemas. In Marjolijn Verspoor, Kee Dong Lee & Eve Sweetser (eds.), Lexical and syntactical constructions and the construction of meaning, 315. Amsterdam: John Benjamins. Damasio, Antonio. 1994. Descartes error. Emotion, reason and the human brain. New York: Grosset/Putnam. Deane, Paul. 2005. Multimodal spatial representation: On the semantic unity of over. In Beate Hampe (ed.), From perception to meaning. Image schemas in Cognitive Linguistics, 235282. Berlin & New York: Mouton de Gruyter. Dewell, Robert. 2005. Dynamic patterns of CONTAINMENT. In Beate Hampe (ed.), From perception to meaning. Image schemas in Cognitive Linguistics, 370393. Berlin & New York: Mouton de Gruyter. Deignan, Alice. 2007. Image metaphors and connotations in everyday language. Annual Review of Cognitive Linguistics 5. 173191. Fauconnier, Gilles & Mark Turner. 1998. Conceptual integration networks. Cognitive Science 22(2). 133187. Fauconnier, Gilles & Mark Turner. 2002. The way we think: Conceptual blending and the minds hidden complexities. New York: Basic Books. Finke, Ronald. 1989. Principles of mental imagery. Cambridge: MIT Press. Finke, Ronald, Thomas Ward & Steven Smith. 1992. Creative cognition: Theory, research, and applications. Cambridge: MIT Press. Finke, Ronald, Steven Pinker & Martha J. Farah. 1989. Reinterpreting visual patterns in mental imagery. Cognitive Science 13. 5178. Gallese, Vittorio. 2005. Embodied simulation: From neurons to phenomenal experience. Phenomenology and the Cognitive Sciences 4. 2348. Gibbs, Raymond W. Jr. 2006. Embodiment and Cognitive Science. New York: Cambridge University Press. Gibbs, Raymond W. Jr., Dinara Beitel, Michael Harrington & Paul Sanders. 1994. Taking a stand on the meaning of stand: Bodily experience as motivation for polysemy. Journal of Semantics 11. 231251. Gibbs, Raymond W. Jr. & Herbert L. Colston. 2006. The cognitive psychological reality of image schemas and their transformations. In Dirk Geeraerts (ed.), Cognitive Linguistics: Basic Readings, 239268. Berlin: Mouton de Gruyter. Gibbs, Raymond W. Jr., Jessica J. Gould & Michael Andric. 2006. Imagining metaphorical actions: Embodied simulations make the impossible plausible. Imagination, Cognition and Personality 25(3). 221238.


J. M. Urena and P. Faber

Gibbs, Raymond W. Jr. & Marcus Perlman. 2006. The contested impact of cognitive linguistic research on the psycholinguistics of metaphor understanding. In Gitte Kristiansen, Michel Achard, Rene Dirven & F. J. Ruiz de Mendoza (eds.), Cognitive Linguistics: Current applications and future perspectives, 211228. Berlin & New York: Mouton de Gruyter. Grady, Joseph. 1997. Foundations of meaning: Primary metaphors and primary scenes. Berkeley: University of California PhD dissertation. Grady, Joseph. 1999. A typology of motivation for conceptual metaphor: correlation vs. resemblance. In R. W. Jr. Gibbs, Raymond W. Jr. & Gerard Steen (eds.), Metaphor in Cognitive Linguistics, 79100. Amsterdam & Philadelphia: John Benjamins. Kimmel, Michael. 2005. Culture regained: Situated and compound image schemas. In Beate Hampe (ed.), From perception to meaning: Image schemas in Cognitive Linguistics, 285 342. Berlin & New York: Mouton de Gruyter. Ko vecses, Zoltan. 2002. Metaphor: A practical introduction. Oxford: Oxford University Press. Ko vecses, Zoltan. 2005. Metaphor in culture: Universality and variation. New York: Cambridge University Press. Lako, George. 1987. Women, re, and dangerous things. Chicago: University of Chicago. Lako, George. 1990. The Invariance Hypothesis: Is abstract reason based on imageschemas? Cognitive Linguistics 1(1). 3974. Lako, George. 1993. The contemporary theory of metaphor. In A. Ortony (ed.), Metaphor and thought, 202251. Cambridge: Cambridge University Press. Lako, George & Mark Johnson. 1980. Metaphors we live by. Chicago: University of Chicago Press. Lako, George & Mark Johnson. 1999. Philosophy in the esh: The embodied mind and its challenge to Western thought. New York: Basic Books. Lako, George & Mark Turner. 1989. More than a cool reason: A eld guide to poetic metaphor. Chicago: Chicago University Press. Lima, Paula Lenz Costa. 2006. About primary metaphors. D.E.L.T.A. 22. 109122. Paivio, Alan. 1971. Imagery and verbal processes. New York: Holt, Rinehart and Winston. Paivio, Alan. 1986. Mental representations: A dual coding approach. New York: Oxford University Press. Popova, Yanna. 2005. Image schemas and verbal synaesthesia. In Beate Hampe (ed.), From perception to meaning: Image schemas in Cognitive Linguistics, 395419. Berlin & New York: Mouton de Gruyter. Pena, Mara Sandra. 2003. Topology and cognition. What image-schemas reveal about the metaphorical language of emotions. Munich: Lincom Europa. Pena, Mara Sandra. 1999. Subsidiarity relationships between image-schemas: An approach to the force schema. Journal of English Studies 1, 187207. Pylyshyn, Zenon. 2003. Return of the mental image: Are there really pictures in the brain? Trends in Cognitive Science 7(3). 113118. Radden, Gu nter. 2002. How metonymic are metaphors? In Rene Dirven & Ralf Porings (eds.), Metaphor and metonymy in comparison and constrast, 407435. Berlin & New York: Mouton de Gruyter. Ruiz de Mendoza Ibanez, Francisco & Mara Sandra Pena Cervel. 2008. Grammatical met onymy within the action frame in English and Spanish. In Mara de los Angeles Gomez Gonzalez, J. Lachlan Mackenzie & Elsa M. Gonzalez-Alvarez (eds), Current trends in contrastive linguistics: Functional and cognitive perspectives, 251280. Amsterdam & Philadelphia: John Benjamins.

Reviewing imagery


Rohrer, Tim. 2005. Image schemata in the brain. In Beate Hampe (ed.), From perception to meaning: Image schemas in Cognitive Linguistics, 165196. Berlin & New York: Mouton de Gruyter. Talmy, Len. 1999. Fictive motion in language and ception. In Paul Bloom, Mary A. Peterson, Lynn Nadel & Merrill F. Garrett (eds.), Language and space, 211276. Cambridge, MA: MIT Press. Thomas, Nigel J. T. 1999. Are theories of imagery theories of imagination? An active perception approach to conscious mental content. Cognitive Science 23. 207245. Thomas, Nigel J. T. 2009. A note on schema and image schema. http:/ / (accessed 20 May 2009). Watt, Roger. 1991. Understanding vision. New York: Academic Press. Weisberg, Robert. 1986. Creativity: Genius and other myths. New York: Freeman.

Book reviews

Monica Gonzalez-Marquez, Irene Mittelberg, Seana Coulson & Michael J. Spivey (eds.). Methods in Cognitive Linguistics. Amsterdam & Philadelphia: John Benjamins, 2007, xxviii 452 pp. Paperback ISBN 978 90 272 2372 2 / EUR 25.00 / USD 37.95. Reviewed by Zhuo Jing-Schmidt, University of Oregon, USA. E-mail: 3zjingsch@uoregon.edu4 There was a time when the cognitive linguistic study of language and its implications for the mind generally relied on introspection. As is probably the case with any other discipline, new theories and methods would not have evolved without the insights gained from intuitions and introspections in the earlier stages of the quest. Leonard Talmys foreword to Methods in Cognitive Linguistics (MCL) drives home this very point. Today, post-introspection trends are emergingperhaps more or less too rapidly for the unpreparedin response to the demand for increased empiricism in cognitive linguistics. Just as Raymond Gibbs points out in the beginning chapter, methodological reliability and theoretical falsiability are called for to re-prole cognitive linguistics as an empirical science. Its time to move on! The message delivered by MCL, an unprecedented blend of good intentions and patient instructions, is acute. The books primary target audience is cognitive linguists. Nevertheless it will serve to give linguists of all theoretical orientations an updated perspective on methodological developments in linguistic analysis. MCL consists of 17 papers organized in ve chapters. Chapter I, Methods and motivations, provides by way of four introductory articles a view of the challenges and chances confronting cognitive linguists in the post-introspection era. Gibbs alerts cognitive linguists of the skepticism from people outside their eld and urges them to develop reliable and replicable methods and construct falsiable hypotheses. He points to the
Cognitive Linguistics 211 (2010), 151179 DOI 10.1515/COGL.2010.005 09365907/10/00210151 6 Walter de Gruyter


Book reviews Cognitive Linguistics 211 (2010)

fundamental dierences between unconscious cognitive processes and the conscious mind. Such dierences, he argues, render the cognitive unconscious unavailable to conscious introspection. Drawing on functionalist traditions of discourse analysis and corpus linguistics in an expansive attempt to fuse the old and the new, Mittelberg, Farmer and Waugh show the utility of usage data for cognitive linguistics. They then oer a useful guide to the terminology of corpus linguistics, a handy list of popular corpora and an informative discussion of the research potentials of Internet search engines. Gonzalez-Marquez, Becker and Cutting oer a practical thirty-page crash course of experimentation, taking us through the steps and procedures towards the mastery of skills and knowledge necessary for experimental research. Gonzalez-Marquez et al. contrast with Gibbs in their eagerness to turn cognitive linguists into experimenters. While Gibbs suggests that cognitive linguists are not expected to turn away from what they do best and try to be something that they are not, that is, experimental psychologists or neuroscientists, Gonzalez-Marquez et al. intend to get you on your feet about doing experiments. Nunez discusses inferential statistics as another tool to narrowing the gap between cognitive linguistics and hard science. No doubt with largescale corpora becoming ever more accessible to linguists, the ability to run statistical tests to determine rigorously the probability of hypotheses is crucial. Cognitive linguists with the conviction that meaning and grammar originate in usage cannot live up to that conviction unless they are able to quantitatively distinguish probability from chance with regard to ob served uses. In addition to providing technical empowerment, Nunez also points out that mutual feeding of knowledge and insights between cognitive linguistics and its neighboring elds is a vital process for progress. Chapter II, Corpus and discourse analysis, consists of two case studies that illustrate the importance and analytical potentials of usage data and especially corpus data in revealing and helping to explain the complexity of patterns of language use in communication. Waugh, Fonseca-Greber, Vickers and Erozs study of pronouns in spoken French discourse illus trates the analytical power of employing multi-layer empirical evidence as motivated by an ecological perspective on discourse. Grondelaers, Geeraerts and Speelmans analysis of the Dutch presentative sentence in two dialectal regions further demonstrates that corpus data and adequate analytical tools are needed to construct convincing descriptive models of usage. Since these are not easy tasks for any individual linguist, cognitive and otherwise, it is likely, as Grondelaers et al. predict, that linguistic research will become a collaborative, cumulative if only more slow-going science.

Book reviews Cognitive Linguistics 211 (2010)


Chapter III, Sign language and gesture, articulates the immediate relevancy of research on two explicitly embodied symbolic systems, signed language and gesture, to the study of conceptual structures in cognitive linguistics. Wilcox and Morford recognize the conceptual junctures at which signed language and cognitive linguistics convene and are mutually illuminating. The newfound link between the gestural-visual systems and speech provides a multimodality perspective on the one hand, as illuminated by Mittelberg. On the other hand, the discovery of such a profound connection uncovers a potential data source invaluable to cognitive linguists, which Sweetser keenly perceives and points out. Chapter IV, Behavioral research, oers six articles to demonstrate how a variety of specialized task-based experiments which are common in cognitive and developmental psychology can be applied to test cognitive linguistic hypotheses. Carlson and Hill discuss diverse task-based experiments used for exploring the linguistic conceptualization of space. Bergen proposes a model of semantic processing based on mental simulation in terms of neural-level activations of perceptual and motor representations. He surveys a number of experimental methods that support the simulation theory, thus providing empirical evidence of the notion of embodied cognition as articulated by cognitive linguists. Hasson and Giora introduce the basic psycholinguistic measures of mental representations prompted by a sentence. They explain the rationale of each measure and illustrate its particular utility with examples, thereby providing beginners with a users manual. Richardson, Dale and Spivey join the quest by showing cognitive linguists what visual attention can reveal about cognition and language. They measure visual attention by tracking eye movements in reading and listening comprehension. This particular angle of research oers unique clues to real-time mental activities related to language processing and use. Brandone, Golinko, Pulverman, Maguire, Hirsh-Pasek and Pruden take a developmental perspective by studying concept formation as it occurs with preverbal infants. Their experimental paradigms help to reconstruct the transitional path along which preverbal concepts are transformed by early language acquisition. Gor is concerned with the cognitive mechanisms behind inectional morphology processing. She shows how two task-based experiments can be conducted across dierent languages and populations to yield strong data on the eects of frequency and phonological similarity across two types of processing. It is exciting to see the analytical potentials of such experiments in testing something as elusive as the intuition of a continuum between lexicon and grammar, an intuition captured in the foundational work of introspective cognitive linguistics. Chapter V, Neural approaches, introduces cognitive linguists to the organic world of the brain and its neural activities as the biological basis


Book reviews Cognitive Linguistics 211 (2010)

of language and cognition. The eort described in this chapter highlights the beginning of what Lieberman (2002:17) envisages as a new era of biological linguistics, one involving cooperative research among linguists, cognitive scientists, and neurobiologists. From Coulsons introduction, the reader gets an idea of how the ERP technique could shed light on cognitive semantic hypotheses by providing a neural-level depiction of the conditions and processes of meaning comprehension as it takes place in the brain. On the other side, the reader is guided by Edelman into the more nascent domains of computation and algorithm in relation to neurobiology. The reader gets a faint taste of abstract computational models of language that are presumed to simulate distributed knowledge representations in terms of cognition-general principles compatible with neural mechanisms and statistical inference. Given the rapid development in brain science and given the relative lack of communication between neurolinguists and cognitive linguists despite the rapid development in brain science, to use Ahlsens (2006:vii) words, the message conveyed in this chapter is particularly valuable to cognitive linguists. As the book unfolds through the chapters, the conventional but inquisitive cognitive linguists may nd themselves entering ever less familiar territories. At the same time, wading through every last chapter, they may nd that the boundaries of what is possible in cognitive linguistics get stretched more or less to allow them to perceive somewhat richer sceneries that are thus far unseen. No doubt anxieties about ones own methodological inadequacies increase accordingly. Those looming anxieties, let us hope, will inspire cognitive linguists to acquire one of the tools revealed to them in this big bundle of a toolbox, or nd someone who already possesses some tools, to collaborate. Recent years have witnessed desirable advances resulting from such integrative eort, as exemplied by Dabrowska (2004), Gallese and Lako (2005), Pulvermuller (2005), Boroditsky and Gaby (2006), Glenberg (2008), and Barsalou et al. (2008), among many others. The most remarkable thing about MCL is that none of these contributions would have been as powerful if they were scattered in separate issues of discipline-specic journals. Yet together they send one startling message: Lets give each other a hand, for together we can leap over chasms!

Ahlsen, Elisabeth 2006 Introduction to Neurolinguistics. Amsterdam/Philadelphia: Benjamins.

Book reviews Cognitive Linguistics 211 (2010)


Barsalou, L. W., A. Santos, W. K. Simmons, and C. D. Wilson 2008 Language and simulation in conceptual processing. In: Symbols and Embodiment, M. De Vega, A. M. Glenberg and A. C. Graesser (eds.). Oxford: Oxford University Press. Boroditsky, L. and A. Gaby 2006 East of Tuesday: Representing time in absolute space. Proceedings of the 28th Annual Meeting of the cognitive Science Society, Vancouver, Canada. Da browska, Ewa 2004 Language, Mind and Brain. Washington, D.C: Georgetown University Press. Gallese, V. and G. Lako 2005 The brains concepts: The role of the sensory-motor system in conceptual knowledge. Cognitive Neuropsychology 22, 455479. Glenberg, Arthur M. 2008 Toward the integration of bodily states, language, and action. In: Embodied Grounding: Social, Cognitive, Aective, and Neuroscientic Approaches, G. R. Semin and E. R. Smith (eds.), 4370. Cambridge: Cambridge University Press. Lieberman, Philip 2002 Human Language and Our Reptilian Brain. Cambridge, MA: Harvard University Press. Pulvermu ller, Friedemann 2005 Brain mechanisms linking language and action. Nature Reviews Neuroscience 6(7), 576582.

Mark Johnson. The meaning of the body: Aesthetics of human understanding. Chicago & London: The University of Chicago Press, 2007, xiii 308 pp. ISBN: 9780226401928. Cloth $ 32.00. Paper $ 22.50. Reviewed by Heli Tissari, University of Helsinki, Finland. Email 3heli.tissari@helsinki.4 Mark Johnsons new book continues to develop themes which he has already tackled in books well-known to cognitive linguists: conceptual metaphors (Lako & Johnson 1980), the bodily basis of meaning (Johnson 1987), and the implications of the conceptual metaphor theory (henceforth CMT) to philosophy (Lako & Johnson 1999). Its rst part deals with bodily meaning and felt sense (pp. 17110), the second part relates the notion of embodied meaning to the sciences of the mind (pp. 111206), and the third part discusses aesthetics and art (pp. 207 283). The rst chapter (pp. 1932) elaborates on a claim which it begins with, that [l]ife and movement are inextricably connected (p. 19). Considering CMT, this is not a new claim, but potentially a shift in emphasis. Movement has been present in discussions of time and temporal aspects


Book reviews Cognitive Linguistics 211 (2010)

of language, but there are also a number of metaphors, e.g., metaphors of containment, which I would rather associate with relative stability. More precisely, we might say that Lako and Johnson have been interested in a persons interaction with her or his surroundings (e.g., in Lako & Johnson 1999), but in practice, discussions of conceptual metaphors often solidify metaphors through labelling and classifying them, and through using nouns for both the target and source domains of metaphors (e.g., love is fire). The second chapter discusses the development of a sense of self in infants, in particular their interaction with their surroundings through movement (pp. 3351). A synopsis of this chapter can be found on page 36, which introduces the notions of communication, object perception and manipulation, and bodily motion. Johnson discusses these by referring to psychological literature, aiming to show that objects are minddependent and individuated relative to our conceptual systems and structures for meaning-making. The world does not come to us prepackaged with determinate objects with their determinate properties (p. 46). This agrees with what he and George Lako claimed as early as 1980. Chapter three (pp. 5268) presents a movement onwards in CMT, claiming in its heading that feeling is rst. Johnson makes explicit tribute to Antonio Damasio and Joseph LeDoux, whom he calls two of the most distinguished researchers in the eld of cognitive neuroscience today (p. 54). He emphasises the importance of our unconscious processing of what is happening as a basis both of emotion and action (p. 66). In other words, he claims that we usually do not make sense of the world by consciously reecting on it. Rather, we feel it. Chapter four introduces the concept of qualities of life (pp. 6985), with particular focus on Deweys pervasive qualities of experiences (pp. 7178), and Gendlins felt situations (pp. 7985). Johnson sums up the former by stating that entire situations are characterized by pervasive qualities, and we pick out particular qualities for discrimination within this unied situational whole (p. 72). His discussion of Gendlin then introduces the concepts of the felt sense and the formal expression, which he claims are two dimensions of a single, ongoing activity of meaning-making (p. 82). Putting it all together, we can both feel a particular quality and name it, although the latter task may be much more dicult than the former. Continuing from this, chapter ve claims that concepts emerge at the meeting-point between the felt and the named (pp. 86110). Johnson bases this claim on the authority of William Jamess writings on thinking, underlining the dierence between such an understanding of meaning and traditional accounts of the relationship between words and concepts.

Book reviews Cognitive Linguistics 211 (2010)


In agreement with his earlier views, he says that [h]uman thinking is a continuous feeling-thinking process that is forever tied to our bodys monitoring its own states (p. 98). These considerations conduct us to the second part of the book which, according to Johnson, is required because we must explain conceptual thinking without introducing immaterial mind or a transcendent ego (p. 113). Chapter six begins this task by outlining a nonrepresentational view of mind (pp. 113134), or what he also calls an interactionalist (or transactional or enactionist) view (p. 117, original emphasis). In this context, he briey discusses the biology of multicellular organisms and the way they develop neural maps as reactions to new kinds of situations (pp. 123130). Johnson begins chapter seven by restating that humans, like animals, have neural maps (p. 135). What he then goes on to say about image schemas is quite familiar from his previous writings (such as Johnson 1987). Towards the end of the chapter he discusses the social, intersubjective character of embodied cognition (pp. 147152), dening social phenomena as those phenomena arising out of recurrent structural couplings that require the coordinated participation of multiple organisms (p. 148). The eighth chapter focuses on the brains role in meaning (pp. 155 175); in other words, the sensorimotor system and its relationship to reasoning and language. In this chapter, Johnson denes concepts as patterns of interaction that are important enough for the ongoing experience of a person (or an animal) to merit being selected from the ow of experience (p. 159); cf. pervasive qualities of experiences, felt situations above. What I found particularly intriguing in this chapter is Johnsons statement that not only image schemas and conceptual metaphors and metonymies, but [a]ll aspects of grammar . . . and all aspects of logical relations need to be accounted for through ties to body-based meaning (p. 170). If a feeling report of this book is allowed, I felt that this claim, following the description of qualities of life and the emergence of concepts in chapters four and ve led to considerable expectations as to what would follow. In that respect, the ninth chapter, with its traditional Lakoand-Johnson-type account of embodied meaning and abstract thought in terms of CMT, was a disappointment. It did not seem that much was added to what appeared in Lako and Johnson 1999. The third part of the book, on embodied meaning, aesthetics, and art (pp. 207283), felt more welcome in these terms, bringing in more novel considerations. Chapter ten is a critique of the devaluation of aesthetics in the Western tradition and of Kants subjectivizing of


Book reviews Cognitive Linguistics 211 (2010)

aesthetics (pp. 209219), followed by examples of meaning as occurring in poetry and visual images (pp. 219234). Remembering Johnsons visual interests from his The Body in the Mind (1987), chapter eleven on music and the ow of meaning is even more interesting (pp. 235262), describing the roles image schemas and metaphors play in pieces of music. In his chapter on music, Johnson beautifully returns to his original interest in movement, discussing, among other things, what he calls the moving music metaphor. In his explication of the metaphor, the target domain musical motion involves such issues as musical passages, their beginnings and ends, rests, and repeats (p. 248). What I missed in this chapter was a deeper involvement with the notion and intentions of a composer. Johnson concludes with a chapter entitled the meaning of the body (pp. 263283), which includes a similarly, if not more, ambitious subheading, the meaning of meaning (pp. 265274). Here he outlines the objectivist theory of meaning (p. 272), and an embodied, experientialist view of meaning (p. 273), in a manner already familiar from Lako and Johnsons earlier books. To compare, a similar list can be found in the preface to Lako (1987), where Lako lists objectivist views (pp. xiixiii), and a very dierent view (pp. xivxv). To contextualise the book in question, it may pay to compare it with two other books, because I wonder whether it is also a spiritual book. Many of the ideas in it are also present in Varela, Thompson and Roschs The Embodied Mind: Cognitive Science and Human Experience (1995 [1991]), which promotes an embodied view of human existence and Buddhist meditation. However, Johnsons aesthetics may also be seen to relate to many of the issues discussed by Evans in his Preserving the Person: A Look at the Human Sciences (1979 [1977]) which, apart from promoting a specically Christian world-view, discusses the then current situation of the human sciences. It is of interest to note that Varela, Thompson and Rosch say that the inspiration for [their] book began in the late seventies (1995: xi). In other words, it seems that some of the views which Johnson discusses in his book were already being debated at that time. I am not saying that they were not debated prior to that, but oering one way of seeing where Johnson comes from and where he may be going. In a footnote, Johnson writes:
When I was in graduate school in the mid-1970s, it was a commonplace prejudice of the culture of analytic philosophy to assume that the smartest, most serious students would do the intellectually rigorous work . . . while those who werent up to this exalted task could entertain themselves with the mushier, subjective value elds. (p. 214.)

Book reviews Cognitive Linguistics 211 (2010)


Compare with Varela, Thompson & Roschs question: How can . . . an attitude of all encompassing, decentered, responsive, compassionate concern be fostered and embodied in our culture? (1995: 252.) Evanss book is of interest particularly in respect to the objectivist theory of meaning criticised by Johnson (p. 272), and by Lako, who summarises it by stating, among other things, that [i]t is . . . incidental to the nature of meaningful concepts and reason that human beings have the bodies they have and function in their environment in the way they do (1987: xiii). Evans wrote in the 1970s that [t]he Zeitgeist has denitely been against the dualist. To many, including many philosophers, dualism just does not seem a live option today. (1979: 105.) Compare also Evanss presentation of the French Catholic Gabriel Marcels perspectivalism with Johnsons aesthetics. According to Evans, Marcel sees lived experience as something in which
the person is seen as a presence not as object, and the diculties encountered are mysteries to be explored, rather than problems to be solved. Marcel believes that the mind-body problem is not a problem but a mystery. My relationship to my body is too intimate to allow me to objectify it as a thing and then ask, How am I related to this thing called my body? (1979: 107, original emphasis.)

Although Johnson, quite unsurprisingly, does not use the word mystery, he refers to the oh of wonder (p. 71). One of his chapter titles comes from a poet who is known to have talked to his God (since feeling is rst, E.E. Cummings, quoted on p. 52). Varela, Thompson and Rosch enumerate folk meanings of the word meditation, among these a mystical state in which higher realities or religious objects are experienced (1995: 23). To round up, what I think Johnson does most beautifully in his The Meaning of the Body is to describe what living and thinking in a human body feels like, and I would recommend it to anyone wanting to relish such a text. Simultaneously, I doubt whether just anyone is really as aware as Johnson is of his embodiment, unless they have trained themselves, whether through meditation (or prayer), through the arts, or through sport. Not that he claims this, but I simply wanted to emphasise it. Varela, Thompson and Rosch advocate one form of such an awareness, claiming that [e]xperience and scientic understanding are like two legs without which we cannot walk (1995: 14). In this light, Johnson seems to be going from the embodiment of meaning to coupling embodied meaning with values and, potentially, towards spiritual experience. Indeed, Johnson reaches the notion of embodied spirituality on the last pages of the book under review (pp. 281282). For him, it means


Book reviews Cognitive Linguistics 211 (2010)

horizontal transcendence, as opposed to what he calls vertical transcendence. Such horizontal transcendence
. . . recognizes the inescapability of human nitude and is compatible with the embodiment of meaning, mind, and personal identity. From this human perspective, transcendence consists in our happy ability to sometimes go beyond our present situation in transformative acts that change both our world and ourselves. (P. 281.)

As a linguist, I would nally like to pick up three issues from this book. The rst is the notion that [l]ife and movement are inextricably connected (p. 19). This is no news to historical linguists and, in general, we know that language changes. It is nevertheless still a challenge for CMT and cognitive linguistics to take this into account, and this is something that Johnson does not discuss in his book. The second issue is the claim which I summarised by saying that concepts emerge at the meeting-point between the felt and the named (pp. 86110). In my view, cognitive linguists should pay more attention to the relationship between word and concept, and Johnson is thus right to pinpoint it as a critical issue. It is something that is too often mentioned in passing without considering what it means. For example, what does it mean to use a notation for conceptual metaphors which couples two nouns with a form of the verb be (e.g., love is fire)? Or, why do we say source-path-goal schema (as on pp. 141142) rather than, for example, the linear schema, or the move schema? Or, what does it mean that the felt sense of an experience is not vague, mushy, empty, or chaotic, but extremely precise, as Johnson claims, going on to say that it is carried forward only by quite specic words or forms (p. 82, original emphasis). (Consider the prototype theory.) The third issue is the role of the brain and the sensorimotor system in reasoning and language. There are potentially myriads of issues to research if we wish to come close to the goal set by Johnson in stating that [a]ll aspects of grammar . . . and all aspects of logical relations need to be accounted for through ties to body-based meaning (p. 170). It is a challenge for cognitive linguists to pose the right new questions to begin with. References
Evans, C. Stephen. 1979 [1977]. Preserving the person: A look at the human sciences. Downers Grove, IL: InterVarsity Press. Johnson, Mark. 1987. The body in the mind: The bodily basis of meaning, imagination and reason. Chicago: The University of Chicago Press. Lako, George. 1987. Women, re and dangerous things: What categories reveal about the mind. Chicago & London: The University of Chicago Press.

Book reviews Cognitive Linguistics 211 (2010)


Lako, George & Mark Johnson. 1980. Metaphors we live by. Chicago: The University of Chicago Press. Lako, George & Mark Johnson. 1999. Philosophy in the esh: The embodied mind and its challenge to Western thought. New York: Basic Books. Varela, Francisco J., Evan Thompson & Eleanor Rosch. 1995 [1991]. The embodied mind: Cognitive Science and human experience. Cambridge, MA & London: The MIT Press.

June Luchjenbroers (ed.). Cognitive Linguistics investigations. Across languages, elds and philosophical boundaries. (Human Cognitive Processing 15). Amsterdam & Philadelphia: John Benjamins, 2006, 334 pp., ISBN 978 90 272 2368 5. Hardbound. EUR 120.00 / USD 180.00. Reviewed by Ignacy Nasalski, Institute of Oriental Philology, Jagiellonian University of Cracow, Poland. E-mail: The book in review has no particular leading theme. It arose from a workshop held at the University of Queensland during the 4th Australian Linguistics Institute, in July 1998, in which researchers from around the world oered papers drawn from a number or areas from within the cognitive sciences. As usually in the case of such compilations, the volume presents accordingly a range of topics and captures a diversity of research activities from various parts of the world and across a range of European, Asian, Bantu, Austronesian and Australian Aboriginal languages. Despite dierences in philosophical approach and applied methodology they all share a commitment to the view that human categorization involves mental concepts that have fuzzy boundaries and are culturally and situation-based (p. 3). Apart from the introduction by the editor which gives the reader a general outline of the volume, the book contains fourteen papers (there are in fact fteen chapters, but I exclude the Introduction from the calculation, because it should not be numbered as chapter 1) that are supposed to illustrate how otherwise separate areas of linguistic concern can present a better clarication of the linguistic distributions in which units are produced in talk; as well as provide a deeper appreciation of the semantic richness of those linguistic units, not captured by Formalist approaches (p. 2). The book is divided into three parts. Part one Cultural models and conceptual mappings consists of four papers presenting investigations into cultural schemata, gesture, mental spaces manoeuvres and the like. It shows how conceptual mapping builds on specic types of knowledge and how cultural models sculpt the verbal communication.


Book reviews Cognitive Linguistics 211 (2010)

Gary Palmer, in a starting paper When does cognitive linguistics become cultural? Case studies in Tagalog voice and Shona noun classiers (p. 1345), uses data from two unrelated languages to supply further evidence for the idea that grammar is governed by cultural schemata rather than universal cognitive schemata. Two case studies from Tagalog and Shona illustrate how lexical domains and grammatical construction link to linguistically determinant cultural models such as scenarios and polycentric categories and end with the conclusion that understanding the grammar and lexicon of a language requires grasp of cultural models and culturally dened imagery (p. 39). In a very interesting and instructive paper Purple persuasion. Deliberative rhetoric and conceptual blending (p. 4765) Seana Coulson and Todd Oakley make use of Blending Theory to demonstrate how cognitive linguistics can contribute to better understanding of some political processes, particularly those involving inuencing people. The authors show how simplied input models and categories are blended to produce categories with its own distinct internal structure and form integrated event scenarios. The conclusion is that an apposite choice of input frames can serve as an ecient persuasive means to encourage a particular construal of events that will result afterwards in the target actions and transform recipients into political activists. Depicting ctive motion in drawings (p. 6785) is a paper by Teenie Matlock who analyzed so called ctive motion sentences such as The road goes/runs along the coast in order to nd out whether ctive motion plays a role in their comprehension. Drawing from picture experiments she provided evidence for a link between motion verbs and the mental stimulation of the action conveyed by them. The presented results challenge standard psycholinguistic accounts for how words are represented and proceeded. They prove that ctive motion sentences include dynamic construal as mentally simulated motion or linear extension, and that comprehension rests broadly on the embodied experience. June Luchjenbroers in the next paper Discourse, gesture and mental spaces manoeuvres (pp. 87105) investigated the dynamics of conversational gesture in the physical space (F-space) in which they occur during discourse. She argues that the boundaries of this space either amplify the information conveyed by the lexical component or provide additional aspects of speaker meaning in the form of clues about speaker cognition. The second part Computational models and conceptual mappings is comprised of studies dealing with computational models that hypothesize dierent features of the cognitive programming and discuss how they can be used to describe cognitive processes associated with the mental lexicon in relation to morphology (Ping Li, pp. 110137), to grammar (Joost

Book reviews Cognitive Linguistics 211 (2010)


Schilperoord and Arie Verhagen, pp. 139168) and the phonological system (Paul Warren, pp. 169186). Ping Li (In search of meaning: The acquisition of semantic structures and morphological systems) argues that the evolution and development of semantic representations as acquired by children result from simple probabilistic procedures as embodied in connectionist networks and analogous statistical learning mechanisms. These connectionist networks can capture, as demonstrated by Li, the representation of semantic structures which can be best viewed as emerging out of a continuously developing dynamical system that operates on statistical computations of the various form-form and form-meaning constraints. Schilperoords and Verhagens interest (Grammar and language production) concentrates on the organizational features of the mental lexicon and mental grammar, and in particular on the question of how function words, prepositions and articles are selected during language production. The authors apply a usage based consideration of function words in order to explain how these words are cognitively processed. The traditional assumptions that function words are stored independently of their lexical heads and that individual content words are retrieved from the lexicon and assembled into larger structures by means of grammatical computation, are here questioned and presented as incorrect or at least incomplete. The nal paper in this section by Paul Warren (Word recognition and sound merger) deals with comprehension in the form of psycholinguistic models of spoken word recognition as well as with the need for a more cognitive account for variation arising from sound change. The particular case under consideration is the merger-in-progress of the front centring diphthong in New Zealand English as in ear/air neutralization. Warren argues that it is the sentential or extralinguistic context that resolves homophone ambiguity as in the case of merged ear/air form just like they do for other homophones. The last part Linguistic components and conceptual mappings focuses on three areas of linguistic description: semantics, grammar and discourse. In the rst paper in this section Verbal explication and the place of NSM semantics in cognitive linguistics (p. 189218) Cli Goddard argues that cognitive linguistics cannot approach verbal explication in a casual manner, because its familiar devices such as diagrams representing image schemata or conceptual metaphors often rely on complex culturespecic iconographic conventions which are smuggled in without the necessary acknowledgement or explanation (p. 209). Goddard makes therefore use of the natural semantic metalanguage framework (NSM),


Book reviews Cognitive Linguistics 211 (2010)

because he sees it as the only well developed and empirically grounded theory of verbal explication that allows additionally to avoid the terminological ethnocentrism. The result is a convincing demonstration that NSM can be successfully employed within the cognitive linguistics paradigm. How do you know shes a woman. Feature, prototypes and category stress in Turkish kadn and kz is a scintillating paper presented by Robin Turner (p. 219234) in which he analyzes Turkish words for girl and woman, and investigates the semantic content of them with their relation to culture and personal interaction. He comes to the conclusion that the inconsistency between the feature-based and prototype-based categorization results in what he calls category stress. This concept is his original ideait can be seen as a kind of cognitive dissonance (p. 221) which results from either contextual factors, such as humour and alternative categorization, or from cultural factors such as social change. The next paper by Iraide Ibarretxe-Antunano Cross-linguistic poly semy in tactile verbs (pp. 235253) complements to some extent Turners research. The author shows how the semantic content of the tactile verb touch in three genetically unrelated languages, viz. Basque, Spanish and English, interacts and contributes to the creation of each semantic extension. From these examples Ibarretxe-Antunano draws more general conclusions and argues that dierent polysemes of a lexical item are obtained through the interaction of the semantic content of both the lexical item itself and its dierent co-occurring elements (the phenomenon she calls compositional polysemy). Maarten Lemmens in a paper How experience structures the conceptualization of causality (pp. 255270) deals with variations in the conceptualization of causative verbs of killing such as suocate, choke or kill in Old English corpus data. He argues that the syntactic choice between two dierent models of causative events, i.e. the transitive or the ergative model, shapes the way the event is experienced. The conclusion, not immediately obvious from introspection, is that the more volitional a participant seems to be whilst engaged in some causative process, the more likely it is that s/he will surface as a volitional Actor in a transitive con` struction; the more autonomous the process, vis-a-vis its cause, the more likely an ergative conception will be triggered (p. 267). In the next paper Internal state predicates in Japanese: a cognitive approach (pp. 271291) Satoshi Uehara develops Langackers framework on subjectivity and tries to explain the use of particular grammatical elements in discourse, such as the nominative particle ga being used for object marking. The author argues that internal state predicates in Japanese can best be characterized as deictic, since they prole the ob-

Book reviews Cognitive Linguistics 211 (2010)


ject of speaking from the speakers point of view. Uehara argues that internal state predicates denoting speakers internal states, feelings and emotional reactions, contribute to the subjective event construal and provide thus support for the Cognitive Grammar theory of subjectivity. The last two papers in the volume deal with a very interesting aspect of human communication, to which cognitive linguistics has contributed too little, namely discourse and narration. Dave Goughs paper Figure, ground and connexity: evidence from Xhosa narrative (pp. 293303) is a study of folk narrative discourse. Gough advocates an orientation to language which holds that its nature can and should be explained in terms of factors outside of language, i.e. in terms of discourse and cognitive factors rather than seeing it as a taxonomic and internally describable separate modules. He argues that grammatical terms like mood and tense refer to quite dierent verbal categories and that the discourse concepts of grounding and connexity provide more coherent explanation of the structure of the verbal system than the traditional analysis. Ming-Ming Pu in her paper Discourse organization and coherence (pp. 305324) explores some narrative constraints that are usually subject of textual analysis. She is primarily interested in the question of how the events in an episodic structure are related to each other. Using narrative data from English and Mandarin Chinese, drawn from a childrens picture book, Pu comes to aquite obvious, I must addconclusion, that conversations and written texts are almost never unordered strings of utterances, but they are usually structured, more or less complex stories, the coherence of which is achieved in a systematic and even automatic process through establishing story frame, focusing on the central character, systematically tracking references and maintaining topic continuity (p. 323). Interestingly her paper provides strong support for an interactional character of communication. She shows namely that the process of text production, whether spoken or written, is guided by cognitive constraints upon speaker to accommodate their addressees processing needs by signalling discourse units and prompting the retrieval of information. Prepared by researchers from universities in Australia, New Zealand, Spain, France, USA, Turkey and Holland, this volume constitutes a signicant contribution to the eld of cognitive and cultural linguistics. Just as the subtitle Across languages, elds and philosophical boundaries suggests, the fteen chapters cover an extensive selection of concepts and notions that are of interest for everyone dealing with such elds as language acquisition, video data analysis, gesture, Blending Theory, ctive motion and the like. The volume presents well documented data from a spectrum of languages that empirically validate or challenge some of the hypotheses or theoretical models. Both those who prefer to focus on one


Book reviews Cognitive Linguistics 211 (2010)

language or a variety at a time will nd the collected texts equally attractive. An important merit of the book is the fact that some papers go beyond mere linguistic investigations, and provide revealing insights into some cultural (Palmer; Goddard; Turner), socio-political (Coulson & Oakley) and psychological (Uehara; Pu) phenomena. Thus, the volume can also be recommended to philosophers, anthropologist and even political scientists.

nter Radden, Klaus-Michael Kopcke, Thomas Berg & Peter Siemund Gu (eds.). Aspects of meaning construction. Amsterdam & Philadelphia: John Benjamins, 2007, x 287 pp. EUR 110.00 / USD 165.00. Reviewed by JoAnne Ne-van Aertselaer, Department of English Studies, Universidad Complutense de Madrid, Spain. E-mail: 3nejoanne@hotmail.com4 A central presupposition of cognitive linguistics is that human beings construct meanings by accessing cognitive structures, which are themselves the result of embodied, encultured and imaginative dimensions of meaning (Fesmire 1994:150). Of the various cognitive structures used in meaning construction, this volume deals specically with metonymy and metaphor (Part I) and mental spaces and conceptual blending (Part II). Most of the articles reect, in dierent ways, concern with the research area of the continuum between metonymy and metaphor and consequently the volume provides a further conceptual up-date on recent work done in these areas. Thus, some of the thematic divergence found in festschrift volumes has been avoided here. The volume was dedicated to Klaus-Uwe Panther and, given his important work on metonymy, many of the articles included in Part I provide excellent links to Panthers signicant body of research (Panther 2005; Panther and Thornburg 1998; Panther and Thornburg 2003a,b; Panther and Thornburg 2004). The Introduction begins with a useful overview of the types of linguistic underspecicationinvolving implicitness, indeterminacy, and incompatibilitywhich lead language users to mutually construct meanings. To be sure, the interpretative nature of language meaning is not a new idea, and not even particular to cognitive linguistics. In the sub-elds of both cognitive psychology (Ausubel et al 1978; Bruner 1990; Mandler 1984; Rumelhart 1980; Shank 1977) and social psychology (Vygotsky 1978), the role of conceptual schemata, built up from real world knowledge, in utterance understanding has long been the focus of constructivist theories. However, most of these theorists have dealt with comprehension and the

Book reviews Cognitive Linguistics 211 (2010)


building of schemata upon the reception of information and not with meaning making between interlocutors, i.e. two conceptualizers, which is the focus of this volume. However, for some time now, both psychology (Mandler 1992; Neisser 1976) and cognitive linguistics (Johnson 1987; Lako 1987; and many others) have begun to coalesce in viewing conceptualization as webs of relations emerging from the union of perception and environment. The remainder of the Introduction refers briey to the work presented in each of the 13 articles which the volume comprises. The editors also provide a practical index of metonymies and metaphor. A brief examination shows that of the metonymies most analyzed by the dierent authors are cause for effect, whole for part, part for whole and place for event; among the conceptual metaphors most examined are objects are human and time is space. Part I of the book deals with metonymy and metaphor, for the most part, from a macroscopic view (emphasis on large webs of systems of these two cognitive patterns). The number of articles that deal with the two conceptual patterns as underlying grammatical phenomena show how far cognitive linguistics has progressed, from microscopic perspectives (presenting and discussing specic examples), toward the explanation of metonymic and metaphorical explanations of grammar from a fundamentally dierent approach (Thornburg, Panther and Barcelona 2006; Verhagen 2005). The chapter by Raymond Gibbs, Experimental tests of gurative meaning construction, explores more extensively central questions which he has previously examined (Gibbs 1993), namely the question of whether claims of cognitive linguistics aord psychologically plausible accounts of how people construe meaning in everyday discourse situations. More specically, Gibbs studies how interlocutors integrate pragmatic knowledge with conceptual metonymies in order to create specic, contextually appropriate inferences. Gibbs rst question is whether conceptual metonymies (e.g., part for whole and place for event) really do have psychological reality, as claimed by Panther and Thornburg (2003b). In order to shed light on this issue, Gibbs turns to recent corpus studies and notes that contrary to what is usually thought (that place for event metonymies are very frequent), Markert and Nissum (2003) actually found that place for people was much more common. He, therefore, cautions that cognitive linguists must be very cautious before claiming that any particular gure of speech is common (p. 22). Noting that much more psycholinguistic research has been carried out on peoples use of conceptual metaphors, he calls for psycholinguists to empirically study conceptual metonymies through priming paradigms, i.e., using a primed conceptual


Book reviews Cognitive Linguistics 211 (2010)

metonymy (i.e., object for user) to see if the priming facilitates peoples reading of metonymic phrases supported by similar conceptual metonymies, such as place for event. In the remainder of the chapter, Gibbs provides empirical support for an important claim made by Panther and Thornburg (2003b), namely, that metonymies make targets accessible (in the scalpel was sued for malpractice, the instrument for person metonymy activates the referent of surgeon) and thus allow for further elaboration. Finally, Gibbs also supports another claim made by Panther and Thornburg (2003b) regarding the adjustments in interpretations made by interlocutors through the use of pragmatic knowledge. In High-level metaphor and metonymy in meaning construction, Francisco Ruiz de Mendoza and Ricardo Mairal, argue that inferential activity constrains the possible interpretations made through these conceptual patterns and that said constraints are also operational in many grammatical processes, in particular in the domain of transitivity examined here. The authors review of various types of constraints on metaphoric and metonymic interpretation (the Invariance Principle, the Extended Invariance Principle, the Correlation Principle, and the Mapping Enforcement Principlethe rst is Lako s and the last three are Ruiz de Mendozas), but given the extensive polysemy of natural language, even these principles may not be enough to make actual language use fully predictable without the inferential activity emerging from speakers rhetorical construals during usage events, a statement with which these authors would not disagree. One of the chief contributions of Ruiz de Mendoza and Mairals paper is their discussion of high-level metaphoric and metonymic activity performed on non-situational cognitive models (here, action and process), which underly grammatical patterns such as transitivity. Starting from a discussion of the high-level-propositional idealized cognitive model, which they call action frame, and Vendlers Aktionsart (action-process distinction), they propose additional criteria for the classication of modes of action (resultative, eectual, experiental and communicative) and subsume some processes (although not all) into the subdomain of resultative actions. The high-level metaphoric activity signaled here is that there may be shifts between action frames and process frames, in the sense that in Peter laughed John out of the oce, the verb has undergone a metaphorical mapping of the actor-goal transitive relationship (from laugh at someone to laugh someone). The authors are somewhat less successful in dealing with other examples (namely 10, She laughed herself out to silence) of metaphorical mappings of transitivity onto intransitive verbs, mainly because, at least as far as example 10, the sentence seems highly unlikely to be a real token of language use. This

Book reviews Cognitive Linguistics 211 (2010)


problem reminds one of Gibbs request for more use of corpora in cognitive linguistics. Still within the action frame, the nal section of the paper deals with high-level metonymy (process for action and process for action for result). These metonymic operations may prole dierent aspects of transitivity constructions, such as the bread cuts easily (process for action) and the bread cuts well (process for action for result) and, thus, the authors explanations provide links to the notion of coercion (Michaelis 2003), a concept which is challenged in Chapter 5 of this volume. In Antonio Barcelonas chapter on The role of metonymy in meaning construction at discourse level is a welcome eort to lend more psychological validity to the function of metonymy in discovering readers textlevel implications. Barcelona rst oers an explanation of the role of metonymy in triggering pragmatic reasoning, as also reected in the work of Panther and Thornburg (2003a, b). Although in other works Barcelona (2002: 227228) oers descriptions of three dierent types of metonymy (schematic, typical, and prototypical ), here he denes only schematic metonymy, because it exhibits all of the properties shared by every type of metonymy. He oers an up-dated denition of this type of metonymy as: . . . an asymmetrical mapping of a conceptual domain, the source, onto another domain, the target. Source and target are in the same functional domain and are linked by a pragmatic function, so that the target is mentally activated (p. 53). Whether one agrees or not with Barcelonas denition of schematic metonymy, which would include almost all uses of a linguistic expression (The book is very large whole for [physical] part), what he presents in this article is not a discussion of this type of metonymy, but of text-level implicatures which are guided by conceptual metaphors (i.e., those with a higher degree of metonymicity than the schematic type). In this study, he extends an earlier analysis of brief conversational exchanges to implicatures generated by a longer text (a six-sentence narrative on hiking in Colorado), which he uses to carry out an informal experiment with seven native speakers in order to check the inferences which Barcelona himself had already generated from his reading of the text. He then ties the conceptual metonymies (action for purpose, category for member, cause for effect, etc.) to the 16 textual implicatures, 14 of which were conrmed by the native-speaker participants. Although this experiment might have been more psychologically convincing if Barcelona had had at least two other people besides himself generate the initial inferences and if he had also added distractors, still Barcelona has provided a uncomplicated on-line method for verifying the inferences which can thus prevent over-intellectualizing linguistic comprehension.


Book reviews Cognitive Linguistics 211 (2010)

Chapter 4 of the volume deals with Chained metonymies in lexicon and grammar by examining body part terms from a cross linguistic perspective. By using data from bilingual dictionaries to explore semantic extensions of body part terms, Martin Hilpert argues that, in dierent languages, there are systematic dierences between lexical and semantic extensions involving chained metonymies. Hilpert also contributes to the literature on the interplay of metonymy and metaphor by showing that metonymic mappings precede the metaphoric ones. The plausibility of the chained metonymies that he studied is veried by experiential motivation, the existence of polysemous links and cross-linguistic attestation of serial metonymic mappings in 76 languages, representing all the known language families. These data contribute signicantly to the reliability of the authors interpretations. However, Hilpert also rightly acknowledges that to validate his assumptions, he would need historical data in order to discuss grammaticalization processes and that some polysemous terms for body parts are very vague, e.g., the term for foot referring also to leg, which makes it dicult to identify which is the more basic term. He presents various strategies for associating body part terms with lexical meaning extensions and gives a detailed account of each type with interesting subsequent extensions: body parts associated with perceptual functions (as an instrument, as a container, as part of physical objects). As far as grammatical meaning extensions, Hilpert claims that all of these have in common the metaphor objects are human as the rst conceptual mapping, and then body part terms are extended to mark spatial and temporal relations. He shows that although extensions onto grammatical meanings are less frequent than the lexical ones, most of the former extensions consist of a series of mappings, constituting grammatical processes (back with its extensions behind and after). Hilperts study conrms the notion put forward in previous research (Goosens 2002) that metonymies based on metaphors are infrequent. But, importantly, his data also suggest that there is a preference for initial metaphoric extension while chained extensions onto lexical meanings favour mainly metonymic processes. In her study of coercion in the construction of meaning, Debra Ziegler proposes that this notion might be more fruitfully thought of as falling within the scope of pragmatic reasoning and reanalysis (echoing some concerns about the notion of coercion already expressed by Elizabeth C. Traugott 2007), and thus as a factor contributing to grammatical change. In order to show that the concept of coercion may actually be superuous, Ziegler focuses on three representative groups of coercion: nominal; complement or subject coercion; and, aspectual coercion. She then provides a complete and compelling diering account for each of the three.

Book reviews Cognitive Linguistics 211 (2010)


For example, for nominal coercion, Ziegler oers explanations such as grammaticalization of the indenite article, which, from Middle English onwards, began to encroach upon contexts in which a mass noun would have been used previously. As far as mass-to-count coercion, she oers a plausible account of metonymy at work (She had a beer.) and count-tomass coercion (There was rat all over the place.), aided by dierences in construal of quantity boundedness, or the absence of determiners signaling boundedness. With equally competent arguments, Ziegler tackles the notion of coercion of complements and shows how the meaning is metaphorically constructed due to activation of an ICM in each case. To nalize this section, she examines aspectual coercion and concludes that this type of so-called coercion is the synchronic recognition of a prolonged and developing diachronic process (p. 117), whereby imperfective aspectual functions sometimes take on stable functions of some OE participles; she then compares these uses to McDonalds slogan Im lovin it. As noted by both Ziegler and Traugott (2007), there appears to be a present-day expansion of stative verbs used with a progressive aspect with a durative interpretation. Throughout her discussion, Ziegler examines not only what happens but also why certain processes occur, i.e., the uses that real speakers make of metonymical adaptation. For this, she convincingly appeals to processes of metonymic changes (including a discussion of the limits of constructive meaning according to the ease of retrieval) and diachronic changes. Since the notion of coercion cannot be used to interpret metonymy because such models often involve conversational implicatures, Ziegler concludes that the term of coercion is redundant. In chapter 6, Brdar and Brdar-Szabo return to a topic previously examined by Barcelona (2004)that of metonymic meanings of proper nouns denoting humans, such as Sarkozy is the Zidane of Finance. They attempt to identify the metonymic processes, involving several tiers, and then the metaphoric steps in the construction of meaning with the use of proper names. They base part of their work on Barcelonas previous studies on the paragon as a metonymic model and come to the conclusion that this model is not suciently motivated. They argue that these meanings are not rigidly xed but rather depend on complex matrices of domains and are also conditioned by the context of use. This comment again brings to mind Gibbs (2006) advice to cognitive linguists on the need for independent evidence from experimental studies on the accessing and interpretation of metonymies and metaphors. In regard to which properties are metaphorically projected between semantic domains, Anatol Stefanowitschs paper, titled Collocational overlap can guide metaphor interpretation, provides statistical data on


Book reviews Cognitive Linguistics 211 (2010)

language processing. He examines the collocational overlap between literally and metaphorically used lexical items that co-occur in metaphorical expressions. He argues that the collocates with the highest combined probability show which properties (in this case, adjectives which may constitute the conceptual core of the node word, such as man, as metaphorically associated with wolf , gorilla and pig) would most likely be transferred from the source to the target domain. In order to ascertain the highest ranking lexical items, Stefanowitsch uses a three-part series of meticulous methodological steps to search in 3.uk4 and 3.us4 websites, with the results being given dierent weightings for the overlaps according to three variants: i) symmetrical (equal weighting to collocations in the source and target domains); ii) source-dominant collocate weighting; and, iii) target-dominant collocate weighting. He rst investigates with a frequency-based denition of collocation, and then with an association strength-based denition, which provided a slight improvement over the previous method. Even after this painstaking labor, Stefanowitsch readily admits to some remaining methodological dilemmas: a) the possibility that the association-based model may not be suciently sophisticated to be able to eliminate irrelevant adjectives as collocates; b) perhaps the strengths used in the comparison (based on Leech et als study of word frequencies in the BNC, 2001) did not reect the frequencies found in the Internet corpus collected by this researcher; and, c) the text types collected may not have been representative (i.e. balanced). Furthermore, perhaps the most problematic drawbacks of the collocation overlap model is that even such laborious studies do not prove that this model has correctly identied the same attributes as human speakers would. To remedy the possible lack of psychological reality, Stefanowitsch then carried out a small-scale informant test in which the 12 native-speaker participants chose the most relevant adjectival attributes metaphorically associating man to wolf , gorilla, pig and one non-conventional association that man is a dolphin. Stefanowitsch unassumingly admits that there still exist many problems with the application of both the collocational overlap model, whether based on frequencies or associationstrength. However, given the all too frequent assumptions of some authors in associating their own cognitive processes with those of the population at large, i.e., without taking into account the question of psychological reality, Stefanowitsch seems to be almost too modest about his methodological procedures, which, at the very least, can be said to attend to a host of criticisms that have long plagued studies in cognitive linguistics (Gibbs 2006). Section II of the volume groups together papers on mental spaces and conceptual blending. It opens with Ronald Langackers Constructing the

Book reviews Cognitive Linguistics 211 (2010)


meanings of personal pronouns, in which he returns (Langacker 1991a: 226230) to the notion of subjectivity/objectivity, i.e. dierent possible construals of the speaker as included in the scene. Langacker rst alludes to a supercial account of I as designating the speaker, you as designating the hearer and they as designating a group which does not include either of the former. He then builds up his arguments towards the much more sophisticated and dynamic model of high-level blending (as when rst and second person pronouns incorporate both the speech scenario and the general viewing scenario (p. 183), i.e. both speaker and hearer are on stage in the blended spaces. To explain this blending, he refers to reference point phenomena by showing how for the rst and second person pronouns the reference point (salient entity) is also the target and for the underspecied third person pronouns the reference point has been identied by the current discourse space, all the information available to the interlocutors. Langacker also provides an interesting discussion of how personal pronouns functioning as impersonal pronouns have essentially the same meanings as they do in personal uses. After a brief discussion of we, you and they, Langacker turns to the underspecication of the pronoun it, whose vague reference comprehends the scope of awareness as an undierentiated abstract whole, as in expressions like it is obvious that . . . . The nal section of the paper returns to the proled referents of I and you and the construal of objectivity/subjectivity. Langacker suggests that the ambivalent nature of the two pronouns in regard to subject vs. object conception is actually part of their meaning. Thus, he provides a compelling account of how cognitive linguistics can explain intersubjectivity (apprehension of other minds and what they apprehend, p. 182). His contribution to this volume provides the necessary additional linguistic account of sense boundaries to Rumelhards psychological explanation of the role of schemata in the construction of intersubjective understanding. Kiki Nikiforidous article on The construction of meaning in relative clauses also deals with underspecication regarding Greek relative clauses introduced by pu, which frequently show indeterminacy in the way(s) the meaning of the head is integrated with the content of the relative clause (p. 190), i.e. I am the woman that the ght took place [for]. Since the relativizer here should function as the complement of a preposition, which is absent, the hearer must engage in a certain amount of semantic/pragmatic interpretation. The missing preposition might be for, over, with or against and, although these missing prepositions are lexically determined by the verb ght, the context is crucial in resolving which of the possible prepositions might be the adequate interpretation of the blended space, which allows for multiple construals.


Book reviews Cognitive Linguistics 211 (2010)

Nikiforidou analyzes her examples from informal speech in terms of blending, in which the relative construction requires a conceptual integration of the meaning of the antecedent with that of the relativizer. She also examines a type of gapless relative clause, i.e., those in which there is no gap corresponding to the head noun, but instead a pronoun which cannot be considered resumptive. The example Shes preparing in the kitchen some dishes that you are crazy [for] might be analyzed as simply missing a preposition, but a better explanation of how people make sense of these two clauses is to consider the relativizer pu as strongly priming a pragmatically evoked causative relationship emergent from the blended space. In the nal section of the article, Nikiforidou discusses the dierent types of constraining principles which may aect possible interpretations and she plausibly concludes, at least as far as the Greek pu relative clause is concerned, the constraints are predominantly those that concern interpretability. In the article comprising chapter 10, Christian Koops addresses the cross-linguistic variability of inferential constructions (it is that) and nds, as have other researchers, that dierent languages licence the use of this construction in a much less restricted way than in English, for example, inferences can be activated by the environment rather than triggered by the linguistic context, as is the case with the English construction. These restrictions for English lead him to investigate, with the use of real spoken data instead of constructed examples, how discourse constraints reduce the applicability of this construction to contexts in which inferences are highly accessible. Further constraints for this construction in English involve certain types of modiers such as the adverb just or certain grammatical contexts such as negation (its not that), which act as facilitators for the hearers presuppositions that a unmentioned value (more specic reason) is relevant to the current discourse. In the case of negation, the it is not that-construction presents a presupposed reason that is then rejected and when this construction occurs with an epistemic modal (it may be that), the clause is presented as the result of a process of deduction (p. 219). Koops predicts that languages like English, in which the unmarked form of the construction rarely occurs, will also show the it is that-construction in combination with modiers and grammatical constructions which reinforce its specicational function (p. 223). Languages which do not show such a restrictive use of the construction, like Spanish, will not make use of the modiers and specic grammatical contexts. He nishes his interesting discussion of the it is that-constructions in various languages by proposing that more research be carried out on a wide range of languages so that his hypothesis can be conrmed.

Book reviews Cognitive Linguistics 211 (2010)


In The construction of vagueness: Sort of expression in Romance languages, Wiltrud Mihatsch uses data from two Germanic languages (English and German) and four Romance languages (French, Italian, Portuguese and Spanish) to study two paths of pragmaticalization involved in emerging stages of taxonomic nouns towards approximative ` modiers, i.e., loose meaning constructions, such as espece de from es` pece, species. With the use of dictionaries, corpora and internet sources, the author provides detailed information on the implicatures which have triggered the pragmaticalization of discourse particles, which, up to now, have only been studied more comprehensively for English and French. She meticulously traces the historical determinologization, caused by dierent communicative needs and background knowledge of speakers and hearers in everyday contexts as opposed to scientic communities (p. 229). The more peripheral members of subcategorization are those for which there arises a need for a qualier in order to permit inclusion into a specied class. As this use becomes more and more entrenched, the original meaning of the taxonomic nouns leads to enhancement of the possible meanings of the NP, whereby negative quantication and free choice may be reinforced and therefore, serve as a basis for grammaticalization processes of indenite pronouns; these, then, may become universal quantiers. Mihatsch uses dictionary entries (Corde, DHLF, OED, etc.) to tie all these data to a productive type of metonymy in which the scalar endpoint is used to emphasize the whole scale (p. 230). The sort of expressions (meaning species) developed a new approximative function and began to work as modiers of the following noun, which is usually a specic or abstract noun, not a central basis-level noun. There is one other path of pragmaticalization of modiers in Romance languages (but not in Germanic languages); this one involves two nouns, the rst, generic one, postmodied by a specifying taxonomic NP. These equivalents of genus become prepositional-like elements (in French genre, in Spanish, del genero de . . . , like). The study shows how diachronic data can adequately motivate the pragmatic changes in sort of expressions over time. Chapter 12, titled Communication or memory mismatch? Towards a cognitive typology of questions, proposes that questions are reexes of memory mismatches, a process which occurs when one is processing stimulus from the outer world. Wolfgang Schulze oers an account of the heterogeneity of interrogative constructions and reects on the claim of cognitive linguistics, according to him, that many variations in form and function are variants of a single cognitive strategy (p. 248). Conceptualization is considered as inherently dynamic (Langacker 1991b, 2001) and although the processes may involve many emergent factors, Schulze


Book reviews Cognitive Linguistics 211 (2010)

hypothesizes that, as far as interrogativity, both the linguistic and conceptual variation ultimately derive from a supercially simple cognitive strategy that has to do with memory mismatch (p. 250), more specically with memory and linearization (i.e., resolution of mismatch and alignment). That is, when one is construing a given world stimulus, there also occurs an activation of stored analogies, which can coincide with the new stimulae or not. If there is coincidence, there is an assertion; noncoincidence results interrogativity, involving various mechanisms (phonological, morphosyntactic, etc.) for constructing either polar questions (address the givenness of the world stimulus) or constituent questions (address a specic part of the stimulus). Unfortunately, the article seems a bit dis-jointed, jumping from a series of issues presented in one section, the up-take of which is not always clear in the following sections. For example, in one of the nal sections, the author gives a brief account of the bleaching of questions words, including tags, but it is not clear exactly what connection holds between this account and the discussions presented in previous sections. Schulze, from the framework of Radical Experientialism (p. 247), concludes that questions are not really part of a communication strategy but rather reactional patterns that reect the given state of cognition (p. 262). The nal chapter of the book, Brutal Brits and persuasive Americans. Variety specic meaning construction in the into-causative by Stefanie Wul, Anatol Stefanowitsch and Stefan Gries, aims at revealing the cultural dierences which may be reected in the lexical llers used in grammatical constructions. To do so, they use corpora from two prominent newspapers to examine the collexemes of verbs pairs which occur in the two varieties of English in into-causative constructions (talk into accepting or bully into accepting). In order to show the strength of association for both the cause slots and the results slots, the authors use a very promising statistical measure, called distinctive collexeme analysis (previously proposed and tested in various other articles, Stefanowitsch and Gries, 2003, 2005), and then, for the 3,908 pairs of cause-result predicates, they have applied the Fischer-Yates exact test for added reliability of signicance. They then group the distinctive collexemes into larger semantic categories (communication, negative emotion, physical force, etc., p. 273274). Their most signicant ndings, apart from testing the validity of the meticulous use of various measurement tools for strength of association, support notions proposed by Construction Grammar, namely the parallel between the pragmatic notions of interpretation and context and the lexico-syntactic notions of verb and construction (p. 277); that is, constructions seem to oer meaning potential in the nuances that are expressed by the lexical llers, which may be inuenced by the

Book reviews Cognitive Linguistics 211 (2010)


speakers cultural input (p. 278). In the verbs studied, the analysis shows the following for the two varieties. In British English, the data showed a predominance of the use of acts causing negative emotions (pressurize, bounce, panic, bully, etc.), denoting cause set into motion, and with no real verb of communication in this slot), and more verbs of communication in the result slot. For American English, the analysis revealed that the cause predicates denote a restriction of movements for the causee (talk, pressure, prod, coax, etc.), with a predominance of light verbs for the result predicates. The authors propose an interpretation, albeit acknowledging the need for more research to verify their ndings: the causeslot appears to indicate a distinctive use of threatening actions in British English, while the result-slot predicates denote a confession frame (p. 279); in American English, the cause-slot appears to indicate a distinctive preference for verbs of communication or verbal persuasion, while the result-slot predicates indicate some type of unspecied action (light verbs). The sophisticated analytical tools used by the various authors provide a welcome addition to ways of attending to some of the criticism that cognitive linguistics has received. These extra-linguistic measures are called for by Gibbs in chapter one, although these authors use linguistic (corpus analysis) and statistical measurements and not psychological ones. And by showing that speakers cultural inputs can be reliably traced, they provide a bridge to work being carried out in critical discourse analysis, thereby establishing a possible common ground for these two rapidly developing elds in modern linguistics (Stockwell n/d: 1). In summary, Radden, Kopcke, Berg and Siemund have edited a useful body of research on metaphor, metonymy and blends, which is valuable in bringing readers up-to-date on recent developments. However, Raymond Gibbs point in the initial chapter is well taken: there is still a lag between theoretical and empirical studies. Without corpus studies, including historical ones, we cannot be sure of how ubiquitous metonymic and metaphoric expressions are; our speaker intuitions are often notoriously incorrect. Corpus studies, besides providing real examples and not constructed ones, may also reveal signicant speaker dierences in patterns of use across language varieties. And, without experimental studies, we cannot be sure that our after-the-fact cognitive modeling of speakers rhetorical construals actually do correspond to their on-line processes, be they in comprehension or production. Thus, the contribution of this book may not lie so much in the aspects studied (although these certainly add to our knowledge of the production and reception of meaning), but rather this volume distinguishes itself in that many of the articles presented here provide a clear pathway to more reliable measurement of cognitive processes.


Book reviews Cognitive Linguistics 211 (2010)

Ausubel, D., J. Novak & H. Hanesian. 1978. Educational psychology: A cognitive view, 2nd edn. New York: Holt, Rinehart & Winston. Barcelona, A. 2002. Clarifying and applying the notions of metaphor and metonymy within cognitive linguistics: An up-date. In R. Dirven & R. Porings (eds.), Metaphor and meton ymy in comparison and contrast, 207278. Berlin & New York: Mouton de Gruyter. Barcelona, A. 2004. Metonymy behind grammar: The motivation of the seemingly irregular grammatical behavior of English paragon names. In G. Radden & K-U Panther (eds.), Studies in linguistic motivation, 357374. Berlin: Mouton de Gruyter. Bartlett, F. C. 1932. Remembering: An experimental and social study. Cambridge: Cambridge University Press. Bruner, J. 1990. Actual minds, possible worlds. Cambridge MA: Harvard University Press. Fesmire, S. 1994. What is cognitive about cognitive linguistics? Metaphor and Symbolic Activity 9(2). 149154. Gibbs, R. 1993. Process and products in making sense of tropes. In A. Ortony (ed.), Metaphor and thought, 2nd edn, 252276. Cambridge: Cambridge University Press. Gibbs, R. 2006. Cognitive linguistics and metaphor research: Past successes, sceptical questions, future challenges. DELTA 22. Goosens, L. 2002. Metaphtonymy: The interaction of metaphor and metonymy in expressions for linguistic action. In R. Dirven & R. Porings (eds.), Metaphor and metonymy in comparison and contrast, 349377. Berlin & New York: Mouton de Gruyter. Johnson, M. 1987. The body in the mind. The bodily basis of meaning, imagination, and reason. Chicago & London: Chicago University Press. Lako, G. 1987. Women, re and dangerous things. What categories reveal about the mjind. Chicago: Chicago University Press. Langacker, R. 1991a. Concept, image and symbol. The cognitive basis of grammar. Berlin & New York: Mouton de Gruyter. Langacker, R. 1991b. Foundations of Cognitive Grammar, vol. 2, Descriptive application. Stanford: Stanford University Press. Langacker, R. 2001. Dynamicity in grammar. Axiomathes 12. 733. Leech, G., P. Rayson & A. Wilson. 2001. Word frequencies in written and spoken English: Based on the British National Corpus. London: Longman. Mandler, J. 1984. Stories, scripts, and scenes: Aspects of schema theory. Hillsdale, NJ: Erlbaum. Mandler, J. 1992. How to build a baby, II, Conceptual primitives. Psychological Review. 587604. Markert, K. & M. Nissum. 2003. Corpus-based metonymy analysis. Metaphor and Symbol 18. 175188. Michaelis, L. 2003. Headless constructions and coercion by construction. In E. Francis & L. Michaelis (eds.), Mismatch: Form-function incongruity and the architecture of grammar (Lecture Notes 163), 259310. Stanford: CSLI Publications. Neisser, U. 1976. Cognition and reality. San Francisco: W. H. Freeman. Panther, K-U. 2005. The role of conceptual metonymy in meaning construction. In F. Ruz de Mendoza & S. Pena Cervel (eds.), Cognitive Linguistics: Internal dynamics and interdis ciplinary interaction, 353386. Berlin & New York: Mouton de Gruyter. Panther, K-U. & L. Thornburg 1998. A cognitive approach to inferencing in conversation. Journal of Pragmatics 30. 755769. Panther, K-U. & L. Thornburg (eds.). 2003a. Metonymy and pragmatic inferencing. Amsterdam: John Benjamins.

Book reviews Cognitive Linguistics 211 (2010)


Panther, K-U. & L. Thornburg 2003b. Metonymies as natural inference and activation of schemas. In K-U. Panther & L. Thornburg (eds.), Metonymy and pragmatic inferencing. Amsterdam: John Benjamins. Panther, K-U. & L. Thornburg 2004. The role of metonymy in meaning construction. 6. 91116. Rumelhart, D. E. 1980. Schemata: The building blocks of cognition. In R. J. Spiro, B. Bruce & W. F. Brewer (eds.), Theoretical issues in reading and comprehension. Hillsdale, NJ: Erlbaum. Shank, R. C. 1977. Scripts, plans, goals and understanding. Hillsdale, NJ: Erlbaum. Stefanowitsch, A. & S. Gries. 2003. Collostructions: On the interaction between verbs and constructions. International Journal of Corpus Linguistics 8(2). 209243. Stefanowitsch, A. & S. Gries. 2005 Covarying collexemes. Corpus Linguistics and Linguistic Theory 1(1). 143. Stockwell, P. n/d. Towards a critical cognitive linguistics? Thornburg, L., K-U. Panther & A. Barcelona (eds.). 2006. Metonymy and metaphor in grammar. Amsterdam: John Benjamins. Traugott, E. C. 2007. The concepts of construction mismatch and type-shifting from the perspective of grammaticalization. Cognitive Linguistics 18 (4). 523557. Verhagen, A. 2005. Constructions of intersubjectivity: Discourse, syntax and cognition. Oxford & New York: Oxford University Press. Vygotsky, L. 1978. Mind in society. Cambridge, MA: Harvard University Press.

Space, language, and cognition: New advances in acquisition research1


Abstract In this introductory chapter to the present special issue about Space, language and cognition: developmental perspectives, we introduce some of the main questions that are currently debated concerning the relationships between cognitive and linguistic representations in the domain of space. This collection of papers addresses these questions by bringing together contributions from dierent disciplines, theoretical perspectives, and methodological approaches. All papers start out with the assumption that spatial cognition is not indierent to spatial language and aim at specifying how the two might be best related by examining the development of spatial representations in children and adults through language use and acquisition. Keywords: spatial language, spatial cognition, co-verbal gestures, language acquisition, cognitive development, cross-linguistic comparisons, motion, typology.


Why space?

The ability to represent and to communicate spatial information accurately and rapidly is vital for the survival of all species, for example

1. The project to put together this special issue initially arose during a paper session on Spatial cognition and its expression in language and gesture: a developmental view organized in Munich by Katrin Lindner (LMU Munich) and Heike Behrens (University of Basle) in the context of the Second International Conference of the German Cognitive Linguistics Association (57 October 2006 check ICCLS). The names of the three editors are listed in alphabetical order. * Address for correspondence: H. Hendriks Research Centre for English and Applied Linguistics University of Cambridge English Faculty Building, 9 West Road, Cambridge CB39DP UK. Email: Cognitive Linguistics 212 (2010), 181188 DOI 10.1515/COGL.2010.006 09365907/10/00210181 6 Walter de Gruyter


H. Hendriks, M. Hickmann and K. Lindner

enabling individuals or groups to ee from danger, to nd food or to return home. A growing number of studies have examined this domain in various disciplines of the cognitive sciences that have brought complementary contributions stemming from dierent scientic traditions and methodologies (linguistics, cognitive psychology, developmental psychology, psycholinguistics, philosophy, neurosciences . . . ). This research has aimed at studying spatial behaviour in order to uncover the underlying processes involved in the construction of internal spatial representations, the resulting spatial categories that guide individuals decisions, and the neural substrata that underlie spatial cognition. Studies have also examined how spatial representations evolve over time from dierent perspectives: across species from a phylogenetic perspective, across languages from a diachronic perspective, and during child development from an ontogenetic perspective. In comparison to other species, a specic and remarkable feature of our spatial cognition is the potential role of human language in determining properties of our mental spatial representations. Language provides a unique and powerful symbolic system that constitutes one of the means whereby we construct and categorize space. One of the central questions currently addressed in the cognitive sciences has been to determine the properties of spatial systems across languages of the world and the dierent ways in which they might contribute to how we construct space. In this respect, the last ten or twenty years have witnessed a blooming of studies in this domain focusing on major debates concerning the relationship between spatial language and spatial cognition. These debates revolve around a number of questions, for example: What are the universals of spatial language and cognition? Do various typological properties of linguistic spatial systems constrain our cognitive spatial representations? What is the degree to which dierent (perceptual, cognitive, linguistic) components of spatial behaviour are autonomous or interact? This collection of papers addresses some of these questions by bringing together contributions from dierent disciplines and perspectives. They all start out with the assumption that spatial cognition is not indierent to spatial language and aim at specifying how the two might be related. All papers are framed within a developmental perspective in which they examine the development of spatial representations through language use and acquisition. Taken together, this collection presents contributions that are representative of new advances in acquisition research and discuss dierent languages (English, French, German, Japanese), dierent phases of language acquisition (children, adolescents, adults), types of learners (rst and second language acquisition), and types of spatial representations (verbal productions, co-verbal gestures).

Space, language and cognition 2. What can spatial language tell us about spatial cognition?


Spatial language forms a system comprising means of talking about different aspects of the spatial universe that is available to our perception. This universe can be roughly described in terms of situations that dene two main sub-spaces: static space and dynamic space. The rst sub-space concerns the localization of entities in relation to other entities in space. It involves, for example, the expression of varied types of information in answer to questions such as Where is X?. This information varies across languages: spatial relations (X is in, on, under, above . . . Y ), posture (X lies, stands, sits . . . on Y ), and other related information (gure and ground properties, functional properties . . . ). The second sub-space concerns dierent types of motion. Motion may occur within a given general location (to run around in the garden) or it may imply a change from one location to another (to run away) and it can be carried out voluntarily by an agent (to run, to leave), be involuntary (to fall ) and/or result from the energy produced by an external force (to push something up). Irrespective of dierences among them, all languages provide means of encoding static and dynamic information, both of which are central to our spatial representations. The study of spatial language therefore constitutes one means of approaching the general nature of human spatial representations and the processes whereby they are constructed for varied purposes. Linguistic organization in any semantic domain inherently involves distinctions and relations among them that enter into a coherent network of discrete categories available to speakers when they construct their spatial representations. Several questions therefore arise in the context of current debates concerning the relationship between language and cognition. One fundamental set of questions that is presently debated concerns the relatively autonomous or interactive nature of dierent dimensions of our spatial representations. Research has begun to examine whether all languages follow similar organizational principles and the extent to which such common properties of linguistic systems might reect deeper universal properties of human spatial cognition. Various claims in linguistics have revolved around whether or not all languages share a common clear distinction between two autonomous domains: reference to entities and spatial reference. Research in neurosciences has even provided evidence in support for these separate systems in the brain (What/Where systems). In addition, cognitive psychology and psycholinguistics have typically studied either the perceptual/cognitive processes involved in the construction of spatial representation or the verbal processes that might contribute to these representations, both of which are presumed to be universal, but little is known about how these two sets of processes might be related.


H. Hendriks, M. Hickmann and K. Lindner

In this respect, the study of spatial language addresses a second related question concerning the universals of spatial representations and the typological constraints that might partially determine them. Linguistic research has indeed uncovered wide dierences across the spatial systems of languages that touch on varied dimensions. Thus, languages provide dierent markers of spatial relations, make dierent distinctions along several spatial dimensions, rely on dierent spatial reference systems, and display dierent grammaticalization and lexicalization patterns for the encoding of varied information relative to location or to motion. A growing number of psycholinguistic studies has begun to show striking cross-linguistic dierences in how speakers express spatial information, for example dierences in the types of information they select and in the ways in which they organize this information in speech. On the basis of such results, some authors have revived the Whoran hypothesis, according to which such dierences reect deeper dierences in cognitive organization. According to this view, behavioural dierences result from the ways in which each language (or language family) lters and channels incoming spatial information, thereby making some of this information more salient and more accessible to cognitive functioning. In contrast, others view such dierences as merely reecting supercial languagespecic properties that do not aect deeper universal properties underlying cognitive organization. In this second view, then, cross-linguistic behavioural dierences are only apparent in speech and do not tap underlying cognitive representations. Clearly, claims concerning the impact of language-specic factors on human spatial cognition must go beyond the study of speech behaviour in order to avoid the pitfalls of circularity. Although available results concerning speech production are highly revealing in suggesting the possibility that particular types of linguistic organization might determine particular types of cognitive organization, this possibility remains still somewhat hypothetical in the absence of complementary evidence directly concerning non-linguistic representations. Going beyond linguistic production requires the study of both verbal and non-verbal behaviours in order to determine whether both types of behaviours dier across language groups and whether both could be accounted for by languagespecic or typological constraints. In this respect, one way of supporting the Whoran claim might be to study the relationship between speech and co-verbal gestures. Gestures during speech production provide another window onto speakers representations in that they are produced on line in close temporal proximity to speech as well as through a dierent modality. The similarities and dierences between simultaneous verbal and co-verbal behaviour might therefore uncover properties that are

Space, language and cognition


common to both modalities and that both follow from predictions based on language properties. 3. Why acquisition?

In addition to descriptive studies of spatial systems in linguistics and to experimental studies of spatial cognition in general cognitive psychology, space is of great interest in several respects from a developmental point of view. Spatial cognition has been a central domain of study for cognitive developmental psychology, serving as a primary foundation of childrens intelligence in ontogeny. For example, during the 20 th century Piagets most inuential developmental theory proposed that cognitive development is based on sensori-motor activity through endogenous and biologically determined cognitive processes (accommodation, adaptation) that drive children through a series of universal stages ( pre-operational, concrete operations, formal operations). Thus, it is by grounding his representations on displacements and locations in space, as well on the manipulation of entities in the surrounding environment, that the child comes to construct the world in such a way as to reach rational forms of thought. Similarly, much previous developmental psycholinguistic research in the spatial domain has focused on presumably universal properties of semantic systems that might be reected in recurrent and gradual progressions during childrens rst language acquisition. For example, some dimensions in the spatial domain seem to be perceptually most salient to children and this salience seems to determine the ways in which they learn to use spatial devices (such as the order in which spatial prepositions are acquired). Since the end of the 20 th century, the development of new methodologies in psychology has led to a revolution in our knowledge about child development. In particular, infancy has been the locus of much debate concerning components of childrens knowledge that seems to be more precocious than expected by previous theories. Thus, we now know that infants during the pre-linguistic period display some surprisingly precocious knowledge about the world, comprising knowledge in a variety of domains that include space but also categories of entities, number, temporality, causation, and agentivity. In the spatial domain, infants as young as three months of age know about the properties of entities, about their spatial relations, and about the physical laws that account for their displacements in space. The interpretation of these results is still controversial and has led to several views of child development. Some view this precocious knowledge as organized in autonomous modules that are pre-programmed in infants biological heritage, while others view it as


H. Hendriks, M. Hickmann and K. Lindner

stemming from active perceptual and cognitive processes that occur very rapidly from birth on. With respect to language acquisition, all of these views propose more generally that children either inherit or develop percepts and concepts during the pre-linguistic period, on the basis of which they look for the adequate linguistic means of expression. This cognitive determinism, then, is the major force driving children to acquire language. At least two dierent views have been proposed, both of which consider that language plays a structuring role during cognitive development. Also during the 20th century, Vygotsky proposed that the advent of language results in a major reorganization of cognition whether on a phylogenetic or an ontogenetic scale. Since Vygotskys writings, more recent studies also conclude that language input and acquisition results in particular types of cognitive functioning, leading young children to search for certain regularities that they would not discover otherwise. Yet a dierent view has been that children are sensitive not only to the general properties of language, but also to the specic properties of their language from the earliest age on, leading them to construct particular categories and/or particular ways of organizing them. Another development in psycholinguistics focuses on the later stages of child development, increasing the temporal span beyond early stages. Thus, studies have begun to examine later stages of language acquisition (pre-adolescence, adolescence), beyond the point of biological maturation considered to be for some approaches a major determinant of early acquisition. This research examines language use either in natural situations (for example conversations) or in tasks involving dierent discourse types (narratives, argumentation). Results are enlightening in understanding complex language skills that are acquired during later phases of acquisition and nonetheless fundamental for learners to become competent native speakers. Finally, during the last twenty years, a growing number of studies have examined dierent types of learners in order to shed further light on the process of language acquisition. The most common type of comparison has involved childrens rst language acquisition and adults second language acquisition. Such comparisons make a contribution to our understanding of language acquisition by allowing us to separate factors that are normally confounded during child development. Thus, the child displays an increasing maturity in all domains (growing cognitive and linguistic maturity), whereas the adult learner comes to the task of second language acquisition with a mature cognitive system. By way of implication, such comparisons also allow us to understand how language and cognition relate across linguistic systems.

Space, language and cognition 4. Language and cognition: present and future directions


The dierent papers in this special issue touch on all of these questions and open new directions for future research. A rst series of papers examines rst language acquisition in a cross-linguistic perspective. In their paper on Typological constraints on the acquisition of spatial language in French and English, Hickmann and Hendriks compare childrens early spontaneous utterances about motion events in English and in French on the basis of experimental and longitudinal corpora collected from the emergence of language on. The paper by Ochsenbauer and Hickmann on Childrens verbalizations of motion events in German presents comparative analyses of experimentally elicited productions about voluntary motion. It discusses German data (adults and children between 3 and 11 years) in comparison to previous results in French and English. The paper by Gullberg and Narasimhan on What gestures reveal about the development of semantic distinctions in Dutch childrens placement verbs examines the gestures produced by Dutch adults and children (of 45 years) in parallel with their uses of placement verbs. Three papers then focus on late acquisition phases. In their paper on Changes in L1 encoding of path after exposure to an L2, Brown and Gullberg analyze how adult Japanese speakers acquiring English as a second language with dierent levels of prociency talk about motion, discussing the eects of L2 on the expression of path in L1. In Im fed up with Marmite, Im moving on to Vegemite Graf discusses childrens (6 to 10 years of age) uses of spatial language in spontaneous conversations, including static and dynamic references, as well as literal and non-literal uses. Finally, the article by Lemmens and Perrez On the use of posture verbs by French-speaking learners of Dutch: a corpus-based study presents the results of a quantitative and qualitative corpus study of the use of the Dutch posture verbs staan (stand), liggen (lie) and zitten (sit) by French-speaking learners of Dutch. As implied by the discussions across these papers, at least two such questions will clearly deserve attention in future research. The rst one concerns the relationships that may exist among processes that take place across semantic domains during language acquisition. In this respect, although this collection of papers focuses specically on spatial language and cognition, many questions concerning the relationships between space, time, and the properties of entities remain surprisingly unattended. Second, perhaps the most dicult challenge to meet in the future concerns the extent to which language has an impact on speakers internal representations. Much more research is still necessary to address this question, using a variety of situations and tasks (categorization of events and of relations, eye movements during the exploration of scenes, verbal


H. Hendriks, M. Hickmann and K. Lindner

and visual memory). It is only through a variety of such behavioural measures that we can hope to uncover whether and how our spatial representations are inuenced by language (and by specic languages) both in communication (verbal representations) and in problem-solving situations (mental representations). As shown by the present volume, studying the relationship between language and space is a task that requires multiple methods in an interdisciplinary perspective. It is our hope that the volume will contribute to promoting future studies of this kind by bringing together papers that cover both typological and developmental questions, by studying a variety of languages, and by relying on varied methodologies. It is only through the joint contribution of multiple perspectives and tools that we will ultimately make significant advances in the study of spatial cognition, of spatial language, and of the relationship between the two.

Typological constraints on the acquisition of spatial language in French and English


Abstract Typological analyses (Talmy 2000) show that languages vary a great deal in how they package and distribute spatial information by lexical and grammatical means. Recent developmental research suggests that childrens language acquisition is constrained by such typological properties from an early age on, but the relative role of such constraints in language and cognitive development is still much debated (Bowerman 2007; Bowerman and Choi, 2003; Slobin 1996, 2003a, 2003b, 2006). In the context of this debate, we compare the expression of motion in two data bases of child English vs. French: 1) experimentally induced productions about caused motion (adults and children of three to ten years); 2) spontaneous productions about varied types of motion events during earlier phases of acquisition (18 months to three years). The results of both studies show that the density of information about motion increases with age in both languages, particularly after the age of ve years. However, they also show striking cross-linguistic dierences. At all ages the semantic density of utterances about motion is higher in English than in French. English speakers systematically use compact structures to express multiple types of information (typically MANNER and CAUSE in main verbs, PATH in other devices). French speakers rely more on verbs and/or distribute information in more varied ways across parts of speech. The discussion highlights the joint impact of cognitive and typological factors on language acquisition, and raises questions to be addressed in further research concerning the relation between language and cognition during development.
* Address for correspondence: M. Hickmann, CNRS Laboratoire Structures Formelles du Langage, UMR 7023, 59 rue Pouchet, 75017, Paris, France. Email: maya.hickmann@ H. Hendriks, Research Centre for English and Applied Linguistics, University of Cambridge, English Faculty Building, 9 West Road, Cambridge CB39DP, UK. Email: Cognitive Linguistics 212 (2010), 189215 DOI 10.1515/COGL.2010.007 09365907/10/00210189 6 Walter de Gruyter


M. Hickmann and H. Hendriks rst language acquisition, French and English, longitudinal and experimental data, typology, path, manner, and cause.




A highly debated question in current research on child language concerns the relative impact of universal versus language-specic determinants in rst language acquisition. In this respect, the spatial domain has been of particularly interest. Space is a most basic domain of human cognition, traditionally presumed to be governed by universal principles and capacities, but nonetheless showing considerable variations across human languages. Such variations concern the linguistic expression of static space (e.g., spatial relations, location) and of dynamic events (e.g., motion, changes of location). Previous developmental research in this domain has led some researchers to postulate that particular typological properties may inuence the rhythm and course of childrens language acquisition, and perhaps even the ways in which they develop and organize spatial concepts. In the context of this debate, we present below two studies concerning the expression of motion events in English versus French, focusing on how adults and children between three and ten years of age express caused motion in experimentally controlled situations (Study 1), as well as on how young children express all kinds of motion events in early spontaneous productions between the ages of two and three years (Study 2). Both studies examine the semantic content denoted in relation to motion and the devices used to express this information, with particular attention to the impact of both typological constraints and cognitive determinants on both aspects of childrens utterances.

2. 2.1.

Motion in language Universals and cross-linguistic variability

Linguistic research on spatial systems shows that languages vary a great deal in how they represent spatial information. With respect to motion events, Talmy (2000) proposes that they fall in dierent families depending on how they package and distribute spatial information by lexical and grammatical means. Satellite-framed languages (e.g., Germanic) typically encode the manner of motion in the verb root and its path in verbal satellites, whereas verb-framed languages (e.g., Romance) typically encode path in the verb root and express manner (if at all) by additional means at the periphery of the clause (examples (1) and (2)).

Typological constraints (1) (2)


She runs/crawls . . . up, down, away, across, into, out of . . . ` Elle monte, descend, part, traverse, entre, sort . . . en courant/a quatre pattes . . . . (She ascends, descends, leaves, crosses, enters, exits . . . by running/ on all fours . . . )

These typological dierences correspond to strong paradigms that run through the language of motion leading to speakers strong preferences for particular means of expression in default cases, despite a number of other available options, such as options that are more marked pragmatically (e.g., those emphasizing manner and/or path in French (3)), that result from borrowings (e.g., English Latinate (4)), or that are remnants from diachronic linguistic evolution (e.g., French verbal prexes such as (5)).1 (3) (4) (5) ` ` Elle a couru jusqua la maison a cloche-pied. (She ran all the way home on one foot.) to enter, to ascend, to descend . . . accoster (to reach the coast), ecremer (to take cream o ), em boter (to in-t) . . .

Furthermore, such preferences also have an impact on speakers preferences beyond the spatial domain, for example aecting the expression of causal relations, that are typically encoded by means of compact causalresultative constructions in Germanic languages (such as English (6)) vs. more distributed constructions in Romance languages (such as French (7)). (6) (7) She kicked the door open and kicked the dog out. ` Elle a ouvert la porte et fait sortir le chien a coups de pied. (She opened the door and made the dog go out by kicking). Motion in rst language acquisition


A large body of research has brought strong evidence showing that very young infants possess much knowledge about space. On the basis of an extensive review, Pruden et al. (2008) conclude that the body of evidence currently available suggests that infants are sophisticated observers of actions and relations, [ . . . ] and that the inherent problem in learning

1. As shown by Kopecka (2006), verbal prexes such as those in (5) are remnants from an earlier stage in the diachronic development of French which evolved from a predominantly satellite-framed system (old French) to a predominantly verb-framed system (contemporary French).


M. Hickmann and H. Hendriks

[spatial] relational terms therefore appears not to be with conceptualizing events and actions in the world, but rather with mapping words onto actions (id: 10). However, as mentioned above, recent developmental research shows that childrens performance in language production and comprehension is constrained by typological properties from an early age on. The role of such constraints in language acquisition and more generally in cognitive development is presently vividly debated. According to some authors, language-specic properties have a deep impact on cognition, inuencing the rhythm and course of language acquisition, as well as the ways in which speakers of all ages from infancy to adulthood select and organize information when planning their utterances (Slobin 1996, 2003a, 2003b, 2006) and/or when constructing their cognitive categories (Bowerman 1996, 2007; Bowerman and Choi 2003; Choi and Bowerman 1991; Lucy and Gaskins 2001). Other authors propose that language-specic effects on cognition occur only at a later age (around 7 years), when the language is assumed to be fully settled as a system (Hohenstein 2005). Yet others view such eects as relatively supercial, eventually aecting verbal behaviors in some situations, but with no deeper implications for non-verbal cognition, itself assumed to follow universal progressions governed by language-independent perceptual and cognitive constraints (Clark 2003; Landau and Lakusta 2006; Munnich and Landau 2003). In the context of these debates, little information is available concerning the expression of motion in French child language, despite the interesting status of this language from a typological point of view (Kopecka 2006, see Note 1). Our previous research (Hendriks et al. 2008; Hickmann 2003; Hickmann and Hendriks 2006; Hickmann et al. 1998; Hickmann et al. 2009; Ochsenbauer and Hickmann 2008) compared how adults and children (between three and ten years) expressed motion in French vs. other languages in controlled experimental situations. General results show rst that utterances about motion are semantically less dense in French than in other languages (English, German, Chinese). In addition, although all languages show an increase with age in the semantic density of utterances about motion (voluntary or caused), this developmental progression is much more striking in French than in other languages. Finally, French utterances about motion are much more variable as a function of age and of event type in comparison to other languages where response patterns are rather systematic. When describing voluntary motion (Hickmann et al. 2009), French adults express both manner and path in some situations (e.g., with crossing events, such as a path verb and a manner phrase (8)), but they do not do so systematically. French children typically focus either on path or on manner, expressing one or

Typological constraints


the other type of information in verb roots (e.g., path in (9), manner in (10)), but rarely expressing both simultaneously within the same utterance. A notable exception concerns some descriptions of upwards motion at all ages (e.g., in (11) the verb grimper to climb up lexicalizes manner and path). In sharp contrast, most utterances in other languages simultaneously encode manner (in verb roots) and path (in satellites), regardless of age and of event type. (8) ` ` Alors le petit garcon traverse la riviere a la nage . . . (FAD01) (then the little boy crosses the river by swimming [at/with a swim] . . . ) et qui ensuite redescend le long du brin dherbe . . . (FAD01) (and who then descends back along the blade of grass . . . ) il y a un garcon qui nage (F0618) (theres a boy who is swimming) ` alors le petit ecureuil grimpe a larbre . . . (FAD01) (then the little squirrel climbs up [at] the tree . . . )

(9) (10) (11)

The dierence that was observed with respect to overall semantic density follows from language-specic properties. As suggested by previous studies (cf. Slobin 1996), it is easier to stack semantic components in satellite-framed languages such as English, which encode path in satellites thereby freeing the verb to express manner, than in verb-framed languages such as French, which reserve the main verb for path information, making it necessary to use more linguistic tools to encode manner at the sentence periphery. However, the fact that density increases with age across all languages shows the role of cognitive developmental factors. It is cognitively less demanding for young children to process only one type of information than to process and to combine more. Finally, the combined eects of cognitive and language-specic factors explain the fact that utterance density undergoes a slower developmental progression in a verb-framed language such as French as compared to satellite-framed languages. French children must not only develop their general cognitive capacities, but also solve the problem of having to use additional linguistic tools in order to combine multiple semantic components. With respect to caused motion, two patterns emerge. In a rst experiment (Hickmann and Hendriks 2006), subjects had to describe object displacements carried out in front of them by the experimenter. In this situation English speakers of all ages typically used a general verb indicating caused motion combined with particles or prepositions encoding spatial information (e.g., to put on/into). In contrast, French adults used a great variety of specic verbs to express simultaneously caused motion and several other types of information, particularly manner of attachment, either


M. Hickmann and H. Hendriks

using no preposition at all or relying on general prepositions (e.g., accrocher le manteau [au portemanteau] to hook the coat [at the coathanger]). Although children generally followed the same pattern as the adults in their language group, young French children used fewer specic verbs and more specic prepositions than older children or adults (e.g., mettre le manteau [sur le portemanteau] to put the coat [on the coathanger]). In a second experiment (Hendriks et al. 2008) adult speakers of French and of English (as well as adult English-speaking learners of French) were asked to describe animated cartoons showing a man that displaced various entities. Native English speakers systematically encoded cause manner in the verb root and path in satellites (to push the ball down, to roll the ball across), while French speakers expressed all kinds of information in dierent parts of speech, showing little systematicity in their responses. These results are in line with the typological properties of French and with previous studies concerning other Romance languages (e.g., Spanish, see Berman and Slobin 1994), suggesting that French presents a dicult system to learn when it comes to simultaneously expressing multiple types of information about motion. The two studies reported below follow up on these previous studies. Study 1 used the same experimental paradigm as Hendriks et al. (2008) to elicit utterances about caused motion among children between the ages of three and ten years. The aim was to determine whether the language dierences that had been observed among adult speakers could also be observed among children. Study 2 then further examined childrens spontaneous productions at earlier ages (about two to three years) in order to pursue the question of language-specic determinants of acquisition from the emergence of language onward.

3. 3.1.

Study 1: Experiment on caused motion Method

3.1.1. Subjects. Adults and children of four age groups participated in the study (12 subjects per group, total N 60). Children were approximately three years (French mean 3;4, range 3;0 to 3;10, English mean 3;3, range 2;11 to 3;6), four years (French mean 4;5, range 3;11 to 4;10, English mean 4, range 3;7 to 4;5), ve years (French mean 5;5, range 4;11 to 5;10, English mean 5;1, range 4;7 to 5;5), and ten years (French mean 10;2, range 9;8 to 11;1, English mean 10;0, range 9;6 to 10;6). Children were tested in kindergartens and schools in Paris and Cambridge. Adults were university students (Universities of Paris 5 and of Cambridge).

Typological constraints
Table 1. Summary of main variables in test items for caused motion (Study 1) I. II. III. IV. V. Manner of A motion Cause Manner of cause Manner of O motion Path of motion


Manner in which Agent moved (walking in all items) Causal relation between Agent and Object (in all items) Action of Agent causing Object to move (push, pull) Manner in which Object moved (roll, slide) Trajectory followed by Agent and Object (up, down, across, into)

3.1.2. Materials. A total of 32 test items was designed (see summary in Appendix 1). They consisted of short animated cartoons in colour, all of which showed the same human agent (called Popi in French and Hoppy in English) in motion (hereafter A) carrying out an action that caused the displacement of an object (hereafter O). Table 1 summarizes all information components that were relevant to motion, two of which were constant across all items (I, II), while others systematically varied across items (III to V). The crossing of variables III to V resulted in 16 possible combinations (2 2 4), each of which was presented by means of two exemplars (resulting in a total of 32 items). The session began with an additional training item of the same type which ensured that subjects would be comfortable with the task and showed them the dierent types of information that were most relevant (corresponding to variables III to V), namely As action (manner of cause), the manner of Os motion (manner of o-motion), and the path followed by A and O (path). Finally, a set of seven distractor items also showed dierent types of situations without any scenery (grey background) nor any human agent. In these items a ball caused the motion of an inanimate entity (e.g., rolling into a book and causing it to move forward), sometimes also causing another result (e.g., a ball rolling into a vase, causing it to fall over and to break). 3.1.3. Procedure. Subjects were seen individually in their school or university setting. They were shown a series of animated cartoons and asked to tell what had happened in each. Target items were presented in four random orders to which subjects were attributed randomly. In all orders distractor items were interspersed at regular intervals among test items (one after every four test items). The following procedure invited participants to be as complete as possible. Young children (three to ve years) were presented a blindfolded doll and asked to tell her everything that had happened so that she could tell the story back. Older children (ten years) and adults were asked to narrate the cartoons for a ctitious listener who would not have seen the stimuli and would have to retell the stories only on the basis of the recordings.


M. Hickmann and H. Hendriks

3.1.4. Coding. The analyses below focus on responses that were elicited with test items (32 responses per subject, a total of 384 responses per age group in each language). The coding procedure aimed at providing three complementary measures of each response: 1) information focus, i.e., the particular types of information expressed in relation to motion within a given utterance, with particular attention to the dimensions that dened our stimuli (variables I to V in Table 1 above); responses that did not express any of these dimensions (e.g., verbs such as to go or to move, that merely expressed motion, but not any particular path, manner, or cause of motion) were coded as not expressing any relevant information; 2) the resulting overall utterance density, i.e., the total number of semantic components expressed within the utterance, which corresponded to one of four density categories (none, one, two, three or more components, hereafter SD0, SD1, SD2, SD3); 3) information locus, i.e., the particular linguistic devices used to express this information (main verbs vs. other devices). The examples below show the uses of main verb roots vs. other devices (in italics) to express dierent information types (in brackets) in responses that varied in semantic density: (12) expresses motion but contains none of the semantic components that were coded (I to V in Table 1) and it was therefore coded as a SD0 utterance; (13) and (14) contain one component, (15) to (17) two components, (18) and (19) three components. (12) (13) (14) (15) (16) (17) (18) (19) Thats a horsey. [ . . . ] Hes moving [none]. (E0302) Its a car [ . . . ] going up and up and up and up [path]. (E0301) Et bah il est monte [path] avec la bouee. (F0305) (And well he ascended with the swimming ring.) He was bringing [cause] the trunk down [path]. (E0507) Il a fait rouler [causemanner-of-o-motion] la roue. (F0305) (He made roll the wheel.) ` Et apres il a monte [causepath] le cadeau. (F0306) (And then he ascended the package.) He was pushing [CauseManner-of-Cause] the present up [Path] onto [Path] the house. (E1009) Il la avance [causepath] avec ses pieds [manner-of-cause]. (F0306) (He moved it forward with his feet.) Results


3.2.1. Utterance density. The overall semantic density of a given age group within each language was dened as the percentage of responses that fell in each density category (SD0, SD1, SD2, SD3) calculated

Typological constraints


Figure 1. Utterance density (Study 1)

over the total number of responses for that group.2 Figure 1 shows utterance density within each age group in English (Figure 1a) and in French (Figure 1b). SD0 utterances were rare at all ages (in either language). As expected, utterances were denser in English than in French and this language dierence could be observed at all ages. Almost all of the utterances produced by English adults contained three or more information components (SD3 92%). In comparison, French adults used such utterances to a lesser extent (SD3 79%) and produced more utterances of lower density (SD2 19%). As for children, the density of their utterances clearly diered across languages, particularly between three and ve years. At these ages English responses were frequently of density SD3 (between 42% and 54%) or at least of density SD2 (between 25% and 36%) and almost all of the ten-year-olds responses were of density SD3 (91%). In French, SD3 utterances mostly occurred at the age of ten years (58%), but SD2 utterances were frequent even at this age (32%). Furthermore,

2. The total possible number of responses for each age group in each language was 384 (12 subjects 32 items), except in rare cases where children did not respond appropriately (e.g., static location).


M. Hickmann and H. Hendriks

SD1 utterances were quite frequent at three years (61%) and only gradulally decreased thereafter (four years 40%, ve years 49%, ten years 10%). 3.2.2. Information focus. Each response was further examined with respect to the particular types of information components that were expressed. Information focus for a given age group within each language was calculated as the percentage of each type of information that was expressed over all semantic components found among all responses within that group. Figure 2 shows the distribution of information components expressed as a function of age in English (Figure 2a) and in French (Figure 2b). No notable dierence occurred in the adults utterances across languages. In both languages adults systematically expressed cause (English 32%, French 33%), path (English 34%, French 30%), and manner of cause (English 28%, French 27%), and they expressed manner of motion less frequently (English 6%, French 9%), mostly focusing in these cases on manner of O-motion and rarely on manner of A-motion. At all ages children generally followed the same pattern as the adults, especially in English. In French, however, a notable exception concerned the three-year-old children, who did not express manner of cause as often as

Figure 2. Overall information expressed (Study 1)

Typological constraints


other groups (3 years 5%, 4 years 15%) and who tended to express path more often than other groups (especially at 3 years 48%). 3.2.3. Information locus. For each response analyses also examined information locus in order to determine the means that were used in the utterance to express information relevant to motion (among the variables shown in Table 1). Particular attention was placed on whether cause, path, manner-of-cause, manner-of-motion (collapsing manner for A-motion and for O-motion) were expressed in main verb roots versus other devices (particles, prepositions, adverbials, subordinate verbs). Figure 3 rst provides an overall summary of how semantic information was distributed across these parts of speech overall (collapsing across all ages within each language). For each type of information expressed, this gure shows the percentage of cases where it was encoded in main verbs versus other devices. English presents a clear complementary distribution whereby cause and both types of manner (manner of cause and manner of motion) were almost always encoded in main verbs (between 95% and 98%), but path in other devices (99%). In contrast, French speakers frequently

Figure 3. Distribution of information across main verbs vs. other devices collapsing over all age groups (Study 1)


M. Hickmann and H. Hendriks

Figure 4. Information expressed by verbs vs. other devices in English (Study 1)

relied on main verbs to express motion information, but also distributed all information types across verbs and other devices including cause (72% and 28%), manner-of-cause (63% and 37%), manner-of-motion (80% and 20%), and path (62% and 38%). Figures 4 and 5 (for English and French, respectively) show in more detail which particular types of information (if any among the components in Table 1) were expressed in main verbs vs. other devices within each age group. For each age within each language information locus in main verbs was dened as the percentage of cases where each information type was expressed over all components found in main verbs for that group (also shown are cases where no component was found). The same procedure served to dene information locus in other devices for a given age group within each language, i.e., as the percentage of each information type expressed over all components found outside of main verbs for that group (also including cases where no component was expressed). English speakers (adults and children at all ages) relied on a systematic strategy, illustrated in (20), whereby they used main verbs to express manner-of-cause and/or cause (overall 38% and 50%), while using other devices to express path (overall 79%). In contrast, French speakers expressed all types of information in dierent parts of speech, despite

Typological constraints


Figure 5. Information expressed by verbs vs. other devices in French (Study 1)

a tendency to use main verbs to encode cause and path (overall 35% and 30%), as illustrated in (21) to (23). (20) (21) Hoppy pushed the suitcase down the hill. (E1004) Popi descend la colline en faisant rouler le ballon jusquen bas. (FAD02) (Popi descends the hill making roll the balloon all the way to the bottom). Il a tire le sac jusqu en haut du toit. (FAD02) (He pulled the bag all the way to the top of the roof.) Popi a fait rouler le gros pneu pour le rentrer dans le garage. (FAD02) (Popi made roll the big tyre to make it enter in[to] the garage.)

(22) (23)

The same patterns occurred across all age groups within each language, with the notable exception of the younger children (ages three to ve years), who did not always express any information outside of the main verb. In English, utterances of this type were more frequent at three years (40%) than at four and ve years (25% and 23%), and they practically disappeared at ten years and at adult age (2% in both cases). In French most utterances were of this type between three and ve years (three 89%, four


M. Hickmann and H. Hendriks

78%, ve 56%), and such cases were not rare even at ten years (24%), decreasing substantially only at adult age (9%). Utterances in which no coded information occurred outside of the main verb root were of two types, illustrated below (relevant information in italics and in brackets, also see more examples in (14) and (16) above): either 1) no additional devices were used at all (examples (24) and (25)); or 2) prepositional phrases occurred but they did not encode information about motion per se, for example referring to general locations (loc in (26) and (27)) or to accompanying entities (ae in (26)). (24) (25) (26) He is pushing [causemanner-of-cause] it. (E0303) Il a pousse [causemanner-of-cause] la table. (E0305) (He pushed the table.) ` Il est monte [path] sur le sable [loc] avec le sac a pommes de terre [ae]. (F0502) (He ascended the sand (dune) with the bag of patatoes.) Hes pulling [causemanner-of-cause] it on the road [loc]. (E0303) Summary



Speakers descriptions of caused motion events show dierent patterns in English vs. French. First, from the earliest age on, utterances were semantically denser in English than in French. In English, speakers (children and adults) expressed multiple types of information in compact structures when denoting motion. In contrast, French speakers of all ages produced utterances that were semantically less dense, although this density dierence was most striking among the younger children. Second, English speakers produced utterances that followed a very systematic pattern from the youngest age on, whereby they expressed cause and manner information in the main verb root and path information in satellites outside of the main verb. In comparison, French speakers were much less systematic with respect to the locus of spatial information in their utterances. At all ages they produced much more varied types of utterances in which they distributed information across dierent parts of speech. Although utterance density increased with age in both languages, this developmental progression was more striking in French than in English. These cross-linguistic dierences were expected and they indicate the impact of typological properties in English (satellite-framed) vs. French (verb-framed) on how children and adults expressed motion. We now turn to analyses of corpus-based spontaneous productions in order to determine whether the same language-specic eects can be observed during initial phases of acquisition.

Typological constraints 4. 4.1. Study 2: Longitudinal corpus of spontaneous productions at two to three years Method


4.1.1. Data base. The data base (summarized in Table 2) consisted of early spontaneous productions from four children, two French-speaking (Clara, Gregoire) and two English-speaking (Sarah, Adam from the Brown corpus in CHILDES, see MacWhinney, 2000). Children were recorded in natural settings during several developmental periods (P1 to P4), that were dened by mean length of utterance (mean MLU for P1 < 2,5; for P2 2,5-4; for P3 4,1-5; for P4 > 5).3 The analyses below focus on all utterances that expressed motion of any kind. For this purpose all utterances that contained an explicit motion verb were selected, resulting in subsamples of utterances that varied in size across children (see Table 2), i.e., very large for three children (over a thousand for Clara and Gregoire, close to two thousands for Adam) and smaller for one child (close to 300 for Sarah). 4.1.2. Coding. Utterances referring explicitly to motion were coded with respect to their semantic content. The list in (28) shows all of the information components that were expressed by children in relation to motion: path of motion, manner of motion, cause of motion, and manner of causing motion. General location was the only other type of information expressed in addition to this list.4 (28) Path of motion (path): direction (up, monter ascend), boundaries (into, entrer enter), deixis (come, venir) ` courir run, a quatre pattes on all fours bring, faire tomber make fall pull, pousser push

Manner of motion (manm): Cause of motion (cse): Manner of cause (manc):

3. All available French recordings were included in the corpora. The English corpora (borrowed from CHILDES) did not include any session corresponding to French period P4, but provided many more sessions for P1 to P3. A comparable subset of English recordings was therefore selected by 1) choosing the most comparable (lowest) ages in English within P1 to P3, since age was notably lower in French than in English for the corresponding mean MLUs of each period; 2) by randomly choosing every second or third recording among the remaining les of each period. Verb-less utterances were excluded (and decreased after P1 for all children). 4. Childrens spontaneous productions also contained some occurrences of verbs denoting specically changes of posture (e.g., sit down, sasseoir) which were not elicited in the experimental data.


M. Hickmann and H. Hendriks

Table 2. Early spontaneous productions: data base for all children (Study 2) FRENCH Clara Period 1 Mean MLU Mean Age* # of sessions Period 2 Mean MLU Mean Age # of sessions Period 3 Mean MLU Mean Age of sessions Period 4 Mean MLU Mean Age # of sessions Total utterances about motion Gregoire Sarah ENGLISH Adam

1,9 1;9 12 3,3 2;9 8 4,7 3;3 8 5,3 4;1 12

1,9 1;9 4 3,2 2;3 5 4,6 2;7 8 4,9 3;4 8

1,9 2;6 12 3,1 3;5 12 4,1 4;6 6 n.a.**

2,1 2;4 6 3,4 3;2 12 4,5 4;6 12 n.a.**





* Ages are shown in years;months. ** Not applicable (see Note 4).

Each utterance about motion was also coded with respect to its overall semantic density. Overall utterance density was dened as the total number of dierent information types that were expressed within each utterance (among the types listed under (28) above).5 Finally, each utterance was coded with respect to information locus in such a way as to identify whether the information components listed under (28) were expressed in main motion verbs or in any other co-occuring device within the utterance (particles, prepositional phrases, adverbial phrases). Examples (29) to (37) illustrate the coding of motion information in dierent loci (indi-

5. Although general locations were coded, they were not considered to encode information that is inherent to motion per se. Since the measure of semantic density aimed at determining which dierent types of information were expressed, utterances expressing the same information content more than once were coded for this content only once, e.g., ` manner in Il court [manm] a quatre pattes [manm]. He is running on all fours or path in Je passe [path] par [path] les eurs I am passing by/through the owers).

Typological constraints


cated in italics and in brackets) within utterances of varying levels of semantic density6. SD1 (29) They all dance [verbmanm]. (Sarah P3) (30) Elle est partie [verbpath]. (Clara P1) (She has left.) (31) Et pis dehors [otherloc] jai fait du poney [verbmanm] dehors [otherloc]. (Gregoire, P2) (And then outside I did [rode] pony outside.) ` (32) Il va tres vite [othermanm]. (Clara P3) (He is going very fast.) (33) I climb [verbmanm] up [otherpath] the ladder. (Sarah P2) (34) Im going y [verbcsemanm] a kite. (Sarah P3) ` (35) Elle rentre [verbpath] a la pointe des pieds [othermanm]. (Clara P4) (Shes entering at/on the tip of her feet.) (36) Faut faire nager [verbcsemanm] ma voiture. (Gregoire P3) (Must make swim my car.) (37) I just pushed [verbcsemanc] it down [otherpath]. (Sarah, P3) Results




Appendix 2 shows all verb roots that denoted motion in the corpora (with a frequency of 10 occurences or more collapsing over all developmental periods). Verb types that were by far the most frequent (450 occurences or more) were English go and French mettre (put), followed by a few very frequent verbs (English come, fall, put, take and French aller go, tomber fall, prendre take, donner give) some of which are considered to be light verbs in the literature because they express little or no semantic information (e.g., go to mean sheer motion). In both languages most verb roots denoted either voluntary motion or explicitly caused motion, and fewer verb types denoted involuntary motion, i.e., motion that was not voluntarily carried out by an agent and did not imply any explicit cause (mostly English intransitive fall and its French equivalent tomber).7

6. Rare cases included in this count consisted of utterances that come from nursery rhymes, the status of which may dier from spontaneous utterances, e.g., The mouse went up [Path] the clock. (Sarah P3). 7. Some transitive and intransitive uses of the same verb roots occurred (e.g., y voluntary vs. caused), particularly in English (nine verb types) as compared to French (one verb type).


M. Hickmann and H. Hendriks

Figure 6. Overall utterance density for motion events (Study 2)

As shown in Figure 6, the overall semantic density of utterances in relation to motion was higher in English than in French. English utterances were frequently of density SD2 (52%) and fewer were SD1 (38%), while SD3 utterances also occurred (10%). In comparison, fewer French utterances were SD2 (20%), most were SD1 (78%), and very few were SD3 (2%). This density pattern held for each developmental period from P1 onwards. It also held for both French children (SD1 Clara 82%, Gregoire 74%; SD2 Clara 17%, Gregoire 22%), but some individual dierences occurred in English between Adam (35% SD1, 56% SD2, 9% SD3) and Sarah (21% SD1, 62% SD2, 17% SD3). A further glance at the data shows that this dierence in utterance density was related to the fact that English-speaking children frequently associated motion verbs with other devices (particles, prepositions, adverbial phrases). In this respect, as illustrated in examples (24) to (37) above, childrens utterances fell into three types corresponding to the locus of motion information, which could be encoded in several ways: only in
Table 3. Utterance types expressing motion information in verbs and other devices (Study 2)* Developmental period English P1 P2 P3 French P1 P2 P3 P4 VerbOther VX Verb only VX Other only VX

36% 34% 33%

56% 54% 53%

8% 12% 14%

0% 1% <1% 2.5%

100% 99% 99.5% 97.5%

0% 0% 0% 0%

* Percentages are calculated within each period collapsing across children within each language, who showed no notable individual dierences.

Typological constraints


Figure 7. Motion information expressed in early spontaneous productions (Study 2)

verbs (hereafter VX, e.g., They all dance, Elle est partie She has left), ` only in other devices (VX, e.g., The mouse went up, Il va tres vite He is going very fast.), or in both (VX, e.g., I just pushed it down, Elle ` rentre a la pointe de pieds Shes entering at/on the tip of her feet). Table 3 summarizes the distribution of childrens utterances among these three utterance types. In French the locus of motion information was clearly in verbs (VX 99%). In English this was less frequently the case (54%), because children also encoded information either jointly in verbs and in other devices (VX 35%) or in other devices only (VX 11%). This pattern held for all developmental periods in both languages. Figure 7 shows in more detail the information components that were expressed in verb roots (Figure 7a) versus other devices (Figure 7b). Concerning verbs, in both languages cause was most frequent (Clara 59%, Gregoire 40%, Sarah 38%, Adam 42%), while manner-of-cause was least frequent (8% to 12% for all children). path was equally frequent in both languages (English 20%, French 24%). manner-of-motion tended to be more frequently encoded in English verb roots than in French ones (Gregoire 24%, Clara 12%, Sarah 33%, Adam 26%). As for other devices, they rarely expressed any information directly relevant to motion itself in


M. Hickmann and H. Hendriks

French. When they were used at all, they mostly encoded general loca tions (Clara 97%, Gregoire 96%). In contrast, other English devices expressed path more frequently (Sarah 61%, Adam 52%) than general locations (Sarah 31%, Adam 43%). 4.3. Summary

The longitudinal analyses of Study 2 focused on young childrens early spontaneous productions in English versus French from under two years onwards. The results are consistent with those of Study 1. They rst show that utterances were denser in English than in French with respect to the dierent types of motion information that were expressed. Second, children relied almost exclusively on verbs to express motion information in French, but frequently associated verb roots with other devices in English. In both languages verb roots expressed cause and path, but they tended to express manner more often in English. As for other devices, they expressed mostly general locations in French, but they encoded path more frequently than locations in English. These results held for all periods, notwithstanding other changes over developmental periods and individual dierences within a given language.


General discussion

Two complementary studies examined how speakers expressed motion events in English versus French. Particular attention was placed on the impact of two main factors: typological constraints that might result from language-specic properties (satellite-framed and verb-framed languages, respectively) and developmental changes across several age groups (adults and children between approximately two and ten years) that might result from universal cognitive factors. Study 1 elicited productions from adults and four groups of children (three, four, ve, and ten years) in a controlled experimental situation focusing on caused motion. Study 2 further examined the expression of all types of motion events at earlier ages in order to determine whether any cross-linguistic dierences could be observed from the emergence of language onward. Longitudinal analyses focused on all spontaneous utterances about motion events that were produced by four children (two learners of French, two learners of English) across several developmental periods between approximately the ages of two and three years. Study 1 rst showed some striking cross-linguistic dierences at all ages. From three years on, utterances were semantically denser and showed a much more systematic distribution of expressed information in

Typological constraints


English than in French. In English, adults and children in all age groups expressed multiple types of information (predominantly two to three or more components among children) using compact structures that typically encoded causemanner in main verbs and path in other devices (mainly particles). In contrast, French speakers produced utterances that were less dense and showed more varied structural patterns. At all ages French children expressed less information in their utterances about motion (predominantly one to two components) than same-aged children in English. They also tended to rely on verb roots to express motion information and/or to distribute this information in varied ways across parts of speech within their utterances (main verbs and other devices). Following Talmys typology, these results show the strong impact of language-specic factors on childrens productions. As a verb-framed language, French typically uses main verbs to express the path of events that imply changes of location, thereby making it dicult to express this information together with cause and manner within a single clause and frequently requiring the use of complex structures that contain subordinate clauses. In contrast, satellite-framed languages such as English typically express path in satellites thereby leaving room for cause and manner information to be expressed in main verbs. These compact structures therefore allow more information to be distributed within simple clauses and they are systematically used by children from the youngest age onward. Second, regardless of language, responses also showed an increase in semantic density with age. In both languages childrens utterances increasingly encoded more types of information (cause, manner, path) between the ages of three and ten years. A striking increase in density was observed after ve years, particularly with the advent of frequent utterances encoding three or more information components at ten years. In addition, the density of childrens utterances increased in both languages between the ages of three and ve years, roughly going from one to two components in French and from two to three or more components in English. This common developmental progression indicates the impact of universal cognitive determinants of language acquisition that might include memory and processing limits, as well as more general discourse organizational skills. From a cognitive point of view, it is simpler for children to produce basic structures expressing fewer types of information than to produce more compact and more complex structures expressing more information types. However, despite such common cognitive determinants, density was strikingly higher in English than in French at all ages. Furthermore, the increase in utterance density was more striking in French. In this respect, note that, when French children expressed only one type of information,


M. Hickmann and H. Hendriks

they tended to encode the cause and/or the path of motion, typically in main verbs, at the expense of other information types, which requires the use of subordinate clauses when expressed together with these components. These results are in line with our previous studies focusing on the expression of voluntary motion in French and English (Hickmann 2003; Hickmann et al. 1998) that showed a higher utterance density, as well as a greater focus on manner information in English than in French. They are also consistent with previous studies comparing a variety of verb-framed and satellite-framed languages (Berman and Slobin,1994; Slobin 2003b, 2006). Study 2 further examined early spontaneous utterances that denoted all types of motion events (voluntary, caused, involuntary) produced by four children between approximately two and three years of age. The results also showed a higher utterance density in English (mostly two information components) than in French (mostly one component) during all developmental periods. This dierence in density was shown to follow largely from the dierent ways in which children distributed information about motion across linguistic devices within their utterances. In particular, children learning French most frequently relied on main verbs alone to express all types of information relevant to motion, reserving prepositional phrases for general locations (if they used them at all). In contrast, children learning English used both main verbs and other devices, thereby expressing more varied information about motion, namely cause together with manner (particularly in main verbs), as well as path (typically in satellites). These results show that typological constraints on acquisition, such as the ones that were expected on the basis of Talmys typology, hold from the earliest age onwards. These studies raise several questions. Among them a rst technical question concerns our methodology in Study 2, and particularly the way in which the data base was constituted for the cross-linguistic comparison among children across developmental periods from P1 onwards. Recall that the recordings from French and English children in this study were matched by their MLU within each period (with a mean MLU of <2,5 for P1, between 2,5 and 4 for P2, between 4,1 and 5 for P3, and >5 for P4). In this respect, note that chronological age was systematically lower in French than in English for the corresponding mean MLU within each period (see Note 4). This discrepancy existed regardless of how MLU was calculated and it is presumably due to the properties of the two compared languages. It therefore cannot be excluded that the higher semantic density observed in the early spontaneous productions of English-speaking versus French-speaking children was at least partially due to their higher chronological age (English > French) in addition to typological con-

Typological constraints


straints resulting from language-specic properties (satellite-framing in English versus verb-framing in French). Nonetheless, a similar dierence in density was found in Study 1, in which children were matched by chronological age, as well as in our previous experimental studies including French in other experimental situations (Hickmann 2003; Hickmann et al. 1998, 2008; Ochsenbauer and Hickmann 2008) and in a number of studies involving other languages either in controlled experimental studies (Berman and Slobin 1994; Slobin 2003a, 2003b, 2006) or in longitudinal studies of early spontaneous productions (Bowerman 2007; Bowerman and Choi 2003). Second, a number of other more general questions remain open. In particular, a potential implication of our results concerns the extent to which the cross-linguistic dierences that were observed in speakers utterances might reect deeper dierences in their underlying conceptual representations. Although the present research does not allow us to draw any conclusions in this respect, its results are in line with those of a number of other studies (e.g., Slobin 2003, 2006) suggesting that language-specic properties may draw speakers attention to dierent aspects of the space surrounding them. From a developmental point of view, childrens exposure to these language-specic properties may lead them to construct different represensations during the active process of acquiring their native language. Such a hypothesis constitutes a major challenge that will require future studies examining the extent to which typological constraints may aect non only speech production, but also non-verbal behaviors and more generally the processes that might lead to dierent types of cognitive organization. At the same time, future research should investigate in more detail the precise nature of general and universal cognitive capacities (for example working and long term memory, processing speed, categorization, and planning) that may drive the development in the expression of space in all child languages. 6. Conclusion

This research showed striking cross-linguistic dierences in how speakers express motion in English versus French. Such dierences follow from typological constraints related to how satellite-framed (English) versus verb-framed (French) languages grammaticalize or lexicalize information relevant to motion events, particularly when these events involve changes of location. They therefore indicate the impact of language-specic properties on language acquisition. Some common developmental progressions were also observed in both languages, particularly an increase with age in the semantic density of utterances denoting motion. This result


M. Hickmann and H. Hendriks

suggests the impact of cognitive factors that are presumably universal and independent of language, such as the development of childrens processing or memory capacities necessary for the joint expression of multiple semantic components. However, typological constraints were also observed at all ages from the youngest age (18 months) to adult age. Furthermore, developmental progressions were shown to be much more striking in French than in English, suggesting the joint impact of cognitive and typological factors on language acquisition. Young learners of verb-framed languages have more problems to solve when expressing multiple types of information about motion as compared to same-aged learners of satellite-framed languages. Future research is necessary to address further questions raised by these results concerning the relationship between language and cognition during development. Received 1 March 2009 Revision received 25 November 2009 University of Paris 8/ University of Cambridge

Berman, Ruth and Dan I. Slobin, 1994. Relating events in narrative: A crosslinguistic developmental study. New York: Lawrence Erlbaum Associates. Bowerman, Melissa. 1996. The origins of childrens spatial semantic categories: cognitive versus linguistic determinants. In John J. Gumperz and Stephen Levinson (eds.), Rethinking Linguistic Relativity, 145176. Cambridge: Cambridge University Press. Bowerman, Melissa. 2007. Containment, support, and beyond: Constructing typological spatial categories in rst language acquisition. In Michel Aurnague, Maya Hickmann and Laure Vieu (eds.), The categorization of spatial entities in language and cognition, 177203. Amsterdam: Benjamins. Bowerman, Melissa and Soonja Choi. 2003. Space under construction: language-specic categorization in rst language acquisition. In Dedre Gentner and Susan Goldin-Meadow (eds.), Language in Mind: Advances in the study of language and thought, 387427. Cambridge, MA: MIT Press. Choi, Soonja and Melissa Bowerman. 1991. Learning to express motion events in English and Korean: the inuence of language-specic lexicalization patterns. Cognition 41. 83 121. Clark, Eve V. 2003. Language and representations. In Dedre Gentner and Susan GoldinMeadow (eds.), Language in Mind: Advances in the study of language and thought, 1724. Cambridge, MA: MIT Press. Hendriks, Henriette, Hickmann, Maya, and Annie-Claude Demagny. 2008. How adult En glish learners of French express caused motion: a comparison with English and French ` natives. Acquisition et Interaction en Langue Etrangere 27. 1541. Hickmann, Maya. 2003. Childrens Discourse: Person, Space and Time across Languages. Cambridge, Cambridge University Press. Hickmann, Maya and Henriette Hendriks. 2006. Static and dynamic location in French and in English. First Language 26 (1). 103135.

Typological constraints


Hickmann, Maya, Henriette Hendriks and Christian Champaud. 2009. Typological con straints on motion in French child language. In Jiansheng Guo, Elena Lieven, Susan Ervin-Tripp, Nancy Budwig, Keiko Nakamura, Seyda Ozcaliskan (eds.), Crosslinguistic approaches to the psychology of language: research in the tradition of Dan Isaac Slobin, 209224. Hillsdale, NJ: Lawrence Erlbaum. Hickmann, Maya, Francoise Roland and Henriette Hendriks. 1998. Reference spatiale dans les recits denfants francais: perspective inter-langues. Langue Francaise 118: Numero special sur lAcquisition du francais langue maternelle. 104123. Hickmann, Maya, Pierre Taranne and Isabelle Bonnot. 2009. Motion in rst language acquisition: Manner and Path in French and English child language. Journal of Child Language, 36 (4). 705742. Hohenstein, Jill. 2005. Language-related motion event similarities in English- and Spanish speaking children. Journal of Cognition and Development, 6. 402425. Kopecka, Annette. 2006. The semantic structure of motion verbs in French: typological per spectives. In Maya Hickmann and Stephane Robert (eds.), Space across languages: linguistic systems and cognitive categories, 83101. Amsterdam: John Benjamins. Landau, Barbara and Laura Lakusta. 2006. Spatial language and spatial representation: autonomy and interaction. In Maya Hickmann and Stephane Robert (eds.), Space in languages: linguistic systems and cognitive categories, 309333. Amsterdam: John Benjamins. Lucy, John and Suzanne Gaskins. 2001. Grammatical categories and the development of classication preferences: A comparative approach. In Stephen Levinson and Melissa Bowerman (eds.), Language Acquisition and Conceptual Development, 257283. Cambridge: Cambridge University Press. MacWhinney, Brian. 2000. The Childes Project: Tools for Analyzing Talk. Mahwah, NJ: Lawrence Erlbaum Associates. Munnich, Edward and Barbara Landau. 2003. The eects of spatial language on spatial representation: setting some boundaries. In Dedre Gentner and Susan Goldin-Meadow (eds.), Language in mind: Advances in the study of language and thought, 113155. Cambridge, MA: MIT Press. Ochsenbauer, Anne-Katherine and Maya Hickmann. 2008. Voluntary motion in French and German child language. Paper presented at the Conference of the International Association for the Study of Child Language. Edinburgh, 28 July1 August. Pruden, Shannon M., Kathy Hirsch-Pasek and Roberta M. Golinko. 2008. Current events: How infants parse the world and events for language. In Thomas F. Shipley and Jerey M. Zacks (eds.), Understanding events: How humans see, represent, and act on events, 160 192. New York: Oxford University Press. Slobin, Dan I. 1996. From thought to language to thinking for speaking. In Gumperz, John J. and Stephen C. Levinson (Eds.), Rethinking linguistic relativity, 7096. Cambridge: Cambridge University Press. Slobin, Dan I. 2003a. Language and thought online: cognitive consequences of linguistic relativity. In Dedre Gentner and Susan Goldin-Meadow (eds.), Language in Mind: Advances in the study of language and thought, 157191. Cambridge, MA, MIT Press. Slobin, Dan I. 2003b. The many ways to search for a frog. In Sven Stromqvist and Ludo Verhoeven (eds.), Relating events in narrative: Typological and contextual perspectives, 219257. Hillsdale, NJ: Erlbaum. Slobin, Dan I. 2006. What makes manner of motion salient? Explorations in linguistic typol ogy, discourse, and cognition. In Maya Hickmann and Stephane Robert (eds.), Space across languages: linguistic systems and cognitive categories, 5981. Amsterdam: Benjamins. Talmy, Leonard. 2000. Towards a cognitive semantics. Harvard: MIT Press.


M. Hickmann and H. Hendriks

Appendix 1. Summary of main features in the stimuli used to test caused motion (Study 1)*
Combination Ground in exemplar a roof roof roof roof snow snow snow snow road road road road cave cave cave cave Ground in exemplar b sand dune sand dune sand dune sand dune grass hill grass hill grass hill grass hill street street street street barn barn barn barn Figure object swimming ring package toy car bag balloon suitcase wheelbarrow trunk wheel apple basket pram rocking horse tyre table trolley chair Manner of cause push push pull pull push push pull pull push push pull pull push push pull pull Manner of O-motion roll slide roll slide roll slide roll slide roll slide roll slide roll slide roll slide Path of motion up up up up down down down down across across across across into into into into

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

hill hill hill hill

* The O gures changed across the 16 combinations. Subjects saw two exemplars (a and b) of each combination, both with the same O but with a dierent scenery and ground referent. Path was always the same for A and O. Two other information components were held constant across all test items: cause (causal relation between A and O) and manner of A-motion (walking).

Typological constraints


Appendix 2. Verb roots used in early spontaneous productions (Study 2)*

English: Level I: go Level II: come, fall, put, take Level III: drive, turn, walk, y, run, jump, sit, stand, dance, step, swim, get, roll, move, drop, spill, throw, turn, push, bring, park, knock, hang, pick. French: Level I: mettre (put) Level II: aller (go), tomber (fall), prendre (take), donner (give) Level III: partir (leave), monter (ascend), venir (come), rouler[intr] (roll), sortir (exit), marcher (walk), passer (pass), sauter (jump), rentrer (enter), descendre (descend), revenir (come back), voler (y), bouger (move), sasseoir (sit down), faire du ski (ski), couler (drip), nager (swim), danser (dance), rattraper (catch up with), courir (run), sen aller (leave), enlever (take o); faire (make[Inntive]), ranger (put away), tourner (turn), remettre (put back), tirer (pull), coller (glue), melanger (mix), jeter (throw), pousser (push), ramasser (pick up), amener (bring), poser (place). * Levels of token frequencies are shown collapsing all developmental periods within each language. Verbs are shown in decreasing order of frequency levels (I:>450, II: 100449, III: 1099), as well as within each frequency level. Other verb roots were much less frequent (less than 10 occurences overall).

Childrens verbalizations of motion events in German


Abstract Recent studies in language acquisition have paid much attention to linguistic diversity and have begun to show that language properties may have an impact on how children construct and organize their representations. With respect to motion events, Talmy (2000) has proposed a typological distinction between satellite-framed (S) languages that encode PATH in satellites, leaving the verb root free for the expression of MANNER, and verb-framed (V) languages that encode PATH in the verb, requiring MANNER to be expressed in the periphery of the sentence. This distinction has lead to the hypothesis (Slobin 1996) that MANNER should be more salient for children learning S-languages, who should have no diculty combining it with PATH, as compared to those learning V-languages. This hypothesis was tested in a corpus elicited from German children and adults who had to verbalize short animated cartoons showing motion events, and the results are compared with previous analyses of French and English corpora elicited in an identical situation (Hickmann et al. 2009). As predicted, and as previously found for English, German children from three years on systematically express both MANNER (in the verb root) and PATH (in particles), in sharp contrast to French children, who rarely package MANNER and PATH together. These results suggest that, when they are engaged in communication, children construct spatial representations in accordance with the particular properties of their mother tongue. Future research is necessary to determine the extent to which cross-linguistic dierences in production

* Address for correspondence: A.-K. Ochsenbauer, Ludwig-Maximilians-Universitat, In stitut fur Deutsche Philologie, Schellingstrae 3/RG, 80799 Munchen. Email: anne. M. Hickmann, CNRS Laboratoire Structures Formelles du Langage, UMR 7023, 59 rue Pouchet, 75017 Paris, France. Email: maya.hickmann@s. Cognitive Linguistics 212 (2010), 217238 DOI 10.1515/COGL.2010.008 09365907/10/00210217 6 Walter de Gruyter


A.-K. Ochsenbauer and M. Hickmann

may reect deeper dierences in the allocation of attention and in conceptual organization. Keywords: cognitive linguistics, language acquisition, language typology, space, thinking for speaking, Whoranism.



During the last twenty years linguists and psycholinguists have postulated dierent ways of relating the process of childrens language acquisition to their cognitive development. Predominant theories in psychology have put forth the existence of universal perceptual and cognitive constraints in language acquisition determining childrens verbal production and comprehension (e.g., Piaget and Inhelder 1947, or Spelke 2003). However, recent studies (for example, Slobin 1996, 2003a, 2003b, 2006) indicate that our language seems to inuence how we think when we speak, for example inviting us to focus on particular aspects of reality. These results suggest that children learn to verbalize situations in a certain way, which is most typical of their mother tongue, and that they organize incoming information accordingly. The study described below further tests this hypothesis by examining how German children and adults express voluntary motion events in controlled experimental situations. A comparison of our results with those of previous comparable studies concerning English and French supports the claim that the linguistic properties of spatial systems inuence how children construct their spatial representations. 2. Space across languages

Talmy (1983, 1985, 1991, 2000) has shown that languages show strikingly dierent lexicalization patterns in the expression of motion events, that are reected in dierent ways of combining semantic information in surface structure. For example, as illustrated in (1) and (2), satellite-framed languages (e.g., Germanic) encode manner in the verb stem (English/ German swim/schwimmen, run/rennen) and path in verbal satellites1 such as particles (across/durch, away/weg). In contrast, as shown in (3),

1. In their draft, Croft et al. (2008) propose to expand Talmys typology, taking into account lexicalization patterns that are less typical but occur regularly in a large number of languages, particularly three symmetrical types: coordination, serialization and compounding.

Childrens verbalizations of motion events


verb-framed languages (e.g., Romance) encode path in the verb stem (traverser to cross, partir to leave) and manner by peripheral constructions ` such as adverbial phrases (a la nage with a swim) or gerunds (en courant by running). (1) (2) (3)
The Das The L The child Kind child enfant child swims across durchschwimmt through-swims traverse crosses the den the la the river Fluss river ` ` riviere a la nage river with a swim and und and et and runs rennt runs part leaves away. weg. away. en courant. by running.

When verbalizing a motion event, speakers choose among several means of expression those which are most typical for their language. One implication is that, while speaking, they are invited to focus on different aspects of reality, and therefore to foreground and background incoming information in dierent ways across languages. Slobin (1996, 2003a, 2003b, 2006) further tested some cognitive implications of Talmys typology, pointing out three factors which increase the likelihood that speakers will express and/or combine particular semantic components. The rst factor is niteness. In satellite-framed languages (see German (4) and (5)) manner is normally expressed in the main inected verb, or more precisely in that part of the main verb that carries verbal morphology. In contrast, speakers of verb-framed languages (see French (6) and (7)) have to use peripheral constructions that may include a non-nite verb. As a result, German (4) and (5) are of the same complexity, whereas (7) is more complex than (6) in French. (4) (5) (6) (7) Das The Das The La The La The Madchen girl Madchen girl lle girl lle girl rennt runs geht goes traverse crosses traverse crosses uber across uber across die the die the la the la the Strae. street. Strae. street. rue. street. rue en courant. street by running.

The second factor is lexeme frequency. In satellite-framed languages verbs expressing simultaneously motion and manner are extremely frequent and often used, even by young children. In verb-framed languages this kind of verb is less frequent. Finally, the last factor is the possibility of expressing information by means of a single (complex) morpheme rather than by a phrase or clause. Examples (8) to (10) illustrate several verb equivalents for some types of motion events in English, German and French, showing that German has many manner-verbs which have


A.-K. Ochsenbauer and M. Hickmann

no monolexematic equivalent in French and sometimes not even in English: (8) (9) (10) schlurfen, to shue along, traner les pieds ` stapfen, to plod, marcher a pas lourds tappen, to go falteringly, marcher dun pas maladroit

Each of these three factors makes it easier for speakers of satelliteframed languages to express manner and path together in one single clause as compared to speakers of verb-framed languages. As a result, one implication is that manner should be more salient in these languages than in verb-framed languages. In contrast, no dierence in salience across languages is predicted for the semantic component path. From a developmental point of view, it might also be predicted that these typological dierences should result in dierent developmental progressions during the acquisition of spatial language. Thus, Slobin suggests that each language should invite children to focus on some specic aspects of spatial representations. As a result, they may gradually take a particular perspective on the world, which may inuence not only how they verbalize motion events, but perhaps also their cognitive organization more generally. Although German stands among other satellite-framed languages, few studies have examined in detail how motion is expressed in this language. One study (Tschander 1999) shows that available classications of motion verbs (Talmy 1985; Landau et al. 1993) are too simplistic. Apart form German verbs containing either manner or path, this study postulates a third category, namely path-manner-verbs (e.g., humpeln to hobble) which describe a combined movement (kombinierte Bewegung). These verbs are used with dierent auxiliaries depending on the speakers focus: with the auxiliary haben (to have) they focus on manner; with the auxiliary sein (to be) they focus on path. Example (11) taken form Tschanders article demonstrates this phenomenon: (11) Debbie hat/ist gehumpelt. Debbie has/is hobbled.

According to Tschander, these two concepts of movement, manner and path, must not constitute separate entries in the lexicon, so that these verbs should correspond to only one entry in the lexicon. Weber (1983) also proposes a more detailed classication of German motion verbs based on several recurrent semantic components. On the basis of a sample of 90 motion verbs, he extracts 35 semantic components, 20 of which actually correspond to some aspect of manner that characterizes most

Childrens verbalizations of motion events


verbs (94%). This analysis shows again that German encodes manner in the vast majority of its motion verbs. With respect to satellites, there is no consensus as to the nature of this class in German. According to Talmy (1991: 486), satellites belong to
[ . . . ] the grammatical category of any constituent other than a nominal complement that is in a sister relation to the verb root. Satellites can be either a bound ax or a free word, and encompass very diverse grammatical forms (English verb particles, German separable and inseparable verb prexes, [ . . . ].).

Haggblades (1994) analysis of German satellites includes a variety of devices, among which the following four classes will be most relevant below: 1) prexes, e.g., uber- in uberqueren (to cross); 2) particles, e.g., rauf- in raufklettern (to climb up) 3) prepositional phrases, e.g., auf den Baum (on the tree); 4) adverbs, e.g., hinauf (up).2 A more detailed discussion of the problematic distinction between German prex- and particle-verbs can also be found in Altmann and Kemmerling (2005: 63 ). For example, they propose a particle type called double-particle (e.g., drauf- up there or herunter- down from there), which add some deictic (and sometimes local) information to their directional component. 3. Universal and language-specic determinants of childrens spatial language

With respect to the relation between language and cognition during child development, one of the most important research questions is whether children construct universal pre-linguistic concepts that underlie language acquisition or whether their concepts are substantially structured or transformed with the emergence of language. During the last years, this question has been approached by linguists and psycholinguists by and large in three dierent ways. Proponents of a rst position claim that the language ability is innate, modular, and domain-specic. In this view neither general cognitive faculties nor language acquisition have any substantial inuence on this initial knowledge. According to Spelke (2003), human language only provides the opportunity to combine knowledge from dierent modules, allowing humans to build representations that are more complex than

2. In some verbs stress is a criterion to distinguish between prex and particle verbs, e.g., uberfahren (ton run over) and uberfahren (to cross over), where accents on vowels mark stressed syllables in the verb.


A.-K. Ochsenbauer and M. Hickmann

those of other species. The second position is perhaps best illustrated by Piagetian theory, which argues that perceptual and cognitive constraints determine language acquisition. Many studies have indeed shown that universal perceptual and cognitive factors inuence concept formation and determine which spatial dimensions are most salient (Antell et al. 1985, Mandler 1996). Such factors account for the recurrent order in which linguistic procedures are acquired and related concepts constructed by children across languages. For example, in a review of studies on the acquisition of spatial prepositions across several languages, Johnston and Slobin (1979) showed that all children rst learn prepositions that encode containment (in), support (on) and occlusion (under), then those that encode proximity (next to), and at last those that refer to distinctions on the sagittal axis (behind, in front of ). This recurrent order reects the relative complexity of spatial markers and suggests that universal cognitive constraints inuence acquisition. Finally, according to the position known as linguistic determinism (Whorf 1956; Bowerman 1996; Slobin 1996), our language inuences how we think when we talk, getting us to focus our attention on particular aspects of reality. Thus, children learn to verbalize situations in a certain way, that is most typical for their mother tongue. In particular, dierent lexicalization patterns across languages (e.g., pre- and postpositions, particles, morphologically complex forms or synonyms) inuence how children acquire spatial language. Several studies (Bowerman 2003; Choi and Bowerman 1991; Hickmann 2006, 2007; Hickmann et al. 2009) have shown that children talk about space more like adults who speak the same mother tongue than like children of the same age learning a typologically dierent mother tongue. For example, from very early age on, English-speaking children express the manner and path of motion together in one single clause because their language possesses very compact structures allowing them to do so easily. In contrast, although it is possible to express manner and path together in French, and although French adults do combine these two types of information in some situations, they do so less frequently and less systematically than English adults. In addition, French children (three to ten years) rarely express both components together, presumably because this kind of response requires more complex structures in French than in English. Finally, at all ages French speakers responses vary with event types: although they typically focus on path with most events, they also focus on manner with crossing events (children) or combine path and manner with upward motion (mostly using the verb grimper to climb up, that lexicalizes both). These results directly follow from the typological properties of English and French, suggesting that children learn very

Childrens verbalizations of motion events


early to express the types of information that are salient in their native language. Therefore, both general and language-specic determinants inuence childrens cognition and language in the domain of space. As noted above, despite some particular properties of German, its lexicalization pattern for motion events is similar to the one in English since most German motion verbs conate motion and manner, while path is typically expressed in a wide range of satellites (particles, adverbs etc.). From a developmental point of view, German children should therefore talk about motion more like English-speaking children (frequent manner verbs and path satellites) than like French-speaking children (frequent path verbs, infrequent manner). This hypothesis is partially supported by some scant available evidence. One study (Bamberg 1994: 221) notes that German children make heavy use of varied motion verbs and satellites in narrative discourse, but provides no further information concerning how these devices are used. Evidence from a study (Gentner 1979) concerning early child English shows the frequent use of light verbs in combination with satellites, which may also be expected to occur among German children. Light uses need not involve the full meaning of verbs, which can be frequently reduced to sheer motion (e.g., gehen to go rather than to walk). Such uses presumably also involve a lower level of grammatical complexity since children often learn the nite forms of these verbs by rote and therefore do not actively inect them. Example (12) illustrates a construction of this type (light verb gehen, verb particle rauf up) which is very frequent among young German children. (12) Er geht rauf. He goes up.

With respect to German, surprisingly little is still known concerning childrens uses of other devices outside of the main verb root, such as those illustrated in (13) to (17) below: spatial adverbs, spatial particles, prexed verbs, and full prepositional phrases which govern either Dative or Accusative case to distinguish general locations from changes of locations, respectively. (13) (14) (15) (16) (17) Der rennt hier. He runs here. Die geht rauf. She goes up. Die Frau uberquert die Strae. The woman crosses the street. Der Ae klettert auf den Baum. The monkey climbs on the[Acc] tree. Das Kind spielt in der Kuche. The child is playing in the[Dat] kitchen.


A.-K. Ochsenbauer and M. Hickmann

The present study aimed at further examining how German children represent motion events in a controlled experimental situation that was similar to the one previously used for English and French (Hickmann 2006; Hickmann et al. 2009). Given the properties of German, the following predictions were made. First, from the youngest age tested (three years) onwards, German children should express both manner and path, relying on structures that encode manner in the verb and path in other devices such as particles, which represent the typical typological pattern of satellite-framed languages. Second, they also should use a great number and variety of motion verbs expressing manner, since such verbs are frequent in the adult input (Talmy 1985, 2000; Weber 1983). Third, their uses of devices outside of the verb root should show some change with age as a function of grammatical complexity. In particular, children should produce these devices in the following order: rst particles and adverbs, which are least dicult because they do not require any inection3; then prexed verbs, which are possible means of expressing motion; and nally, full prepositional phrases, which govern dierent case markings. 4. 4.1. Method Subjects

The results reported below concern 60 monolingual Germans in ve age groups (12 subjects per age). Four groups of children, boys and girls, were tested in kindergartens and primary schools of Augsburg. Their ages were approximately three years (mean 3;8, range 3;4 to 4;4), four years (mean 4;7, range 4;6 to 5;4), six years (mean 6;7, range 6;4 to 7;2), and ten years (mean 10;5, range 10;4 to 10;11). A control group of adults involved students from the University of Munich. 4.2. Materials

Two sets of animated cartoons were constructed (see Appendix). In all cartoons characters carried out a displacement in a particular manner (e.g., walking, running, jumping, etc.), then left the scene. One set of target items (six up-targets and six down-targets) showed a scene with a vertical ground referent, along which displacements took place (e.g., a squir-

3. Particles are used very early in German (e.g., Auto rauf car up). Later in development they are integrated into particle verbs and then form part of the Satzklammer (sentence bracketing) which is syntactically more complex and thus more dicult to learn (e.g., Ich schieb das Auto rauf. I push the car up.).

Childrens verbalizations of motion events


rel running up/down a tree and away). In another set of items (six control items) the characters entered onto one side of the scene against a blank screen, moved to the other side, and left. manner corresponded to the types of actions that took place in the target items during the characters departure (e.g., walking). These displacements were carried out in the absence of any scenery that could provide specic relevant ground entities for the expression of path. Upward and downward motion was selected as targets for the stimuli because they correspond to events that are most familiar to children4. Furthermore, the addition of control items provided a direct contrast between two conditions. Target items focused subjects attention on location changes that involved relevant manner and path information, whereas control items minimized path information and highlighted manner. It was expected that German subjects should 1) express manner with both types of items, but 2) combine manner and path with target items and 3) do so more often with increasing age. Control items also provided a way of determining whether children were able to produce some manner information, particularly if they had not spontaneously mentioned this information when describing target items. 4.3. Procedure

Subjects were seen individually in their school or university setting. They were presented the cartoons on a computer screen and asked to narrate each cartoon as completely as possible. The entire session was audiotaped. Primary school children and adults were told that a future addressee, who would not be shown the cartoons, would have to reproduce the stories on the basis of the recordings. Younger children were introduced to a doll and were asked to blindfold her as part of a game in which they would be telling her secrets. They were reminded throughout to tell her everything that had happened because she could not see and would also like to tell the story. This procedure ensured that subjects produced full descriptions. Cartoons were presented in six dierent random orders in which target items always occurred before control items. A training item began the session.

4. The stimuli actually included six other cartoons that were interspersed among the target items, but are not discussed in the present paper because of space limitations. This additional set of stimuli showed events that involved crossing a boundary (e.g., a baby crawling across a street, a boy swimming across a river). All results concerning these events are entirely in line with those reported here for up/down events.

226 4.4.

A.-K. Ochsenbauer and M. Hickmann Coding

The analyses focused on utterances that described motion. These utterances contained several types of information relevant to motion that were encoded by a variety of linguistic devices, grouped below into two classes: main verbs vs. all other devices. With respect to main verbs, the coding rst distinguished those that were potentially light, particularly all verbal forms of gehen (to go), from all others that had their full semantic meaning (e.g., rennen to run, hupfen to jump). Most uses of gehen (to go) presumably expressed sheer motion, rather than a particular manner of motion, given that the corresponding experimental item did not at all show walking (e.g., cyclist in example (18)). Motion verbs other than gehen were further coded with respect to path (e.g., kommen to come) and manner (e.g., hupfen to jump). Devices outside of the main verb were of three types: 1) prepositional phrases (e.g., auf den Baum on the tree); 2) spatial particles (e.g., ruber across); 3) other relevant expressions like ad verbs (e.g., hier here). They were further coded in terms of whether they expressed path (e.g., weg away), manner (e.g., auf allen Vieren on all fours), and other types of information, for example locations (e.g., da there).5 (18) Die The Results Up and down motion Fahrradfahrerin, die woman-cyclist, she geht da ruber. goes there across.

5. 5.1.

5.1.1. Main verbs. As expected, young children frequently produced the light verb gehen (to go) in combination with various other devices, particularly at ages three to six (34%). Light verbs then decrease sharply at ten years (8%) and practically disappear in the adult group (2%). As illustrated in example (19), the meaning of gehen (to go) as light-verb can be reduced to sheer motion. (19) Der geht dahin und geht da hoch, frisst den ganzen Honig, wieder runter und geht dann weiter. (3 years) He goes there and goes there up, eats all the honey, again down and goes then along.

5. Our data showed no verbal root conating both manner and path, nor any subordinated clauses expressing motion information.

Childrens verbalizations of motion events


Figure 1. Semantic content of main verbs for target events as a function of age

Figure 1 shows how frequently path and manner were expressed in the main verb as a function of age.6 Main verbs expressed manner more frequently than path at all ages (77% vs. 23% overall). Examples (20) and (21) show two typical sentences from German speakers who combine manner-verbs with path-devices. However, path-verbs were not infrequent among young children (29% at three years, 30% at four years, 37% at six years, in comparison to 15% at ten years), as illustrated in (22). With increasing age, manner-verbs became clearly most frequent, particularly at ten years (87%) and in the adults group (94%). Post-hoctests (Bonferroni) comparing uses of manner verbs showed a signicant dierence between ages six and ten years (p < .05), but none between ages three to six years, nor between 10-year-old children and adults. (20) (21) Die hupft da, krabbelt auf zum Kase rauf. (3 years) She bounces there, crawls on to the cheese up. Das Eichhornchen springt zu dem Baum und klettert hinauf und kriecht in das Loch 3und geht4 [/ und hupft wieder raus und run/] ter. (6 years)7 The squirrel jumps to the tree and climbs up and creeps into the hole 3and goes4 [/ and bounces then out and down. /] Die geht auf den Stangel und geht auf das Blatt und esst es ein bisschen und geht wieder runter. (6 years) She goes on the stick and goes on the leaf and eats it a little bit and goes again down.


6. Very few German verbs express Manner and Path simultaneously, as for example steigen (to climb up) or tauchen (to dive [down]) . Since these verbs were extremely scarce in our data, we did not distinguish them from simple Manner-verbs. 7. The symbol [/ marks a self-correction concerning the passage shown between pointed /] brackets 3 4.


A.-K. Ochsenbauer and M. Hickmann

5.1.2. Other devices. We now turn to all linguistic procedures that expressed spatial information outside of the main nite verb (included among other devices, see coding above). Table 1 shows the mean number of these devices that were used in each age group within one response. This number increased with age, particularly after six years. Age comparisons in this respect showed signicant increases between ages six and ten (p < .05), as well as between ten-year-olds and adults (p < .05), but no signicant dierences between ages three and six.
Table 1. Number of devices outside of main verbs for target events as a function of age* Age groups 3 years 4 years 6 years 10 years Adults Total number of devices 14,58 14,42 13,67 17,08 19,27 Number of devices per event 1,21 1,20 1,14 1,42 1,75

* Number of devices per utterance.

In their verbalizations of up- and down-motion, participants used either one satellite as in example (23) or more as in (24) and (25). On average, adults used 1,75 devices per utterance, as compared to 1,21 at three years. Nevertheless, even some of the children at three years produced as many devices as adults. (23) (24) (25) Die klettert auf die Blume. (3 years) She climbs on the ower. Das krabbelt hoch zum Baum. (3 years) It crawls up to the tree. Wir haben eine Raupe, die sich im Garten bewegt, und sich dann auf einen Stangel raufhangelt [ . . . ]. (adult) We have a caterpillar that moves in the garden, and then clings on[on] a stripe.

Figure 2 further shows the distribution of other devices within each age. In all age groups particles were most frequent, but tended to decrease with age (from 75% at three years to 57% at adult age). Prepositional phrases tended to increase between ages three (9%) and six (24%), then to decrease until adult age (12%). However, no age dierences were signicant for either of these types of devices. As for the third residual category, consisting above all of adverbs, it was rather infrequent until six years (8% to 14%), but increased thereafter until adult age (31%). Age

Childrens verbalizations of motion events


Figure 2. Types of devices outside of main verbs for target events as a function of age

comparisons showed signicant increases in these devices between six and ten years (p < .05), as well as between age ten and adults (p < .05). No other age dierences were signicant. With respect to the semantic information encoded in these linguistic devices, overall 80% expressed path (as in (18) to (20) above), as compared to only 1% manner and 19% other relevant information (for instance information about the setting). As expected, path devices were signicantly more frequent than manner devices (T-Test, df 59, p .001) and the remaining category mostly concerned information about the setting and the ground, e.g., hier (here) or auf einer Wiese (in a meadow). As shown in Figure 3, the same pattern was observed within each age group.

Figure 3. Semantic information expressed in other devices for target events as a function of age

Age comparisons showed no signicant dierences in the frequencies of path devices, despite a slight increase from ages three to six (82% to 90%) and decrease thereafter (78% at age ten, 67% among adults). The


A.-K. Ochsenbauer and M. Hickmann

residual class of devices, for example those that provided general locations (other than particles and prepositional phrases) increased with age, showing more frequent uses by adults (33%) than by children in any age group (three years 15%, 4 years 12%, six years 10%, ten years 21%). As illustrated in examples (26) and (27), this dierence between young children and adults mainly concerned information about the setting and the ground. (26) Das krabbelt da hoch, dann holt sie den Kase und dann geht sie runter. (4 years) It crawls up there, then it takes the cheese and then it goes down. Die Katze ahm springt an einem Telefon(mast), nein an einem Strommast und ahm krabbelt dann hoch und klaut sich ein Ei aus dem Nest, also das stubst das Ei runter und das Ei fallt auf den Boden, das Ei bricht entzwei und die Katze springt runter und leckt dann das Ei auf. (adult) The cat ehm jumps at a telephone (pole), no at a power pole and ehm then crawls up and nicks an egg from the # nest, well it nudges the egg down and the egg falls to the ground, the egg breaks in two and the cat jumps down and then licks the egg. Control items



Figure 4 shows the semantic information that was expressed in motion verbs with control items across the dierent age groups. At all ages the great majority of verbs expressed manner rather than path (overall 91% vs. 9%). Despite slight variations in this respect, no age dierences were signicant. Examples (28) to (31) show a great variety of dierent manner-verbs across ages. In comparison, regardless of age, very few responses contained any other device outside of the main verb, except for occasional locative expressions. No analysis of light verbs is presented for control items, since these verbs were very rare with these items.

Figure 4. Semantic information in main verbs used for control items as a function of age

Childrens verbalizations of motion events (28) (29) (30) (31) Die springt. (3 years) She jumps. Die ist so gekrabbelt. (4 years) She has crawled like this. Die Robbe, die robbt halt so ja, die Raupe, da. (10 years) The seal, it crawls just like this, yes, the caterpillar, there. Der Bar tapst. (adult) The bear lumbers.


A nal analysis compared the types of motion verbs that were used in relation to target and control items. Table 2 shows the number of dierent lexemes (types) that were used for each item type. In both cases the number of lexeme types remained stable between ages three and six (89 types), then increased slightly with target items (1011 types) and drastically with control items (17 types). However, although childrens productive lexicon seems to undergo an explosion between ages six and ten, note that they produced some verbs that were not used by adults, e.g., robben (to crawl) or schrubben (to scrub) for the motion of the caterpillar. Moreover, three neologisms were found in the data: raupen, which might be derived from the noun Raupe (carterpillar) to mean to move as a caterpillar; kraupen, which probably fuses the verb krabbeln (to crawl) and the noun Raupe (carterpillar); tappeln, which probably combines the verbs tappen (to go falteringly) and tippeln (to trip).

Table 2. Number of dierent motion verb lexemes (types) as a function of age Age groups 3 years 4 years 6 years 10 years adults Target items 9 9 9 11 10 Control items 8 9 9 17 17

6. 6.1.

Discussion Lexicalization patterns in German and other child languages

Our experiment examined how German children and adults described voluntary motion events, with particular attention to the expression of path and manner in relation to up and down motion. As expected, the results showed that speakers mostly encoded manner in nite verbs and path in other linguistic devices outside of the verb. This rst result was


A.-K. Ochsenbauer and M. Hickmann

observed at all ages from the youngest age (three years) to adult age. As predicted, these observed patterns are in line with the proposal that German (like other satellite-framed languages) invites speakers to simultaneously focus on both manner and path. Additional qualitative information shows that German speakers were greatly concerned with manner of motion. First, as illustrated in (32) and (33), self-corrections were observed at all ages from the youngest to the oldest age group, showing that in the majority of cases speakers were searching for the motion verb that exactly corresponded to particular motion events. (32) [ . . . ] ahm jetzt die Raupe, die ist 3auf den Stangel hochgegangen hat das4 [/ auf den Stangel hochgekrabbelt 3und die4 [/ und /] /] hat das Blatt angebissen. (4 years) [ . . . ] eh now the caterpillar she 3went up on the stalk, has4 [/ she /] climbed up on the stalk 3and she4 [/ and took a bit of the leaf. /] Ein Eichhornchen krabbelt, also klettert 3einen Berg4 [/ ah einen /] Baum rauf. (Adult) The squirrel crawls, so climbs up 3on a mountain4 [/ eh on a /] tree.


Second, speakers used a wide range of manner verbs in all age groups, sometimes involving very subtle nuances, for instance tappen, tippeln or trippeln (all of which may be translated into English as to go falteringly or to trip). Furthermore, they used a great number of dierent verbal particles, most of which expressed dierent aspects of path, for example contracted particles such as drauf or hinunter containing up to three different types of semantic information, as illustrated in (34): (34) drauf: da (there): her (towards): auf (up): hinunter: hin (towards): unter (down): d(a) (he)r auf general location in which motion takes place deixis direction (along a vertical axis) hin unter deixis direction of the movement (vertical axis)

Finally, speakers encoded path information not only in particles, but also in many other linguistic devices, such as prepositional phrases or adverbs, thereby producing very detailed path descriptions, as illustrated in (35). (35) [ . . . ] von rechts kommt eine Raupe ins Bild, bewegt sich auf einen Halm zu, klettert hinauf bis zum ersten Blatt [ . . . ]. (Adult)

Childrens verbalizations of motion events


[ . . . ] a caterpillar comes from the right into the screen, moves towards a blade, climbs up to the rst leaf [ . . . ]. A comparison of these results with those that were previously obtained in the same experimental situation for French and English (Hickmann et al. 2009) follows our hypotheses. As predicted on the basis of Talmys typology, our results concerning German are similar to those reported for English and dier signicantly from those reported for French. Like German speakers, English-speaking adults and children predominantly express manner and path together within their utterances. They encode manner in the main verb and path in particles and prepositions (e.g., to crawl up/down). However, children also produce some path-only responses in which they express sheer motion in the verb and path in satellites (e.g., to go up/down). In contrast, although French adults frequently express manner and path together, they do so less frequently and less systematically than English-speaking adults. French children tend to focus on path alone, encoding this information in the main verb (e.g., monter to ascend, descendre to descend) and they either do not express manner or express this information outside of the verb (in gerunds e.g., descendre en courant to descend by running or in adverbials e.g., monter avec les pattes to ascend with the paws). Dierences also occur at all ages as a function of event type. manner-only responses are rare in both languages and at all ages, with the exception of boundary-crossing events (see above). In addition, French provides a very frequent verb that simultaneously encodes manner and upwards direction (grimper climb up) inviting French speakers to produce more mannerpath responses with upward motion than with downward motion. A couple of additional points arose during our analyses of some German spatial devices and remain open. First, with respect to motion verbs, previous linguistic analyses of German (particularly Haggblade 1994; Weber 1983 among others) may need substantial qualications, particularly in relation to light verbs. Although the classication of gehen (to go) as a light verb is probably uncontroversial in most cases, its semantic content may dier across contexts (also see Di Meola 1994, for a study of kommen and gehen). For example, as illustrated in (36), this verb may encode manner information, particularly when used to describe motion in relation to the control items. In other contexts such as (37), however, it can encode information about deixis, particularly in descriptions of departures from the screen (which were not analyzed in the present study). (36) Experimenter: Child: Und die kleine Maus? Die geht. (4 years) And the mouse? She goes [walks].

234 (37)

A.-K. Ochsenbauer and M. Hickmann Die Raupe frisst ein bisschen vom Blatt wo sie hochgegangen ist, dann geht [?] sie wieder runter und geht [geht weg]. (4 years) The caterpillar eats a bit of the leaf where it went up, then it goes down under and goes [goes away].

Second, following Haggblade (1994: 43), our analysis included information about the setting and the ground among the semantic information that was encoded by subjects uses of linguistic devices outside of the verb. However, as suggested by some authors (e.g., Talmy 1985, 2000 or Slobin 1996, 2003b), this type of information is of a dierent nature and should not be included as part of the semantics of motion per se in the class of satellites. Excluding such spatial devices would imply a more conservative coding resulting in fewer satellites overall but in proportionally more satellites expressing path. It would therefore not invalidate our analysis and on the contrary increase the satellite-framed properties of German observed at all ages in our data. 6.2. Developmental progressions in German

Our results show the same lexicalization patterns among children and adults. From the earliest age tested onward (three years), German speakers express manner and path in compact utterances, encoding manner in the nite verb and path mostly in verbal particles. As predicted, manner is as salient to them as path, a result that follows from the typological properties of German as a satellite-framed language. Since German systematically encodes manner in the main verb, children seem to pay attention to this information from three years on and also encode it in their motion event descriptions. Nonetheless, several developmental progressions also occur, revealing a leap particularly between the ages of six and ten years. A rst progression concerns an increase in the semantic and syntactical complexity of childrens utterances. For instance, the complexity of linguistic devices encoding information outside of the main verb increased with age. As expected, young children most often used adverbs (e.g., da there, hier here) to locate the motion event and simple particles (e.g., rauf up, runter down) to describe path. It is only at around six years that prepositional phrases (e.g., auf dem Boden on the ground, auf den Baum on the tree) are used more frequently and with relative ease. This developmental progression was observed in relation to the devices that were used outside of the main verb. Although the semantics of these devices do not change over time, they are used increasingly with age. Children mostly use path particles that can be easily combined with a great number of dierent verbal stems. As children learn other devices,

Childrens verbalizations of motion events


they rst acquire particles as undierentiated and whole linguistic entities that are not analysed (neither morphologically nor semantically) and therefore produce frequent contracted forms (drauf onto, nunter down etc.). Qualitative analyses show that each child uses only one single form within a given set of particles diering with respect to the deictic element hin/her and their corresponding reduced forms n-/r- (e.g., nauf, nunter, nuber; rauf, runter, ruber; herauf, herunter, heruber). From very early on, young children use semantically complex satellite forms, even though we cannot assume that they know all semantic contrasts within a given paradigm (e.g., hinauf/herauf ). Second, the data show some changes across age groups with respect to verb use. Although motion verbs are quite diverse in all age groups, reecting in particular the highly salient nature of manner in German (as in other satellite-framed languages), children between three and six years also make frequent uses of light motion verbs (particularly gehen to go), thereby producing utterances that are morphologically and semantically simpler. Finally, as children get older, they show an increasing ability to organize discourse, as shown by the fact that they gradually learn to specify information about the setting and the ground. Unlike adults, children often do not provide sucient information for their listener to reconstruct the spatial universe of discourse. They typically only describe motion itself, without any spatial anchoring, for example without specifying the general location in which these events occurred, nor the source and goal locations implied by some of these events, making it dicult for their listener to interpret changes of location. A very similar developmental progression was also observed in previous analyses of childrens narratives (Hickmann 2003) across several dierent child languages (English, French, German, Chinese). Thus, although language-specic factors strongly contribute to shaping German childrens lexicalization patterns when they verbalize motion events, these factors alone cannot account for the fact that their responses show an increase in semantic density and in syntactic complexity. Other factors must clearly play a role in how these childrens spatial language changes with age. Thus, childrens cognitive system matures during language acquisition and some of these changes presumably underlie some changes in their linguistic system, for example an increase in their memory, processing, reasoning, and planning capacities, all of which are involved in complex discourse activities. In addition, the most striking developmental changes were observed at around six years, which corresponds to the age at which German children start school and are challenged in the domain of language.

236 7.

A.-K. Ochsenbauer and M. Hickmann Conclusion

The pattern found for German children is consistent with the one reported in other satellite-framed languages such as English and quite different from the one reported in verb-framed languages such as Spanish or French. When describing motion events, young children learning satellite-framed languages systematically express both manner and path within their utterances. The typological properties of their mother tongue simplify this task by compactly packaging these two types of information in constructions comprising linguistic devices that are among the rst morphemes to be mastered during language acquisition (verbal particles). Depending on their language, then, speakers within a given speech community choose to talk about or to ignore particular aspects of denoted situations. This process of selection presumably leads them to build up spatial representations that are partially characteristic of their language. In this sense, our results support the view that children partially construct the semantics of space in accordance with the language-specic characteristics of their mother tongue. Future research needs to address further questions concerning the depth of such typological constraints on speakers representations beyond language use. Received 1 March 2009 Revision received 25 November 2009 University of Munich/ University of Paris

Altmann, Hans and Silke Kemmerling. 2005. Wortbildung furs Examen. Gottingen: Vander hoeck and Ruprecht. Antell, Sue Ellen and Albert Caron. 1985. Neonatal perception of spatial relations. Infant Behaviour and Development 8. 1523. Bamberg, Michael. 1994. Development of linguistic forms: German. In Ruth Berman and Dan I. Slobin (eds.). Relating events in Narrative: A crosslinguistic developmental study, 189284. Hove: Erlbaum. Bowerman, Melissa. 1996a. Learning how to structure space for language. A crosslinguistic perspective. In Paul Bloom (ed.). Language and space, 385436. Cambridge: MIT Press. Bowerman, Melissa. 1996b. The origins of childrens spatial semantic categories: Cognitive versus linguistic determinants. In John J. Gumperz and Stephen Levinson (eds.). Rethinking linguistic relativity, 145176. Cambridge: Cambridge University Press. Bowerman, Melissa and Soonja Choi. 2003. Space under construction. Language-specic categorization in rst language acquisition. In Dedre Gentner and Susan Goldin-Meadow (eds.). Language in Mind. Advances in the study of language and thought, 387427. Cambridge: MIT Press. Choi, Soonja and Melissa Bowerman. 1991. Learning to express motion events in English and Korean: the inuence of language-specic lexicalization patterns. Cognition 41. 83121.

Childrens verbalizations of motion events


Croft, William, Johanna Baraddal, Willem Hollmann, Violeta Sotirova and Chiaki Taoka. 2008. Revising Talmys typological classication of complex events. Unpublished draft: http:/ / Di Meola, Claudio. 1994. Kommen und gehen. Eine kognitiv-linguistische Untersuchung der Polysemie deiktischer Bewegungsverben. Tubingen: Niemeyer. Gentner, Dedre. 1978. On relational meaning. The acquisition of verb meaning. Journal of Child Development 49. 988998. Haggblade, Elisabeth. 1994. Die Lexikalisierung von semantischen Komponenten in den Bewegungsverben. Berlin: microche version. Hickmann, Maya. 2006. The relativity of motion in rst language acquisition. In Hickmann, Maya and Stephane Robert (eds.), Space across languages: linguistic systems and cognitive categories, 281308. Amsterdam/Philadelphia: John Benjamins. Hickmann, Maya. 2007. Static and dynamic location in French: developmental and crosslinguistic perspectives. In Michel Aurnague, Maya Hickmann and Laure Vieu (eds.), Spatial entities in language and cognition, 205231. Amsterdam/Philadelphia: John Benjamins. Hickmann, Maya, Pierre Taranne and Philippe Bonnet. 2009. Motion in rst language acquisition: manner and path in French and English child language. Journal of Child Language, 36 (4). 705741. Johnston, Judith and Dan Isaac Slobin. 1979. The development of locative expressions in English, Italian, Serbo-Croatian and Turkish. Journal of Child Development 6, 530 545. Landau, Barbara and Ray Jackendo. 1993. What and Where in spatial language and spatial cognition. Behavioral and Brain Sciences 16(2). 21738. Mandler, Jean M. 1996. Preverbal Representation and Language. In Paul Bloom (ed.), Language and space, 365384. Cambridge: MIT Press. Piaget, Jean and Barbel Inhelder. 1947. La representation de lespace chez lenfant. Paris: Presses Universitaires de France. Slobin, Dan I. 1996. From thought and language to thinking for speaking. In John J. Gumperz and Stephen Levinson (eds.), Rethinking linguistic relativity, 7096. Cambridge: Cambridge University Press. Slobin, Dan I. 2003a. Language and thought online: cognitive consequences of linguistic relativity. In Dedre Gentner and Susan Goldin-Meadow (eds.), Language in mind: Advances in the study of language and thought, 157191. Cambridge, MA: MIT Press. Slobin, Dan I. 2003b. The many ways to search for a frog. In Sven Stromqvist and Ludo Verhoeven (eds.), Relating events in narrative: Typological and contextual perspectives, 219257. New Jersey: Erlbaum. Slobin, Dan I. 2006. What makes manner of motion salient: Explorations in linguistic typol ogy, discourse, and cognition. In Maya Hickmann and Stephane Robert (eds.), Space across languages: Linguistic systems and cognitive categories, 5981. Amsterdam: John Benjamins. Spelke, Elizabeth S. 2003. What makes us smart? Core knowledge and natural language. In Dedre Gentner and Susan Goldin-Meadow (eds.), Language in mind: Advances in the study of language and thought, 277311. Cambridge, MA: MIT Press. Talmy, Leonard. 1983. How language structures space. In Herbert Pick and Linda Acredolo (eds.), Spatial Orientation, 225282. New York: Plenum. Talmy, Leonard. 1991. Path to Realization. A Typology of Event Conation. In J. Sutton and C. Johnson (eds.), Proceeding of the seventeenth annual meeting of the Berkeley Linguistic Society, 480519. Berkeley. Talmy, Leonard. 2000. Towards a cognitive semantics: Concept structuring systems. Cambridge: Cambridge University Press.


A.-K. Ochsenbauer and M. Hickmann

Talmy, Leonard. 1985. Lexicalization Patterns. Semantic structure in lexical forms. In Timothy Shopen (ed.), Language typology and syntactic description: Grammatical categories and the lexicon, Vol. 3, 57149. Cambridge: Cambridge University Press. Tschander, Ladina. 1999. Bewegung und Bewegungsverben. In Ipke Wachsmuth and Bernhard Jung (eds.), Proceedings der 4. Fachtagung der Gesellschaft fur Kognitionswissen schaft, 2530. Bielefeld: Sankt Augustin. Weber, Gerhard. 1983. Untersuchungen zur mentalen Reprasentation von Bewegungsver ben. Merkmale, Dimensionen und Vorstellungsbilder. Dissertation. Hannover. Whorf, Benjamin Lee. 1956. Language, thought and reality. Cambridge, MA: MIT Press.

Appendix Stimuli T: Target items up and down (T1) (T2) (T3) (T4) (T5) (T6) A squirrel runs to a tree, up into and out of a hole in the tree, down, and away. A caterpillar crawls to a plant, up the stalk to eat a piece of leaf, down, and away. A bear walks to a tree, climbs up to a beehive to get some honey, climbs down to eat it, and walks away. A cat runs to a telephone pole, jumps up to a birds nest, drops an egg, jumps down to lick the egg, and runs away. A mouse tiptoes to a table, climbs up to take a piece of cheese, slides down, and tiptoes away. A monkey walks to a banana tree, climbs up to take a banana, then slides down and walks away.

C: Control items manner maximally salient (C1) squirrel running; (C2) caterpillar crawling; (C3) bear walking; (C4) cat running; (C5) mouse tiptoeing; (C6) kitten running.

What gestures reveal about how semantic distinctions develop in Dutch childrens placement verbs

Abstract Placement verbs describe every-day events like putting a toy in a box. Dutch uses two semi-obligatory caused posture verbs (leggen lay and zetten set/stand) to distinguish between events based on whether the located object is placed horizontally or vertically. Although prevalent in the input, these verbs cause Dutch children diculties even at age ve (Narasimhan and Gullberg, accepted). Children overextend leggen to all placement events and underextend the use of zetten. This study examines what gestures can reveal about Dutch three- and ve-year-olds semantic representations of such verbs. The results show that children gesture dierently from adults in this domain. Three-year-olds express only the path of the caused motion, whereas ve-year-olds, like adults, also incorporate the located object. Crucially, gesture patterns are tied to verb use: those children who over-use leggen lay for all placement events only gesture about path. Conversely, children who use the two verbs dierentially for horizontal and vertical placement also incorporate objects in gestures like adults. We argue that childrens gestures reect their current knowledge of verb semantics, and indicate a developmental transition from a system with a single

* Address for correspondence: M. Gullberg, Centre for Languages and Literature, Lund University, PO Box 201, 221 00 Lund, Sweden. Email: B. Narasimhan, University of Colorado at Boulder, Department of Linguistics, Hellems 290, 295 UCB, Boulder, CO 80309, U.S.A. Email: Acknowledgements: We gratefully acknowledge the cooperation of the teachers and students of the Kindercentrum Dribbel (Molenhoek, the Netherlands). We are also grateful for funding from the Max Planck Institute for Psycholinguistics. We wish to express our thanks to Judith Bindels, Bregje Esmeijer, Marieke Hoetjes, Anke Jolink, Ilonka Petal, Anne Rijpma, Femke Uijtdewilligen, Sonja Wichert, and Arna Van Doorn for help with data collection, coding, and establishment of interrater reliability, and to Melissa Bowerman, Asifa Majid, Leah Roberts and two anonymous reviewers for feedback and helpful suggestions. Any remaining errors are solely ours. Cognitive Linguistics 212 (2010), 239262 DOI 10.1515/COGL.2010.009 09365907/10/00210239 6 Walter de Gruyter


M. Gullberg and B. Narasimhan

semantic component(caused) movementto an (adult-like) focus on two semantic components(caused) movement-and-object. Keywords: gesture, verb semantics, Dutch, language development, placement.



How adult-like are childrens verb meanings in the early stages of development? Prior work on how children tune in to semantic patterns in the input has investigated childrens comprehension of verb meaning (e.g., Gentner 1978; Thomson and Chapman 1977) as well as their production of verbs in both elicited and spontaneous contexts (e.g., Choi and Bowerman 1991; Fisher et al. 1994; Gropen et al. 1991; Naigles and HoGinsberg 1998; Pye et al. 1996). However, with some notable exceptions (e.g., Anglin 1970; Bowerman 1978), surprisingly few studies have asked how adult-like childrens semantic systems are once forms are in use in production. We therefore know remarkably little about the nature of the semantic representations children operate with, what changes take place in the system over the course of development, and when such changes occur. This study explores childrens development of verb meaning and what semantic distinctions may underlie their extension patterns in the semantic domain of object placement by looking across modalities. More specically, we ask what childrens gestures about putting things in places can tell us about their developing semantic systems. 1.1. Placement, caused motion verbs, and their development

Children and adults talk frequently about the placement of objects such as putting a toy in a box. Object placement can be dened as events of caused motion where an object (a located or gure object) is moved to a location (a reference object or ground) with (typically manual) control exerted over the located object until it reaches its end location (cf. Bowerman et al. 2002; Bowerman et al. 2004). Placement (put) has long been a popular candidate for a cognitive and linguistically basic notion (Goldberg 1995; Pinker 1989), and children are assumed to acquire light verbs such as put early and easily (Clark 1978; Pinker 1989). But there is also crosslinguistic variation, for instance in the number of verbs that populate this domain and their level of semantic granularity (cf. papers in Ameka and Levinson 2007; Kopecka and Narasimhan to appear; Levinson and Wilkins 2006). Patterns range from single, light, all-purpose verbs like English put, via systems with a small number of (caused posture) verbs

Gestures and the development of semantic distinctions in Dutch


with more specic semantics and constrained extensions like set, stand and lay, to large inventories of very specic, classicatory placement verbs as in the Mayan languages. Moreover, languages sometimes have mixed systems with optionality between the use of light and more specic verbs, as in English where put co-exists with the rarer set, stand, and lay (cf. David 2003; Pauwels 2000). The acquisition of verbs in this domain also displays variation crosslinguistically (e.g., Chenu and Jisa 2006; Hansson and Bruce 2002; Hickmann and Hendriks 2006; Slobin, et al. in press). But interestingly, neither number of semantic distinctions made in a given semantic domain, nor optionality of use in the input seem to signicantly delay verb acquisition (e.g., Brown 1998; Narasimhan and Gullberg 2006). Based on a broad crosslinguistic comparison of the acquisition of placement verbs in languages that lexicalise path in verbs (verb-framed) vs. in satellites (satellite-framed, Talmy 1985), it has instead been suggested that acquisition is determined by many factors, including the interaction between semantic distinctions made in the verb and other non-verbal forms (e.g., case marking, adpositions) expressing relevant spatial information (Slobin et al. in press). Dutch uses a small set of caused posture placement verbs, zetten set and leggen lay. In addition to caused change of location, these monomorphemic verbs encode information about gure objects and their end conguration in that location or ground. Among other factors, the choice of verb for a given event depends on the properties of the object being located: its shape, its orientation, and its disposition with respect to the ground. Specically, the semantic distinctions concern the presence of a functional base and whether the gure object is resting on it, and whether the spatial extension or projected axis of the object is vertical or horizontal (Lemmens 2002, 2006; van Staden et al. 2006). For gure objects resting on their base, often extending vertically, zetten set is typically used, as in example (1). For gure objects lacking a functional base and/or extending horizontally, leggen, lay, is preferred, as seen in (2). (1) (2) zij zet de kop/de es op tafel she sets the cup/the bottle on the table zij legt de bal/de es op tafel she lays the ball/the bottle on the table

Dutch caused posture verbs are semi-obligatory and frequent in adult usage, and they are also ubiquitous in the input to Dutch children (Narasimhan and Gullberg accepted). In line with claims that children tune in very early to the habitual patterns of encoding in their language (Choi and Bowerman 1991; Slobin et al. in press), Dutch children might


M. Gullberg and B. Narasimhan

therefore be expected to acquire these verbs early, easily, and uniformly. However, these verbs cause unexpected diculties for children as old as four and ve leading to non-adult-like verb use (Narasimhan and Gullberg accepted). When Dutch childrens verb use is compared to that of adults for the same set of scenes, children are found to over-extend leggen lay and under-extend zetten set, seemingly picking one default verb to apply to all placement events. The question arises as to what semantic distinctions children who use leggen lay for all placement events actually operate with. One novel way to examine this question is to consider other available vehicles of meaning, namely speech-associated gestures, along with speech. 1.2. Gestures and language-specic meaning

Speech-associated gestures are closely linked to speech and language. Generally, speech and gesture are semantically, temporally and pragmatically coordinated such that the most meaningful part of a gesture, the stroke, typically is temporally coordinated with a part of speech expressing closely related meaning (Kendon 1980; McNeill 1992). Although theories about the speech-gesture relationship dier in their views on the locus and nature of the connection, the connection itself is undisputed (for a review, see De Ruiter 2007). Adults gestural practices dier crosslinguistically for various reasons (cf. Kendon, 2004). Recent research suggests that the variation is partially related to linguistic variation. Although gestures convey information in a dierent format from speech, they reect the linguistic choices speakers make: what information is considered newsworthy and when (McNeill 1992; McNeill, Levy and Pedelty 1990). Insofar as languages select dierent information for expression, gestural forms and their timing relative to speech thus dier crosslinguistically. For instance, gestures have been shown to be inuenced by how semantic components like path and manner of motion are lexicalised and packaged syntactically in a given lan guage (e.g., Duncan 1996, 2005; Gullberg et al. 2008; Kita and Ozyurek zyurek et al. 2005). Languages like 2003; McNeill and Duncan 2000; O Turkish, which expresses path and manner of motion in separate spoken clauses (e.g., descend [path] while rolling [manner]), also tend to be accompanied by gestures which express each component separately: one separate gesture for the path and another for the manner (e.g., Kita and Ozyurek 2003). Gestures also appear to be inuenced by verb semantics alone when information structure and syntactic packaging are kept constant. For instance, French and Dutch organise placement descriptions similarly

Gestures and the development of semantic distinctions in Dutch


(agent-action-object-location) and the simple transitive placement verbs project similar structures. However, the semantics of the placement verbs dier. French has a general placement verb mettre put, which chiey encodes the caused motion. French adults predominantly accompany placement descriptions by gestures expressing only the direction or path of the movement (Gullberg in press, submitted). This is in contrast to Dutch adults who instead chiey accompany their caused posture verbs by gestures incorporating the gure object with the direction of the gestural movement in hand shapes that reect the imagined object. These object-incorporating gestures are not restricted to a specic verb, but occur with both caused posture verbs (viz. leggen as well as zetten). Since the information structure and syntactic packaging of placement descriptions is similar across the two languages, the dierence in gesture patterns arguably stems from the dierent semantic specicity of the placement verbs. The Dutch gestural focus on objects seems to be prompted by the semantic distinction based on the object and its properties, viz. leggen lay for objects without a base extended horizontally, and zetten set for objects resting on their base, extending vertically. Conversely, the absence of a French gestural interest in objects seems to be inuenced by the relatively less specic verb semantics in French. The observed coordination between speech and gesture, which includes crosslinguistic dierences in semantic and syntactic distinctions, suggests that gestures can be seen as vehicles of language-specic meaning on a par with speech. They can therefore provide an additional window onto speakers event-related, semantic representations. 1.3. Gestures in language development

A growing body of research indicates that gestures and speech develop in parallel in childhood (e.g., Bates and Dick 2002; Capirci and Volterra 2008; Nicoladis et al. 1999; Volterra et al. 2005). However, despite the integration of the modalities, a number of studies also show that gestures serve as precursors to speech (e.g., Bates and Dick 2002; Tomasello et al. 2007), carrying more communicative weight in younger children (e.g., Guidetti 2005; Stefanini et al. 2008). A particular research tradition focuses on how gestures foreshadow speech such that non-redundant meaning is expressed in gesture before it can be expressed in speech in so called mis-matches (Church and Goldin-Meadow 1986). The presence of such gestures has been seen as an indication of transitional knowledge states and of a readiness to learn both language (e.g., Capirci et al. 1996; Goldin-Meadow 2007; Ozcaliskan and Goldin-Meadow 2005), and learning more generally (e.g., Alibali and Goldin-Meadow 1993; GoldinMeadow 2003; Pine et al. 2004) even beyond younger childhood.


M. Gullberg and B. Narasimhan

There is also evidence that gestures can be informative about the development of semantic representations in general (Capone, 2007), and about the development of language-specic semantics in particular. This latter aspect has been examined in the domain of motion where the realisation of semantic components like path and manner has been explored in speech and gesture. A series of studies investigating descriptions of motion and causal events in English and Turkish have shown that children between three and nine learning these languages overall display general (universal) patterns in younger childhood, and language-specic pat terns later on (e.g., Allen et al. 2003; Ozyurek and Ozcaliskan 2000). At age three, the children examined often display similar patterns crosslinguistically, conating elements of path and manner, or cause and path of motion, in gesture and in speech. Language-specic patterns emerge around age ve or six, depending on the study and construction examined. Particularly interesting is the observation that speech and gesture often express the same meaning even if neither modality is adult-like. For instance, the youngest Turkish childrens gestures diered from those of Turkish adults in that they conated cause and path of motion more often than Turkish adults did, but they were consistent with their own spoken descriptions which also conated these components more than adults (Furman et al. 2006). Similar ndings come from a study of the expression of path and manner of motion in French. French children aged four and six were adult-like in their tendency to both talk and gesture predominantly about path (Gullberg et al. 2008). In sum, these studies of childrens speech and gesture generally suggest that childrens gestures reect the meanings that they express in speech. This in turn suggests that childrens gestures can be informative about their semantic representations at a given point in time.


This study

The aim of the present study is to examine the nature of childrens semantic knowledge of placement verbs in more detail. We do this by considering how Dutch three- (N 5) and ve-year-olds (N 7) use gestures in parallel with Dutch caused posture verbs to describe object placement events compared to Dutch adults (N 10). We ask the following two questions: (1) Do Dutch children gesture like adults in the domain of placement? (2) If not, do Dutch childrens placement gestures dier depending on their patterns of use of placement verbs, and if so, how? Previous research leads us to expect Dutch adults to produce the caused posture verbs leggen lay and zetten set/stand in the description

Gestures and the development of semantic distinctions in Dutch


of placement events, and also to produce gestures that reect the semantic importance of gure objects through a preference for object-incorporation with the direction or path of gestural motion. Our rst analysis examines whether childrens gestures accompanying placement descriptions look overall adult-like. Second, to determine whether childrens use of placement verbs is adult-like, we investigate childrens deployment of verbs to a set of target events, which systematically vary the orientation of gure objects, and compare them to adults. Finally, we examine whether gesture use diers between those children whose verb use is adult-like versus those whose verb use is not in order to explore whether the information expressed in gesture can shed light on the semantic representations underlying the usage of placement verbs.



To elicit natural speech and gesture data while maintaining control over the extensions of placement verbs, we used a referential communication task (Yule, 1997) in the form of a Director-Matcher game. One participant, the Director, describes video clips depicting placement events to a confederate, the Matcher, who must then select the picture corresponding to the description from a set of possible options. The dyadic set-up as well as the information gap between the participants is conducive to gesture production despite the short, simple descriptions. We rst examine children and adults overall gesture production and the frequencies of use of object-incorporating versus path-only gestures. We then compare children and adults verb use to describe the same scenes in a subset of contrastive target placement events. Finally, we explore the connections between gesture production and verb use in individuals. 3.1. Participants

Participants were 29 children acquiring Dutch (aged 3;1 to 6;0) recruited through a Dutch preschool (Molenhoek, the Netherlands). For the purposes of this analysis, we excluded all children who produced fewer than three gestures during the task, leaving 12 children in total for analysis. The children fell naturally into two groups of children aged 3;14;5 (M 3;6, N 5), and children aged 5;16;0 (M 5;4, N 7). For ease of exposition, the child groups are referred to as three-year-olds and veyear-olds. Additionally, 29 adult native speakers of Dutch were tested as


M. Gullberg and B. Narasimhan

controls, 10 of whom produced more than three gestures and were therefore retained for analysis. 3.2. Materials

The stimuli, developed for a crosslinguistic comparison of placement event descriptions (cf. Narasimhan and Gullberg 2006; accepted), consisted of a set of video clips showing a female actor manually placing gure objects (henceforth simply objects) on a shelf or a table top. Sixteen target events showed eight objects (a doll, a monkey, a bear, a dog, a can, a book, a ashlight, and a picture frame) being placed either in a vertical or horizontal position at a location (see the Appendix, target events listed in boldface). Twenty ller events and 3 warm-up items showed a range of other objects being dropped, squeezed, etc. These were not expected to elicit placement verbs. The stimulus clips were randomized and organized into two orders. The presentation of the stimulus order was counterbalanced. A set of still photos of the objects in their end location was also produced. 3.3. Procedure

Participants were tested individually and given oral instructions that they were going to play a game where they had to help one person (Experimenter2) put a set of pictures in the right order. Participants saw one video clip at a time on a laptop screen manipulated by Experimenter1. Experimenter2, who could not see the video screen, asked the participants What did the woman do?. Based on the participants descriptions, Experimenter2 chose the correct still image from the set of stills depicting the placement scenes. If participants gave a simple locative expression or an intransitive description (e.g., the book is/lies on the table), then Experimenter2 asked What happened or What did the woman do? Adults controlled the computer themselves. The testing procedure was otherwise identical for adults and children. The session started with three warm-up items. The entire testing session was audio- and video taped. 3.4. Data treatment

3.4.1. Speech. Native speakers of Dutch transcribed the rst spontaneous transitive description of each video clip (cf. Plumert et al. 1995). An (adult) example is given in (3), with the rst transitive description in boldface. (3) ze pakt een dingetje . . . zon knuelbeertje en die zet ze op tafel she takes a thingy . . . a little teddy bear and that she sets on [the] table

Gestures and the development of semantic distinctions in Dutch


The placement verbs were selected for further analysis. Where two utterances described the same scene with dierent object labels, the rst one was selected. Finally, in cases of self-corrections, the rst immediately following complete and/or interpretable description was retained. A similar procedure was applied to uninterpretable utterances. 3.4.2. Gesture. The narrow focus on the rst descriptions is particularly important for the gesture analysis. Gestures are sensitive to information structure and tend to co-occur with the most newsworthy element. In the rst description of the placement event, that information is the placement act itself in conjunction with the ground. In contrast, in elaborations prompted by questions, other spatial information is often targeted such as specic locations like at the right-hand corner on top. Gestures accompanying such elaborations are often deliberately demonstrative, sometimes aligning with spoken deictic expressions referring to the gesture (like this). These gestures therefore target other information and are potentially driven by other mechanisms than gestures performed without any particular demonstrative intent. Also excluded from analysis, and for similar reasons, were gestures occurring with disuencies or multiple hesitation phenomena (cf. Gullberg 1998). Using frame-by-frame analysis of digital video in video annotation software (ELAN, http:/ /, we identied gestures occurring with the spontaneous rst descriptions of the placement events. Specically, we identied gestural strokes, that is, the expressive part of the gestural movement where the spatial excursion of the limb reaches its apex, and post-stroke holds, or cases where the hands are temporarily immobile in gesture space before moving on (Kendon 1972, 2004: 111112; Kita et al. 1998; Seyfeddinipur 2006). All gestures thus identied were then coded for whether they encoded (gure) object information, or only direction or path of movement. This coding was done with sound turned o and was based on the structural properties of the gestures alone to avoid circularity when gesture information was compared to speech information. Gestures were coded as expressing object information when they displayed a hand shape that reected and incorporated the gure object into the movement. Gestures were coded as expressing only path of movement when they expressed a spatial excursion (cf. Kendon 2004) laterally, vertically or sagittally from the speakers body and displayed no particular hand shape, that is, a relaxed, oppy hand or a pointing hand shape. Examples of these categories are displayed in Figure 1a (Object-incorporation) and Figure 1b (Path-only). Finally, in the same annotation software with sound turned back on, we also transcribed the speech that co-occurred exactly with the gesture


M. Gullberg and B. Narasimhan

Figure 1a. Example of gesture coded as Object-incorporating displaying a hand shape indicating the presence of a gure object.

Figure 1b. Example of gesture coded as Path-only displaying a at hand with no hand shape indicating the presence of a gure object.

stroke, although no detailed speech-gesture alignment analysis was performed for this study. Interrater reliability of the gesture coding was established by having a second coder judge the data. The interrater reliability for gesture identication was .94 (N 235) and for form coding (object-incorporation vs. path-only) .92. In cases of discrepancy, the judgement of the second coder was retained. Table 1 summarises the total number of gestures per age group.
Table 1. Number of gestures per age group # speakers 3-year-olds 5-year-olds Adults Total 5 7 10 22 # gestures 66 70 99 235



The dependent variables are proportions of gestures per participant expressing object-incorporation vs. path-only, and proportion of verb types used per participant. Because the dependent variables are proportions, they were arcsine transformed for statistical analysis (Howell 2002); however, non-transformed values are reported in tables, gures and text. Analyses of gesture data draw on non-parametric statistical tests, specically Kruskal-Wallis for comparisons of multiple independent samples and Mann-Whitney for comparisons of two independent samples. Speech

Gestures and the development of semantic distinctions in Dutch


data are analysed with parametric one-way ANOVAs followed by Tukey HSD tests for post-hoc comparisons. 4. 4.1. Results Overall gesture use

We rst examine whether Dutch children produce the same gestures and to the same extent as Dutch adults, excluding warm-up items but including both target and ller events. Figure 2 summarises the mean proportion of gestures that express object-incorporation in the form of objectrelated hand shapes or path-only as a function of age (3 years, 5 years, adults).

Figure 2. Mean proportion of gestures expressing object-incorporation in hand shape (Obj) or path-only (Path) as a function of age. (Error bars standard error).

Adult Dutch speakers show a clear preference for incorporating object information in gestures that accompany placement descriptions. They produce gestures with hand shapes that incorporate objects in the gestural movement. Moreover, the occurrence of these object-incorporating gestures is not restricted to a specic verb, but they occur with both verbs across the board. These data replicate previous ndings showing a robust adult Dutch gestural preference for object-incorporation with placement descriptions (Gullberg in press, submitted). The child data look strikingly dierent. The youngest children in particular almost exclusively produce gestures that express only path. In order to investigate whether there was a dierence in the overall pattern of gesture usage across the three age groups a Kruskal-Wallis test was run on the mean proportion of object-incorporating gestures


M. Gullberg and B. Narasimhan

(Obj 1) with age (3, 5, adults) as the between-subject factor. The groups diered signicantly in the mean proportion of object-incorporating gestures ( w 2 (2, N 22) 11.46, p < 0.01). Specically, 3-year-olds produced signicantly fewer object-incorporating gestures (M 1%, SD 2%) than both 5-year-olds (M 47%, SD 27%; z 2.44, p 0.02) and adults (M 62%, SD 15%; z 3.09, p < 0.001), who did not dier from each other (z 1.52, p 0.13). The youngest children clearly prefer to express only path in their gestures accompanying placement descriptions, and only very rarely do they express object information. 5-year-olds express considerably more object-incorporation, although their preferences do not numerically match those of adult speakers. Children thus gesture dierently from adults. They do not appear to imitate the adult gestural input, nor to imitate the practical placement actions by enacting a placement event with a symbolised, imagined object (cf. Capirci et al. 2005). 4.2. Verb use

We next investigate whether children use the same verbs to describe the same scenes as adults, that is, whether they have the same extension patterns as adults or convey the same meaning with the verbs as adults do. We focus on verb use for the 16 target items, which systematically vary object orientation. We group the target scenes by orientation into two groups of 8 scenes each: horizontal and vertical placement. All verb responses, including inappropriate forms for a given orientation, went into the analysis. For each age group, the mean proportion of responses per verb type (leggen, zetten, and OTHER) was computed (cf. Narasimhan and Gullberg accepted).1 Figure 3 summarises the mean proportion of verbs used to describe horizontal (Figure 3a) and vertical items (Figure 3b), respectively, as a function of age. For horizontal items the typical adult verb choice is leggen lay. All age groups overwhelmingly used the verb leggen for items placed horizontally. The three-year-olds also used a sprinkling of OTHER verbs. Oneway ANOVAs for each verb type with age group as the between-subject factor2 revealed no dierence between the groups in use of leggen lay (F(2,19) 1.66, p 0.22), or zetten set/stand (F > 1). However, the

1. This analysis is similar to the one performed in Narasimhan and Gullberg (accepted), but is performed here on a sub-set of those data, viz. only on speech data from those participants who also gesture. 2. Because an items analysis on as few items as 8 is dicult to interpret, no items analysis was performed.

Gestures and the development of semantic distinctions in Dutch


Figure 3a. Mean use of leggen, zetten, and OTHER in Dutch for 8 horizontal target scenes across age groups (error bars standard error).

Figure 3b. Mean use of leggen, zetten, and OTHER in Dutch for 8 vertical target scenes across age groups (error bars standard error).

groups did dier in their use of OTHER (F(2,19) 4.94, p 0.02), with three-year-olds using signicantly more OTHER verbs (M 13%, SD 21%) than ve-year-olds (M 2%, SD 6%; Tukey HSD p 0.04) and adults


M. Gullberg and B. Narasimhan

(M 1%, SD 4%; Tukey HSD p 0.02), who did not dier.3 Critically, the groups did not dier in the use of leggen. Dutch children thus use leggen for horizontal items as often as adults already by the age of three and half. For vertical items, the standard verb choice should be the verb zetten set/stand. This is the only verb used by adults, but children behave surprisingly dierently. The youngest children use leggen lay for more than half of the vertical items and only rarely use zetten. One-way ANOVAs for each verb type with age group as the between-subject factor revealed a dierence between the age groups in the use of zetten set/stand (F(2,19) 11.02, p < 0.001), with three-year-olds using zetten signicantly less (M 21%, SD 39%) than adults (M 100%; Tukey HSD p < 0.001), and ve-year-olds also using zetten signicantly less than adults (M 73%, SD 33%; Tukey HSD p 0.03). Despite the numerical dierence, the child groups did not dier statistically from each other. In contrast, the two child groups did dier in their use of leggen lay for vertical items (F(1,11) 6.31, p 0.03), with the three-year-olds using signicantly more leggen (M 61%, SD 31%) than ve-year-olds (M 9%, SD 15%). The child groups did not dier in their use of OTHER, however (F(1,11) 3.19, p 0.11). The Dutch three- and ve-year-olds in this sample thus dier from adults in their under-use of zetten set/stand for vertical items, and both child groups dier from adults in that they use leggen lay to describe vertical scenes, three-year-olds signicantly more so than ve-year-olds. Some children thus use leggen across the board for all placement scenes. 4.3. Gesture use with zetten and leggen

We nally turn to the question whether gesture use diers between those children who use both leggen lay and zetten set/stand and those who over-extend leggen to all placement regardless of orientation of the object. Figure 4 summarises the mean proportion of gestures that express objectincorporation in the form of object-related hand shapes or path-only as a function of whether children over-extend leggen and chiey use one verb (N 6), or whether they use two verbs (N 6) to describe the 16 target scenes. The adult data are included for ease of comparison. When gestures are considered in parallel with speech, a binomial distribution is found such that children who mainly use only one verb, leggen

3. We could not perform an omnibus (repeated measures) ANOVA on all verb types across age groups since not all groups used all verbs. The same argument holds for the analysis of vertical items.

Gestures and the development of semantic distinctions in Dutch


Figure 4. Mean proportion of gestures expressing object incorporation in hand shape (Obj), or path-only (Path) as a function of whether children use one verb or two verbs to describe placement verbs. (Error bars standard error).

lay, to describe all 16 target events also predominantly produce gestures that express path-only. In contrast, children who use two verbs, both leggen lay and zetten set/stand, to describe horizontal and vertical placement, respectively, also produce object-incorporating gestures, even if their proportions do not quite match those of adults. Again, a KruskalWallis test (Obj 1) revealed that the groups diered signicantly in the mean proportion of object-incorporating gestures (w 2 (2, N 22) 12.15, p < 0.01). Specically, children using mainly one verb produced signicantly fewer object-incorporating gestures (M 8%, SD 16%) than both children using two verbs (M 46%, SD 26%; z 2.25, p 0.03) and adults (M 62%, SD 15%; z 3.28, p < 0.001), who did not dier from each other (z 1.47, p 0.15). Finally, a correlation analysis was run on the mean proportion of object-incorporating gestures and the mean proportion of leggen lay used for vertical scenes across the age groups. The analysis revealed that with decreasing use of leggen for vertical placement, the proportion of gestures expressing object-incorporation increases signicantly (r(20) 0.7475, t(20) 5.03, p < 0.001). This indicates that as children cease to label vertical scenes with leggen lay and shift to using zetten set/stand, they are also more likely to produce object-incorporating gestures. 5. General discussion

This study examined how Dutch three- and ve-year-olds use gestures in parallel with caused posture placement verbs to describe object placement events compared to Dutch adults. There are two main ndings. First,


M. Gullberg and B. Narasimhan

children in these age groups gesture dierently from adults. Dutch adults show a robust preference for expressing objects and direction or path of movement simultaneously in placement gestures, a result replicating previous ndings (Gullberg in press; submitted). In contrast, three-year-olds show a strong bias towards gestures that express only the path of the (caused) movement, and although ve-year-olds are more likely to produce object-incorporating gestures, they are numerically still not adultlike in their preferences. Second, children who use the placement verbs in non-adult-like ways in speech also gesture in non-adult-like ways. That is, children who overextend leggen lay to all placement scenes express only the path of the movement in gesture. In contrast, children who use both leggen lay and zetten set/stand dierentially for horizontal and vertical placement respectively also incorporate objects in gestures like adults. They are adultlike in both speech and gesture, targeting information about objects and movement in both modalities. What can these ndings tell us about the semantic distinctions that underlie childrens extension patterns in the semantic domain of object placement? The gesture data suggest that children who use leggen lay to label all placement events are only targeting one semantic component, namely the movement or motion component of the caused motion verbs, as seen in their gestures expressing only path. These children do not seem to care about the object being moved and its properties. Recall that in the adult system some attention to the object is necessary for the choice of a specic placement verb to a given scene, which is arguably what prompts adults to also gesture about objects. There is no evidence in the childrens gestures that the objects matter at this stage. Consequently, it is as if the verb leggen has an over-general meaning for children, cause to move or put. Similarly, the object-incorporating gestures produced by children who use both leggen and zetten dierentially suggest that they have tuned their attention to encompass the object. Specically, the gestural incorporation of the object suggests that these children have included the objects in their semantic representations of the caused posture verbs. We therefore argue that speech and gesture together indicate a developmental transition from a system with a single semantic component based on (caused) movement-only, reected in the use of one single verb (leggen), to an (adult-like) focus on (caused) movement-and-object mirrored in adult-like use of two verbs (leggen and zetten). Notice that the crucial issue is not whether children notice the objects more generally, but whether the object is included in the representations of the transitive caused posture verbs. Even children who use leggen for all transitive placement descriptions occasionally talk and gesture about

Gestures and the development of semantic distinctions in Dutch


the objects. However, the objects are not included in the compact transitive caused posture verb descriptions, but appear elsewhere. In (4) a child comments on the object outside the transitive description for a scene where a monkey is placed in a standing position: (4) nou die [staat] hij legt hem op de tafel well it [stands] he lays him on the table (DuCh18 aged 4;5)

The child rst describes the end state of the object using a correct intransitive posture verb, staan stand. She immediately follows this by a description of the placement event itself using leggen lay. Interestingly, the gesture accompanying the intransitive verb staan, indicated in square brackets, still expresses path-only, and more specically, movement towards the ground. There is no hand shape indicating an object-incorporation. The example nevertheless highlights (a) that the child is not confused about the objects orientation (vertical, standing), and (b) that she attempts to express both information about the object in its end state in staan and the caused motion in leggen. This may be a precursor stage to bringing the two elements together in one single adult-like representation. One might wonder why not all children gesture about objects. Given the nature of placement events, children could have been expected to imitate the practical action as perceived and enact the placing of an object with a symbolised, imagined object in hand (cf. Capirci et al. 2005; McNeill 2005). They might also have been expected to imitate the gestural input provided by Dutch adults as they talk and gesture about placement. Finally, children could even have been expected to gesture about objects on theoretical grounds, given the documented tendency for speakers to make more ne-grained distinctions at the goal of a path of motion (e.g., Lakuta and Landau 2005; Regier and Zheng 2007). The object in its end conguration is arguably a goal-related part of the caused motion. The fact that children do not, and that there is a developmental trend from gesturing about movement-only towards adult-like gesturing about movement-and-object simultaneously suggest that childrens gestures in this domain are inuenced by their linguistic activities, and, more specically, by the semantic distinctions they operate with, even at young ages. The converse question is why younger children so overwhelmingly target path or the direction of movement alone in their placement gestures. One possibility is that this is a reection of communicative development, as suggested by Clark and Grossman (1998). The youngest children may interpret the communicative goal of the situation dierently from older children, and focus on direction or path towards the goal ground. However, this does not explain the strong co-occurrence of such gestures with


M. Gullberg and B. Narasimhan

childrens overextension patterns with the verb leggen lay. A compensatory account of gestures might suggest that children gesture about path because they do not talk about it.4 Although they use caused motion verbs, it could be argued that the path element is not explicit in these verbs. However, the data contain examples of children using the frequent path-prexed forms of the Dutch caused posture verbs, such as neerleggen down.lay and inleggen in.lay. There is no dierence between gestures accompanying such explicitly path-prexed verbs and the bare caused posture verbs. It therefore does not seem to be a matter of pathcompensation in gesture. A third possibility is that the dierence in gesturing has nothing to do with the semantics of Dutch specically. Instead, the path component may be a more universally basic motion element, as suggested by Talmy (1985), which all children therefore target initially. More crosslinguistic data is needed in the placement domain to address that issue. An additional option is that children do not target path information per se but rather that the path gestures reect a more basic focus on change.5 Kamp (1980) has argued that change is the fundamental basis for any event structure. In the study at hand, both a focus on change in general or a more specic focus on path would yield path gestures. It remains an issue for future research to disentangle these options. Finally, there is no evidence in the data that gestures foreshadow speech such that children who use leggen lay for both vertical and horizontal placement produce object-incorporating gestures, using gesture to indicate an interest in objects that they are not yet able to express in speech. These results seem to run counter to ndings in the literature on cognitive development where children are found to express aspects of mathematical reasoning in gesture not yet accessible to them in speech (e.g., Alibali and Goldin-Meadow 1993; Goldin-Meadow 2003; Pine et al. 2004). The absence of such mis-matches in the current data does not invalidate the basic observation that gestures reect childrens current knowledge of placement verb semantics. First, their absence can simply be a sampling accident. Given the small number of children in this study, we cannot exclude the possibility that some children may gesture about objects while still over-extending leggen to all placement events. However, an alternative explanation is that children engaged in reasoning tasks have more room to express alternative, additional, or dierent meanings in gesture than do children who talk and gesture about as mundane and

4. This was suggested by an anonymous reviewer. 5. We thank an anonymous reviewer for this suggestion.

Gestures and the development of semantic distinctions in Dutch


simple things as putting toys on tables. The nature of the task and the age range examined here may both contribute to the patterns observed. Moreover, support for non-compensatory gesture production in language development comes from other developmental studies in the domain of voluntary motion. Turkish three-year-olds have been shown to dier from Turkish adults in gesture and speech, with the childrens gestures critically matching their own speech (Furman et al. 2006). Similarly, French children aged four and six talking about path and manner overwhelmingly gesture and talk about the same elements (Gullberg et al. 2008). This is despite the fact that the complex constructions for expressing manner in French might have led children wanting to express all aspects of motion to talk about path and instead gesture about manner. However, even the youngest children look adult-like and gesture about path when talking about path, and gesture about manner only when speaking about manner. Similar consistency is reported for bilingual French-English children (Nicoladis and Brisard 2002). The view of childrens gestures as mainly compensatory at these ages thus receives little support here. In conclusion, this study suggests that Dutch childrens knowledge of placement verb semantics moves from a focus on (caused) movementonly to a focus on (caused) movement-and-object in conjunction. Part of childrens diculties with the Dutch caused posture placement verbs seem to be related to understanding the role of the object as a necessary semantic component in these monomorphemic, portmanteau verbs that conate cause, motion, and properties of the object in one form (cf. Narasimhan and Gullberg accepted). The transition from a system based on one single semantic distinction to a system with an adult-like focus on movementand-object is not necessarily complete by age ve, but seems to continue to develop in later childhood. How and exactly when this transition takes place is an empirical question. More generally, the study highlights the value of studying childrens gestures as a means of examining semantic representations in development. Gestures, as vehicles of language-specic meaning, provide a window on the details of what semantic elements underpin non-adult-like use of forms in inappropriate contexts, as well as what elements undergo change in switches towards more adult-like speech. Gestures allow us to go beyond error analysis of speech, stating merely that childrens verb meanings differ from those of adults, and allow us to explore how they dier. Received 1 March 2009 Revised received 25 November 2009 Max Planck Institute for Psycholinguistics/ University of Colorado


M. Gullberg and B. Narasimhan

Alibali, Martha W., and Susan Goldin-Meadow. 1993. Gesture-speech mismatch and mechanisms of learning: What the hands reveal about a childs state of mind. Cognitive Psychology 25(4). 468523. Allen, Shanley, Asli Ozyurek, Amanda Brown, Reyhan Furman and Tomoko Ishizuka. 2003. Early speech about manner and path in Turkish and English: Universal or language-specic? In Barbara Beachley, Amanda Brown and Frances Conlin (Eds.), Proceedings of BUCLD 27, 6372. Somerville: Cascadilla Press. Ameka, Felix K., and Stephen C. Levinson (eds.). 2007. Locative predicates. [Special Issue]. Linguistics 45 (5/6). Anglin, Jeremy. 1970. The growth of word meaning. Cambridge, MA: MIT Press. Bates, Elizabeth and Fred Dick. 2002. Language, gesture, and the developing brain. Developmental psychobiology 40(3). 293310. Bowerman, Melissa. 1978. Systematizing semantic knowledge: Changes over time in the childs organization of word meaning. Child Development 49(4). 997987. Bowerman, Melissa, Penelope Brown, Sonia Eisenbeiss, Bhuvana Narasimhan, and Dan I. Slobin. 2002. Putting things in places. Developmental consequences of linguistic typology. In Eve V. Clark (ed.), Space in Language. Location, Motion, Path, and Manner, S1S122. Stanford: CLS. Bowerman, Melissa, Marianne Gullberg, Asifa Majid and Bhuvana Narasimhan. 2004. Put project: The cross-linguistic encoding of placement events. In Asifa Majid (ed.), Field Manual (Vol. 9, pp. 1018). Nijmegen: Max Planck Institute for Psycholinguistics. Brown, Penelope. 1998. Childrens rst verbs in Tzeltal: Evidence for an early verb category. Linguistics 36(4). 715753. Capirci, Olga, Annarita Contaldo, Maria C. Caselli and Virginia Volterra. 2005. From action to language through gesture: A longitudinal perspective. Gesture 5(1/2). 155177. Capirci, Olga, Jana M. Iverson, Elena Pizzuto and Virginia Volterra. 1996. Gestures and words during the transition to two-word speech. Journal of Child Language 23(3). 645 675. Capirci, Olga, and Virginia Volterra. 2008. Gesture and speech. The emergence and development of a strong and changing partnership. Gesture 8(1). 2244. Capone, N. C. 2007. Tapping toddlers evolving semantic representation via gesture. Journal of Speech and Hearing Research 50. 732745. Chenu, Florence and Harriet Jisa. 2006. Caused motion constructions and semantic generality in early acquisition of French. In Eve V. Clark and Barb F. Kelly (Eds.), Constructions in acquisition, 233261. Stanford: CLS. Choi, Soonja, and Melissa Bowerman. 1991. Learning to express motion events in English and Korean: The inuence of language-specic lexicalization patterns. Cognition 41(13). 83 121. Church, R. Breckinridge, and Susan Goldin-Meadow. 1986. The mismatch between gesture and speech as an index of transitional knowledge. Cognition 23(1). 4371. Clark, Eve V. 1978. Discovering what words can do. In Farkas, Donka, Wesley Jacobsen and Karol Todrys (eds.), Papers from the parasession on the lexicon, 3457. Chicago: CLS. Clark, Eve V., and James B. Grossman. 1998. Pragmatic directions and childrens word learning. Journal of Child Language 25(1). 118. David, Caroline. 2003. Les verbs of putting: Typologie, schema syntaxique et organisation semantique des constructions prepositionnelles en anglais contemporain. Poitiers, Universite de Poitiers, PhD thesis.

Gestures and the development of semantic distinctions in Dutch


De Ruiter, Jan-Peter. 2007. Postcards from the mind: The relationship between speech, gesture and thought. Gesture 7(1). 2138. Duncan, Susan. 1996. Grammatical form and thinking-for-speaking in Mandarin Chinese and English: An analysis based on speech-accompanying gesture. Chicago, University of Chicago, PhD dissertation. Duncan, Susan. 2005. Co-expressivity of speech and gesture: Manner of motion in Spanish, English, and Chinese. In Charles Chang et al. (eds.), Proceedings of the 27th annual meeting of the Berkeley Linguistic Society, 353370. Berkeley: BLS. Fisher, Cynthia, D. G. Hall, S. Rakowitz and Lila Gleitman. 1994. When it is better to receive than to give: Syntactic and conceptual constraints on vocabulary growth. Lingua 92. 333375. Furman, Reyhan, Asli Ozyurek and Shanley Allen. 2006. Learning to express causal events across languages: What do speech and gesture patterns reveal? In David Bamman, Tatiana Magnitskaia and Colleen Zaller (eds.), Proceedings of the 30th BUCLD, 190201. Somerville: Cascadilla Press. Gentner, Dedre. 1978. On relational meaning: The acquisition of verb meaning. Child Development 49(4). 988998. Goldberg, Adele. 1995. Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Goldin-Meadow, Susan. 2003. Hearing gesture: How our hands help us think. Cambridge, MA: The Belknap Press. Goldin-Meadow, Susan. 2007. Pointing sets the stage for learning languageand creating language. Child Development 78(3). 741745. Gropen, J., Steve Pinker, M. Hollander and R. Goldberg. 1991. Syntax and semantics in the acquisition of locative verbs. Journal of Child Language 18(1). 115151. ` Guidetti, Michele. 2005. Yes or no? How young French children combine gestures and speech to agree and refuse. Journal of Child Language 32(4). 911924. Gullberg, Marianne. 1998. Gesture as a communication strategy in second language discourse. A study of learners of French and Swedish. Lund: Lund University Press. Gullberg, Marianne. in press. Language-specic encoding of placement events in gestures. In Eric Pederson and Ju rgen Bohnemeyer (eds.), Event representations in language and cognition. Cambridge: Cambridge University Press. Gullberg, Marianne. Submitted. Linguistic representations inuence speech-associated gestures in the domain of placement. Gullberg, Marianne, Henriette Hendriks, and Maya Hickmann. 2008. Learning to talk and gesture about motion in French. First Language 28(2). 200236. Hansson, Kristina and Barbro Bruce. 2002. Verbs of placement in Swedish children with SLI. International Journal of Communication Disorders 37(4). 401414. Hickmann, Maya and Henriette Hendriks. 2006. Static and dynamic location in French and English. First Language 26(1). 103135. Howell, David C. 2002. Statistical Methods for Psychology (5th ed.). Pacic Grove, CA: Duxbury. Kamp, Hans. 1980. Some remarks on the logic of change, part 1. In Christian Rohrer (ed.). Time, tense, and quantiers, 135180. Tuebingen: Niemeyer. Kendon, Adam. 1972. Some relationships between body motion and speech: An analysis of an example. In Aron W. Siegmann and Benjamin Pope (eds.), Studies in dyadic communication, 177210. New York: Pergamon. Kendon, Adam. 1980. Gesticulation and speech: Two aspects of the process of utterance. In Key, Mary R. (Ed.), The relationship of verbal and nonverbal communication, 207227. The Hague: Mouton.


M. Gullberg and B. Narasimhan

Kendon, Adam. 2004. Gesture. Visible action as utterance. Cambridge: Cambridge University Press. Kita, Sotaro and Asli Ozyurek. 2003. What does cross-linguistic variation in semantic coordination of speech and gesture reveal?: Evidence for an interface representation of spatial thinking and speaking. Journal of Memory and Language 48(1). 1632. Kita, Sotaro, Ingeborg Van Gijn and Harry van der Hulst. 1998. Movement phases in signs and co-speech gestures, and their transcription by human coders. In Ipke Wachsmuth and Martin Frohlich (eds.), Gesture and Sign Language in human-computer interaction, 2335. Berlin: Springer. Kopecka, Annette and Bhuvana Narasimhan (eds.). To appear. Events of putting and taking: A crosslinguistic perspective. Amsterdam: Benjamins. Lakuta, Linda and Barbara Landau. 2005. Starting at the end: The importance of goals in spatial language. Cognition 96(1). 133. Lemmens, Maarten. 2002. Tracing referent location in oral picture descriptions. In Andrew Wilson, Paul Rayson and Tony McEnery (eds.), A rainbow of corpora. Corpus linguistics and the languages of the world, 7385. Munchen: Lincom-Europa. Lemmens, Maarten. 2006. Caused posture: experiential patterns emerging from corpus research. In Anatol Stefanowitsch and Stefan Gries (eds.), Corpora in cognitive linguistics. Corpus-based approaches to syntax and lexis, 263298. Berlin: Mouton. Levinson, Stephen C. and David Wilkins (eds.). 2006. Grammars of space. Explorations in cognitive diversity. Cambridge: Cambridge University Press. McNeill, David. 1992. Hand and mind. What the hands reveal about thought. Chicago: University of Chicago Press. McNeill, David. 2005. Gesture and thought. Chicago: University of Chicago Press. McNeill, David and Susan Duncan. 2000. Growth points in thinking-for-speaking. In McNeill, David (Ed.), Language and gesture, 141161. Cambridge: Cambridge University Press. McNeill, David, Elena Levy and Laura Pedelty. 1990. Speech and gesture. In Hammond, Georey R. (Ed.), Cerebral control of speech and limb movements, 203256. Amsterdam: North Holland. Naigles, Letitia R., and Erika Ho-Ginsberg. 1998. Why are some verbs learned before other verbs? Eects of input frequency and structure on childrens early verb use. Journal of Child Language 25(1). 95120. Narasimhan, Bhuvana and Marianne Gullberg. 2006. Perspective-shifts in event descriptions in Tamil child language. Journal of Child Language 33(1). 99124. Narasimhan, Bhuvana and Marianne Gullberg. Accepted. The role of input frequency and semantic transparency in the acquisition of verb meaning: Evidence from placement verbs in Tamil and Dutch. Journal of Child Language. Nicoladis, Elena and Frank Brisard. 2002. Encoding motion in gestures and speech: Are there dierences in bilingual childrens French and English? In Clark, Eve V. (Ed.), Space in Language. Location, Motion, Path, and Manner, 6068. Stanford: CLS. Nicoladis, Elena, Rachel I. Mayberry and Fred Genesee. 1999. Gesture and early bilingual development. Developmental Psychology 35(2). 514526. Ozcaliskan, Seyda and Susan Goldin-Meadow. 2005. Gesture is at the cutting edge of early language development. Cognition 96(3). B101B113. Ozyurek, Asli, Sotaro Kita, Shanley Allen, Reyhan Furman and Amanda Brown. 2005. How does linguistic framing of events inuence co-speech gestures? Insights from crosslinguistic variations and similarities. Gesture 5(1/2). 219240. Ozyurek, Asli, and Seyda Ozcaliskan. 2000. How do children learn to conate manner and path in their speech and gestures? In E. V. Clark (Ed.), Proceedings of the 30th Stanford Child Language Research Forum, 7785. Stanford: CLS.

Gestures and the development of semantic distinctions in Dutch


Pauwels, Paul. 2000. Put, set, lay and place: A cognitive linguistic approach to verbal meaning. Munchen: Lincom Europa. Pine, Karen J., Nicola Lufkin and David Messer. 2004. More gestures than answers: Children learning about balance. Developmental Psychology 40(6). 10591067. Pinker, Steven. 1989. Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Plumert, J., M., K. Ewert and Spear, S. J. 1995. The early development of childrens communication about nested spatial relations. Child Development 66(4). 959969. Pye, Clifton, Diane F. Loeb and Yin-Yin Pao. 1996. The acquisition of breaking and cutting. In Clark, Eve V. (Ed.), Proceedings of the 27th Annual Child Language Research Forum, 227236. Stanford: CLS. Regier, Terry and Mingyu Zheng. 2007. Attention to endpoints: A cross-linguistic constraint on spatial meaning. Cognitive Science 31(4). 705719. Seyfeddinipur, Mandana. 2006. Disuency: Interrupting speech and gesture. Nijmegen, Radboud University, PhD Dissertation. Slobin, Dan I., Melissa Bowerman, Penelope Brown, Sonia Eisenbeiss and Bhuvana Narasimhan. In press. Putting things in places: Developmental consequences of linguistic typology. In Eric Pederson and Jurgen Bohnemeyer (eds.), Event representations in language and cognition. Cambridge: Cambridge University Press. Stefanini, Silvia, Martina Recchia and Maria Cristina Caselli. 2008. The relation between spontaneous gesture production and spoken lexical ability in children with Down syndrome in a naming task. Gesture 8(2). 197218. Talmy, Leonard. 1985. Lexicalization patterns: Semantic structure in lexical forms. In Timothy Shopen (Ed.), Language typology and syntactic description, 57149. Cambridge: Cambridge University Press. Thomson, Jean, and Robin Chapman. 1977. Who is daddy revisited: The status of 2-yearolds over-extended words in use and comprehension. Journal of Child Language 4(3). 359375. Tomasello, Michael, Malinda Carpenter and Ulf Liszkowski. 2007. A new look at infant pointing. Child Development 78(3). 705722. van Staden, Miriam, Melissa Bowerman and Mariet Verhelst. 2006. Some properties of spatial description in Dutch. In Levinson, Stephen C. and David Wilkins (Eds.), Grammars of space. Explorations in cognitive diversity, 475511. Cambridge: Cambridge University Press. Volterra, Virginia, Maria Christina Caselli, Olga Capirci and Elena Pizzuto. 2005. Gesture and the emergence and development of language. In Michael Tomasello and Dan I. Slobin (eds.), Beyond nature-nurture: Essays in honor of Elizabeth Bates, 340. Mahwah, NJ: Erlbaum. Yule, George. 1997. Referential communication tasks. Hillsdale, N.J.: Erlbaum.

Appendix A:

Materials (target items in bold)

Warmup item 1 Warmup item 2 Warmup item 3 Agent_put_bear_lying Agent_put_ashlight_lying Agent_put_book_lying


M. Gullberg and B. Narasimhan

Agent_put_doll_standing Agent_put_paper_envelope Agent_squeeze_wet_cloth Agent_put_book_standing Agent_put_can_lying Agent_put_ashlight_standing Agent_put_monkey_lying Agent_put_can_standing Agent_spin_disc Agent_put_picframe_standing Agent_put_bear_standing Agent_drop_can_accidentally Agent_put_doll_lying Agent_drop_pencils_table Agent_put_mouse_vase Agent_drop_book_lying Agent_drop_can_lying Agent_put_napkin_oor Agent_drop_doll_lying Agent_put_cookiebatter_tray_spoon Agent_ick_coin Agent_put_piece_puzzle Agent_put_dog_standing Agent_put_rice_table Agent_put_picframe_lying Agent_put_pillowcase_pillow Agent_put_arm_frame Agent_put_monkey_standing Agent_put_ring_pole Agent_put_dog_lying Agent_put_tomato_bag Agent_drop_matchsticks_table Agent_drop_monkey_lying

Changes in encoding of PATH of motion in a first language during acquisition of a second language

Abstract Languages vary typologically in their lexicalization of PATH of motion (Talmy 1991). Furthermore, lexicalization patterns are argued to aect syntactic packaging at the level of the clause (e.g., Slobin 1996b) and tend to transfer from a rst (L1) to a second language (L2) in second language acquisition (e.g., Cadierno and Ruiz 2006). Crosslinguistic and developmental evidence suggests, then, that typological preferences for PATH expression are highly robust features of a rst language. The current study examines the robustness of preferences for PATH encoding by investigating (1) whether Japanese follows patterns identied for other verb-framed languages like Spanish, and (2) whether patterns established in an L1 can change after acquisition of an L2. L1 performance of native speakers of Japanese with intermediate-level knowledge of English was compared to that of monolingual speakers of Japanese and English. Results showed that monolingual Japanese speakers followed basic lexicalization patterns typical of other verb-framed languages, but with dierent realizations of PATH packaging within the clause. Moreover, native Japanese speakers with knowledge of English displayed mixed patterns for lexicalization and expressed signicantly more PATH information per clause than either group of monolinguals. Implications for typology and second language acquisition are discussed. Keywords: motion events, PATH, Japanese, English, second language acquisition, crosslinguistic inuence, attrition.

* Address for correspondence: A. Brown, Syracuse University, Department of Languages, Literatures and Linguistics, Oce 323C, 340 H.B. Crouse Hall, Syracuse, N.Y. 132441160, U.S.A. Email: M. Gullberg, Centre for Languages and Literature, PO Box 201, 221 00 Lund, Sweden. Email: Cognitive Linguistics 212 (2010), 263286 DOI 10.1515/COGL.2010.010 09365907/10/00210263 6 Walter de Gruyter

264 1.

A. Brown and M. Gullberg Introduction

In human understanding of motion, the notions of Source (point of origin), path (trajectory), and Goal (destination) are core (Johnson 1987). All languages encode such concepts, and the ways in which these elements are mapped onto lexical items pattern remarkably systematically across languages (Talmy 1991). Typological preferences particularly for lexicalization of path appear so robust that they aect syntactic packaging at the level of the clause (Slobin 1996b, 1997) and tend to transfer from a rst language (L1) to a second (L2) in second language acquisition (e.g., Cadierno 2004; Cadierno and Ruiz 2006; Navarro and Nicoladis 2005; Negueruela et al. 2004; Stam 2006). The current study examines the robustness of preferences for path encoding by investigating whether Japanese follows patterns identied for other verb-framed languages such as Spanish, and whether patterns established in an L1 can change after acquisition of an L2. Distinctive patterns in this crosslinguistic and developmental data would underscore the importance of taking individual language experiences into account in characterizations of languages on the basis of usage data, and would have further implications for our understanding of the relationship between languages in the multilingual mind. 2. Background

In inuential work, Talmy (1991) has suggested that languages can be divided into two typological groups depending on how path of motion is lexicalized: in the verb (verb-framed) or outside the verb (satelliteframed). To illustrate, examples are given below for Japanese (verbframed) and English (satellite-framed), with path expressions underlined. (1) Tama-ga saka-o kudaru Ball-Nom hill-Acc descend1 The ball descends the slope The ball rolls down the hill


In (1), a prototypical example from Japanese, path is lexicalized in the verb kudaru descend. In (2), a corresponding prototypical example from English, path is lexicalized in the so-called satellite (verb particle) down. Renements of the typology (e.g., Slobin 2004b) notwithstanding, support for the prevalence of basic typological distinctions in lexicaliza1. Abbreviations used in examples are Nom Nominative Case, Acc Accusative Case, Gen Genitive Case, Top Topic Marker, Con Connector.

Changes in encoding PATH of motion


tion of path has been found in many empirical studies on dierent languages (e.g., Gennari et al. 2002; Naigles et al. 1998; Slobin 1996b). Talmys typology (1985, 1991, 2000) reects characteristic preferences in a language, but there are often several options for path lexicalization in both satellite-framed and verb-framed languages. In addition to the preponderance of satellites, English, for example, possesses several path verbs such as descend, ascend, etc., although, as Talmy observed, most of these are borrowings from Latin, representing a more formal register, which is not characteristic of English. Japanese, however, has a number of rather more frequent options for path expression besides simple main verbs. Example (3) illustrates several of these. (3) Tama-ga toi-kara detekite bouringu-jyou-made ball-Nom pipe-from exit.come.Con bowling-alley-to haitte itte enter.Con go.Con Lit: The ball comes exiting the pipe, and goes entering the bowling alley

Example (3) displays three dierent kinds of possibilities for path expression in Japanese other than simple main verbs: postpositions, e.g., made until/to, kara from; complex motion predicates, e.g., haitte itte go entering, consisting of hairu enter and the deictic verb iku go; and compound verbs, e.g., detekite come out, a combination of deru exit and kuru come. Such possibilities are not necessarily unique to Japanese. Spanish, for example, employs directional adpositions, which can be stacked within the clause,2 as well as complex motion predicates.3 Compound verbs are also seen in other verb-framed languages such as Korean (Slobin 2004b).
2. Use of directional adpositional phrases in combination with verbs of manner of motion in verb-framed languages is argued to be restricted such that they cannot be used for telic events (Aske 1989) or events involving state changing boundary crossing (Slobin and Hoiting 1994). To some extent, Japanese may be similarly constrained, which may explain the ungrammaticality of *John-ga gakkoo-ni/e hashitta/aruita John walked/ran to school (Tsujimura 1994, cited in Inagaki 2002:119), although see Inagaki (2002: 191, footnote 11) for comments on variations in native speaker judgments of sentences such as these. However, John-ga gakkoo-made hashitta/aruita John walked/ran to school (Inagaki 2002:191) is commonly accepted, which may reect semantic dierences concealed in translation equivalents. 3. In Japanese, Matsumoto (1991; 1996) claims that such complex motion predicates are mono-clausal and contain a motion verb, either manner or path, with a connective -te sux followed by a main tensed verb. He restricts the verbs that can appear in tensed/ nal positions in such constructions to deictic motion verbs, e.g., iku, go; kuru, come; irassharu, go; kaeru, return.


A. Brown and M. Gullberg

Basic dierences in lexicalization patterns have been argued to have consequences at the level of the clause. In a corpus of literary translations, Slobin (1996b) illustrates possibilities in English and Spanish. He found that English texts tended to encode more information about path than Spanish texts through numerous mentions of Ground within individual clauses describing motion.4 (4) I went into the hall and through to the dining room. Entre en el hall y pase al comedor. I entered the hall and passed to the dining room (Du Maurier 1938: 243, cited in Slobin 1996b: 216)

In the English sentence above, there are two Ground elements associated with a single path verb (went into the hall / through to the dining room). In Spanish, on the other hand, comparable information is spread across two clauses, each associated with dierent path verbs (entre en el hall; pase al comedor). Slobin hypothesized that, as English generally locates path outside the verb root, many more path elements (that is, path particles and Ground elements expressing trajectory information such as into the hall ) can be concatenated within a clause, thereby yielding a more extended path description. For Spanish speakers to do the same, each path expression would require a separate verb clause. And indeed, the analysis of novels revealed that English-speaking writers on average mentioned 2.24 Ground elements in each description of a motion event, in contrast with the 1.52 elements mentioned by Spanish-speaking writers. Thus, although they employed fewer clauses, writers of English ultimately added more path detail to their motion event descriptions than their Spanish-speaking counterparts. These observations lie at the heart of the concept of thinking for speaking (Slobin 1996a), that is, the idea that speakers typically attend to the aspects of an event that their language has the readily available linguistic means to express, and that over time, this habitual attention leads to certain rhetorical styles. Thinking for speaking, then, would predict generally compact expression of complex trajectories in English. The existence of crosslinguistic dierences in lexicalization and encoding of path have also prompted the question of what happens when indi4. Observations about depiction of Ground and its relationship to path here should be distinguished from other observations in the literature regarding descriptions of Ground in the process of scene setting, i.e., descriptions of the context in which the motion took place prior to descriptions of the motion itself, which allow information about path to be inferred (cf. Slobin 1996b).

Changes in encoding PATH of motion


viduals acquire knowledge of a competing system, for example, in the case of second language learning. Studies of both intermediate and advanced L2 speakers have found traces of properties from the L1 in L2 production, generally known as transfer from the L1. In the domain of path expressions, examples of transfer include non-target-like use of path verbs, redundant use of path satellites, and acceptance of ungrammatical combinations of manner and path constructions (e.g., Cadierno 2004; Cadierno and Ruiz, 2006; Inagaki, 2001; Navarro and Nicoladis 2005; Negueruela et al. 2004; Stam 2006). Diculties with such seemingly simple lexical items as up, down, enter, exit in English, even at high levels of L2 prociency, is rather striking. Although this kind of data overwhelmingly suggests that typologically determined preferences for expression of path in the L1 are resistant to change, there is a small body of evidence indicating that patterns may shift in an L1 under the inuence of presence of an L2even during L2 acquisition and in L2 speakers who are not functional bilinguals. To date, studies have focused on manner of motion in speech and gesture (Brown and Gullberg, 2008) and gesture perspective in the expression of motion (Brown 2008), but little is known about whether an L2 inuence on the L1 can also be found in the expression of path of motion. In sum, given the variety of available morphosyntactic resources in Japanese outlined above, we may question whether Japanese really patterns like other verb-framed languages such as Spanish in terms of preference for expressing one path constituent per clause as opposed to concatenating several such expressions within the clause. Moreover, since expression of path is moderated by preference rather than governed by grammar in both English and Japanese, there is potential for eects of one language on another in the context of second language acquisition. While eects of the L1 on the L2 have been found in L2 production in this domain, no study has examined concurrent eects of an L2 on the L1 (although see Hohenstein, Eisenberg and Naigles 2006 and Tatsumi 1997 for a discussion of bidirectional crosslinguistic inuence in bilingualism in the domain of motion), especially at modest levels of prociency in the L2. 3. This Study

The aim of this study is twofold. The rst goal is to examine the extent to which Japanese conforms to the typical verb-framed pattern in language usage. If it does, monolingual speakers of Japanese should lexicalize path primarily in simple, main verbs, which diminish the possibility of stacking expressions within the clause. On the other hand, if speakers make use


A. Brown and M. Gullberg

of the full range of morphosyntactic resources available in Japanese, e.g., postpositions, compound verbs, and complex motion predicates, they may actually encode more information about path than speakers of other verb-framed languages, e.g., Spanish, through concatenation of expressions. The second aim is to test the robustness of typological preferences for expression of path by investigating whether acquisition of an L2 can inuence patterns established in an L1. Since Japanese and English dier typologically in this domain, we observe native speakers of Japanese with knowledge of English as an L2 and compare performance in their native L1 to that of monolingual speakers of each language. If inuence of an L2 on an L1 exists and is a normal part of L2 acquisition and not L1 loss, these non-monolingual Japanese speakers are predicted to display properties of English in fully grammatical production in Japanese, for example in lexicalization and concatenation of path.

4. 4.1.

Methodology Participants

A total of fty-seven adults aged between 18 and 48 participated in this study, distributed across four groups: monolingual Japanese speakers resident in Japan (16 speakers), monolingual English speakers resident in the USA (13 speakers), and native Japanese speakers with knowledge of English resident in Japan (15 speakers) or the USA (13 speakers). Biographical information and information on general language usage was gathered using a detailed questionnaire developed by the Multilingualism Project at the Max Planck Institute for Psycholinguistics (Gullberg and Indefrey 2003). The monolingual speakers of each language had had minimal exposure to an L2, were not engaged in active study of an L2, and did not use an L2 in their everyday lives; therefore, they were considered functionally monolingual. Further, all native Japanese speakers with knowledge of English were engaged in active use of their L2. Crucially, the L2 speakers in Japan had never lived in an English-speaking country, while those in the USA had been residents for between one and two years. This contrast in residence was designed to control for possible eects of L1 loss. Changes in path expression seen only in the L1 of those in the USA might be explained by attrition of the L1 due to residence in the L2 community. However, similar L1 patterns in both groups would render an explanation based on L1 attrition less likely. Even though this study is only concerned with L1 production, learners L2 knowledge was carefully measured to ensure uniform prociency in

Changes in encoding PATH of motion


English. Participants rst rated their own prociency in speaking, listening, writing, reading, grammar, and pronunciation. They then completed the rst grammar section of the Oxford Placement Test (Allan 1992). Third, their oral prociency was evaluated by consensus judgment of two certied examiners using the University of Cambridge Local Examinations Syndicate (UCLES) oral testing criteria for the First Certicate in English (FCE).5 Both the Oxford Placement and the FCE criteria placed the native Japanese speakers with knowledge of English resident in Japan and the USA within intermediate range. The groups did not signicantly dier in prociency as measured by the Oxford Placement Test (t (25) 0.795, p 0.434), but marginally diered in prociency as measured by the Cambridge FCE criteria (t (26) 1.982, p 0.058), with the learners resident in Japan scoring slightly higher than those resident in the USA. Participants biographical and language usage data as well as English prociency data are summarized in Table 1.
Table 1. Summary of biographical and language usage/prociency data Language background Mean AoEa : English Mean usageb : English Mean self-ratingc : English Mean Oxford Score Mean FCEd Score
a d

Non-monolingual Japanese (Japan) (n 15) 11.9 (range 913) 3 hrs (range .58.5) 2.97 (range 24.17) 78% (range 6088%) 4.27 / 5 (range 25)

Non-monolingual Japanese (USA) (n 13) 12.8 (range 1214) 6 hrs (range 112) 3.27 (range 1.84.3) 75% (range 5885%) 3.69 / 5 (range 2.35)

Age of exposure; b Hours of usage per day; c A composite score of individual skill scores; Cambridge First Certicate in English



Short narrative descriptions were elicited based on the six-minute, animated Sylvester and Tweety Bird cartoon, Canary Row (Freleng, 1950), used in several studies on expression of motion in speech and ges ture (e.g., Kita and Ozyurek, 2003; McNeill 1992; Stam 2006; inter al.). The cartoon contains numerous motion events, centering around Sylvesters repeated but failed attempts to catch Tweety. In order to get maximal information from participants and increase the likelihood of mention of motion events, the entire cartoon was broken down and shown in
5. More information can be found at http:/ /


A. Brown and M. Gullberg

manageable scenes following McNeill (1992). Two dierent sequences of scenes were systematically varied in the presentation of the stimulus. From the stimulus material, four motion events consistently described by participants were selected for coding and analysis, yielding four dierent paths: climb through, roll down, clamber up, swing across. 4.3. Procedure

All participants narrated in their L1. The native Japanese speakers with knowledge of English also produced narratives in their L2, but only the L1 data are reported here. Note, however, that the language order in which the second language speakers gave descriptions was counterbalanced across participants with a minimum of three days between appointments. This minimized the likelihood of both the L1 and L2 being fully active at the same time, therefore controlling for the eects of language mode (Grosjean 1998). Depending on the language of the experiment, participants were tested individually by either a native English- or native Japanese-speaking confederate. The participant and experimenter rst engaged in a brief warm-up, consisting of small talk in the target language, in order to put participants in monolingual mode. Next, the experimenter told participants that they would be watching a series of animated scenes from a cartoon on a computer screen and should retell what they had seen to the experimenter in as much detail as they could remember. The experimenter was trained to appear fully engaged in the participants narratives, but to avoid asking questions and crucially to avoid supplying the target path. 4.4. Speech segmentation and coding

Narrative descriptions were transcribed from digital video by a native speaker of the relevant language. Descriptions were divided into clauses, dened as any unit that contains a unied predicate . . . (expressing) a single situation (activity, event, state), following Berman and Slobin (1994: 660). Clauses sometimes contained more than one verb. Innitives or participles functioning as complements of modal or aspectual verbs, for example, were not segmented separately, e.g., [He wants to go], and neither were predicates that were narrator comments, e.g., [I think he went]. In Japanese, clausal segmentation presented some challenges due to the status of the connector morpheme, -te, which can connect a whole series of verbs. Linguists ascribe various semantics to -te, which might affect the placement of clausal boundaries (see, for example, Hasegawa 1996; Kuno 1973; Nakatani 2003). Following Kuno (1973) and Nakatani

Changes in encoding PATH of motion


(2003), in this analysis -te was considered primarily a simple connector of temporal sequence. Thus, all such inected verbs were segmented as individual clauses, with the exception of those occurring in mono-clausal complex motion predicates, dened by Matsumoto (1991; 1996) as consisting of a motion verb, -te sux, and a deictic verb. Examples of clausal segmentation of individual narratives by an English speaker and a Japanese speaker respectively are shown in (5) and (6). (5) 1[okay so Sylvester decides to crawl inside the drainpipe up to the windowsill ] 2[Tweety sees] 3[him coming] 4[and puts a bowling ball down the drainpipe] 5[and it ts] 6[and it meets Sylvester] 7[who ends up with a the ball inside of his stomach] 8[and he runs] 9[and rolls down the hill with it into a bowling alley] 10[when you hear a strike] 1[amamizu-no kou ochiru] rainwater-Gen like descend (the thing) the rainwater goes down like this 2[toi-ga arundesukedo] pipe-Nom exist.but there is a drainpipe 3[soko-kara naka-ni neko-ga haitte-itte] there-from inside-to cat-Nom enter.Con-go.Con from there, the cat went inside and 4[sono hiyoko-no tokoro-made ikouto-shitandesukedo] that bird-Gen place-to and tried to reach the place where that chick is 5[hiyoko-wa booringu-no booru-o soko-no toi-ni bird-Top bowling-Gen ball-Acc there-Gen pipe-to ue-kara otoshite] up-from drop.Con the chick drops the bowling ball on the drainpipe from the top and 6[ee sono neko-ga haitteiru] um that cat-Nom is.inside where that cat is inside 7[naka-ni otoshitande] inside-to drop.Con (the bird) dropped (it) inside of (the drainpipe)



A. Brown and M. Gullberg 8[kou nanka neko-ga sore-o sono booringu-no booru-wo like like cat-Nom that-Acc that bowling-Gen ball-Acc nonde-shimatte] drink.Con-nish.Con something like the cat swallowed that bowling ball and 9[de saka-o kou kudaru-youni ] and hill-Acc like descend-like and like goes down the slope like this 10[kou ochite-itte] like fall.Con-go.Con (the cat) is falling down like this 11[booringu jyou-ga choudo atta-node] bowling place-Nom precisely existed-so and there was a bowling alley just there and so 12[soko-ni haitte-shimaimashita] there-to enter.Con-nished (he) got in there

Next, clauses describing the four target motion events were identied and coded using Elan, a digital video tagging software program developed at the Max Planck Institute for Psycholinguistics (Wittenburg et al. 2006). In example (5), clauses 1 and 3 relate to the climb through event and 8 and 9 relate to the roll down event. In example (6), clauses 3 and 4 relate to the climb through event, and 9, 10 and 12 relate to the roll down event. Clause 10 illustrates a complex motion predicate, in which two verbs are joined within a single clause, ochiru fall and iku go. A coding scheme was employed whereby all lexical elements encoding information about the trajectory followed by the Figure object were coded as path, including directional adpositional phrases indicating source and goal of motion and deictic verbs indicating motion. This coding scheme largely followed schemes outlined in previous studies on motion events (e.g., Jensen 2002; Kita and Ozyurek 2003; Slobin 1996b, 1997, 2004b; 6. In addition, the following language-specic guideWeingold 1992, 1995) lines were employed. Morphologically complex words in Japanese composed of a manner component and path component e.g., tobi-komu, or two path components, e.g., toori-nukete go through-come

6. Although they are included here in order to be inclusive with respect to specication of a trajectory, many coding schemes do not include source, goal or deictic expressions as path. Although this may appear controversial, the reader is reminded that the crucial comparisons in this study are within-language for which exactly the same coding scheme was applied.

Changes in encoding PATH of motion


out were divided, and each path component was coded separately since each part of the lexical compound contributes independently to the meaning of the construction. Complex motion predicates consisting of a progressive motion participle with a deictic motion verb were treated similarly. The Japanese verbs, hairu enter and deru exit, were not coded as motion verbs at all unless they were combined with kuru come or iku go as auxiliaries or adpositional phrases such as ni to, following Kitas (1999) claim that these verbs in their bare forms express discrete changes of state without motion semantics (although see Tsujimura 2002, for an alternative analysis of Japanese enter and exit verbs). Furthermore, in Japanese, we excluded all spatial nouns, e.g., ue top/upness in ue-ni agaru rise to the top, as we considered these to encode location more than trajectory. We excluded comparable cases of locative expressions in English, e.g., climbed on the drainpipe, climbed the inside of the drainpipe unless these were used adverbially to express motion, e.g., went in/inside/into. The rst level of analysis investigated lexicalization of path. Here, the repertoire of lexical items used and the distribution of path semantics across morphosyntactic resources were identied. Two possible morphosyntactic patterns were distinguished in this analysis: verbal and adverbial. The second level of analysis addressed concatenation of path by examining the number of path expressions of any type per clause. Examples of analysis of lexicalization and concatenation of path in descriptions of the roll down event in Japanese and English appear in (7) and (8), with clause boundaries marked by brackets and path expressions underlined. (7) [Neko-wa sakamichi-o korogatte ikimashita] cat-Top hill-Acc roll.Con went Lit: The cat went rolling on the hill [The ball rolled out of the drainpipe, down the hill and into the bowling alley]


Example (7) from Japanese contains only one overt path expression, a verb, embedded in a complex motion predicate with a manner and path component: korogatte iku go rolling.7 In example (6) from English, however, there are three path expressions, all adverbials: out, down and into.
7. Native Japanese speakers may argue that this utterance contains directional information other than that conveyed by iku go. This may be due to the special status of korogaru roll, which in combination with a Ground phrase, e.g., saka-o hill-Acc, without a directional particle, may encode implicit directional semantics e.g., saka-o korogatte roll on/down the hill. However, since any additional directional information expressing descent in (5) is regarded as implicit, it has not been included in the coding, gloss or translation.

274 4.5.

A. Brown and M. Gullberg Reliability of speech coding

To establish reliability of data coding, 15% of the entire data set was segmented and coded by an independent second coder. 95% agreement was reached on selection of relevant clauses for coding, and of these, 100% agreement was reached on coding of lexicalization and concatenation. Disagreements were settled by accepting the judgment of the initial coder. 4.6. Analysis

Two dierent analyses were conducted to investigate the expression of path in L1 narrative production: rst, we identied lexicalization patterns in each group, and second, we assessed concatenation patterns. For all quantitative analyses, the native Japanese speakers with knowledge of English resident in Japan were compared to their counterparts resident in the USA. When no dierences were found between them, the data were collapsed to form a single group of non-monolingual speakers. Non-parametric statistical tests were employed throughout, specically Kruskal-Wallis for multiple group analyses and Mann-Whitney for between group analyses.

5. 5.1.

Results Lexicalization of PATH

In order to investigate lexicalization patterns, we rst identied the lexical repertoire for path expression employed by speakers in each group to describe the four target motion events. Table 2 shows the range of verbal and adverbial path types used by monolingual Japanese, non-monolingual Japanese and monolingual English speakers. In this qualitative analysis, native Japanese speakers with knowledge of English resident in Japan are displayed separately from those in the USA in order to balance participant numbers across groups and minimize the likelihood of dierences in the size of lexical repertoires arising from simple dierences in group size. As Table 2 shows, all groups employed both verbs and adverbials to lexicalize path. However, the diering number of lexical types appearing in each language is a clear indication that lexicalization patterns vary crosslinguistically. As expected, monolingual speakers of English employed a greater variety of adverbial expressions, whereas lexical diversity in both monolingual and non-monolingual Japanese discourse was chiey observed in verbs. In contrast to clear dierences between languages, within-language patterns appeared rather more uniform, regardless of

Changes in encoding PATH of motion

Table 2. Lexical repertoire for PATH expression Mono Japanese n 16 Non-mono Japanese: Japan n 15 agaru rise hairu enter idou-suru move iku go komu (only in compound form) into kuru come mezasu go toward mukau go toward noboru climb nukeru go through ochiru fall oriru decend shinnyuu-suru invade tadoritsuku arrive tooru go through tsutau go along tsutawaru be passed along ugoku/ugokasu move/be moved utsuru move Non-mono Japanese: USA n 13 agaru rise chikazuku approach hairu enter iku go komu (only in compound form) into kuru come noboru climb ochiru fall shinnyuu-suru invade tadoritsuku arrive tooru go through toutatsu-suru arrive tsutau go along tsutawaru be passed along utsuru move


Mono English n 13

path verb types

agaru rise hairu enter iku go komu (only in compound form) into kudaru descend kuru come noboru climb8 noru (only in compound form) onto ochiru fall shinnyuu-suru invade tai-suru go toward tooru go through tsutau go along tsutawaru go through utsuru move wataru cross

come get go

8. Japanese linguists (e.g., Matsumoto 1996) consider noboru climb.ascend as a path verb because it can only encode upwards trajectory (ue-ni noboru climb up / *shita-ni noboru climb down), in contrast to its closest translation equivalent in English, climb, which is considered a manner verb as it can be paired with both upwards and downward trajectories (climb up / climb down). In Japanese, noboru also occupies the position of a path verb (second position) in a manner-path verb compound. However, Sugiyama (2005) discusses the problematic nature of this verb, explaining that it can be represented by three dierent Chinese characters, only two of which have a clear path reading. The third character, she argues, has a much stronger suggestion of manner, indicating use of ones hands or feet. Moreover, there is no clear way of knowing which meaning the speaker intended. However, as she observes, the addition of yojiru clamber with noboru in the compound construction, yoji-noboru clamber.ascend more clearly expresses the semantics of manner. Thus, all cases of noboru have been coded here as path.


A. Brown and M. Gullberg

Table 2. (Continued) Mono Japanese n 16 Non-mono Japanese: Japan n 15 he to kara from made until/to ni to Non-mono Japanese: USA n 13 he to kara from made until/to massigura toward ni to Mono English n 13

path adverbial types

he to kara from made until/to ni to

across along back behind beyond down from in inside into on out of over through to up

language experience. There was complete overlap in adverbial types employed by monolingual and non-monolingual Japanese speakers and comparable numbers of verb types with a large degree of overlap. However, in order to fully explore lexicalization patterns given the possibilities for dierent verb constructions in Japanese, e.g., compound verbs and complex motion predicates, we calculated the number of path verbs versus adverbials per clause. Figure 1 shows the mean number of verbs expressing path per clause in all clauses containing path information in each language group. As there was no signicant dierence between the non-monolingual Japanese speakers resident in Japan versus the USA (z 1.322, p 0.186), the data for the two groups were collapsed. There was a signicant dierence between the groups in mean number of path verbs per clause ( w 2 (2, N 57) 29.826, p < 0.001). Specically, monolingual English speakers produced signicantly fewer path verbs per clause than both monolingual Japanese speakers (z 4.572, p < 0.001) and native Japanese speakers with knowledge of English (z 5.111, p < 0.001), who did not signicantly dier from each other (z 0.356, p 0.722). Figure 2 shows the mean number of adverbials expressing path per clause in all clauses containing path information. Again there was no sig-

Changes in encoding PATH of motion


Figure 1. Mean number of path verbs per clause: J (monolingual Japanese speakers), J (E) (L1 of native Japanese speakers with knowledge of English), and E (monolingual English speakers)

Figure 2. Mean number of path adverbials per clause: J (monolingual Japanese speakers), J (E) (L1 of native Japanese speakers with knowledge of English), and E (monolingual English speakers)

nicant dierence between the non-monolingual Japanese speakers resident in Japan versus the USA (z 0.278, p 0.781), so the groups were collapsed. The groups again diered in mean number of path adverbials per clause ( w 2 (2, N 57) 26.775, p < 0.001). This time, monolingual English speakers produced signicantly more path adverbials per clause


A. Brown and M. Gullberg

than native Japanese speakers with knowledge of English (z 4.306, p < 0.001), who in turn produced marginally signicantly more path adverbials per clause than monolingual Japanese speakers (z 1.895, p 0.058). In sum, these results illustrate between-language but also to some extent within-language dierences. In line with previous crosslinguistic research showing dierences between lexicalization patterns in satelliteversus verb-framed languages (e.g., Gennari et al. 2002; Naigles et al. 1998; Slobin 1996b), the native English speakers observed here lexicalized path in a wide range of adverbials, whereas the native Japanese speakers lexicalized path in a comparably wide range of verbs. Yet analyses also show that both English and Japanese speakers were not grammatically constrained by their typological classication and expressed path in alternative ways. Most striking, however, is the nding that native Japanese speakers with knowledge of English used marginally signicantly more adverbials than their monolingual Japanese counterparts, which suggests an inuence of knowledge of English. Crucially, given that performance among non-monolingual Japanese speakers did not dier according to their country of residence and lexicalization was fully grammatical, the higher adverbial usage did not appear to arise from loss of the L1. 5.2. Concatenation of PATH

From the analyses above, we see that Japanese and English speakers employ both verbs and adverbials for lexicalization of path. Given that adverbials can be concatenated, this may have repercussions for path expression at the level of the clause. Moreover, as noted previously, even stacking of path verbs is an available option in Japanese. Example clauses (9)(13) from descriptions of the climb through and clamber up events demonstrate this range of options in monolingual and nonmonolingual Japanese discourse as compared to monolingual English discourse. (9) [Neko-ga amadoi-no naka-o tsutatte] cat-Nom drainpipe-Gen inside-Acc go.along.Con Lit: The cat goes along the inside of a drainpipe [Tori-no tokoro-ni ikouto] bird-Gen place-to Lit: (The cat) tries to go to the birds place [Haisuikan-no naka-o toori-nukete] drainpipe-Gen inside-Acc go.through-go.through Lit: (The cat) goes along going through the inside of the drainpipe



Changes in encoding PATH of motion (12) [Chiyou-kara Tweety-no tokoro-made nobotte ground-from Tweety-Gen place-to climb.ascend.Con itta] went Lit: (He) went climbing from the ground to Tweetys place [And goes rolling down the street into a bowling alley]



Example (9) from a monolingual Japanese speaker illustrates the typical verb-framed pattern, a clause with one path expression in the main verb tsutau go along.9 Examples (8) and (9) from non-monolingual Japanese speakers with knowledge of English present clauses with two path expressions in each: in the rst, the verb iku go and the postposition ni to, and in the second, the compound verb combining tooru go through and nukeru go through. The example in (10), also from a nonmonolingual Japanese speaker, however, contains four path expressions in a completely grammatical clause: two postpositions, kara from and made to, and a complex motion predicate consisting of two verbs, noboru climb.ascend and iku go. The nal example in (11) from a monolingual English speaker contains three path expressions: one verb, go, one adverb, down, and one preposition, into. These examples demonstrate clearly that with the full range of morphosyntactic devices, Japanese speakers can concatenate path expressions grammatically within the clause as easily as English speakers can. The remaining question is whether they actually do. Figure 3 shows the mean number of path expressions of all types (verbs and adverbials) per clause in all clauses containing path information in each language group. Once again, there was no signicant dierence between the non-monolingual Japanese speakers resident in Japan versus the USA (z 0.723, p 0.470), so the data for the two groups were collapsed. There was a signicant dierence between the groups in mean number of path expressions per clause (w 2 (2, N 57) 16.193, p < 0.001). Native Japanese speakers with knowledge of English stacked signicantly more path expressions per clause than monolingual Japanese speakers (z 2.010, p 0.044), who packed signicantly more path expressions per clause than monolingual English speakers (z 2.079, p 0.038). In sum, results on concatenation of path expressions within the clause revealed surprising between- and within-language dierences. First, not only did speakers of Japanese in general stack more path expressions per clause than would be expected from a verb-framed language, but they
9. The spatial noun naka inside was not coded as path in Japanese for reasons outlined in the section on coding of speech.


A. Brown and M. Gullberg

Figure 3. Mean number of path expressions of all types per clause: J (monolingual Japanese speakers), J (E) (L1 of native Japanese speakers with knowledge of English), and E (monolingual English speakers)

also packed signicantly more path expressions per clause than monolingual speakers of a satellite-framed language, English. Second, in their L1, Japanese speakers with intermediate knowledge of English concatenated signicantly more path expressions per clause than their monolingual Japanese counterparts. Again, there was nothing ungrammatical about non-monolingual L1 production, as can be seen in examples (7)(10), and non-monolingual speakers in the USA patterned in the same way as those in Japan, implying that this pattern was not the result of L1 loss. 6. Discussion

This study investigated the robustness of typological preferences for path expression by examining (1) the extent to which expression of path in Japanese follows patterns demonstrated in other verb-framed languages with respect to lexicalization and concatenation, and (2) whether patterns in an L1 can change after acquisition of an L2. Regarding the rst research question, analyses of monolingual expression of path both conrm and challenge previously found typological differences in lexicalization patterns. In line with previous research, monolingual Japanese speakers encoded path primarily in a wide range of verbs, whereas monolingual English speakers encoded path primarily in a wide range of adverbials. However, the full range of morphosyntactic devices available in Japanese meant that monolingual speakers of this

Changes in encoding PATH of motion


verb-framed language were not restricted to the one path expression per clause seen in other verb-framed languages and instead concatenated signicantly more path expressions than monolingual speakers of a satelliteframed language, English. Important to note, however, are dierences in the semantics of path expressions used. In English, adverbials can encode all components of a trajectorythe source, the goal and the intervening movement. Therefore, the stacking of adverbials within a clause, e.g., down the street into a bowling alley, can actually encode separate trajectories within a journey. In Japanese, on the other hand, adverbials only encode the source and goal of a trajectory. Hence, the stacking of adverbials, e.g., chiyou-kara Tweety-no tokoro-made from the ground to Tweetys place, only encodes dierent components of a single trajectory, specically the starting and ending points. Thus, with a greater stacking of path expressions, native Japanese speakers were not necessarily encoding more complex trajectories within a clause than native English speakers, just greater specications of a single trajectory. In addition to these semantic dierences, methodological dierences between this and prior studies might help account for the disparity between Japanese and other verb- or satellite-framed languages. For example, there is variation between studies in the number and nature of motion events described, and similar patterns may not hold for all motion events. Indeed, Matsumoto (p.c.) suggests that some of the motion events analyzed in this study may not have involved a journey complex enough to elicit maximal concatenation of path expression in English speech. Moreover, there also may have been dierences in the biographical proles of participants. Data in previous studies came from native speakers of the languages, who may or may not have had varying degrees of prociency in another language, whereas the monolingual participants employed in the current study were carefully selected on the basis of their limited foreign language experience. With the results of this study indicating that use of ones L1 can be subtly altered with even intermediate prociency in an L2, it becomes crucial to control for second language knowledge in any investigation of the native speaker baseline. The above dierences in semantics and methodologies notwithstanding, the fact remains that the existence of postpositions as well as compound verbs and complex motion predicates allow Japanese speakers to concatenate path expressions to a surprisingly high degree. Furthermore, we must keep in mind that since speakers of satellite-framed languages typically reserve the verb slot for a manner verb and have few options for compound manner-path verbs, adverbials oer the best option for accumulation of path expressions. In short, it may not have been the


A. Brown and M. Gullberg

monolingual English speakers in this study who patterned dierently from native English speakers in previous studies, but the monolingual Japanese speakers who did not pattern in a way generally predicted for speakers of verb-framed languages. This supports ndings from at least one other verb-framed language, Basque, which also pays a lot of attention to source and goal of path in a range of morphosyntactic devices, and thus behaves rather like a satellite-framed language (Ibarretxe-Antunano 2004). We conclude that in contrast to previous claims, typological classication does not necessarily restrict concatenation of path information within the clause. Moreover, these ndings highlight the importance of distinguishing between what a language allows and what speakers of that language actually do. Regarding the second research question, L1 preferences for path expression do not appear to be impervious to change. Native Japanese speakers with intermediate knowledge of English employed a mixed strategy for path lexicalization in their L1, Japanese, with frequent use of both verbs, like their monolingual Japanese counterparts, but also adverbials, like monolingual English speakers. These same speakers then produced the most greatly specied trajectories of all, with signicantly more path expressions per clause than either monolingual group. These results suggest that established typological patterns in the L1 might be inuenced by patterns being acquired in the L2, even at intermediate levels of L2 prociency. More specically, non-monolingual speakers of Japanese with some knowledge of English appear to combine both Japanese and English lexicalization strategies for expression of path in their L1. In all likelihood, this strategy accounts for the highly specied and compact encoding of path, since a combination of verbs and adverbials can be easily stacked within the clause. Importantly, as there were no dierences between Japanese speakers residing in the L1 versus the L2 community and as increased L1 expression of path was completely grammatical, these results do not seem to indicate any kind of language loss. Instead, in this arena, characterized by linguistic preference as opposed to grammaticality, such patterns suggest a fully grammatical process of convergence between the L1 and L2, much as has been proposed for the linguistic systems of bilinguals (e.g., Bullock and Toribio 2004; Colantoni and Gurlekian 2004; Montrul 2004; Sanchez 2004; Tatsumi 1997). If the patterns observed here do reect the eects of acquisition of L2 English on use of L1 Japanese, the nature of the inuence is rather more complicated than a simple matter of translation. As noted above, directionals function dierently in English and Japanese. For example, the Japanese equivalent of the English adverbial up, which does not

Changes in encoding PATH of motion


specify an end point, would be ue-ni to the top/upness, which species a spatial noun as the goal of motion. Therefore, in using comparable morphosyntactic resources as an English speaker to lexicalize path, a non-monolingual speaker of Japanese communicates slightly dierent semantic information, e.g., the source and goal of motion as opposed to the intervening trajectory. These ndings have several theoretical and methodological implications. First, with respect to linguistic typology, monolingual baseline results reveal that the relationship between typology and discourse is not as simple as has been predicted, and that there is still a need for further empirical testing in a wider range of languages of predictions for language usage on the basis of typological distinctions. Furthermore, multilingual results suggest that studies of language usage should consider the impact of individual language experiences, particularly with respect to second languages as common as English, the eects of which might be seen across entire groups of speakers. Moreover, the synchronic changes observed here may oer predictions for more systemic diachronic shifts, for example of the kind seen after language contact between speech communities (see Slobin 2004b for a discussion of the impact of German on Italian in the domain of motion event language). Second, in the eld of second language acquisition, the relationship between a rst language and a second is generally considered to be unidirectional with features of the L1 inuencing the L2. However, we argue instead that the relationship may be bidirectional with features of the L2 concurrently inuencing the L1. While this has long been acknowledged in the functional bilingualism literature, where prociency levels in both languages are high (e.g., papers in Cook 2003; Dussias 2001; Hohenstein et al. 2006; Pavlenko and Jarvis 2002), eects of an L2 even at intermediate levels of prociency on a supposedly stable L1 question the validity of benchmarks used in research on and assessment of second language acquisition. L2 production is typically compared to and assessed against that of a native speaker, whose established language is seen as a xed target. The stability, unity and invariability of this standard is likely to be an over-simplication (cf. Davies 2003). If another language, however imperfectly mastered, also inuences the native language, this suggests that the native L1 is not an invariable entity, but rather a moving target. For this reason, we should be more wary of the term non-target-like in regard to L2 production. As a consequence, we may then have reason to question ndings on the limits on ultimate attainment in an L2 (cf. Birdsong 2005). In conclusion, this paper argues that expression of path in monolingual Japanese does not completely follow patterns established in other


A. Brown and M. Gullberg

verb-framed languages and that encoding of path in the L1 may change after even partial acquisition of an L2. We need more usage data in a range of languages in order to fully explore typological preferences and their eects on discourse. We also need data from other L1L2 pairings in order to distinguish more clearly between patterns arising from convergence of knowledge of particular languages and those arising from general eects of bilingualism. Much work thus remains to be done. Nevertheless, at this point we may conclude that although the linguistic expression of path exhibits considerable crosslinguistic dierences among monolingual speakers, it does not seem to be as robust as expected and impervious to change. Received 1 March 2009 Revision received 25 November 2009 Syracuse University/ Max Planck Institute for Psycholinguistics, Nijmegen

Allan, David. 1992. Oxford Placement Test. Oxford: Oxford University Press. Aske, Jon. 1989. path predicated in English and Spanish: A closer look. Proceedings of the Fifteenth Annual Meeting of the Berkeley Linguistics Society. 114. Berman, Ruth and Dan I. Slobin. 1994. Relating Events in Narrative: A Cross-linguistic Developmental Study. Mahwah, NJ: Lawrence Erlbaum. Birdsong, David. 2005. Nativelikeness and non-nativelikeness in L2A research. International Review of Applied Linguistics 43(4). 319328. Brown, Amanda. 2008. Gesture viewpoint in Japanese and English: Cross-linguistic interactions between two languages in one speaker. Gesture 8(2). 256276. Brown, Amanda and Marianne Gullberg. 2008 Bidirectional crosslinguistic inuence in L1 L2 encoding of manner in speech and gesture: A study of Japanese speakers of English. Studies in Second Language Acquisition 30(2). 225251. Bullock, Barbara E., and Almeida Jacqueline Toribio. 2004. Introduction: Convergence as an emergent property in bilingual speech. Bilingualism: Language and Cognition 7(2). 91 93. Cadierno, Teresa. 2004. Expressing motion events in a second language: A cognitive typological perspective. In Michel Achard and Susanne Niemeier (eds.). Cognitive Linguistics, Second Language Acquisition, and Foreign Language Teaching, 1349. Berlin: Mouton de Gruyter. Cadierno, Teresa and Lucas Ruiz. 2006. Motion events in Spanish L2 acquisition. Annual Review of Cognitive Linguistics 4. 183216. Colantoni, Laura and Jorge Gurlekian. 2004. Convergence and intonation: Historical evidence from Buenos Aires Spanish. Bilingualism: Language and Cognition 7(2). 107 119. Cook, Vivian (ed.). 2003. Eects of the Second Language on the First. Clevedon, UK: Multilingual Matters. Davies, Alan. 2003. The Native Speaker: Myth and Reality. Clevedon: Multilingual Matters. Freleng, Friz. 1950. Canary Row [Film, animated cartoon]. New York: Time Warner.

Changes in encoding PATH of motion


Dussias, Paola, E. 2001. Sentence parsing in uent Spanish-English bilinguals. In Janet Nicol (ed.), One Mind, Two languages: Bilingual Language Processing, 159176. Cambridge: Blackwell. Gennari, Silvia P., Steven A. Sloman, Barbara C. Malt, and W. Tecumseh Fitch. 2002. Motion events in language and cognition. Cognition 83(1). 4979. Grosjean, Francois. 1998. Studying bilinguals: Methodological and conceptual issues. Bilin gualism: Language and Cognition 1(1). 131149. Gullberg, Marianne and Peter Indefrey. 2003. Language Background Questionnaire. Nijmegen, Max Planck Institute for Psycholinguistics. http:/ / Multilingualism/questionnaire. Hasegawa, Yoko. 1996. The (nonvacuous) semantics of TE-linkage in Japanese. Journal of Pragmatics 25(6). 763790. Hohenstein, Jill, Ann Eisenberg and Letitia Naigles. 2006. Is he oating across or crossing aoat? Cross-inuence of L1 and L2 in SpanishEnglish bilingual adults. Bilingualism: Language and Cognition 9(3). 249261. Ibarretxe-Antunano, Iraide. 2004. Motion events in Basque narratives. In Sven Stromqvist and Ludo Verhoeven (eds.), Relating Events in Narrative Volume 2: Typological and Contextual Perspectives, 89111. Mahwah, NJ: Lawrence Erlbaum. Inagaki, Shunji. 2001. Motion verbs with goal PPs in the L2 acquisition of English and Japanese. Studies in Second Language Acquisition 23(2). 153170. Inagaki, Shunji. 2002. Motion verbs with locational/directional PPs in English and Japanese. Canadian Journal of Linguistics 47(3/4). 187234. Jensen, Anne. 2002. Expressing motion events in a second language. Paper presented at EUROSLA, Basel, 1821 September. Johnson, Mark. 1987. The Body in the Mind: The Bodily Basis of Meaning, Imagination and Reasoning. Chicago: University of Chicago Press. Kita, Sotaro. 1999. Japanese enter/exit verbs without motion semantics. Studies in Language 23(2). 307330. Kita, Sotaro and Asli Ozyurek. 2003. What does cross-linguistic variation in semantic coordination of speech and gesture reveal?: Evidence for an interface representation of spatial thinking and speaking. Journal of Memory and Language 48(1). 1632. Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge, Mass: MIT Press. Matsumoto, Yo. 1991. On the lexical nature of the purposive and participial complex motion predicates in Japanese. Proceedings of the 17th Annual Meeting of the Berkeley Linguistics Society 180191. Matsumoto, Yo. 1996. Subjective motion and English and Japanese verbs. Cognitive Linguistics 7(2). 183226. McNeill, David. 1992. Hand and Mind. What the Hands Reveal about Thought. Chicago: Chicago University Press. Montrul, Silvina. 2004. Subject and object expression in Spanish heritage speakers: A case of morphosyntactic convergence. Bilingualism: Language and Cognition 7(2). 125142. Naigles, Letitia R., Ann R. Eisenberg, Edward T. Kako, Melissa Highter and Nancy McGraw. 1998. Speaking of motion: Verb use in English and Spanish. Language and Cognitive Processes 13(5). 521549. Nakatani, Kimiko. 2003. Analyzing -te. In William McClure (ed.), Japanese/Korean Linguistics, 377387. Stanford: CSLI. Navarro, Samuel and Elena Nicoladis. 2005. Describing motion events in adult L2 Spanish narratives. In David Eddington (ed.), Selected Proceedings of the 6th Conference on the Acquisition of Spanish and Portuguese as First and Second Languages, 102107. Somerville, MA: Cascadilla Proceedings Project.


A. Brown and M. Gullberg

Negueruela, Eduardo, James P. Lantolf, Stephanie R. Jordan and Jaime Gelabert. 2004. The private function of gesture in second language speaking activity: A study of motion verbs and gesturing in English and Spanish. International Journal of Applied Linguistics 14(1). 113147. Pavlenko, Aneta and Scott Jarvis. 2002. Bidirectional Transfer. Applied Linguistics 23(2). 190214. Sanchez, Liliana. 2004. Functional convergence in the tense, evidentiality, and aspectual systems of Quechua. Bilingualism: Language and Cognition 7(2). 147162. Slobin, Dan I. 1996a. From thought and language" to "thinking for speaking.". In John J. Gumperz and Stephen C. Levinson (eds.), Rethinking Linguistic Relativity, 7096. Cambridge: Cambridge University Press. Slobin, Dan I. 1996b. Two ways to travel: Verbs of motion in English and Spanish. In Masayoshi Shibatani and Sandra A. Thompson (eds.), Grammatical Constructions: Their Form and Meaning, 195219. Oxford: Oxford University Press. Slobin, Dan I. 1997. Mind, code and text. In Bybee, Joan, John Haiman and Sandra A. Thompson (eds.), Essays on Language Function and Language Type: Dedicated to T. Givon, 437476. Philadelphia: John Benjamins. Slobin, Dan I. 2004b. The many ways to search for a frog: Linguistic typology and the expression of motion events. In Sven Stromqvist and Ludo Verhoeven (eds.), Relating Events in Narrative: Typological and Contextual Perspectives, 219257. Mahwah, NJ: Lawrence Erlbaum. Slobin, Dan I. and Nina Hoiting. 1994. Reference to movement in spoken and signed languages: Typological considerations. Proceedings of the Twentieth Annual Meeting of the Berkeley Linguistics Society 487505. Stam, Gale. 2006. Thinking for Speaking about motion: L1 and L2 speech and gesture. International Review of Applied Linguistics 44(2). 143169. Sugiyama, Yuki. 2005. Not all verb-framed languages are created equal: The case of Japanese. In Proceedings of the Thirty-rst Annual Meeting of Berkeley Linguistics Society. Talmy, Leonard. 1985. Lexicalization patterns: Semantic structure in lexical forms. In Timothy Shopen (ed.), Language Typology and Syntactic Description Vol. 3, 57149. Cambridge: Cambridge University Press. Talmy, Leonard. 1991. path to realization: A typology of event conation. Proceedings of the Seventeenth Annual Meeting of the Berkeley Linguistics Society, 480519. Talmy, Leonard. 2000. Toward a Cognitive Semantics Vols. 1 and 2. Cambridge, MA: MIT Press. Tatsumi, Tomoaki. 1997. The bilinguals thinking for speaking: Adaptation of Slobins Frog story experiment to Japanese-English bilinguals. Japan: Sophia University, BA Thesis. Tsujimura, Natsuko. 2002. Japanese enter/exit verbs revisited: A reply to Kita (1999). Studies in Language 26(1). 165180. Weingold, Gotz. 1992. Up and down: On some concepts of path in Korean motion verbs. Ohak Yonku Language Research 28(3). 1543. Weingold, Gotz. 1995. Lexical and conceptual structures in expressions for movement and space: With reference to Japanese, Korean, Thai and Indonesian as compared to English and German. In Urs Egli, Peter E. Pause, Christoph Schwarze, Arnim von Stechow and Gotz Weingold (eds.), Lexical Knowledge in the Organization of Language, 301340. Amsterdam/Philadelphia: John Benjamins. Wittenburg, Peter, Hennie Brugman, Albert Russel, Alex Klassmann and Han Sloetjes. 2006. ELAN: A professional framework for multimodality research. Proceedings of the Fifth International Conference on Language Resources and Evaluation LREC. Genoa: Italy.

Im fed up with MarmiteIm moving on to VegemiteWhat happens to the development of spatial language after the very first years?

Abstract The present article examines childrens spatial language during late phases of development. To this end, the spontaneous speech of American English speaking children between 6 and 10 years of age was analyzed within an analytical framework developed in a previous study for spontaneous speech from speakers between 10 and 19 years of age (Graf 2006). The analysis dened ve basic spatial categories and four levels of abstraction in spatial meaning in order to capture all spatial relations and their literal as well as non-literal uses. The results show that the spatial language of speakers between 6 and 10 years of age diers mainly with respect to speakers preference for literal and metaphorical uses. Compared to the ndings in Graf (2006), the age groups analyzed here may be viewed as the developmental phase during which children and adolescents will reach the end of their journey towards adult use of spatial reference. Keywords: Space, language acquisition, developmental phase, literal meaning, metaphorical meaning.


Introduction: In search of a larger picture of spatial language development

The ubiquity of space in thought and language has inspired a vast amount of research. The ontogenetic path children take to become competent spatial communicators therefore is a prominent topic in cognitive

* Address for correspondence: E-M. Graf, Alpen-Adria Universitat, Institut fur Anglistik und Amerikanistik, North Building, level 0, Zimmer I.0.35, Universitatsstrae 6567, 9020 Klagenfurt, Austria. Email: Cognitive Linguistics 212 (2010), 287314 DOI 10.1515/COGL.2010.011 09365907/10/00210287 6 Walter de Gruyter


E.-M. Graf

science, developmental psychology and language acquisition research. Any informed discussion of the relation between language and space should include an accounting of how young children come to represent the meaning of spatial terms (Quinn 2005: 294). To apply a developmental perspective is particularly rewarding with respect to the close and multi-faceted interrelations of space, language, and cognition: The universal need for spatial knowledge, representation and communication results in an early mastering of basic spatial skills and leads to a high frequency of spatial reference in infants linguistic experience and input. Whereas the formerdue to, among other things, the relatively similar needs of young childrenleads to the construction of universal spatial categories such as location and motion, the latter results in languagespecic encodings of spatial relations and entities on the language level (for an account on the dierent grammars of space see Levinson and Wilkins 2006). (Spatial) language development follows what Gentner and Boroditsky (2001: 248) call a division of dominance continuum with respect to the inuence of cognition and language on spatial representation and communication (see also Bowerman 1996; Bowerman and Levinson 2001; Choi 2006; Gentner and Goldin-Meadow 2003; Gumperz and Levinson 1996). The way children learn to refer to space in language is both inuenced by general cognitive factors that account for universally similar acquisition paths and language-specic factors that account for the particularities found in these developmental paths (e.g., whereas English speakers linguistically categorize spatial events on the basis of support (X is on the table) or containment (X is in the bowl ), Korean speakers focus on tight vs. loose t between objects) (cf. Hickmann 2007: 227; see also Choi and Bowerman 1991; Pruden et al. 2008). These assumptions of the spatial ubiquity in thought and language and of the division of dominance continuum represent the theoretical context into which the present analysis is embedded. Its goal is to analyze the literal and metaphorical1 spatial language development of English speaking


Metaphorical spatial language results from cognitive and socio-communicative processes. According to Conceptual Metaphor Theory, cognitive processes account for our human propensity to conceptualize the abstract, e.g., time, via the concrete, e.g., space, to make it intellectually accessible. Conceptual metaphors result from ontological mapping processes between concrete, bodily and experientially based source domains onto abstract target domains. These mapping processes include the extension of congurations from the source to the target, become entrenched as conventional cognitive patterns in the mind of the speakers and are, in turn, expressed in language via metaphorical items. At the same time, socio-communicative processes such as frequency of use of metaphorical expression account for the entrenchment and (lexical) conventionality of

Im fed up with MarmiteIm moving on to Vegemite


children between 6 and 10 years of age. These age groups represent a research gap as the majority of both inter- and intralinguistic studies on spatial language acquisition concentrate on childrens development during their very rst years. The lack of interest in later stages of spatial language development is paralleled by a general disinterest in language development during school age and puberty (but see e.g., Berman 2004; Berman and Slobin 1994; Hickmann 2003; Karmilo and Karmilo-Smith 2001; Nippold 1998; Romaine 1984). In a previous study, the author therefore focused on English speaking adolescents between 10 and 19 years in order to determine if and how their spatial language developed (Graf 2006). The study showed no clear indication of any developmental changes in spatial language as a whole for the analyzed period: Irrespective of their age, speakers showed great similarities with respect to the spatial categories and their uses at the dierent levels of abstraction. The acquisition of spatial language may therefore be considered nalized after age 10. With respect to the dramatic ongoing changes reported in studies of earliest phases of spatial language acquisition, i.e., the years up to the age of six, and the reported lack of developmental changes after age 10, the years 6 to 10 seem to represent a crucial transition phase from apprentice to master of spatial language. The analysis is based on the categorical framework developed for the previous study. It deviates from existing spatial language research in two respects. Firstly, in accordance with Levinson and Wilkins (2006: 551) nding that . . . we know that spatial language is not fully mastered until late childhood, it adopts a life-span perspective as suggested by Eckert (1998) (for a similar argument see Karmilo and Karmilo-Smith 2001: 1). Spatial ontogenesis is viewed as a three-phase process2, whereby each phasechildrens preverbal spatial development, their early phases of spatial language acquisition (primary spatial language acquisition) and their late phases of spatial language acquisition (secondary spatial language acquisition)builds on and further elaborates what has already been mastered (see Chapter 2 and footnote 3).

spatial metaphors (see Graf in press; Keysar et al. 2000; Svanlund 2007). With time and usage, such lexical metaphors often lose more and more of their transparency andat least in some casesseem to be dying or already dead (Rice, Sandra and Vanrespaille 1999: 124) as is the case with the grammaticalized items going to future or existential there (termed transmetaphorical in the current approach (Graf 2006) (see Table 2)). 2. Reference to dierent developmental phases is found only implicitly in the literature.


E.-M. Graf

Secondly, the study applies an integrative perspective. Most studies concentrate on the acquisition of specic literal tokens of spatial language in proper spatial contexts. Little research, in contrast, has examined childrens use of space in non-spatial, metaphorical contexts. Studies interested in the acquisition of metaphorical language primarily analyze metaphors as instances of gurative language (Gardner 1974; Winner 1976, 1995; Pearson 1990), seldom as a basic human means of intellectually accessing abstract phenomena (but see for example Johnson 1997, 1999, 2001; Ozcaliskan 2003, 2004, 2005). In view of the immense attention space has received as a primary ontological source domain for metaphorical extensions onto abstract domains such as TIME (e.g., Boroditsky 2000; Gentner 2001; Tenbrink 2007), it might be rewarding to bring this aspect more to the forefront in spatial language acquisition research. In this study, non-literal uses are analyzed alongside literal uses of spatial categories. The categorical framework applied in this study follows from its integrative perspective. It concentrates on spatial categories which are understood as linking elements between spatial cognition and spatial language, rather than on the linguistic correlates such as spatial verb spatial preposition that represent these categories, e.g., go down. Such a procedure allows for the integration of the various phases of spatial development given that spatial categories are dependent on pre-conceptual image-schemata as the pre-linguistic representation format of spatial knowledge and constitute more abstract representations on the language level (cf. Quinn 1998: 160; Tomasello 1998: XX). In addition, the focus on spatial categories allows for the consideration of dierent levels of abstraction in spatial meaning at which spatial congurations may be used. For example, motion events may be represented at the literal spatial level (He goes to the station) or at the metaphorical spatial level (He always goes to extremes). This framework aims to capture all instances of spatial reference in the data and focuses neither exclusively on particular spatial relations such as motion events and their linguistic correlates nor on either literal or metaphorical spatial expressions. Instead, the aim is to analyze as wide a variety as possible of spontaneous spatial references in natural, everyday conversations of American English speakers between 6 and 10 years of age with the help of a qualitative analysis. The study will hopefully make a rst contribution towards lling the existing research gap in this area of study. In addition, it will hopefully link the various developmental phases in search of the larger picture of spatial language acquisition.

Im fed up with MarmiteIm moving on to Vegemite 2. Phases of spatial language development3


The focus of this article is on the phases of spatial language development, in particular on the later phase between 6 and 10 years of age. The very rst years of spatial language acquisition, i.e., primary spatial language acquisition, have been extensively studied from a variety of theoretical and methodological backgrounds. These numerous, mainly microanalytic in-depth studies have created islands of diering research interests and knowledge and are characterized by a pervasive methodological heterogeneity (Hickmann 2003: 175). Well aware of such methodological and theoretical heterogeneity, Graf (2006) carried out a meta-analysis of 75 studies in spatial language development from the last 35 years in order to come up with general developmental trends of young childrens spatial language acquisition (see also Hickmann 2003, and Nowak 2007, who gives a detailed outline of the meta-study methodology). Based on this meta-analysis the following features characterize and dene primary spatial language acquisition (in English) (cf. Graf 2006: 147 ):
Children come a long way from their rst spatial holophrases (cf. Dominey 2006: 138)4 such as up to both the comprehension and production of practically all spatial linguistic tokens, however not yet in their full semantic and pragmatic scope and contrast. They show a preference for those spatial congurations already prominent during their pre-verbal spatial development such as dynamic events expressing goal or source, and they encode these with general motion verbs such as go or come. They also produce rst metaphorical extensions of spatial

3. Spatial language development follows the general phases of language acquisition: The pre-linguistic phase thereby lays important cognitive and social foundations for the acquisition of language and forms the gestural basis for symbolic communication (Blake 2000; Lock 1980, 1999; Ozcaliskan and Goldin-Meadow 2005; Tomasello 2003). Early phases of language acquisition mean the onset of verbal communication around a childs rst birthday as a consequence of their universal human predilection for pattern-nding and intention-reading (Tomasello 2003). These rst years of language acquisition witness dramatic changes from context-dependent holophrastic utterances to an impressive linguistic repertoire during kindergarten and pre-school; developmental changes happen on a daily basis and require researchers to work in narrow time frames to trace them. According to the few available studies on late phases of language acquisition, subsequent developments are in turn characterized by an enrichment and sophistication of already acquired systems (cf. Grimshaw and Holden 1976: 41). Such deepening and renement happens at a much slower pace and requires the comparison of widely separate age groups to be able to trace developmental changes (cf. Nippold 21998: 3). 4. In early one-unit utterances (Tomasello 2003: 39) or holophrases, dened as unparsed holistic utterances that correspond directly to a meaning (Dominey 2006: 138), a single lexical item forms an early construction through its xed embedding in a specic context of situation.


E.-M. Graf

structures in language and start combining spatial expressions to refer to complex spatial congurations such as up here. Therefore, spatial language acquisition is clearly on its way in the years leading up to childrens schooling. However, intracategorical and inter-categorical dierences and correlations must still be acquired, metaphorical and other non-literal uses must be fully mastered and the possibilities of combining spatial expressions must be further explored.

In accordance with the general features of later phases of language development, secondary spatial language acquisition should be characterized by a further elaboration and renement of already acquired linguistic spatial competence. Spatial categories and their linguistic correlates should become used in a fully semantically and pragmatically contrastive manner, on all levels of abstraction as well as in combinations where necessary. In Graf (2006), the author assumed the phase of secondary spatial language to cover the adolescent years from 10 to 19 and compared their spatial language development in the context of three dierent age groups (10 to 13 years of age, 14 to 16 years of age, and 17 to 19 years of age)5. Yet, as the ndings showed, there is neither an increase in the complexity of the spatial language nor any variation with respect to the spatial congurations expressed. Furthermore, no development with respect to literal and non-literal uses of spatial categories could be reported. Instead, the speech of the three age groups under scrutiny was strikingly similar: A highly identical distribution of spatial categories, a similar complexity with respect to the use of simple and compound spatial categories, and, nally, a nearly identical picture with respect to literal and non-literal uses of spatial language could be reported (for terminology see below). The following assumption therefore motivates this study: Given that spatial language acquisition is not nalized during the very rst years and that there is hardly any developmental change found in the language of speakers older than 10, the years between 6 and 10 seem to be a decisive and separate period in spatial language development. 3. 3.1. The method The data

The data stems from the American English section of the CHILDES corpus, the Carterette transcripts. Unlike the material in the COLT corpus


Assuming secondary (spatial) language acquisition to cover these years is based on Nippolds (1998) indirect classication.

Im fed up with MarmiteIm moving on to Vegemite


used in Graf (2006), i.e., the 10 to 19-year-olds, where informants wore a microphone and a Walkman and were asked to record all conversations they had for some time, the Carterette material was evoked in a simple social situation. A researcher was sitting around a table together with three children of the same age group at a time and told them that she wanted to nd out what children of their age were interested in. She encouraged them to talk about anything they wanted to with each other, but did not herself participate in the interaction. Whereas the warm-up period was discarded from the transcripts, the speech produced during the rest of the conversation showed all features of natural speech, i.e., a broad range of topics, interruptions, use of slang words and generally a natural turn-taking among the speakers (see the data manual for Carterette and Jones, page 17). Although the two sets of data were gathered in slightly dierent contexts, i.e., naturally occurring conversations in the COLT corpus and evoked conversations in the CHILDES corpus, given that both sets of data share the characteristics of naturally evolving speech among interaction partners, it clearly sets them apart from experimentally elicited responses, produced under strictly controlled co- and contextual conditions. We therefore feel that comparison of the data is possible. The transcribed material contains speech from 54 rst graders, 48 third graders and 48 fth graders from junior college classes of a city college in California as well as the speech of 24 adults. For the present purpose only the childrens parts were used which divided into three age groups: 6 yearolds (Group I), 8 year-olds (Group II), and 10 year-olds (Group III). For each age group, well over 10,000 words were recorded and transcribed6. The aim is to analyze as wide a variety as possible of spontaneous spatial reference in natural, everyday conversations in English7. This approach is in keeping with Ames and Learneds (1948: 63) important insight that observation of spontaneous verbalizations seems in many ways to be

6. One problem related to such a method of eliciting and observing spontaneous language is Labovs (1972) observers paradox. The fact that children are aware of being recorded can be seen in the following utterance taken from the speech of eight year olds: How much longer do we have to talk?. 7. All items identied by the researcher as spatial words were checked against other studies on spatial language for corroboration of their status as prepositions, dimensional adjectives, etc. The Oxford English Dictionary was consulted to assess the spatial semantics of a word whenever the classication was unsure to the researcher. Items that did not occur in other studies, or that could not be identied as spatial in the Oxford English Dictionary did not get considered in the study.


E.-M. Graf

one of the most useful methods for studying the childs concepts of space. 3.2. The analytic framework

The categorical framework, developed in Graf (2006) for the analysis of spontaneous speech of speakers between 10 and 19 years of age, allows for the analysis of spatial language in the broadest sense possible, as it captures all spatial relations and their literal as well as their non-literal uses. At the same time, such an integrative endeavor can only be executed at the cost of detail. Such detail is found in the many microscopic indepth studies of earliest spatial language acquisition, which provide valuable insights into the exact acquisition of certain spatial expressions or the exact onset of language-specic inuences on spatial cognition etc. Bringing these two dierent research approaches together is one of the future research desiderata formulated in Section 5. 3.2.1. The spatial reference act. In accordance with the usage-based approach to language acquisition (Tomasello 2003), the framework centers on the communicative unit of Spatial Reference Act (SRA) and its components. The unit of analysis is dened as follows:
A speaker or Reference Origo (RO) refers to something, the Reference Entity (RE), usually in relation to something else, the Reference Relatum (RR) in space, the Reference Field (RF)8, from a particular point of view, the Reference Perspective (RP). In addition, Spatial Reference Acts are addressed to someone, the hearer or decoder of the spatial communicative act (Graf 2006: 79).

3.2.2. The spatial categories. The spatial categories (and their subtypes) used to analyze the data dierentiate a number of verbally encoded spatial congurations that result from dierent relations between the Reference Entity (RE) and the Reference Relatum (RR) as basic constituents of a Spatial Reference Act. Five simple spatial categories between Reference Entity and Reference Relatum are claimed: Reference Entity Distance, which encodes a deictic relation of distance or proximity (This is


The following denition of the Reference Field is given in Graf (2006: 84): Reference Field (RF) is understood as the implied surrounding space within which the speaker anchors the specic spatial conguration between the Reference Entity (RE) and the Reference Relatum (RR). The concept of Reference Field is closely related to the more frequently used concept of frame of reference (see e.g., Levinson 1996, 2003). The pragmatic focus on spatial language adopted here is however better captured in the concept of the Reference Field.

Im fed up with MarmiteIm moving on to Vegemite


my new bike), Reference Entity Motion, which encodes dynamic relations between RE and RR (He ran down the hallway), Reference Entity Location, which encodes a static relation between RE and RR (I forgot my keys on the table), Reference Entity Dystation9, which encodes intermediate relations between location and motion RE and RR (Peter lives across the street), and Reference Entity Dimension10, which encodes the dimensional Gestalt of the RE in (implicit) relation with RR (I want a big coke) (see Table 1 below). Based on a meta-analysis of existing studies on the acquisition of dimensional spatial expressions such as big or tall during the very rst years, Graf (2006: 172) concluded that certain categories and their linguistic correlates such as proximity/distanceexpressed via the spatial adverbs here and there or the demonstratives this or thatare cognitively simpler, whereas others such as dimension are generally more complex (due to cognitive prerequisites, semantic complexity and contextdependency) and acquired at a later stage. The distribution of the basic categories within and across the three age groups may therefore oer valuable insight into their developmental states.

Table 1. Categories of space in the English language Spatial categories Reference Entity Distance (distance/proximity) Reference Entity Motion (dynamic category D) Reference Entity Location (static category S) Reference Entity Dystation (dystatic category DS) Reference Entity Dimension (dimensional category DIM) Token examples This is my new bike. He ran down the hallway. I forgot my keys on the table. Peter lives across the street. I want a big coke.

9. Such intermediate types are not a matter of language alone, but are also found in perception and conceptualization. See Talmy (1996: 211 ), (2000: 99 ) and 2003 for a detailed outline of the dierent types of ctive vs. factive spatial relations. 10. Whereas the spatial Gestalt is an additional factor in the sub-categorization of dynamic and static relations in language (e.g., The pictures are on the table vs. The pictures are in the drawer) (for detailed discussions of such functional features see Carlson and van der Zee (eds.) (2005) or Aurnague, Hickmann and Vieu (eds.) (2007)), the spatial category REDIM is centered on such spatial Gestalts.


E.-M. Graf

In addition to these ve simple spatial categories that represent one particular spatial conguration on the language level, we nd the representation of complex spatial congurations, i.e., the combination of spatial relations within what is termed here compound category. In the example in this room (age 8) we nd the combination of the simple spatial category Reference Entity Location realized on the language level via the (spatial) verb be spatial preposition in and the simple spatial category Reference Entity Distance realized on the language level via the demonstrative pronoun this. The relation between reference entity and reference relatum comprises of two types of spatial information, one of containment and one of proximity. Compared to in the room which is an example of the simple category containment, in this room is complex in that it also expresses proximity and is therefore considered a compound spatial category. Further examples of compound categories from the data are a place up in the desert (age 10), big long rope (age 10), this big castle there (age 6), close to that time (age 6), these two little girls come over (age 8), or move this one big rock (age 8)11. Thus, the Spatial Reference Act is based on the transformation of two (or more) simple categories into a compound one that transmits the overall spatial information (for similar arguments see Boers 1996; Gapp 1997; Vorweg and Rickheit 1999; Carstensen 2001). According to Stockman and Vaughn-Cooke (1992), the use of compound categories is cognitively and linguistically more complex and thus represents a developmental challenge for younger speakers. At the same time, their use often reects a meta-pragmatic awareness in the speaker that their communicative partner may need more precise spatial information in order to decode the message correctly and understand speakers spatial intention (cf. Gapp 1997: 67; Plumert 1996: 376). The use of compound spatial categories therefore seems to be another important measuring device for (cognitive and) linguistic development in the context of spatial language. 3.2.3. The levels of abstraction in spatial meaning. Part of the authors research interest is to trace 6 to 10 year old speakers spontaneous use of non-literal spatial reference alongside literal uses. Therefore, the four dif-

11. Although some of the examples include the combination of two instances of the same category (e.g., big long rope), whereas others represent a combination of two dierent categories (e.g., these two little girls), both cases are treated here as representing compound spatial categories. As the primary explanation for the use of such compound categories is speakers metapragmatic awareness that more precise spatial information is necessary, the type of categories combined is not considered an indicator of complexity.

Im fed up with MarmiteIm moving on to Vegemite


ferent levels of abstraction in spatial meaning as dened in Graf (2006) are applied to the data (see Table 2 below). This scalarity of spatial meaning can be explained via the categorical framework of the present approach and from a general cognitive and pragmatic perspective. In accordance with the categorical framework presented here, the dierent levels of abstraction are a consequence of the surrounding space or Reference Field into which the spatial congurations between the RE and the RR are embedded by the speaker (RO). The Reference Field is embedded in the ROs Perceptual Space (PS) in case of literal spatial reference (Its over there), in his or her Conceptual Space (CS) in case of literal spatial reference (He lives in Berlin) or metaphorical spatial reference (The meeting is in June), in both, as is the case in metaliteral meaning (Go to bed now!), or even outside ROs Conceptual Space as is the case in transmetaphorical meaning (Its going to rain) (for more detail see Graf 2006: 96 ). At the same time, the scalarity must be explained in the context of communicative interaction which motivates spatial reference in the rst place. Salient information as well as aspects of pragmatic strengthening, i.e., the conventionalization and entrenchment of situated implicatures that arise from communicative experience with a particular lexical item as new meaning components, help create dierent levels of abstraction over time (Evans 2003; Graf in press; Keysar et al. 2000; Svanlund 2007). These levels are dierentiated in terms of degree of transparency and imageability as well as recoverability of the underlying spatial relation and the question of salient information. The ubiquity of spatial language in childrens surrounding input as well as their own frequent use of spatial reference due to the importance of such information for (communicative) interaction set o such conventionalization and entrenchment processes at a very early stage. The following levels of abstraction are dierentiated here: Literal spatial meaning (This is Jim.)metaliteral spatial meaning (Go to bed now!), where the spatial linguistic times convey non-spatial information such as e.g., Its late, you need to sleep alongside the spatial information change of locationmetaphorical spatial meaning (Im moving on to Vegemite)transmetaphorical spatial meaning (Its going to rain), i.e., dead or dying metaphors (Diewald 1997; Kuteva and Sinha 1994; Rice, Sandra and Vanrespaille 1999). Examples from the data include in room eight (age 8), I am in here (age 6), go to Hawaii (age 6), live around the block from me (age 6) or at the Sahara hotel (age 10) for literal spatial meaning, buy something from the store (age 10), go to the show (age 6), get up (age 8), little boy (age 10) or get sth. from somebody (age 8) as examples for metaliteral spatial meaning, before that (age 6), stay over night (age 10), at the usual time (age 6),


E.-M. Graf

on Jeepers Creepers (age 8) or on her birthday (age 8) for metaphorical spatial meaning. Examples for transmetaphorical spatial meaning, i.e., grammaticalized and highly conventionalized items such as going to future or existential there, are found in the transcripts of all age groups12.
Table 2. Scalar nature of spatial meaning (cf. Graf 2006: 96) Level of spatial abstraction literal spatial meaning metaliteral spatial meaning metaphorical spatial meaning transmetaphorical spatial meaning Example This is Jim. Go to bed now! Im moving on to Vegemite! Its going to rain.

Based on the general assumption that ages 610 years represent a proper phase in spatial language acquisition, it is hypothesized that this period shows the following intra-phasal developments and thus diers from earlier (06 years) and later (1019 years) periods: 1) The distribution of simple and compound spatial categories, i.e., the frequency of use of compound spatial categories should be higher in the language of 10 year old speakers than in the language of 6 year old speakers. The distribution of the ve simple spatial categories; i.e., the frequency of use of more complex spatial categories (where hypothesized complexity is based on Graf 2006) such as Reference Entity Dimension should be higher in the language of 10 year old speakers than in the language of 6 year old speakers. The preference for literal or non-literal uses of spatial categories; the frequency of use of spatial categories on non-literal levels of abstraction should be higher in the language of 10 year old speakers than in the language of 6 year old speakers.



3.2.4. The analysis. The method chosen here is a qualitative categorical interpretation, followed by a quantitativeyet not statistical analysis of the classied material. In naturally evolving speech it is harder than, for example, in experimental data, to identify valuable measure points that can be quantied. This problem is even more prominent be12. Due to their grammaticalized status going to future and existential there as instances of transmetaphorical spatial meaning do not suppose a higher cognitive complexity for speakers than e.g., metaphorical spatial meaning. Instead, they are learnt as separate items in their idiomatic meaning from early on (cf. Graf in press on the relationship between syntagmatic co-occurrence of multi-word units, metaphorical strength and levels of abstraction in spatial meaning).

Im fed up with MarmiteIm moving on to Vegemite


cause size of the data is harder to control for. Well conducted qualitative research in the context of these data allows the researcher to highlight many more interesting phenomena and explain them within the context in which they occur. However [w]ell conducted qualitative research is very labour-intensive and therefore qualitative studies typically use, of necessity, much smaller samples of participants than quantitative ones (Dornyei 2007: 38). Due to this relatively small sample size a statistical analysis as a second step would not come up with reliable results. In order to nevertheless illustrate developmental trendswhich need to be corroborated in the future with the help of more datathe current qualitative analysis of spontaneous data is followed by a simple quantitative counting approach. In a rst step, the data of each age group was exhaustively searched for spatial linguistic tokens, which were marked and classied according to the categories spatial category, simple or compound category and level of abstraction. Despite the primary focus on the spatial categories, the categorical interpretation of the corpus data via contextual and co-textual information is based on a qualitative evaluation of the categories linguistic correlates (e.g., spatial prepositions or motion verbs) as their representatives on the language level. However, such an analysis of the linguistic correlates of spatial categories is not straightforward as will be discussed in more detail in Section 5. In the next section, the results are given in relation to the (sub-)hypotheses put forth earlier. In order to give a fuller picture of late stages of spatial language acquisition as a whole, reference will be made to the ndings of the previous study concerning years 10 to 19 where necessary. 4. Results

Does spatial language dier with respect to the distribution of simple and compound spatial categories?

Table 3 below gives the total number of spatial congurations found in the data and, more importantly for the present purpose, the number of simple and compound categories for each age group.
Table 3. Number of spatial categories and distribution across simple and compound categories 6 total number of spatial congurations simple compound 1,894 1,605 (84.7%) 289 (15.3%) 8 2,046 1,693 (82.7%) 353 (17.3%) 10 2,395 1,971 (82.3%) 424 (17.7%)


E.-M. Graf

Whereas six-year-old speakers represent a total of 1,894 spatial congurations in their speech, the speech of eight-year-olds contains 2,046, and that of ten-year-olds contains 2,395. Altogether, 6,335 Spatial Reference Acts were analyzed. Steady increase in SRAs from the youngest to the oldest speakers must be interpreted with great care at this point. According to the Carterette manual, roughly the same amount of language data is coded for the three groups, which would suggest that older speakers refer more often to spatial congurations than younger speakers. Possible explanations for these varying numbers are found in the free choice of topics, howeverone characteristic feature of natural speech. Figure 1 below presents the absolute numbers of simple and compound categories used by each age group in proportions.

Figure 1. Total number of spatial categories and their distribution across simple and compound categories (numbers in percentage)

We nd a rather consistent distribution of simple and compound categories; more than 4/5 of all spatial congurations in the three age groups are represented via simple categories. There is, however, some development: Whereas the simple categories account for 84.7% and the compound categories for 15.3% of all spatial congurations expressed by the six-year-old speakers, the simple categories account for 82.7% and the compound categories for 17.3% in the language of the eight-year-old speakers. Finally, the oldest speakers apply simple categories in 82.3% of all cases and compound categories in 17.7% of all cases. Such clear dominance of the simple spatial categories was also found in the previous study of speakers between 10 to 19 years of age (see Table 4), where simple categories accounted for about 88%, and compound categories for about 12% of all instances of spatial reference throughout the data.

Im fed up with MarmiteIm moving on to Vegemite


Table 4. Total number of spatial categories and their distribution across simple and compound categories from Graf (2006) 1013 total number of spatial congurations simple compound 2,289 2,019 (88.2%) 270 (11.8%) 1416 2,142 1,917 (89.5%) 225 (10.5%) 1719 2,000 1,759 (87.9%) 241 (12.1%)

Contrary to very balanced ndings in Graf (2006) for the 10 to 19 year olds, the distribution of simple and compound categories in the present data is shifting towards a slightly larger proportion of compound categories towards the end of the developmental phase as speakers at the age of ten most often refer to spatial congurations via compound categories, followed by speakers at the age of eight. Considering the higher cognitive and linguistic complexity of compound categories as well as their pragmatic function of clarifying referential ambiguity, it seems expectable that older speakers make more frequent use of such complex spatial reference in the form of compound categories. However, as the analysis of the 10 to 19 year old speakers use of compound spatial categories evinced an overall lower, albeit balanced, proportion (P88% vs. P82 to 85%), the current interpretation, i.e., the frequency of use of compound spatial categories as an indicator of more advanced language development, must be critically evaluated in future research (see Section 5).
Does spatial language dier with respect to the overall distribution of the ve simple spatial categories?

The next possible indicator of spatial development is the distribution of the ve simple spatial categories Reference Entity Distance (proximity/ distance), Reference Entity Motion (D), Reference Entity Location (S), Reference Entity Dystation (DS) and Reference Entity Dimension (DIM) (see Table 5 and Figure 2 below). Speakers of all age groups most often refer to dynamic spatial relations: Around 50% of all simple spatial categories in each age group represent dynamic relations (e.g., ride all around town or go home from school (6 years). This striking dominance of motion events is replicated in the ndings from earliest phases of spatial language development (see Graf 2006: 161 for a summary of studies) and from the pre-linguistic phase where young infants already show a preference for moving visual displays over static displays as reported e.g., by Atkinson (1993). A further conrmation of the exceptional importance of dynamism for human cognition


E.-M. Graf

Table 5. Types of simple spatial categories13 6 S D DS DIM Proximity/distance 321 (20.0%) 791 (49.3%) 81 (5%) 111 (6.9%) 301 (18.8%) 8 349 (20.6%) 901 (53.2%) 51 (3.0%) 61 (3.6%) 331 (19.6%) 10 408 (20.7%) 996 (50.5%) 74 (3.8%) 101 (5.1%) 392 (19.9%)

Figure 2. Types of simple spatial categories (numbers in percentage)

and language is illustrated in Table 6 below, where ndings from the analysis of older children and adolescents use of the ve basic spatial categories corroborate and support this clear trend (cf. Graf 2006: 185 ). Another interesting nding is the proportion of the static category and the deictic category that includes both distance and proximity. Both
Table 6. Types of simple spatial categories from Graf (2006) 1013 S D DS DIM Proximity/distance 315 (15.6%) 992 (49.2%) 100 (4.9%) 85 (4.2%) 526 (26.1%) 1416 344 (17.9%) 983 (51.3%) 72 (3.8%) 59 (3.0%) 459 (24.0%) 1719 332 (18.8%) 827 (47.1%) 80 (4.6%) 47 (2.7%) 472 (26.8%)

13. D Dynamic category (Reference Entity Motion); S Static category (Reference Entity Location); DS Dystatic category (Reference Entity Dystation); DIM Dimensional category (Reference Entity Dimension).

Im fed up with MarmiteIm moving on to Vegemite


amount to around 20% each for each age group. These ndings contrast with those from the language of speakers between 10 and 19 years of age. While we also nd a very consistent distribution, the older speakers used the category distance/proximity more often than the static category (around 25% vs. 1719%). Finally, both the dimensional category and the dystatic category are with only 3 to 7%relatively little used. This trend again is replicated in the previous study.
Does spatial language diers with respect to the preference for literal or nonliteral uses of spatial categories?

The last question focuses on the dierent levels of abstraction in spatial meaning. Table 7 below summarizes the ndings of the distribution of the simple spatial categories across the four levels of abstraction, i.e., literal, metaliteral, metaphorical and transmetaphorical spatial meaning.
Table 7. Levels of abstraction in simple categories 6 literal metaliteral metaphorical transmetaphorical 890 (55.5%) 183 (11.4%) 411 (25.6%) 121 (7.5%) 8 837 (49.4%) 117 (6.9%) 580 (34.3%) 159 (9.4%) 10 884 (44.9%) 123 (6.2%) 762 (38.7%) 202 (10.2%)

We observe that literal use outweighs all other uses in all three age groups, i.e., all speakers use the simple spatial categories most often to refer to truly spatial congurations. The meta-study of more than 75 studies dedicated to early phases of spatial language acquisition (cf. Graf 2006: 176 ) showed that childrens development with respect to the use of spatial categories on non-literal levels is on its way, but that this development is in no way nalized. In the acquisitional phase analyzed here, this developmental trend continues: 6 year old speakers refer in more than 55% of all cases to literal spatial congurations, 8 year old speakers already less than 50% of the time and, nally, the oldest speakers use decreases to 44%. At the same time, the metaphorical uses constantly rise across the three age groups from around 25% in the language of speakers aged 6 towards nearly 40% in the spontaneous speech of the ten year olds. These ndings are especially sound when considered in the larger context of how literal and metaphorical spaces develop throughout the years. Although the majority of spatial acquisitional studies focus on literal space, indirect ndings do conrm the early importance of spatial knowledge in the


E.-M. Graf

conceptualization of abstract domains: Results from Clark and Carpenter (1989, 1994), Friedman and Seely (1976), Johnson (2001), Macrae (1976) and Weist (1991) illustrate that children rst acquire demonstratives, spatial prepositions and spatial verbs in their literal spatial sense, before they beginas early as age 2to metaphorically extend them onto nonliteral, abstract domains such as time. In contrast, looking at Table 8 below, the ndings from speakers between 10 and 19 years of age clearly indicate that they have fully mastered all levels of abstraction in spatial meaning. Furthermore, metaphorical uses not only predominate across all three age groups, but we also witness a very similar distribution of all levels of abstraction in spatial meaning in that data.
Table 8. Levels of abstraction in simple categories from Graf (2006) 1013 literal metaliteral metaphorical transmetaphorical 727 (36.0%) 145 (7.2%) 847 (42.0%) 299 (14.8%) 1416 518 (27.0%) 169 (8.8%) 893 (46.6%) 337 (17.6%) 1719 547 (31.1%) 139 (7.9%) 756 (43.0%) 318 (18.0%)



Spatial language acquisition can be considered from two dierent perspectives. Whereas the rst perspective concentrates on the questions when and how children acquire spatial reference, i.e., adopts a more process-oriented stance, the second perspective presupposes the acquisition of these items and concentrates on their actual use, i.e., on spontaneous spatial reference, thus adopting a more product-oriented stance. The process of spatial language acquisition is primarily tested with the help of experiments where the items under scrutiny are elicited in narrowly dened testing procedures; the product of spatial language acquisition is best investigated with the help of analyzing spontaneous, authentic language data that oers the widest possible co- and contextual variation. The majority of acquisitional studies focus on the procedural aspects, applying experiments in carefully controlled contexts such as for example toy play or picture book reading. The acquisition of spatial items is thereby predominantly tested in purely spatial contexts14 (for a critical evaluation of such
14. Of great interest in these studies has been how language-specic characteristics such as a satellite- or verb-framed typology inuence the acquisition process of the category SPACE, whose importance and ubiquity in human thought and language is universally observable and acknowledged (Choi and Bowerman 1991; Bowerman and Choi 2001; Choi 2006; Hickmann and Hendriks 2006).

Im fed up with MarmiteIm moving on to Vegemite


procedures see Crystal 1997; Yont et al. 2003). Product-oriented studies nd a prominent forerunner in Ames and Learneds study from 1948, but their approach has not been frequently replicated. The present study as well as Graf (2006) take up this approach and look at the natural use of the dierent types of spatial reference in literal as well as non-literal senses. The results stem from a qualitative categorical interpretation of transcripts of spontaneous speech samples. Yet, such method is not an unproblematic endeavor. Due to the characteristics of spoken language such as false starts, repetitions, ungrammatical structures etc., the analyzed material contains instances of spatial representations whose status is problematic. The uncontrollability of conversational topics in natural use of spoken language adds another challenge. Furthermore, the analyses above were based on the transcribed version of the face-to-face interaction that should be complemented by multimodal information (Arndt and Janney 1987). In this respect, the categorical interpretation is challenging in certain cases. A similar argument holds for the uncontrollability of the co- and context, which also renders decisions on category membership sometimes dicult. Last but not least, basing the categorical interpretation on the categories linguistic correlates is not a straightforward endeavor. While closed-class tokens such as spatial prepositions or the demonstratives constitute a clear-cut group and pose no problems for the analysis, open-class items are often less easy to interpret and it is sometimes hard to draw the line between what counts as a spatial verb or noun or not: Whereas come and go and other examples of core spatial expressions are easily classiable and are frequently dealt with in the literature on spatial language, prepositional and phrasal verbs such as write down, speak up or look at are less clear cut cases. The same holds true for spatial nouns, where category membership is often fuzzy. To keep the subjective moment of such a qualitative approach at a minimum, independent coding by more than one researcher is a clear desideratum for further projects in order to achieve inter-coder reliability. Bearing such methodological problems in mind, the presented ndings of this (exploratory) study do raise important issues of spatial language development. For the period under scrutiny here, i.e., the years 6 to 10, the distribution of simple vs. compound categories, the use of the ve simple spatial categories as well as the use of these spatial categories on the four levels of abstraction were hypothesized to function as possible indicators of spatial development (see Chapter 3). Consequently, these years were hypothesized to represent a proper developmental phase within speakers spatial development, bridging the phases of primary spatial language acquisition and the years after age 10, whereaccording to a prior


E.-M. Graf

study in the same framework (Graf 2006)no further spatial development could be reported. As regards the distribution of simple and compound categories in the Carterette data, speakers rather consistently apply compound categories in about 1/5 of all instances of spatial reference. Only a small increase in the use of compound categories can be reported for the older speakers. To adequately apply compound categories in context, children have to come to terms with what Carstensen (2001: 73) calls Konzept-Unvertrag lichkeit (concept incompatibility): Spatial congurations that cannot be perceived and conceived together in a complex conguration cannot be represented together on the language level. At the same time, children have to learn that a certain spatial ambiguity for the communicative partner may be solved with the help of applying compound categories in ones Spatial Reference Act (e.g., Which one is yours? The one to your left). What is more, the use of compound spatial categories such as the tiny little opening through the couch between the wall (age 10) supposes a higher cognitive and linguistic complexity. The slight increase in compound categories from 6 to 10 years of age then possibly implies an ongoing development for the phase under scrutiny based on such questions of cognitive complexity and communicative awareness. The overall dominance of simple categories, in turn, replicates the ndings from the analysis of spatial reference in 10 to 19 year old speakers. However, on average the amount of simple categories was greater in the previous study with around 88% of simple categories across the three tested age groups. Bringing the ndings from the two studies together, the issue of the general comparability of the two sets of data needs to be raised: Whereas experimental settings allow for an exact control of the testing parameters, spontaneous language material dees such control in general. In addition, the Carterette material stems from semi-spontaneous interaction initiated by a researcher, whereas the COLT data represents truly spontaneous interaction. Still, the chosen databased on its co- and (partly) contextual variationis the best and only available material to undertake the endeavor of a qualitative interpretation of spontaneous spatial reference in a developmental context. A critical look at what types of spatial categories are combined within such compound categories as well as an assessment of the communicative co- and context are necessary. What is more, controlled experiment situations in which speakers use of compound categories are tested in relation to their age, cognitive and communicative development etc. are needed to rene and further elaborate on these qualitatively based ndings. As regards the ve simple spatial categories and the question whether their distribution functions as an indicator of development, the following

Im fed up with MarmiteIm moving on to Vegemite


trend can be reported: The most frequently used spatial category is Reference Entity Motion; speakers of all age groups refer to dynamic spatial relations in roughly 50% of all instances of spatial reference. Our human propensity towards dynamism as claimed by Talmy (2003: 12)already conrmed for the pre-linguistic spatial development where infants prefer dynamic over static relations and corroborated for the earliest stages of spatial language acquisition (cf. a meta-analysis in Graf 2006: 147 )is also shaping the phase from 6 to 10 years of age and explains the lack of development. Both distance/proximity relations as well as static spatial relations are consistently used; the same holds true for the categories Reference Entity Dimension and Reference Entity Dystation, which are rather consistently distributed across the data and are of overall minor importance. The slightly dierent proportions of Reference Entity Location and Reference Entity Distance in the present and the previous data may derive from the varying contextual situations of the data collection or the general freedom of topic choice in spontaneous speech. Whereas the speakers between six and ten from the Carterette les were sitting round a table while conversing with each other, the older informants in the COLT corpus were engaged in all kinds of activities such as walking home from school, playing football, going to the movies etc. (see Section 3). Such outdoor activities may require a more frequent use of deictic information; however, again, this assumption remains to be conrmed by further analysis. In addition, it is necessary in the future to focus on the various subtypes of the simple spatial categories, whose diering cognitive complexity may evince, after all, some developmental dierences in the language of speakers of dierent ages. At this point, however, the distribution of the simple spatial categories cannot be considered an indicator for spatial development. Last, but not least, the distribution of the four levels of abstraction was hypothesized as another possible indicator for spatial development. The proportion of literal and non-literal uses of spatial categories in the language of speakers between 6 and 10 years of age indeed points to clear developmental changes: There is a decrease in literal uses from the youngest speakers, where more than half of all instances of spatial categories are literal, to the oldest speakers, were less than half of all instances of spatial categories are literal. The decrease is paralleled by an increase of metaphorical uses across the three age groups, starting with one out of four in the language of the 6 year olds and ending with roughly one out of three in the language of the oldest, i.e., the 10 year olds. This developmental trend in the data is backed up by acquisition studies which claim that children rst acquire spatial expressions in their literal spatial


E.-M. Graf

meaning, before they learn to extend them onto non-literal uses (cf. Clark and Carpenter 1994; Friedman and Seely 1976; Johnson 2001; Macrae 1976; Weist 1991). Moreover, the rise of metaphorical uses andat the same timethe decline of the literal uses is indirectly corroborated by the ndings from the previous study. All age groups consistently showed more metaphorical than literal uses of all spatial categories under scrutiny (cf. Graf 2006: 189 ) (for a more detailed analysis of the types of spatial metaphors in the language of speakers between 10 and 19 years of age and the varying degree of their metaphorical strength see Graf in press). We can observe a clear developmental change then with respect to literal and non-literal uses of spatial categories in the phase analyzed here, i.e., literal and non-literal uses function as true indicators of spatial development. At the same time, this development seems to be nalized at around the age of 10 according to the previous study. This implies a relatively narrow acquisition frame for spatial metaphors, which may best be explained with the ubiquity of spatial reference in the linguistic input with which children grow up as well as their own need to spatialize abstract ideas to make them intellectually accessible. Other metaphorical extensions, based on more complex source domains, may take more time to fully develop: . . . not surprisingly, the ability to understand more complex analogical and metaphorical mappings involving less familiar domains . . . or higher order relations . . . increases with age and achieves adult-like quality somewhere between ages 10;014;0 (Ozcaliskan and Goldin-Meadow 2005: 233). However, as was already mentioned . . . there has been no systematic work on how children learn the metaphorical extensions of motion [and other spatial relations, the author] as they become native speakers of a particular language (Ozcaliskan 2005: 291f ). More research on the acquisition of spatial language in non-spatial contexts is required, especially with respect to those frequent incidents of highly conventionalized spatial metaphors (see Graf in press). What seems especially rewarding for future research then is the concept of conventionalization and how it inuences childrens acquisition of the dierent levels of abstraction in spatial meaning: According to Svanlund (2007: 577f ), the conception of conventionalization as put forth in conceptual metaphor theory thereby . . . underestimates the social nature of conventions. It also underestimates the role of linguistic experience. A more usage-based, i.e., communicatively oriented approach is required. To sum up, the period from 6 to 10 years of age can indeed be considered a proper developmental phase in spatial language acquisition. Particularly with respect to literal and non-literal uses of spatial language important developmental steps take place. The youngest speakers refer twice as much to literal than to metaphorical spaces, but even the eight-

Im fed up with MarmiteIm moving on to Vegemite


year-old children still show a clear preference for such literal uses. However, their metaphorical uses are clearly on the rise, and the oldest group of speakers, the ten year olds, shows a nearly balanced preference for both literal and metaphorical uses, with a slight preference still for the literal ones. This would mean, at the same time, that the period of secondary spatial language development covers the years 6 to 10, not between 10 and 19 years of age as assumed at the beginning of Graf (2006). 6. Conclusion and Future Research Desiderata

A larger picture of spatial language development is slowly taking shape. The following developmental trend emerges: English spatial language development takes place during three separate, but related, phases, up to the age around 10. After these years hardly anything develops, at least according to Graf (2006). The presence of spatial experience from the earliest moments of human life and their importance for human cognition and language could account for this relatively early mastering of linguistic reference to space. However, more integrative studies of spontaneous spatial reference in naturally occurring speech are needed. As assumed at the beginning of this paper, contrary to such a lack of spatial linguistic development after age 10, important developmental steps take place in the language of children between six and ten years of age, especially in the context of literal and metaphorical use of spatial categories. One possible explanation may lie in the dramatic changes that happen in childrens lives around that age. In the language context under scrutiny here, they leave home and start school, they learn to read and to write, i.e., they acquire new media that supply them with additional (spatial) information, etc. All of this supposes new cognitive and communicative challenges that contribute to the consolidation and elaboration of the already acquired spatial linguistic capacities. However, much remains to be done to conrm or challenge and further specify the big picture. Necessary next steps must include the analysis of adult spontaneous language with respect to the use of the various spatial categories as well as their use on the dierent levels of abstraction. At this stage, it can already be hypothesized that we nd a similar predominance of dynamic spatial categories as well as a majority of metaphorical uses of spatial categories, which would manifest, once more, the human propensity towards spatial metaphors and the human propensity towards dynamism. In addition, childrens earliest spontaneous use of spatial language should also be studied from an integrative perspective. As claimed by Tomasello (2003), children pass through a stage of isolated islands of


E.-M. Graf

knowledge that are only linked later on due to more linguistic input and experience, the capacity to draw analogies and make abstractions as well as their general cognitive development. It should be rewarding to trace such linking of the various spatial categories and their linguistic forms with respect to questions of mutual inuence, the acquisition of the various meaning components within certain spatial expressions in diering co- and contexts, as well as their use on the various levels of abstraction. Such assumed mutual inuence of knowledge and use of certain spatial categories and their correlates on other spatial categories should thereby be traced both in spontaneous language and tested in experimental settings. The combination of a qualitative and integrative analysis of spontaneous spatial language use with experimentally based research on e.g., age-preferential encodings of spatial congurations via simple or compound categories is of utmost importance for the reliability of the ndings. Once such product- and process-oriented ndings are available, the spatial development of speakers of English can be documented from the very beginning to its nalization in adolescent years as well as its further characteristics in adult years. Received 1 March 2009 Revision received 25 November 2009 References
Ames, Louise and Janet Learned. 1948. Development of Verbalized Space in the Young Child. Journal of Educational Psychology 39. 101116. Arndt, Horst and Richard Janney. 1987. InterGrammar. Berlin: Mouton de Gruyter. Atkinson, Jean. 1993. A neurobiological approach to the development of where and what systems for spatial representation in human infants. In Naomi Eilan, Rosaleen McCarthy and Bill Brewer (eds.), Spatial representation: Problems in philosophy and psychology, 325339. Oxford: Blackwell. Aurnague, Michel, Maya Hickmann and Laure Vieu (eds.). 2007. The categorization of spatial entities in language and cognition. Amsterdam: John Benjamins. Berman, Ruth (ed.). 2004. Language development across childhood and adolescence. Amsterdam: John Benjamins. Berman, Ruth and Dan I. Slobin. 1994. Relating events in narrative: A crosslinguistic developmental study. New York: Lawrence Erlbaum Associates Blake, Joanna. 2000. Routes to child language: Evolutionary and developmental precursors. Cambridge: Cambridge University Press. Boers, Frank. 1996. Spatial prepositions and metaphor. Tubingen: Gunter Narr. Boroditsky, Lera. 2000. Metaphoric structuring: understanding time through spatial Metaphors. Cognition 75. 128. Bowerman, Melissa. 1996. The origins of childrens spatial semantic categories: cognitive vs. linguistic determinism. In John J. Gumperz and Stephen Levinson (eds.), Rethinking linguistic relativity, 145176. Cambridge: Cambridge University Press.

Alpen-Adria University

Im fed up with MarmiteIm moving on to Vegemite


Bowerman, Melissa and Stephen Levinson (eds.). 2001. Language acquisition and conceptual development. Cambridge: Cambridge University Press. Bowerman, Melissa and Soonja Choi. 2001. Shaping meanings for language: Universal and language-specic in the acquisition of spatial semantic categories. In Melissa Bowerman and Stephen Levinson (eds.), Language acquisition and conceptual development, 475511. Cambridge: Cambridge University Press. Carlson, Laura and Emile van der Zee (eds.). 2005. Functional features in language and space: Insights from perception,categorization, and development. Oxford: Oxford University Press. Carstensen, Kai-Uwe. 2001. Sprache, raum und aufmerksamkeit. Tubingen: Max Niemeyer. Choi, Soonja. 2006. Inuence of language-specic input on spatial cognition: Categories of containment. First Language 26(2). 207232. Choi, Soonja and Melissa Bowerman. 1991. Learning to express motion events in English and Korean: The inuence of language-specic lexicalization patterns. Cognition 41. 83121. Clark, Eve and Kathie Carpenter. 1989. On childrens use of from, by, and with in oblique noun phrases. Journal of Child Language 16. 349364. Clark, Eve and Kathie Carpenter. 1994. The notion of source in language acquisition. In Paul Bloom (ed.), Language acquisition, 251284. Cambridge, MA: MIT Press. Crystal, David. 1997. The Cambridge encyclopedia of the English language. Cambridge: Cambridge University Press. Diewald, Gabriele. 1997. Grammatikalisierung. Eine Einfuhrung in Sein und Werden gram matischer Formen. Tubingen: Max Niemeyer. Dominey, Peter. 2006. From holophrases to abstract grammatical constructions: Insights from simulation studies. In Clark, Eve and Barb Kelly (eds.), Constructions in acquisition, 137161. Standford: CSLI Publications. Dornyei, Zoltan. 2007. Research methods in applied linguistics. Oxford: Oxford University Press. Eckert, Penelope. 1998. Age as a sociolinguistic variable. In Florian Coulmas (ed.), Handbook of Sociolinguistic, 151167. Oxford: Oxford University Press. Evans, Vyvyan. 2003. The structure of time: Language, meaning and temporal cognition. John Benjamins, Amsterdam. Friedman, William and Pamela Seely. 1976. The childs acquisition of spatial and temporal word meaning. Child Development 47. 11031108. Gapp, Klaus-Peter. 1997. Objektlokalisation: Ein System zur sprachlichen Raumbeschreibung.Wiesbaden: Deutscher Universitatsverlag. Gardner, Howard 1974. Metaphors and modalities: How children project polar adjectives onto diverse domains, Child Development 45. 8491. Gentner, Dedre. 2001. Spatial metaphors in temporal reasoning. In Meredith Gattis (ed.), Spatial schemas and abstract thought, 203222. Cambridge, MA: MIT Press. Gentner, Dedre and Lera Boroditsky. 2001. Individuation, relativity, and early word learning. In Melissa Bowerman and Stephen Levinson (eds.). Language acquisition and conceptual development, 215256. Cambridge: Cambridge University Press. Gentner, Dedre and Susan Goldin-Meadow (eds.), 2003. Language in the mind: Advances in the study of language and thought. Cambridge, MA: MIT Press. Graf, Eva-Maria. 2006. The ontogenetic development of literal and metaphorical space in Language. Tubingen: Gunter Narr. Graf, Eva-Maria. In press. Adolescents use of spatial time metaphors: A matter of cognition or socio communicative practice? Journal of Pragmatics. Special Issue: The Language of Space and Time. Gumperz, John J. and Stephen Levinson (eds.). 1996. Rethinking linguistic relativity. Cambridge: Cambridge University Press.


E.-M. Graf

Hickmann, Maya. 2003. Childrens discourse: Person, space and time across languages. Cambridge: Cambridge University Press. Hickmann, Maya. 2007. Static and dynamic location in French. Developmental and crosslinguistic Perspectives. In Michel Aurnague, Maya Hickmann and Laure Vieu (eds.), The categorization of spatial entities in language and cognition, 205231. Amsterdam: John Benjamins. Hickmann, Maya and Henriette Hendriks. 2006. Static and dynamic location in French and in English. First Language 26(1). 103135. Johnson, Christopher. 1997. Learnability in the acquisition of multiple senses: SOURCE Reconsidered. Proceedings of the 22th annual meeting of the Berkeley Linguistic, 469 480. Berkeley: Berkeley Linguistic Society. Johnson, Christopher. 1999. Metaphor vs. Conation in the acquisition of polysemy: The case of see. In Masako K. Hiraga, Chris Sinha and Sherman Wilcox (eds.), Cultural, typological and psychological issues in cognitive linguistics: Current issues in linguistic theory, 155169. Amsterdam: John Benjamins. Johnson, Christopher. 2001. Constructional Grounding: On the relations between deictic and existential thereconstructions in acquisition. In Alan Cienki, Barbara Luka and Michael Smith (eds.), Conceptual and discourse factors in linguistic structures, 123136. Stanford: CSLI Publications. Karmilo, Kyra and Annette Karmilo-Smith. 2001. Pathways to language: From fetus to adolescent. Cambridge: Cambridge University Press. Keysar, Boaz, Shen Yeshayahu, Sam Glucksberg and William Horton. 2000. Conventional Language: How Metaphorical Is It? Journal of Memory and Language 43. 576593. Kuteva, Tania and Chris Sinha. 1994. Spatial and non-spatial uses of prepositions: Conceptual integrity across semantic domains. In David Mark and Andrew U. Frank (eds.), Cognitive and linguistic aspects of geographic space, 419434. Dordrecht: Kluwer. Labov, William. 1972. Sociolinguistic patterns. Oxford: Blackwell. Levinson, Stephen. 2003. Space in language and cognition: explorations in linguistic diversity. Cambridge: Cambridge University Press. Levinson, Stephen and David Wilkins (eds.). 2006. Grammars of space: Explorations in cognitive diversity. Cambridge: Cambridge University Press. Levinson, Stephen and David Wilkins. 2006b. Patterns in the data: towards a semantic typology of spatial description. In Stephen Levinson and David Wilkins (eds.), Grammars of space: Explorations in cognitive diversity, 512552. Cambridge: Cambridge University Press. Lock, Andrew. 1980. The guided reinvention of language. London: Academic Press. Lock, Andrew. 1999. Preverbal communication. In Gavin Bremner and Alan Fogel (eds.), Handbook of infancy research, Oxford: Blackwell. Macrae, Alison. 1976. Movement and location in the acquisition of deictic verbs. Journal of Child Language 3. 191204. Mayer, Mercer. 1969. Frog, where are you? New York: Pied Piper. Nippold, Marilyn A. 1998. Later language development: The school-age and adolescent years. Austin: Pro-Ed. Nowak, Peter. 2007. Meta-Studien Methodik ein neues Methodenparadigma fur die Dis kurs-forschung. Gesprachsforschung 8. 89116. Ozcaliskan, Seyda. 2003. Childrens developing understanding of metaphors about the mind. In Barbara Beachley, Amanda Brown and Frances Conlin (eds.). Proceedings of the 27th Annual Boston University Conference on Language Development, 603614. Somerville, MA: Cascadilla Press.

Im fed up with MarmiteIm moving on to Vegemite


Ozcaliskan, S eyda. 2004. Encoding the manner, path and ground components of a metaphorical motion event. Annual review of cognitive linguistics 2. 73102. Ozcaliskan, Seyda. 2005. On learning to draw the distinction between physical and meta phorical motion: is metaphor an early emerging cognitive and linguistic capacity? Journal of Child Language 32. 291318. Ozcaliskan, Seyda and Susan Goldin-Meadow. 2005. Gesture is at the cutting edge of early language development. Cognition 96. B101B113. Pearson, Barbara 1990. The comprehension of metaphor by preschool children. Journal of Child Language, 17. 185203. Plumert, Jodie M. 1996. Young childrens ability to detect ambiguity in descriptions of locations. Cognitive Development 11. 375396. Pruden, Shannon M., Kathy Hirsch-Pasek, Roberta M. Golinko. 2008. Current events: How infants parse the world and events for language. In Thomas F. Shipley and Jerey M. Zacks (eds.), How humans see, represent, and act on events, 160192. New York, NY: Oxford University Press. Quinn, Paul C. 1998. Object and spatial categorization in young infants: what and where in early visual development. In Slater, Alan (ed.), Perceptual development: Visual, auditory, and speech perception in infancy, 131165. Hove: Psychology Press. Quinn, Paul C. 2005. Developmental Constraints on the Representation of Spatial Relation Information: Evidence form Preverbal Infants. In Laura Carlson and Emile van der Zee (eds.), Functional features in language and space: Insights from perception, categorization, and development, 293309. Oxford: Oxford University Press. Rice, Sally, Sandra Dominiek and Mia Vanrespaille. 1999. Prepositional semantics and the fragile link between space and time. In Masako Hiraga, Chris Sinha and Sherman Wilcox (eds.), Cultural, psychological, and typological issues in cognitive linguistics, 107127. Amsterdam: John Benjamins. Romaine, Suzanne. 1984. The language of children and adolescentsthe acquisition of communicative competence. New York: Basil Blackwell. Stockman, Ida and Fay Vaughn-Cooke. 1992. Lexical elaboration in childrens locative action expression. Child Development 63(5). 11041125. Svanlund, Jan. 2007. Metaphor and convention. Cognitive Linguistics 18/1. 4789. Talmy, Leonard. 1996. Fictive motion in language and ception. In Paul Bloom, Mary Peterson, Lynn Nadel and Merrill F. Garrett (eds.), Language and space, 211276. Cambridge, MA: MIT Press. Talmy, Leonard. 2000. Towards a Cognitive Semantics. Cambridge, MA: MIT Press. Talmy, Leonard. 2003. Fictive Motion in Language and Ception. Paper presented at the LIPP symposium, Ludwig-Maximilians-Universitat, Munich. Tenbrink, Thora. 2007. Space, time, and the use of language: An investigation of relationships. Berlin: Mouton de Gruyter. Tomasello, Michael. 1998. Introduction. In Michael Tomasello (ed.), The new psychology of language: Cognitive and functional approaches to language structure (viixxii). Mahwah, NJ: Lawrence Erlbaum. Tomasello, Michael. 2003. Constructing a language: A usage-based theory of language acquisition. Cambridge: Harvard University Press. Vorweg, Constanze and Gert Rickheit. 1999. Richtungsausdrucke und Heckenbildung beim sprachlichen Lokalisieren von Objekten im visuellen Raum. Linguisische Berichte 178. 152204. Weist, Rudy. 1991. Spatial and temporal location in child language. First Language 11. 253 267.


E.-M. Graf

Winner, Ellen. 1976. New names for old things: The emergence of metaphoric language. Journal of Child Language 6. 469491. Winner, Ellen (ed.). 1995. Developmental perspectives on metaphors. Hillsdale: Lawrence Erlbaum. Yont, Kristine, Catherine Snow and Lynne Vernon-Feagans. 2003. The role of context in mother-child interactions: an analysis of communicative intents expressed during toy play and picture book reading with 12month olds. Journal of Pragmatics 35. 435454.

On the use of posture verbs by French-speaking learners of Dutch: A corpus-based study


Abstract This article presents the results of a quantitative and qualitative corpus study of the use of the Dutch posture verbs staan (stand), liggen (lie) and zitten (sit) by French-speaking learners of Dutch. In addition to providing a quantied insight into which uses of these verbs prove most problematic to the L2 learners, the study has also revealed three important tendencies. Firstly, in line with the typological dierences between French and Dutch (where these verbs behave like noun classiers), our analysis conrms the French-driven tendency of the learners for underusing these verbs. Secondly, seemingly paradoxical to the previous point, is that these learners occasionally overuse these posture verbs in contexts where no such verb is allowed. Thirdly, our qualitative analysis of errors reveals that the learners operate on grammaticised semantic distinctions drawn from the target language. Even if the categories used by L2 speakers may not be the same as those exploited by native speakers, our analysis suggests that the L2 speakers are thus aware of the patterns in the input and exploit them in a fashion that may not dier all that much in kind from those in L1 acquisition. Keywords: posture verbs, Dutch, second language acquisition, learner corpus, lexical, overgeneralisation.

* Address for correspondence: M. Lemmens, Professeur en linguistique et didactique des langues, UFR Angellier, Universite de Lille 3, B.P. 60149, 59653 Villeneuve dAscq CEDEX, France. Email:; J. Perrez, Facultes Universitaires Saint-Louis (FUSL), Boulevard du Jardin botanique 43, 1000 Bruxelles, Belgium. Email:; Acknowledgements: The authors wish to thank the editors of this special issue (Henriette Hendriks, Maya Hickmann and Katrin Lindner), Sabine De Knop, Philippe Hiligsmann, Aliyah Morgenstern and the two anonymous reviewers for their constructive comments on an earlier version of this paper. The authors are responsible for any errors that remain. Cognitive Linguistics 212 (2010), 315347 DOI 10.1515/COGL.2010.012 09365907/10/00210315 6 Walter de Gruyter

316 1. 1.1.

M. Lemmens and J. Perrez Introduction Scope and issues

Anyone familiar with the teaching of Dutch as a foreign language will know that the use of the three cardinal posture verbs zitten (sit), liggen (lie) and staan (stand) are often quite problematic for learners. In this paper, we present the results of a corpus-based study, in which we looked at how these verbs are used by Belgian francophone learners of Dutch.1 We approach the data from a quantitative as well as a qualitative perspective. The quantitative analysis allows us to evaluate in which uses the diculties are mostly situated. The qualitative analysis discusses some of the mechanisms that lead L2 speakers to produce these errors. The present study is but a rst (yet essential) step towards a more systematic analysis of the use of posture and location verbs in learner data (in Dutch as well as other languages) and will be followed by comparative research drawing on more controlled (spoken) data along the lines of earlier research in this domain (cf. Lemmens 2005a). Notwithstanding its modesty in scope, our corpus data support a usage-based model of (second) language acquisition, suggesting evidence for partially unit-based learning strategies as well as for systematic overgeneralisations of acquired patterns, much like what is known to occur in L1-acquisition. While the semantic categories with which the learners operate may not be the same as those of the native speakers, the learners errors show that the learner language is a linguistic system, in which grammaticised semantic distinctions drawn from the target language do play an important role (Klein 2008; cf. also Klein and Perdue 1993; Hiligsmann 1997). If it werent for this (partial) semanticisation, we could not explain the apparent paradox in the L2-data, i.e., the undeniable (typologically determined) underuse of the posture verbs in general, combined with posture verb overkill in many of the sentences in which these verbs do occur. 1.2. Typological background

In earlier work (Lemmens 2002), one of the authors has characterized the diculty that francophone learners of Dutch have with posture verbs as being situated on three interrelated levels: (i) coding exibility, (ii) coding variability and (iii) coding obligation. As the term suggest, coding exibil1. The present study focuses on Dutch as spoken in Flanders, the Dutch-speaking part of Belgium, which is also the variant with which the French-speaking learners in this study will be most regularly confronted (even if not exclusively). There are some interesting dierences between Belgian and Netherlandic Dutch concerning the use of posture verbs (cf. also Lemmens 2006).

On the use of posture verbs


ity refers to the wide range of semantic extensions (semasiological variation) that the posture verbs have in Dutch, since they have grammaticalised to basic locational verbs that are not only used to refer to the basic human postures, but also to the location of any entity in space or metaphorical extensions thereof (cf. section 1.3 below). The second diculty concerns the coding variation, which represents the other side of the coding coin (onomasiological variation), since one and the same spatial conguration, such as for example in De boter in de koelkast (the butter in the refrigerator) may be coded either with staan (in which case it metonymically refers to the butter dish standing on its base), with liggen (in which case it talks about the package typically lying on its longest side), or with zitten (an a-positional usage referring to containment only). Each of those verbs clearly imposes its own semantic prole on the scene; the choice of verb cannot be predicted with absolute certainty based on dimensions of the located object (although these dimensions may play a role in certain contexts). Often, (French) L2 speakers are mislead by these dimensions, saying for example that a bed in a room or a plate on the table (entities with a salient horizontal dimension/orientation) are lying whereas in Dutch staan (stand) is to be used. The third level of diculty, the coding obligation, concerns the fact that the use of a posture verb is obligatory in Dutch whenever an entity is located in space, whereas in English and in French, it is quite common (if not obligatory) to use a verb of EXISTENCE (such as be/etre) in locative predications, as illustrated in example (1) below.2 (1) a. b. c. my keys are on the table / the car is in front of the house mes cles sont sur la table / la voiture est devant la maison mijn sleutels liggen (*zijn) op de tafel / de auto staat (*is) voor het huis

While in English one could still use lie and stand in these two contexts (even if often giving a more stilted formulation), this is quite infelicitous in French: *mes cles sont couchees sur la table / *la voiture est debout devant la maison. The coding obligation in Dutch also holds for many metaphorical uses (even if some leniency is to be attributed to these, cf. Section 3 below). In short, not only do francophone learners of Dutch have to go against their native speaker intuitions and use a posture verb instead of a neutral
2. Examples without a reference have been constructed by the authors (possibly varying on attested uses in other corpora), examples from the corpus will be marked with an IDnumber, and examples found via Google will have a URL reference.


M. Lemmens and J. Perrez

verb, they are also confronted with considerable semasiological and onomasiological variation. While this typological dierence, related to that between Verb-framed and Satellite-framed languages (see Talmy 2000), has been discussed in earlier work (see, e.g., Lemmens 2005a; Lemmens and Slobin 2008 and the references therein), similar observations have been made in some other recent publications. Of note is the special issue of Linguistics edited by Ameka and Levinson (2007), devoted to location and posture verbs in a typologically varied language sample. While French is not included in their comparative study, the typological distinction between French and Dutch would in their terminology be cast as that between a Type I language (using a single locative dummy verb) versus a Type II language using a small set of locative verbs (typically, but not exclusively, posture verbs, as in Dutch). Before we turn to a more detailed discussion of the use of posture verbs in L2 productions, it is essential that we briey review some of the main patterns of use for the three posture verbs in Dutch, presented in the next section. Restricted to patterns that are immediately relevant to the L2 data, the description is but a summary of more elaborate descriptions of the Dutch posture verbs presented elsewhere (Lemmens 2002, 2006). 1.3. A short overview of Dutch posture verbs

In line with the basic assumptions of Cognitive Grammar, the Dutch posture verbs liggen, zitten, and staan can safely be said to be structured around a prototype, the representation of the three basic human positions. As Newman (2002) correctly observes, these prototypes are experiential clusters of attributes and the extended uses can be explained drawing on the notion of image schemata based on our everyday experience of lying, standing, sitting. Classifying the extensive networks in broad strokes, we can distinguish three types of uses: postural uses, referring to human posture; locational uses, referring to the location of any entity in space; and metaphorical uses, referring to location in abstract space or location of abstract entities in concrete space. The following sections will look at some of the extensions in more detail. 1.3.1. Staan. The most important uses of staan can be summarized as in the schema below. (i) (ii) (iii) be on ones feet be on ones base extend upward from base (origin) extend from origin in any direction have a vertical orientation (absence of base or not on base)

On the use of posture verbs (iv) (v) be in canonical position written text as standing


The image of an object on its base, a logical extension of the prototype conguration of a human being on its feet, is undoubtedly the most productive one within the locational domain. In an earlier corpus study (Lemmens 2002), it was shown to account for almost 60% of the locational uses. Its conceptual importance is further reected in the fact that the real dimensions of the object do not play a role anymore: for any object resting on its base, a coding with staan becomes the most likely candidate, even if it is more horizontal than vertical, as is the case for cars, plates or laptops, which are said to be standing when resting on their base. Considering cognitive processing, one could argue, as does Serra Borneto (1996) discussing German stehen (stand), that the conceptualisation of a base triggers a mental verticality, i.e., the mental image of an upward extension of an object taking the base as its origin.3 Typically, the situation involves a vertical extension (e.g., trees or grass growing upwards from their roots and thus standing), but through image schematic transformation (rotation), the verb can also be applied in contexts where non-vertical direction is at issue, as in Er staan geen takken meer aan deze boom There stand no branches to this tree anymore.4 Such uses of staan do not express verticality but a (moderate) form of perpendicularity.5 Verticality only comes in as a determinative factor in the absence of a base, as in (2a), or when the object is not resting upon its base and verticality is needed to identify its orientation, as in (2b). (2) a. Het boek staat in het rek. / De golfstok staat in de paraplubak. the book stands on the shelf / the golfclub stands in the umbrella holder. De borden staan in de afwasmachine / De es stond omgekeerd op tafel. the dishes stand in the dish washer / the bottle stood upsidedown on (the) table


It is particularly in this case that staan provides a maximal opposition with liggen. Before continuing with liggen and zitten, however, we need

3. On German posture verbs, see also Fagan 1991 and Kutscher and Schultze-Berndt 2007. 4. The English glosses are but literal translations of the Dutch originals using as much as possible the English equivalents sit, lie, or stand. 5. Dutch is not isolated in this. Perpendicularity is a notion also important for example to a language as Trumai, a genetic isolate spoken in Brazil (cf. Guirardello-Damian 2002).


M. Lemmens and J. Perrez

to mention a few metaphorical extensions for staan that will be immediately relevant to the L2-data. The rst (extension (iv) in the schema above) concerns a number of different uses that all relate to the idea of standing as the canonical position for human beings (cf. also Van Oosten 1984: 144). There are a number of (non-linguistic) arguments to justify this claim. First, standing upright is the position that most distinguishes the human being (homo erectus) from other species, esp. primates. Moreover, standing is the starting position for the proto-archetype of human, self-propelled movement, viz. walking or running (on two legs). Related to this is that, when standing, humans are physically stronger than when sitting or lying and generally have better control over their body movements. Humans in a standing position are also perceptually more distinguishable from their surroundings (cf. the metaphor stand out and outstanding). In short, human beings physically function best when in a standing position, feeding the idea of canonicity. But also other sources can serve to conrm the canonicity: if you ask someone to quickly draw a human being, they will typically draw a standing gure.6 Finally, returning to the domain of linguistic meaning, it can be seen that many extensions, locational and metaphorical, draw precisely on standing as the canonical position. This is especially true for Dutch where, as explained above, staan has become the conventionalized coding for any object resting on its base, the default position also being the objects optimal position, i.e., the functional position it has been designed for. In short, standing being the canonical position for human beings, motivates the use of staan to refer to a human beings default posture, even when posture is backgrounded or even no longer at issue.7 This is reinforced by the use of the verb to refer to objects in their normal (i.e., functional) position. This pertains to our study in two dierent ways. First, it may help to explain why staan is used in contexts where there may still be a reference to the standing posture as the most typical posture for the activity at hand, but where there is some non-postural reading as well. A typical case is that of working as a teacher or as a shopkeeper, where you would commonly say (at least in Belgian Dutch) ik sta in het onderwijs (I stand in the education) or Ik sta in een herenboetiek (I stand in a clothes

6. Notice that the standing position is also the way in which humans are represented in handbooks on human anatomy. 7. Clearly, motivation does not equate prediction, as the notion of canonical position can be overruled by for example cultural factors. In Ese Ejja, for example, an endangered language (Tacana family) spoken in Peru and Bolivia, for some contexts the default position for men is neki (stand), that for women, ani (sit) (Vuillermet 2008).

On the use of posture verbs


boutique for men). Typically, one stands in front of the class room when teaching or behind the counter when running a shop, but both sentences do more than just refer to that postural conguration, referring to the job as a whole, which involves many more kinds of activities than just standing (walking around, sitting and correcting exams, etc.). Notice that for other types of jobs, if one wants to use a posture verb at all, it will be zitten (e.g., Hij zit in de computerbranche He sits in the computer business), but this is an a-postural use of the verb referring to containment (cf. section 1.3.3 below). Second, and more important than the above cases which are rather limited, there are cases where a standing posture is no longer at issue at all, and staan simply refers to the default position. This has given rise to a wide range of extended uses, as illustrated by the following examples: (3) a. b. De politici staan tegenwoordig veel dichter bij de burgers. politicians these days stand much closer to the citizens Hoe sta jij tegenover de nieuwe spelling? how do you stand against the new spelling ( Whats your position about . . . ) Dit thema staat te ver van de leefwereld van het kind (DL1-S0278)8 this theme stands too far from the world of the child ( is too remote from)


Such uses are clearly no longer postural or locational, as they concern ones ideological position on certain issues or simply the position of one ` entity vis-a-vis another. At the same time, the use of staan is well motivated here, as there still is a link with being in ones default position, particularly since it mostly conceptualises the located entities as being placed there. It cannot be denied, however, that the link that can be construed with ones default position is of variable strength in these uses, suggesting a gradient of metaphorisation. For example, while all uses in the example above are metaphorical, there is arguably a cline ranging from (a) (least metaphorical) to (c) (most metaphorical). The second metaphorical extension that plays an important role in both the L1 and L2 data is that of written text, which in Dutch is invariably coded with staan, as illustrated in the following examples:

8. References to corpus examples such as this one consist of 3 parts: (1) DL1 or DL2 identifying it as taken from the Dutch L1 or L2 corpus respectively, (2) the letters S, Z, L referring to resp. staan, zitten, and liggen, and (3) a number identifying the sentence in question.

322 (4)

M. Lemmens and J. Perrez a. b. Wat staat er op deze pagina? what stands there on this page? Sommigen staan op een wachtlijst. (metonymy: NAME / PEOPLE) some (people) stand on a waiting list

The motivation behind this use is probably no longer transparent even to native speakers; nevertheless, two converging factors can still be attributed some motivating force (and this regardless of whether they are etymologically accurate). First, there is the image of text standing on the supporting paper, as if in relief. In order to be readable, letters must be placed on their at side, which thus becomes their base. The mental scanning vector is thus from the paper upwards to the top surface of the printed letter. Second, letters can be seen as standing on (visible or invisible) horizontal lines on the paper. Hence, you write on the line, the letters have a height. The mental scanning vector is thus dierent, perpendicular to the previous one, going from the bottom of the line to the smallest top of the letter. We thus disagree with Serra Borneto (1996) who analyses similar uses of German stehen (stand) as resulting from the metaphor written text as vertical ordering; we do agree with him, however, when he says that the gurative extension, which started from a perceptual image, has established itself in the conventional knowledge of the speakers and is now active, independently from the original spatial image (1996: 477). Whatever its motivation, it is clear that this usage has become highly entrenched to the extent that it has laid the basis for extensions to all kinds of imprints (including non-textual ones), of either temporal or permanent nature, such as pictures in a book, text or icons on a screen, marks on the body. All of these can, and often must, be coded with staan. Within the prototypically structured radial network encoded by staan, this usage could be characterized as a local prototype from which new uses extend. The discussion in section 3.2 below will consider the importance of this local prototype for the L2 data. 1.3.2. Liggen. The following is an overview of the most important uses of liggen that will be briey discussed here: (i) be on ones sides (human posture) not be on base with horizontal orientation (inanimate entities) not be on ones base (regardless of orientation) location of dimension-less entities geotopographical location (cities, buildings, etc.) location of abstract entities

(ii) (iii) (iv)

On the use of posture verbs


Horizontality is much more important for liggen than verticality is for staan. This horizontality manifests itself in dierent types. Two large categories of horizontal objects can be distinguished, line types and sheet types, which are maximally distinct in their prototypes but share a transitional zone (small boards, for example, are conceivable as wide lines yet also as small elongated sheets). Within the sheet category are also included dierent kinds of tissues (e.g., clothes, towels, etc.) and substances (e.g., liquids, sand, etc.), since they are non-rigid objects that naturally take a horizontal expansion under their own gravitational weight. The dierence between Het zout ligt op tafel and Het zout staat op tafel (The salt stands/lies on the table) is thus metonymical: in the rst case, liggen refers to the salt as substance which, unconstrained by any xed boundaries, will atten out on the table; in the second case, staan shifts the focus from the substance itself to the saltshaker (itself left implicit however), posited on its base, and thus in a standing position. One of the particularities of Dutch (but something one nds in other languages as well) is that it has conventionalized the verb liggen to encode the location of symmetrical entities (balls, cubes, wads, etc.). These can be characterized by a lack of dimensional salience as Serra Borneto (1996) correctly observes for German liegen, perfectly similar to Dutch in this context. He points out how in the absence of dimensional dierentiation there is no mental tracing away from the origin that one has with vertical objects or objects resting on their base. The dimension-less use of liggen motivates a number of metaphorical extensions concerning the location of abstract entities. We are not referring here to the cases where these abstract issues are saliently associated with a particular horizontal form, as may be the case for example with frontiers conceived as lines, or foundations as horizontal supports. The abstract uses that we are concerned with here are those entities that seem to lack such imagery, as for example in De verantwoordelijkheid ligt bij jou The responsibility lies with you. We will not go into detail as to what motivates this extension (see Lemmens 2006); for our present purposes it suces to point out the entrenchment of liggen as the usual encoding for abstract entities. Another particularly well-entrenched usage of liggen in Dutch is that of geotopographical location as Serra Borneto (1996) has called it. This concerns cases where buildings, cities, and the like are located geographically. Even when standing right in front of a quite saliently vertical building, like a church, that typically is thought of as standing (resting on its base), we can still felicitously say, e.g., De kerk lag pal voor ons the church lay right in front of us; in that case, we would obviously not be talking about it as a building, but about its geographical location. As we


M. Lemmens and J. Perrez

will detail below, some interesting (erroneous) patterns for this usage emerge from the L2-data. 1.3.3. Zitten. follows: (i) The most important uses of zitten can be summarized as

(ii) (iii)

be in a sitting posture (considerable postural variation) default posture of small animals default posture of insects (close) containment (locational usage) (close) contact (locational usage)9

Strikingly, zitten shows considerably more variety in the postural domain than do liggen and staan, as it is used for a diversity of positions: (i) resting on the buttocks like on a chair (prototype posture for zitten), or (ii) with the legs crossed (yoga-position), or (iii) with legs stretched out; (iv) a squatting position; (v) on all fours; (vi) on hands and knees; or (vii) on ones knees. Interestingly, some of the extended uses of zitten can be explained from these postural variations. For instance, in a squatting position, the lower legs are bent, the body is close to the ground and often, there is an additional support with our hands on the ground. This postural conguration motivates the use of zitten to express the default position of lower animals such as rabbits, mice, frogs, etc. who usually are not said to stand. In the domain of animal postures, zitten has even gone further in that it is also the default verb for insects that, just as frogs and mice etc., only have a dual postural opposition zitten-liggen (the latter being used, for example, when they are dead). In the absence of postural variation, not much of the notion of posture is probably retained in these uses. The postureless nature of these uses, in combination with the postural variety sketched above, may explain the verbs noncommitment to posture and its productive extensions to other postureless uses. The most important one immediately relevant to the L2-data, is what we conveniently label containment-zitten. In the case of containment-zitten the verb no longer encodes posture but merely situates the entity as (closely) contained by a container. Hence the use of zitten to refer to water in a bottle, money in your pocket, a key in the keyhole, dust in your hair, a CD in a CD-case, etc., but also people sitting in prison or in a hotel room, etc. When used with inanimate enti-

9. While this extension is quite important for zitten, where the verb expresses close contact, as for example in Er zit geen deurknop aan deze deur there sits no doorknob on this door, it turned out to be irrelevant to this study and will thus be ignored.

On the use of posture verbs


ties, the contexts usually concern close containment or cases where the position of the contained entity depends on that of the container. As can be expected, zitten is also often used when metaphorical containment is at issue, such as suspense sitting in a race, or a bug sitting in a computer system, or the meaning sitting in a word or text. As we will show, the latter will be of particular interest for our L2-data. The productivity of containment-zitten is clearly illustrated by the L1-corpus used in our study: 64% of the cases refer to containment.10 As has become clear from the above discussion, the three cardinal posture verbs zitten, staan, and liggen have become basic location verbs in Dutch. While many uses of their extensive semantic networks have not been discussed here, the above summary has revealed the basic semantic mechanisms that underlie their most important uses. Considering their locational uses, and particularly the variations that may exist (such as a building said to lie or stand, or salt on the table as lying or standing, or butter lying, sitting, or standing in the fridge), we could say that the Dutch posture verbs actually function as noun classiers, just as noun suxes may do in more exotic languages, specifying that the noun in question refers to an entity that is liquid, oblong-shaped, pointed, rigid, sand-like, sticky, tubular, etc. Clearly, the Dutch categories are less rened than in many of these languages, yet the parallel with how the Dutch posture verbs indeed categorize the located entities cannot be denied. Interestingly, Gullbergs analysis of gestures conrms this idea, showing that Dutch speakers are signicantly more likely to incorporate gure object information in their gestures than are French speakers (Gullberg to appear; see also Gullberg and Narashimhan, this volume). Using these posture verbs in an idiomatically correct way is quite hard to master for French-speaking learners of Dutch. The study discussed here is a rst attempt at clarifying these diculties in more detail and preparing the ground for further research. Before turning to the actual quantitative and qualitative analysis, it is appropriate to say a few words about the corpora used (L1 and L2) and how we analysed them. 1.4. Corpus and corpus analysis

This study is based on two corpora: a learner corpus (DL2) and a control corpus (DL1). The learner corpus is a selection from the Leerdercorpus Nederlands (Learner corpus Dutch; see Perrez and Degand, in prep.).

10. This percentage lines up nicely with another corpus-based study (Lemmens 2002), where some 50% of the 4,311 sentences with zitten referred to containment.


M. Lemmens and J. Perrez

This corpus is a collection of texts written by learners of Dutch from different L1-backgrounds (French, German, Polish, Indonesian and Hungarian). Our French selection is drawn from (i) a series of argumentative essays written by French-speaking learners of Dutch studying Dutch as a main option and (ii) writing tasks performed by French learners of Dutch in the context of the CNaVT-exam11. The latter texts show a greater diversity, ranging from essays, summaries and reports, to letters and e-mails. For each text, some meta-information has been recorded concerning the author (mother tongue, study level and orientation) and the text itself (type of text, year, CNaVT-prole). In total, the French DL2subcorpus contains 1,247 texts amounting to 323,921 words. The control corpus (DL1) is composed of a range of argumentative essays written by native speakers of Belgian Dutch in the framework of a writing prociency class (rst year university students, Ghent University). The size of the corpus is admittedly rather limited (approximately 52,000 words) but its primary interest lies in its argumentative nature which matches quite well the main type of texts in the learner corpus. Our study of posture verbs on the basis of these corpora is not without limitations. Firstly, argumentative texts are not really representative of the contexts in which posture verbs typically occur, which may result in a limited number of attestations. The planned follow-up studies on spoken data will surely overcome this limitation. Secondly, as indicated before, it is essentially restricted to Belgian Dutch; some of the uses mentioned here may not be common in Netherlandic Dutch. Thirdly, for written corpora it is not always possible to reconstruct the contexts in which they have been produced, which makes it occasionally dicult for the researcher to interpret the learners intentions in having used a given posture verb. Finally, the absence of an objectively determined indication of the individual levels of prociency did not allow a reliable investigation into the evolution over the dierent levels. Despite these limitations, our study has revealed relevant tendencies concerning the use of posture verbs by French-speaking learners of Dutch. In line with the above analysis of staan, liggen and zitten, we coded, for both the L1 and L2 corpus, the dierent semantic categories that these verbs are used in. This has been done at two levels of detail. At the highest level, a distinction was made between postural, locational and metaphorical uses of the verbs. In addition to these three categories, we distin-

11. The Certicaat Nederlands als Vreemde Taal (CNaVT ) is an internationally recognized certicate for students of Dutch comparable to the Cambridge Certicate in Advanced English.

On the use of posture verbs


guished two other categories at the highest level, viz. the use of the posture verb (i) as root of a particle verb construction and (ii) as part of an idiomatic expression. The former category refers to cases where one of the posture verbs is combined with a particle (such as opstaan get up (from bed) or toestaan allow), whereas the latter category refers to xed collocational uses (such as bekend staan be famous or onder stress staan be under stress) as well as cases where a posture verb is used as part of a xed multiword unit (in combination with a preposition, an adverb, an adjective and/or a noun) whose global meaning cannot be derived from the meaning of its components in isolation.12 Examples are voor de hand liggen (lie before the hand be evident), op eigen benen staan (stand on own legs be independent), or het niet meer zien zitten (not see it sit any longer not able to see ones way out of a situation). The motivation for separating these two categories is, rstly, that the meaning of the verb in these constructions is often quite remote from its postural or locational meaning. Secondly, French-speaking learners of Dutch, when using such constructions, arguably do not really intend to use a posture verb, but rather directly translate a French construction (Il est evident que . . . , It is clear that) into a Dutch counterpart that simply happens to be built with a posture verb (Het staat vast dat . . . it stands xed ( is) clear that). In other words, the use by the learners of staan in vaststaan does not say anything about their ability to use the posture verb properly, but would rather be part of a unit-based learning strategy. At a more rened level, additional codes were used to further specify the use of the posture verb within the larger categories described above. This is particularly relevant to the locational and metaphorical uses of the verbs. Such further specication allows us to determine what type of location the posture verb refers to (e.g., geotopographical location, documents on a desk, etc. for liggen or containment for zitten) or what type of metaphorical use was at issue (e.g., abstract entities or scales for liggen; canonical position or written text for staan; containment, or stuckness for zitten). Some of these labels probably deserve some further comments. For instance, the label containment for zitten has been applied to locational as well as metaphorical uses. The dierence between these two lies

12. In order to avoid a subjective labeling, we have used as a reference the Van Dale Groot Woordenboek Nederlands-Frans bilingual dictionary to establish whether a given structure should be considered as an idiomatic expression or not. While obviously the dictionary cannot be taken as a awless norm, we believe that the choice is further justied by the observation that this is the resource that French L2 learners are most likely to turn to. Only for a handful of cases, where it clearly concerned an omission in the dictionary, was the label a decision of the authors.


M. Lemmens and J. Perrez

in the nature of the container, which is concrete in the former, example (5) but abstract in the latter ones, example (6).13 (5) locational:containment Ik zit namelijk op kot en heb geen kabelaansluiting in huis. (DL2-Z0042) I sit ( live) in a student room and dont have any cable connection there. metaphor:containment Na het treinongeval en de reis beseft hij waar hij in zijn leven zit. (DL2-Z-0053) after the train accident and the journey, he realises where he sits ( is, stands) in his life


The further specications for staan also deserve some comments. In the description of the locational use of staan, we specied whether this location was canonical, referring to contexts where the location is concurrently related to the default position of a human being or of an object (resting on its base) as in example (7). (7) Dus de producten zullen op een logische plek in de winkel staan en het zal gemakkelijker voor de klant zijn. (DL2-S-0040) thus the products will stand in a logic location in the store and it will be easier for the customer.

The label canonical has also been used to qualify some of the metaphorical uses of the verb. In these cases, it refers to metaphorical extensions of staan that can still be related to its postural or locational meaning (see example (3b) above). The other metaphorical uses concern cases where staan could not directly be linked with its postural or locational meaning (see example (3c) above). The coding scheme as described here has been applied to all of the sentences in the DL1-corpus and all correct uses in the DL2-corpus. The cases where the L2-users did not use the posture verb correctly were set apart at the highest level via the label error, so as to ensure that statistics on the distribution of usage only applied to correct cases. On a more rened level, we then coded a specication of the dierent types of errors made by the French-speaking learners of Dutch. These dierent types of

13. The examples from the DL2 corpus are reproduced in their original form; it may thus be that there are other language errors than the ones that we are interested in (such as the infelicitous use of in huis in this example). These errors will be ignored here.

On the use of posture verbs


errors will be discussed extensively in the subsequent sections. Before we turn to them, some important methodological observations have to be made concerning the corpus analysis as described here. Firstly, it should be clear that the coding scheme described above has its methodological limitations. The codes are set up as heuristic tools to allow qualitative and quantitative analysis, but the resulting categorization should not be taken as reecting a nal and xed map of the meanings of these verbs. For one thing, it may in some cases not always be easy to distinguish between certain categories, for example, with contexts where both a postural and locational reading could be entertained. The absence of clear-cut distinctions is also nicely illustrated by the degrees of metaphorisation illustrated above. Secondly, while we strongly prefer corpus-based analysis over intuitionbased analysis, the latter cannot be excluded when evaluating L2productions. To assess our data as objectively as possible, all fragments have rst been analyzed by both authors separately, for which, as became apparent in the subsequent comparison, there was a high degree of agreement. Problematic cases were further discussed and submitted to the judgment of minimally two other native speakers of Belgian Dutch.

2. 2.1.

Quantitative analysis of L2 data Overall frequencies

In total, 557 sentences have been extracted in which one of the three posture verbs occurred (407 fragments from the learner corpus and 150 from the control corpus). A rst general observation is that staan is the most frequently used posture verb in both corpora. In the control corpus, it is followed by liggen and zitten respectively, whereas in the learner data, zitten is slightly more frequent than liggen (cf. Table 1)14. Further comparison between the L1 and L2 data show that, in line with the typological dierences between French and Dutch, the learners globally tend to underuse the posture verbs in their L2 productions (62.85 vs. 143.95

14. The DL1 frequency distribution (staan > liggen > zitten) does not exactly replicate the one given in Lemmens (2002) (staan > zitten > liggen), in which zitten even appeared to be almost as frequent as staan. This can probably be explained by the dierent types of texts analysed in the latter study, i.e., newspaper articles versus argumentative essays in the present study.


M. Lemmens and J. Perrez

Table 1. Distribution of liggen, staan and zitten in the learner and control corpora Verbs Control 52,056 words Occ. liggen staan zitten Total 55 73 22 150 Freq./50,000 52.8 70.05 21.1 143.95 Occ. 88 209 110 407 Learner 323,921 words Freq./50,000 13.6 32.25 17 62.85

Figure 1. Distribution of the posture verbs across their categories of use

occurrences per 50,000 words).15 This is more outspoken for liggen and staan than for zitten.16 A more detailed analysis of how the posture verbs are used by the L1speakers and the learners respectively points to some interesting tendencies, as illustrated by Fig. 1. A rst observation is that the learners appear

15. Given the unequal size of the learner and control corpora, the frequencies in Table 1 have been normalized to 50,000 words, the greatest common decimal factor; such a procedure is not uncommon in corpus linguistic studies for frequencies in large data sets (see, e.g., Newman and Rice 2004). In the smaller data sets (e.g., Tables 2 and 3) standard percentages are used. 16. The tendency of Francophone L2 speakers of Dutch to underuse posture verbs and overuse of location verbs (the latter issue is not considered here) is conrmed by a pilot study (Lemmens 2001) comparing picture descriptions: native speakers of Belgian Dutch used posture verbs in 58% of the locational phrases, whereas the French learners only used them in 19% of the cases and resorted to a location verb in 63% of the cases (p < 0.0005).

On the use of posture verbs


to use posture verbs inappropriately in about 11% of the cases; these will be discussed more extensively in sections 2.4 and 3. While we have labelled them as errors which, from a L1 perspective they are, they may be quite motivated (and thus, in a sense correct) within the learner language (see Section 3). A second observation which can be derived from Fig. 1 is that the learners tend to use the posture verbs more frequently in postural and locational contexts (respectively 15% and 17%) than the L1 speakers (respectively 4.7% of postural contexts and 10% of locational).17 Narrowing down these results over the three verbs individually (see Table 2), we see that the tendency towards a greater postural use by the learners especially holds for zitten (42.7 % of the cases vs. 13.6% in the control corpus) and to a lesser extent for staan (7.2% of the cases vs. 4.1 % in the control corpus).18 These cases refer to sentences where the sitting or standing position is prominent. As far as the locational contexts in the learner productions are concerned, they appear to be the most frequent for liggen (38.6% of the cases vs. 7.3 in the L1 corpus), and zitten (23.6% of the cases vs. 9.1% in the control corpus). The latter observation does not apply to staan, however, which is more frequently used in locational contexts by the native speakers (13.7% of the cases vs. 6.2% in the learner corpus). Considering the 34 locational uses encoded by liggen in the learner productions, one observes that in a huge majority of the cases (73.5%), they refer to sentences expressing a geotopographical location. Other examples of this locational use of liggen concern papers (14.7% of the cases) or books (8.8% of the cases) lying on a desk. Locational zitten in the L2 data almost exclusively concerns sentences clearly expressing the notion of containment, as in example (8), or to borderline cases taking an intermediate position between posture and location, as in (9). (8) Aan het begin, zit [ . . . ] de hoofdguur alleen thuis als de telefoon rinkelt. (DL2-Z-0017)

17. One may wonder (as did one of the reviewers) whether the dierences in frequency may not be due to the dierent topics of argumentative essays in the two corpora. As will be recalled, the learner corpus does indeed contain a larger variety of topics and text types (argumentative essays, summaries, emails and business letters), but none of the topics in either corpus are such that they are biased towards postural or locational uses. The variety in topics does not, in other words, invalidate the major claims made in this paper, such as the overall underuse of the posture verbs by L2-speakers as well as their overusing these verbs in some contexts. Furthermore, the dierent text types do not affect the qualitative analysis presented in Section 3. 18. The percentages in Table 2 represent the uses per verb; for example, of the 55 attestations for liggen in the control corpus (see Table 1), only 1 is postural (1.8%).


M. Lemmens and J. Perrez

Table 2. Distribution of staan, liggen and zitten across the L1 and L2 corpora Control Occ. Postural liggen staan zitten Locational liggen staan zitten Metaphorical liggen staan zitten 7 1 3 3 16 4 10 2 69 30 26 13 % 4.7 1.8 4.1 13.6 10.7 7.3 13.7 9.1 46 54.5 35.6 59.1 Occ. 63 1 15 47 73 34 13 26 123 24 75 24 Learners % 15.5 1.1 7.2 42.7 17.9 38.6 6.2 23.6 30 27.3 35.9 21.8


at the beginning, [ . . . ] the main character is sitting home alone as the phone rings Maar roken is niet alleen slecht de mens die rookt, maar ook voor de mensen die erbij zitten. (DL2-Z-0082) but smoking is not only bad for the smoker, but also for the people sitting with him

In sum, the learners tend to use posture verbs, especially liggen and zitten, to a greater extent than the native speakers in more prototypical contexts denoting posture or location. Conversely, the native speakers use the posture verbs more frequently in metaphorical contexts than the learners (46% of the cases vs. 30% in the learner productions). This, again, is more striking with liggen and zitten, showing a far greater proportion of metaphorical uses in the productions of the native speakers than in the learners essays (54.5 % vs. 27.3 % for liggen; 59.1% vs. 21.8% for zitten). This tendency does not hold for staan, whose metaphorical usage is comparable in both corpora (35.9% of the cases in the learner corpus vs. 35.6% in the control corpus). When liggen is used metaphorically by the native speakers, it almost exclusively appears in contexts where it co-occurs with an abstract entity, such as problems, solutions, causes, etc. (93.3% of the cases). The remaining metaphorical uses of liggen include contexts where it refers to a conceptualized scalar entity (6.7% of the cases). Al-

On the use of posture verbs


though metaphorical liggen is less frequent in the L2 data, its use by the learners seems to be similar to the native usage, appearing in the rst place in combination with abstract entities (75% of the cases), and in the second place in contexts where it refers to a scale (25% of the cases). Of the 13 occurrences of metaphorical zitten in the productions of the L1-speakers, 12 refer to some abstract notion of containment (92.3%), whereas in the remaining example zitten is used as part of a progressive construction. Quite surprisingly, despite its lower frequency in the L2 productions, metaphorical zitten is used in a greater variety of ways by the learners. In addition to a majority of containment uses (66.6% of the cases) and two progressive uses (8.3% of the cases), zitten also appears in contexts where it encodes the notion of possession (8.3% of the cases, see (10)) and in contexts where it refers to the idea of being stuck (12.5% of the cases, see (11)). While this discrepancy between the learner and native usage of metaphorical zitten could probably be explained in terms of the relative size and scope of the control corpus, the observed variety of uses of metaphorical zitten among the learners tend to suggest that some learners correctly manage such very specic metaphorical uses of zitten. (10) (11) [ . . . ] ik zit met een klein probleempje (DL2-Z-0043) [ . . . ] I sit with ( have) a small problem. De kans bestaat dat de ambitie van de werknemer te hoog of te laag zit. (DL2-Z-0030) the chance exists that the ambition of the employee sits ( is) too high or too low.

Finally, even though metaphorical staan is as frequent in L2 as in L1, the native speakers and the learners use it in quite dierent ways. First of all, the native speakers use it more frequently in canonical position contexts, i.e., where the metaphorical use of staan can easily be linked up with its postural or locational meaning. This use of metaphorical staan accounts for 53.8% of the cases in the L1 corpus. A second context in which metaphorical staan is commonly used in the L1 corpus (34.6%), is related to the notion of written text as standing entity. Third, 11.5% of the cases in the control corpus concern metaphorical extensions without a clear link to postural or locational uses. The distribution of metaphorical uses of staan in the L2-corpus is slightly dierent: the notion of written text as standing entity accounts for 73% of the cases, followed by the uses referring to canonical position (21.3%) and other metaphorical ones (4%). The dierences internal to the group of metaphorical extensions suggest that L2 speakers do have control of the standing text pattern (a point to which we shall return in 3.2), but their overall semantic map of


M. Lemmens and J. Perrez

the metaphorical extensions clearly diers from that of the L1 speakers, as could be expected. The L2 speakers probably have not yet mastered the semantic motivations linking up the dierent uses, as shown by the low frequency of canonical and other metaphorical uses (i.e., other than the standing text pattern). To conclude the discussion of the overall distributional tendencies, some observations can be made concerning the use of the posture verbs as part of a particle verb and as part of an idiomatic expression. While the former cases are rather limited (4.7% in L1 and 3.4% in L2)all but one example being constructed with staan (vaststaan it is clear that, openstaan be open to, toestaan allow to, etc. but klaarliggen be ready), the group of idiomatic expressions deserves a short discussion. The occurrence of posture verbs as part of an idiomatic expression occurs more frequently in the L1 than in the L2 productions (33.3% of the cases vs. 21.6%). This holds for staan (37% vs. 29.7%), liggen (34.5% vs. 22.7%) and zitten (18.2% vs. 5.5%), even though in both corpora such uses appear more regularly with staan and liggen. Frequent expressions with liggen in the L1 data are voor de hand liggen lie near the hand ( be evident) (31.6%), aan de basis liggen lie at the basis of (21%) and in iemands handen liggen lie in s.o.s hands (10.5%). In the learner data, the most frequent examples are voor de hand liggen, (40% of the cases), ergens aan ten grondslag liggen lie at the basis of (25%) and iemand na aan het hart liggen lie near to s.o.s heart ( be very dear to someone) (15%). As far as zitten is concerned, its idiomatic uses are quite limited. Both the native speakers and the learners use it in hoe zit het met . . . ? how sits it with ( what about . . . ?) (50% of the cases in both L1 and L2 ) and iets zien zitten regard feasible (50% in the L1 corpus vs. 16.6% in the L2 corpus). In addition, the learners also use zitten in iemand in het haar zitten annoy someone (lit. sit s.o. in their hair) (33.3% of the cases). Finally, idiomatic uses of staan, in opposition to zitten and liggen, show a great diversity of examples among the native speakers (27 occurrences distributed across 17 dierent expressions; 0.63 Type/Token ratio), as well as among the learners productions (62 occurrences distributed across 24 dierent expressions; 0.39 Type/Token ratio). The occurrences of these expressions seem to be quite equally distributed in the L1 corpus; more frequent examples including aan het hoofd staan van stand at the head of ( be in charge of ) (11%), centraal staan stand ( be) central (11%) and op eigen benen staan stand on your own legs ( be independent) (11%). On the other hand, in the L2 data, staan is extensively used as part of the expression centraal staan (29%), followed by ter beschikking staan stand ( be) at ones disposal (9.7%) and in contact staan stand ( be) in contact with (8%).

On the use of posture verbs


In sum, even though these expressions are somewhat less frequent in the L2 productions, overall these uses are correct, supporting the idea that these are learnt as xed units. 2.2. Quantitative error analysis

The second step in the quantitative analysis focuses on the dierent types of errors made by the French-speaking learners when they use staan, liggen and zitten; in Section 3 we will then consider some of these errors from a more qualitative perspective. As shown in Fig. 1 above, the learners use the posture verbs incorrectly in approximately 11% of the cases (46 sentences in total). All in all, this is a relatively good result, but this may be attributed to the fact that the corpus consists of written data only, where learners have more time for reection. It is expected that the error rate in spontaneous speech will be much higher. The highest proportion of errors occurs with staan (65.2%), followed by liggen (19.6%) and zitten (15.2%). The dierent types of errors which have been identied are summarized in Table 6. Recall that these errors all concern cases where a posture verb has been used incorrectly, either (i) because the wrong posture verb was chosen (posture verb confusion) or (ii) because a posture verb was not possible in the given context (posture verb panic). These are the two main categories in Table 6; they will be discussed in more detail below. The third group concerns a collection of miscellaneous cases (i) where it was not at all clear what the speaker was trying to say, (ii) where a posture verb was used instead of a phrasal verb (e.g., toestaan allow)19, or (iii) where a given construction has not been reproduced correctly (hence, constructional contamination). The neat subdivisions in Table 3 concern in reality a much more complicated interplay of factors, especially for the miscellaneous group. At the same time, the division allows us to identify the two main error patterns discussed below, i.e., posture verb confusion and posture verb panic. The subdivisions within these two groups represent an onomasiological perspective, as they identify the context to be encoded, and they do so via the verb that would have been used had the situation been coded correctly. For example, an error labelled staan:metaphor:text refers to a sentence expressing the idea of texts located on paper for which

19. These complex verbs probably contribute to the overall posture verb problem, but in line with the decision taken to treat these as a separate category for the correct sentences (see section 1.4 above) we have put these in a separate group here as well.


M. Lemmens and J. Perrez

Table 3. Types of errors in the learner corpus Context of error 1. posture verb confusion liggen-context liggen:metaphor:abstract entity liggen:locational:paper liggen:locational:geotopographical staan-context staan:metaphor:text staan:locational:canonical zitten-context zitten:metaphor:containment zitten:locational:containment zitten:progressive 2. posture verb panic existential verb neutral location copula 3. miscellaneous constructional contamination posture verb instead of particle verb unclear Total 3 4 3 46 6.5% 8.7% 6.5% 100% Occurrences 20 9 1 3 5 6 5 1 5 2 2 1 16 10 5 1 % 43.5% 19.6% 2.2% 6.5% 10.9% 13.1% 10.9% 2.2% 10.8% 4.3% 4.3% 2.2% 34.8% 21.7% 10.9% 2.2%

staan should have been used but for which the learner chose another posture verb. The following section provides a more detailed analysis of the most common patterns in these two error groups. 3. Qualitative analysis

The quantitative analysis above has revealed a number of tendencies that could be summarized as follows: i. ii. the posture verbs are largely underused in L2 productions; the dierent posture verbs are often confused;

On the use of posture verbs iii. iv.


staan is the most frequent verb in the incorrect sentences (30 / 46 or 65.2%); L2 speakers sometimes use posture verbs where a neutral verb is to be used.

Given the typological dierences between Dutch and French, observations (i) and (ii) do not really come as a surprise. The other two observations may not have been intuitively obvious, even if in retrospect they, too, are perhaps not so surprising after all. The high frequency of staan in the set of errors (regardless of the subdivision drawn up above) lies in line with the verb expressing the canonical position of humans (and entities on their base or in their optimal or functional position). Looking at this from the learners perspective then, this can be phrased as follows: when learners have identied a context as a posture verb context, they will most likely choose staan as the default posture verb (especially when they have no idea which posture verb to use). Notice that this corresponds nicely with frequency of exposure: the fact that staan refers to canonical posture also makes it the most frequent posture verb in Dutch (cf. Table 1 and the results in Lemmens 2002, 2005b). While it is often said in L2-pedagogy that what you put in, is not what you get out, the preference for staan as the default verb seems to indicate that L2 speakers do pick up dominant patterns in the target language without being explicitly told. (As a rule, pedagogical grammars do not mention frequency and/or prototypicality.) Finally, there is the somewhat surprising observation that L2 speakers use posture verbs where Dutch does not allow them. This can certainly in part be attributed to what we conveniently call a general posture verb panic, that incites L2 speakers to simply replace any form of locational or existential zijn be with a posture verb (a form of hypercorrection); nevertheless, there is still some semantic logic in their behaviour, not unlike that exploited by native speakers, as will be detailed in section 3.1 below. Sections 3.2 and 3.3 will look at the cases where the wrong posture verb is chosen (group 1 in Table 3 above). A number of these errors actually centre around certain well-entrenched substructures (local prototypes); in these cases, the link with the postural prototype may no longer be transparent, but this well-entrenched usage motivates new extensions. There are two such structures that we consider here, viz. text as a standing entity (section 3.2) and geotopographical location (section 3.3). 3.1. Overuse of posture verbs

Given the strong obligation for using a posture verb in Dutch when one wants to express the location of an entity, L2 learners will most likely


M. Lemmens and J. Perrez

realise the importance of using these verbs at a relatively early stage in their learning process. The high number of metaphorical extensions of these verbs (in the L1-control corpus, about 46% of the cases) will undoubtedly add to the initial confusion and may lead to some kind of posture verb panic, inciting learners to use a posture verb in contexts (often metaphorical ones) where no such verb is allowed. Consider the following cases: (12) a. De vrouw *staat een beetje wanhopig omdat ze wilde dat haar man de tuintrap verft. (DL2-S-0205)20 the woman stands a bit desperate because she wanted her husband to paint the gardensteps Geachte Vrouw, Hier *zit de resultaten van mijn verslag. (DL2Z-0059) dear woman (sic), here sits the result of my report


In both cases, the use of a posture verb is inappropriate; the verb zijn be has to be used. The grammaticalisation of posture verbs has not gone that far (yet) that a pure copular use (X BE ADJ) as in (12a) is generally possible, even if there are cases that come quite close (e.g., het huis staat leeg the house stands empty). Yet even for the latter, a certain locative colour remains, whereas in the example here this is not the case. If a locative complement had been added or even a te V complement (expressing a progressive), staan would actually have been quite possible: zij staat er wanhopig bij she stands there desperately PREP or zij staat wanhopig tegen haar man te roepen she stands desperately to her husband to yell ( is yelling at). For example (12b) on the other hand, the locative hier here (expressing something like enclosed with this letter, herewith) is not locative enough to sanction a posture verb. The use of zitten may have been triggered by the idea of the report being attached to the letter; French joindre join (expressing ATTACHMENT) often takes zitten as the Dutch equivalent. In some cases, the error may be attributed to a confusion of dierent idiomatic constructions: (13) b. In de eerste tekst zoekt men als er een verband *staat tussen de witte massa (DL2-S-0094) in the rst text they (try to) nd whether there stands a connection between the white matter

20. For ease of identication, the verb errors in the cited learner examples have been marked with a *; as said before, other mistakes that may occur in the sentences have not been corrected nor have they been marked.

On the use of posture verbs


The two correct expressions, quite similar to each other, are either X staat in verband met Y (X stands in connection with Y) or Er is een verband tussen X en Y (there is a connection between X and Y); this seems to be a clear case of constructional cross-contamination. However, the latter example (as a handful of others) may also be due to a (phonological) confusion of staan and bestaan (exist). While bestaan is etymologically related to staan, this is no longer obvious (even native speakers are probably not aware of this) and the verb is often interchangeable with existential zijn. However, one of the L2-errors in our data illustrates that the interchangeability does not always hold: (14) De kranten zijn meestal goed maar ik vind dat er ook een nadeel *staat . . . Er zijn bladen die de waarheid niet precies vertellen (DL2S-0130) the newspapers are usually good but I think that there stands also a disadvantage . . . There are papers that do not tell the exact truth

Supposing that the L2-speaker confused staan with bestaan, then this would still yield a coding that at best is highly marked, since the verb zijn is the most appropriate alternative. This intuition is conrmed by a Google search on er zijn/bestaan nadelen there are/exist disadvantages, yielding 17,500 vs. 3 hits respectively. The reason why bestaan is disfavoured is that it too strongly focuses on the idea of existence, whereas this seems uncalled for in the present context. While the dierences between bestaan and zijn appear motivated, a full explanation of these goes beyond the scope of the present paper.21 Let us now look at two important subcases where the wrong posture verb is used, triggered by two well-entrenched uses, staan as used to refer to written text (3.2) and liggen as used to refer to geotopographical location (3.3). 3.2. Text as a standing entity

The L2 speakers seem to be suciently familiar with the Dutch convention of using staan to refer to written text, as it is used correctly in 55 occurrences, which amounts to 30.7% of their correct uses of staan. At the

21. It is to be expected that this dierence will be at least partially similar to that in English between be and exist. Notice that the latter, too, is derived from a Latin verb that referred to standing (ex- out, forth sistere cause to stand). Similarly, Spanish has estar be, which evolved from Latin stare stand, whose uses dier from those of the copula ser be. The evolution of estar actually lines in line with the claim that standing is the canonical position for humans.


M. Lemmens and J. Perrez

same time, there are a number of cases where this context leads to mistakes, either because staan is not being used or because staan is used incorrectly (overextension). Let us begin with the rst case; heres one of the L2-examples: (15) Zotte mensen *zit ook tussen aanhalingstekens omdat het een uitdrukking van het meisje is. (DL2-Z-0016) crazy people sits also between quotation marks because it is the expression of ( used by) the girl

The students choice for zitten is not without motivation, the word being closely contained by the quotation marks; however, talking about the graphemic representation of language renders a coding with staan absolutely compulsory. At the same time, the strong obligation can lead to (subtle) errors as well, as is the case for the following student, inappropriately overextending that use of staan: (16) Er ? *staat een bijbedoeling in de zin die op een verschillende manier genterpreteerd zal worden (DL2-S-0160) there stands a hidden intention in the sentence that will be interpreted in a dierent way

At rst glance, nothing seems wrong with this example, since the student is talking about the text that will be interpreted dierently. However, upon second thought, the formulation just does not seem fully idiomatic, since a hidden intention is not really orthographically expressed, unlike is the case with straightforward meanings, where this metonymy does apply, e.g., Hun ideeen staan in het werkboek their ideas stand in the workbook (Google example).22 A coding with zitten (expressing containment) would thus have been more idiomatic. Despite the fact that basically any text can be said to have a meaning sitting in it, there are certain conventionalised collocations, as becomes apparent from the following L2-error, where the learner is talking about the information in newspapers: (17) Maar als je [die kranten] eens koopt, ontdekt je dat daar niets in *zit (DL2-Z-0055) but when you buy [these newspapers], you discover that there sits nothing in them

Apparently, this learner has mastered the usage of zitten to refer to the meaning inside texts, yet a native speaker of Dutch would immediately

22. http:/ /, last accessed Jan. 12, 2009.

On the use of posture verbs


replace the verb with staan to render the sentence more idiomatic.23 The reason for this is that in newspaper articles, and generally all other types of non-ctional prose, meanings should be directly derivable from the very words themselves since these texts are not supposed to have hidden meanings. In Dutch postural logic, their meaning thus stands on the paper, black on white. It is obviously not impossible for these texts to have a deeper layer of meaning (implications, humour, sensitivity, etc.), yet such is typically not associated with them.24 Rather, these are the things one nds in ctional prose, poems or song lyrics. Notice, however, that if the meaning is suciently evident from the words/text itself, staan remains a preferred coding even in these types of texts. In sum, what the above L2-examples reveal is that the learners are aware of certain common extensions of staan (printed text) and zitten ((close) containment), as further illustrated by the higher frequency of theses uses (see in section 2.1) that are generally also quite frequent in the L1 data. At the same time, the learners may not have fully mastered some of the collocational subtleties of the target language, which themselves are semantically well-motivated (which, unfortunately, we cannot aord to elaborate on here). 3.3. Geotopographical location

One of the extensions that is common for liggen is to express geotopographical location, which concerns the location of entities that are typically conceived of as locations themselves, such as buildings, cities, villages, etc. This usage is quite frequent in the learner data: there are 29 occurrences in the L2-corpus, which amounts to 33% of the total attestations for liggen (88); only 2 of these are incorrect (cf. below). Conversely, there are 5 contexts of geotopographical location where the L2-learner uses staan instead of liggen, of which 4 are given in example (18) below (2 were by the same speaker, so only one of those is given).

23. Notice that this sentence would be appropriate if the speaker were referring to extra things that one may nd in a newspaper, such as loose advertisement brochures, a concert calendar or other loose sections/quires you can take out, free stickers of CDs enclosed with it, etc. Such physical containment is, however, not referred to in this particular context. 24. A simple Google search on in de krant zit yielded only 2 examples (as opposed to 8,160 for in de krant staat) referring to this context (and not to the one mentioned in the previous footnote) where the located entities were niet genoeg diepgang not enough depth and zoveel emotie so much emotion; they essentially conrm the tendencies described here.

342 (18)

M. Lemmens and J. Perrez a. b. c. de landen die vlak bij de zee *staan . . . (DL2-S-0012) the countries that stand close to the sea terwijl Gosselies . . . verder van Charleroi *staat (DL2-S-0114) while Gosselies stands further away from Charleroi25 Daar *staat een beautycenter met sauna, bubbelbad en massages. (DL2-S-0158) there stands a beauty centre with sauna, jacuzzi and massages De universiteit *staat in Luik en ik houd veel van deze stad (DL2-S-0200) ` the university stands in Liege and I love this city very much


For (18a) and (18b) there is absolutely no discussion: the location of land areas and cities must be coded with liggen. For (18c) and (18d), the situation is more complex, since the two buildings could be conceived of as standing (a vertical entity resting on its base). However, in the context at hand, such a focus on the building is rather infelicitous. Interestingly, one erroneous use of liggen in this context misses precisely on this point: (19)

Bauval [beweert] dat de piramiden van Gizeh op een bepaalde wijze *liggen in overeenstemming met het midden van het sterrenbeeld Orion. (DL2-L-0026) Bauval claims that the Gizeh pyramids lie in a certain way in accordance with the middle of the Orion constellation

Overall, this is a context of geotopographical location, and liggen is possibly acceptable; the reason why it has been marked as an error, is that the sentence talks about the pyramids being deliberately positioned in a certain way, which renders their canonical position (resting/put on their base) again salient, and the use of staan would have been more idiomatic. There is one particular case where a learner seems to overextend the geotopographical context to one that is not really one: (20) Voor een steeds betere dienstverlening zal de geldverdeler buiten het kantoor *liggen (DL2-L-0030) to provide an increasingly better service the cash dispenser will lie outside the bank oce

Usually, cash dispensers are built into the wall and do not occupy a land surface of their own, which makes the use of liggen very marked; rather a coding with a more general location verb such as zich bevinden be lo-

25. Gosselies and Charleroi are two cities in the French-speaking part of Belgium.

On the use of posture verbs


cated is preferred. One could say that the use of liggen here creates a Google Maps-eect which is inappropriate for entities not usually conceived of as locations having (x,y) coordinates. We can say that, at least judging on the data used for this study, that L2-learners have mastered this use of liggen fairly well, interferences probably being most common in the context of buildings and the like for which a coding variation liggen/staan is mostly possible, even if the context usually guides the language user to a clear preference.


Conclusions and prospects

Our pilot study of the use of the Dutch posture verbs staan, liggen and zitten by French-speaking learners has unravelled some interesting tendencies. In our quantitative analysis of the data, we rstly observed that the learners underuse these verbs in their productions as could be expected given the typological dierences between French and Dutch regarding the expression of posture and location. Considering the specic contexts in which the posture verbs have been used correctly by the learners, we have further shown that they use them far more frequently in postural and locational contexts, whereas the native speakers tend to use them more frequently in metaphorical contexts. This observation suggests that the learners are more inclined to use the posture verbs in their basic contexts, being less at ease with their metaphorical extensions. These distributions lend support to the idea that the coding exibility is a major diculty for the learners when faced with the wide range of extensions staan, liggen and zitten. Rening our analysis revealed, however, that the learners appear to master some specic patterns of the metaphorical uses of the posture verbs, such as zitten expressing containment or possession, staan referring to text as standing entity or liggen expressing the location of abstract entities. Our qualitative analysis even pointed out that the learners tend to overextend certain of these metaphorical patterns. Similarly, the qualitative analysis has revealed some other cases of overgeneralisation whereby the learners resorted to a posture verb in contexts in which a more general verb would have been more natural (posture verb panic). This overuse of certain patterns not only shows that the L2 user has probably mastered the logic of certain specic uses, but also that they are exploiting these insights to encode similar situations. In doing so, they will inevitably overgeneralise or ignore collocational patterns of the target language that L1 speakers have acquired through massive and repeated exposure to linguistic input. The absence of a L1 acquisition


M. Lemmens and J. Perrez

control corpus motions to extreme caution, yet on the basis of the L2 patterns discussed above (and ignoring issues of cognitive development and maturity), we are inclined to conclude that, in general terms, L2 acquisition strategies may exploit the same principles as what one observes for L1. Clearly, given the high frequency of posture verbs in Dutch and the problems these entail, the language learner may sometimes decide to play safe and blindly apply a posture verb in contexts where their native language might have guided them to using a location verb such as zijn (be) or zich bevinden (be found). However, looking at the errors in question reveals that L2 speakers do follow general overextension strategies that characterise L1 acquisition as well (cf. Brown 1958; Clark 2003: 211212; Tomasello 2003: 1278).26 In other words, these errors are not blind but reveal at least a partial insight into the linguistic system. While some hypercorrection cannot be excluded, their insight is clearly in line with the input data, as shown, for example, by the high frequency of staan in the errors or by the recurrent use of some expressions such as centraal staan (stand central). Considering the (recurrent) use of certain specic metaphorical patterns of the posture verbs has allowed us to evaluate to what extent the learners master the semantic network of the verb in question. In line with a usagebased approach, our claim, partly supported by the observation that the learners seem to have a good control of expressions with posture verbs, is that the learners, when assimilating a new pattern of use of a posture verb, might rather learn it as an separate unit and miss insights as to how the dierent nuances of a given posture verb relate to each other, preventing them to integrate the new pattern into the verbs semantic network. To put it another way, having mastered some specic metaphorical uses of a given posture verb does not mean that the learners master the whole semantic structure of the category. Some of the patterns that the L2-learner has to uncover may be relatively straightforward, such as use staan when coding the location of an entity on its base, or use liggen for a symmetrical object located in space, or use zitten when an entity is closely contained by another. Other cases, some of which were discussed here (but there are many others), remain problematic, as it may not always be clear from an (encoding) point of view whether a neutral verb is to be used or rather a posture verb and if so, which one. This is

26. As we have not yet looked at L1 acquisition of posture verbs, we are not claiming that the actual patterns of extension are necessarily the same; the suggestion is that the general extension mechanisms, as revealed through overextension, are exploited by both.

On the use of posture verbs


particularly true for certain collocational patterns for which the internal motivation may not be so easy to discover. In line with a usage-based view on language acquisition, it is expected that the L2 language learner may eventually unravel these via the same interplay of factors that the L1 language learner operates with, i.e., frequency of input, implicit and explicit negative input and statistical pre-emption (cf. Tomasello 2003; Goldberg 2006). Summing up then, our analysis appears to support two important interrelated claims. The rst one is that it is incorrect to consider the learner system as simply an imperfect version of the target language; rather, it is a linguistic system in its own right that follows a mixed logic: some of the errors are due to interferences from their native language (in our case, the underuse of posture verbs) yet others are due to overextensions of patterns they observe in the target language, as illustrated above. The second observation, which follows logically from the rst and which, moreover, has important pedagogical consequences, is that input plainly matters, also for L2 acquisition: L2 speakers do pick up dominant patterns in the target language without being explicitly told (cf. also Rast 2008) and they apply these creatively. The corpus-study reported on here is obviously but a rst (yet necessary) step to unravel the processes at work in L2 acquisition of Dutch posture verbs. Despite its limitations, mainly related to corpus size and the type of texts, our study has allowed us to discover some general patterns in the errors produced by the learners, which might have been more dicult to observe in a controlled experimental setting. This particularly concerns the metaphorical uses of the posture verbs. Further research is obviously warranted and will be pursued along two paths. Firstly, extending the existing contrastive research for L1 as described in Lemmens (2005a), we will carry out elicitation experiments where francophone L2 speakers describe the location of entities as given by a controlled set of illustrations and compare these (semi-spontaneous) narrations to those produced by native speakers. Secondly, we will do follow-up experiments probing into intuitions of L1 and L2 speakers concerning some of the onomasiological variations described above. It is to be expected that this research will conrm the tendencies outlined here and provide further insight into the L2 language system. Finally, in order to fully evaluate the suggestion that the L1 and L2 acquisition strategies for posture verbs are comparable, a more systematic analysis of the acquisition of these verbs in L1 is warranted. Received 1 March 2009 Revision received 25 November 2009 Universite Lille 3 & CNRS Facultes Universitaires Saint-Louis


M. Lemmens and J. Perrez

Ameka, Felix K. and Stephen. C. Levinson. 2007. The typology and semantics of locative predicates: posturals, positionals, and other beasts. Linguistics 45. 847871. Brown, Roger. 1958. How shall a thing be called? Psychological Review 65. 1421. Clark, Eve V. 2003. First language acquisition. Cambridge: Cambridge University Press. Fagan, Sarah. 1991. The Semantics of the positional predicates liegen/legen, sitzen/setzen, and stehen/stellen. Die Unterrichtpraxis 24.136145. ` Goldberg, Adele E. 2006. Constructions at work: The nature of generalization in language. Oxford: Oxford University Press. Guirardello-Damian, Raquel. 2002. The syntax and semantics of posture forms in Trumai. In Newman, John (ed.), The linguistics of sitting, standing and lying (Typological Studies in Language 51), 141177. Amsterdam and Philadelphia: John Benjamins. Gullberg, Marianne. To appear. Language-specic encoding of placement events in gestures. In Eric Pederson and Ju rgen Bohnemeyer (eds.), Event representations in language and cognition. Cambridge: Cambridge University Press. Gullberg, Marianne and Bhuvana Narashimhan. This volume. Gestures and the development of semantic distinctions in Dutch. Hiligsmann, Philippe. 1997. Lingustische aspecten en pedagogische implicaties van de tussen taal van Franstalige M.O.-leerders van het Nederlands [Linguistic aspects and pedagogical ` ` implications of the interlanguage of French M.O. learners of Dutch]. Geneve and Liege: ` ` Droz, Bibliotheque de la Faculte de Philosophie et Lettres de lUniversite de Liege. Klein. Wolfgang. 2008. From the learners point of view. Paper presented at the International Conference on Language acquisition: Comparative perspectives. Paris, 56 December. Klein, Wolfgang and Clive Perdue (eds.). 1993. Adult language acquisition: Cross-Linguistic perspectives. Cambridge: Cambridge University Press. Kutscher, Silvia and Eva Schultze-Berndt. 2007. Why a folder lies in the basket although it is not lying: the semantics and use of German positional verbs with inanimate Figures. Linguistics 45. 9831028. Lemmens, Maarten. 2001. LOCATION versus POSITION: Coding strategies for referent location Paper presented at the ICLC7, University of California, Santa Barbara, California, 2227 July. Lemmens, Maarten. 2002. The semantic network of Dutch posture verbs. In John Newman (ed.), The linguistics of sitting, standing and lying. (Typological Studies in Language 51), 103139. Amsterdam and Philadelphia: John Benjamins. ` Lemmens, Maarten. 2005a. Motion and location: toward a cognitive typology. In Genevieve Girard (ed.), Parcours linguistiques. Domaine anglais. (CIEREC Travaux 122), 223244. St. Etienne: Publications de lUniversite St Etienne. Lemens, Maarten. 2005b. Aspectual posture verb constructions in Dutch. Journal of Germanic Linguistics 17. 183217. Lemmens, Maarten. 2006. Caused posture: experiential patterns emerging from corpus research. In Anatol Stefanowitsch and Stephan Gries (eds), Corpora in cognitive linguistics: Corpus-based approaches to syntax and lexis. Berlin: Mouton de Gruyter, 261296. Lemmens, Maarten and Dan I. Slobin. 2008. Positie- en bewegingswerkwoorden in het Ne derlands, het Engels en het Frans. In Philippe Hiligsmann, Melanie Baelen, Anne-Lore Leloup and Laurent Rasier (eds), Verslagen en mededelingen van de Koninklijke Academie voor Nederlandse Taal- en Letterkunde 118, 1732. Newman, John. 2002. A cross-linguistic overview of the posture verbs sit, stand, and lie. In John Newman (ed.), The linguistics of sitting, standing and lying. (Typological Studies in Language 51), 124. Amsterdam and Philadelphia: John Benjamins.

On the use of posture verbs


Newman, John and Sally Rice. 2004. Patterns of usage for English SIT, STAND, and LIE: A cognitively inspired exploration in corpus linguistics. Cognitive Linguistics 15. 351396. Perrez, Julien and Liesbeth Degand. In preparation. Het Leerdercorpus Nederlands (Cahiers du Cental X). Presses universitaires de Louvain. Rast, Rebekah. 2008. Foreign language input: Initial processing. Clevedon: Multilingual Matters. Serra Borneto, Carlo. 1996. Liegen and stehen in German: A study in horizontality and verticality. In Eugene Casad (ed.), Cognitive linguistics in the Redwoods, 458505. Berlin: Mouton de Gruyter. Talmy, Leonard. 2000. Towards a Cognitive Semantics (Vol. 1 and 2). Cambridge, MA: MIT-press. Tomasello, Michael. 2003. Constructing a language. A usage-based theory of language acquisition. Cambridge, MA.: Harvard University Press. Van Oosten, Jeanne. 1984. Sitting, Standing and Lying in Dutch: A Cognitive Approach to the Distribution of the Verbs Zitten, Staan, and Liggen. In Jeanne van Oosten & Johan Snapper (eds.), Dutch linguistics at Berkeley, UCB. 137160. Vuillermet, Marine. 2008. Ese Ejja posture verbs do not just sit there: An inquiry into other ways they stand out. Paper presented at the Workshop for American Indigenous Languages, University of California, Santa Barbara, 2324 May.

Book reviews

Steen, Gerard J., Finding Metaphor in Grammar and Usage. A Methodological Analysis of Theory and Research. Series Converging Evidence in Communication and Language Research 10. Amsterdam/Philadelphia: John Benjamins, 2007, 430 pp. ISBN 978 90 272 3897 9. Hardbound EUR 110.00 / USD 165.00 Reviewed by Alice Deignan, University of Leeds, UK. Email: Steens detailed discussion of research into metaphor in language and thought is extremely timely in its tackling of methodological issues that have often been neglected in the excitement of rapid theoretical developments in the eld. Encompassing much of the vast literature on conceptual and linguistic metaphor produced since Lako and Johnsons groundbreaking Metaphors We Live By (1980), the author describes recent theoretical developments and research projects in depth, and relates them to each other through a framework carefully developed in the rst chapter of the book. The length of the book allows space for complex issues to be explored, and for detailed discussion of metaphor research; critical descriptions of single projects often run to a number of pages. The insightful and detailed commentary on a huge range of projects is one of the strengths of the book. The book is divided into three parts, the rst of which, Mapping the Field, sets out theoretical and methodological foundations, and includes a useful review of the currently dominant models of metaphor in language and thought. The second part, Finding metaphor in grammar, takes the reader in detail through the methodological processes involved in identifying metaphor in the language system, beginning with identifying linguistic candidates for metaphor and through the theoretical questions that need to be tackled. These concern domain identication, establishing mappings and considering the nature of processing performed by
Cognitive Linguistics 212 (2010), 349370 DOI 10.1515/COGL.2010.013 09365907/10/00210349 6 Walter de Gruyter


Book reviews Cognitive Linguistics 212 (2010)

language users. The third part, Finding metaphor in usage, describes the steps involved in identifying metaphor and related tropes in language in use. Each chapter in the second and third parts follows the same structure, which has been established in Part 1: rstly, descriptions of research exemplifying the issues at stake are presented, chosen from both synchronic and diachronic perspectives. Secondly, in a section entitled Conceptualisation, the theoretical problems are outlined, again with reference to a selection of studies, and then, in a section named Operationalisation, the methodological challenges of investigating these are explored. In what are usually the longest sections of each chapter, the author then shows how these issues can be tackled, through the three major methodological approaches reviewed in Part 1: introspection, observation and manipulation. Chapter 1 sets up three two-way distinctions in the way that metaphor can be considered: as grammar (language as system) or as usage (language in use, performance); as existing in language or in thought; and as symbol or as behaviour. Grammar as used here and in the books title, may be an initially misleading term for some linguists: here it includes lexicogrammar, and all aspects of language that are part of the socioculturally conventionalized and cognitively entrenched part of the many concrete events of usage that occur in reality (p. 5). Grammar is, as Steen points out, derived and abstracted from usage. In his discussion of the distinction between language and thought, Steen cites cross-linguistic metaphor studies to show how metaphor can vary between languages at the level of thought and at the level of language. He points out that in some studies it is dicult to tell whether claims are about language or about thought. The third distinction raises the problem of whether language is fundamentally symbol or behavior. Steen argues that this distinction has been conated by cognitive linguists, who tend to investigate language as if it were a symbolic system, but then treat their ndings as having reality for human processing, or behaviour. He argues that the two aspects need to be disentangled. Metaphor can be investigated through each of these three prisms, which intersect to lead to eight dierent kinds of question. Steen argues that it is important for researchers to clarify which kind of question they are tackling. He discusses the problems of using evidence from one area of research to investigate hypotheses in another; for example, using evidence from language to investigate thought, and shows how this can lead to circularity. This chapter clears a way through a number of issues that have often been muddled and is in itself a very useful contribution to the eld. Chapter 2 begins with a description of the well-known scene in the lm Mary Poppins, in which the characters become helpless with laughter

Book reviews Cognitive Linguistics 212 (2010)


and gradually oat upwards. Analysts see conceptual metaphors everywhere, and would tend to see this as a non-linguistic realisation of HAPPY IS UP; Steen asks whether they are right to do so. Caution about evidence is advised: evidence that is merely compatible with a hypothesis such as conceptual metaphor theory is not sucient to demonstrate that the hypothesis is true: an important problem for the theory has been its proponents lack of consideration of alternative explanations for the linguistic evidence. A number of developments in metaphor research are discussed as examples of deductive and inductive approaches, and as ways in which theory can be tested. The important message of the chapter is that researchers have often given evidence that is consistent with conceptual metaphor theory as evidence for it; this is not enough- converging evidence is needed. Chapter 3 tackles theoretical denitions of metaphor. Four approaches to metaphor are detailed: metaphor as cross domain mapping (the conceptual metaphor approach of Lako and Johnson); metaphor as involving many spaces (Blending); Glucksbergs view of metaphor as class inclusion, and the career of metaphor approach of Gentner and her co-researchers. These are very usefully overviewed and compared. Metonymy is also dened, using the notions of closeness and intimacy. Steen argues that any pair of domains can be both contiguous (metonymy) and similar (metaphor), and that degrees of each are therefore independent scales, rather than opposite ends of the same scale. In some cases, the relationship between domains will appear to be more strongly one of contiguity, and the resulting trope will be primarily metonymic. In other cases, the domains are similar, resulting in a trope that is primarily metaphorical. This argument allows for the same linguistic expression to be used more metaphorically at some times and more metonymically at others. Detailed examples from language in use would have helped to support this interesting argument. Steen notes that Lako and Johnson attacked the notion of metaphor as similarity; he examines their objections and modies a denition of similarity in this light. However, as he notes, establishing a linguistic or conceptual denition of metaphor in terms of similarity and comparison does not imply that language users process metaphors in this way. Steens rigorous insistence on the separation between process and product is essential here. The chapter then outlines some problems that need to be tackled: the demarcation of source and target domains, whether highly conventionalised expressions can be considered as metaphorical, and whether metaphors have to be processed online. In Chapter 4, Steen argues the need for tight conceptualisation. He develops criteria for metaphor identication with reference to the four


Book reviews Cognitive Linguistics 212 (2010)

theories and eight aspects of metaphor originally dened. He looks at how each of the theories would identify metaphor in usage, and goes on to consider the dierences in their positions. There is detailed analysis of examples to show how the choice of unit of analysis has an impact on eventual decisions about the source domain and nature of conceptual mapping. The pragglejaz procedure aims to identify metaphor in language, which, Steen argues, is only indirectly connected to metaphor in thought, psychological structures and processes in individuals. The pragglejaz procedure attempts to be atheoretical; Steen claims that although the term comparison is used, the procedure could be employed by researchers who reject the notion of comparison, such as class inclusion theorists, to identify metaphor in text, because it makes no underlying claims about the nature of thought. The question of metaphoricity being perceived dierently by dierent language users is mentioned, but this complex issue is not explored in detail. The dictionary and corpora are discussed as tools for metaphor analysis. Steen describes changes in the literal use of words such as ardent and fervent, though changes in the dictionary descriptions of such words may be due to changes in lexicographical practice over the last forty years, rather than, or as well as, etymological developments. Like much of the book, this chapter brings systematicity to the work on metaphor in dierent schools, oering some new perspectives and insights such as the proposed relationship between metaphor and metonymy. Chapter 5 tackles the thorny question of what counts as data in metaphor studies. It begins with David Lodges amusing send-up of a cognitive scientist trying to analyse his own thoughts in the contemporary novel Thinks, illustrating the problems of capturing and using introspective data. Metaphor studies use three kinds of data: verbal, non-verbal and meta. Verbal data, or the analysis of text, are the most familiar to linguists. Non-verbal data are typically behavioural data from language users in action (p. 105), often used in psycholinguistic studies, and consist of studies such as the analysis of eye movements, pausing, gesture, and reading times. Meta data dier from verbal and non-verbal data in that they are not evidence of performance by language users, but reections on performance, and include tests for ambiguity and polysemy. They are typically generated by experts, but in some cases, characterisations produced by non expert users are helpful as evidence in themselves or are used as a stage in the development of experimental materials. Steen describes the collections of these dierent kinds of data in a sound, thoughtful discussion, of great use for research students, as well as for experienced researchers. Data are then reclassied into introspective, observed and manipulated, a division which is used throughout the rest of the book. The

Book reviews Cognitive Linguistics 212 (2010)


advantages and disadvantages of using each of these collection techniques are discussed; these issues are returned to in more depth in later chapters. In this chapter, quantitative and qualitative data are also discussed, as is the use of statistical measures. This is the nal chapter of Part 1 of the book, setting up categories and systems for researching metaphor. Part 2 is concerned with applying these to nding metaphor in grammar. Chapter 6 looks at the rst of the eight potential research questions concerning metaphor that were developed in Chapter 1, seeking linguistic metaphors in grammar. It begins with a systematic critique of some of the metaphors proposed by Lako and Johnson (1980), reiterating the point that while the linguistic evidence presented is consistent with the proposed conceptual metaphors, this is not sucient. Steen shows that a more detailed analysis of the linguistic evidence for TIME IS MONEY, using dictionary data, considerably weakens the claims of Lako and Johnson. In his consideration of introspective evidence, Steen explores the issues of monosemy, polysemy and ambiguity, describing the traditional tests that have been developed to identify these. He argues that these tests are helpful for all but borderline cases- although it could be counter-argued that the borderline cases are exactly where tests are needed, central cases being easily resolved. The observation data that are described are mainly from corpora; the relatively recent access to very large corpora and fast search tools has given metaphor researchers a hugely powerful resource. Manipulation is less often used as a source for data for identifying linguistic metaphor in grammar; Steen explains why this is and reviews some of the few studies that attempt it. Chapter 7 explores the very dicult issue of what constitutes a domain; Steen points out that domains need to be separated from each other in a methodologically responsible way (p. 171), as a preliminary to identifying the relationship between two domains, and then deciding whether cross-domain mapping is involved, which is the topic of the following chapter. Traditionally, the central distinction that has been posited between domains is concrete-abstract, but Steen shows what will be familiar to metaphor researchers who have tried to operationalise this distinction: although at rst sight concrete and abstract would appear radically different, in practice domains seem to be ordered on a continuum from one to the other. Steen also describes the problem of connecting linguistic forms to domains. Further, the analyst needs to decide at what level to specify mappings; Lako and Johnsons domains appear to operate at an intermediate level, while Grady (1997) proposed analysing at a basic level. Steen considers in detail how decisions about domains impact on definitions of metaphor, and shows how this works in practice through discussion of dictionary data and problems such as how to deal with simile.


Book reviews Cognitive Linguistics 212 (2010)

In Chapter 8, Steen follows up his discussion of domain identication with an exploration of the relationships between domains and how these can be classied. Kovecses (2000) series of studies on emotion are dis cussed as examples of cross-domain mapping. It is stressed that it is vital to check that words really belong to the domains that are claimed for them, some previous research having been less than rigorous on this point. Steen shows how issues of identifying relationships between domains are dealt with- or, in a number of cases, ignored- in introspective work, in corpus studies, and in work in the psycholinguistic tradition. At this stage he reformulates the dierence between metaphor and metonymy, managing to avoid using the notion of domain at all, talking instead about perceived similarities. As elsewhere in the book, it is informative even for researchers already familiar with the work reviewed to see it contrasted and discussed methodologically. Steen perceptively nds common problems across dierent traditions and shows how dierent data types and techniques of analysis can complement each other. In Chapter 9, the focus shifts from language to the cognitive, and Steen considers ways of tackling two more of the questions raised in Chapter 1, concerning storing, acquiring and losing linguistic and conceptual metaphors. Two types of cognitive process are considered: usage and acquisition; that is, local and long term. There is a very interesting discussion of the acquisition of linguistic and conceptual metaphors, citing Johnson (1999), who presents evidence from transcripts of childrens talk. It is argued that when children acquire a conceptual metaphor such as UNDERSTANDING IS SEEING, they learn metaphorical and literal meanings of SEEING together; this is not a metaphor for children, and the literal and metaphorical meanings are only separated later. Steen therefore concludes with Johnson that the acquisition process is fundamentally a metonymic one. Earlier views on the topic were not evidencebased and though they seemed commonsensical at the time, have been shown by Johnsons evidence to be incorrect. This underlines the importance of using evidence rather than or in addition to theorising. The discussion of the cognitive aspect of metaphor in grammar for adults is mostly centred round Gibbs work (for example, 2006), partly because he is one of the very few researchers to have tackled this area at all and also because his work is groundbreaking. Steen asks whether introspection can be enough to answer questions about cognition and metaphor acquisition, storage and maintenance and concludes that it is not possible to reliably introspect about long-term processes. Introspection generates hypotheses but not ndings. Observation techniques such as using a corpus are also of limited use, because symbolic patterns that are abstracted from usage do not have a one-to-one relationship with mental representa-

Book reviews Cognitive Linguistics 212 (2010)


tions. Steen argues convincingly that evidence gained by manipulation is needed to test hypotheses generated by introspective and observational data. Part 3 of the book is concerned with nding metaphor in usage, and is divided into three chapters, dealing with metaphorical language, related linguistic gures such as simile, and cognitive processes. In Chapter 10, the study of metaphorical language use, Steen argues that the range of metaphors found in usage is far greater than that found in grammar, because of the enriched contextual meaning, and because ellipted forms can also be included. It is pointed out that shared context makes the target domain well known and specic, and contributes a great deal to the meaning of the metaphor. Interesting and useful contrasts are drawn between diachronic studies which focus on the meanings generated by individual users, such as Chiltons (1996) study of cold war discourse spanning 50 years, and synchronic studies of typical user meaning, such as Kollers (2004) study of business discourse. Steen then returns to the pragglejaz procedure, previously mentioned in Chapter 4, for its explicit steps and listing of the various decisions that have to be made. Pragglejaz does not attempt to specify any underlying cross-domain mapping; Steen shows how Barcelonas (2002) two-step procedure for identifying domains and characterising mappings can take up where pragglejaz nishes. Steen discusses the use of corpora to observe metaphor in usage, and contrasts the approaches of several current writers. It is argued that manipulation, through using large numbers of informants to make judgements about carefully constructed cases, can illuminate issues such as degree of metaphoricity and semantic relatedness which are less reliably assessed through introspection and observation. Chapter 11 deals with other forms of metaphor, mainly simile. Simile does not involve indirect language use but there is nonetheless some form of cross domain mapping. There is a long and interesting discussion of Shakespeares Shall I Compare Thee to a Summers Day, showing how linguistic metaphor is embedded within simile. It is argued that although this text is exceptionally complex, a degree of related complexity is also to be found in ordinary discourse. Chiltons cold war data (1996) also show some very interesting similes and metaphors, which are developed over a stretch of text. Mapping is involved even though not all the resultant language is strictly metaphorical. This kind of cross-domain activity realised through non-metaphorical language is evidence that metaphor is conceptual as well as linguistic. There are also some fascinating worked examples from Tennyson and the Iliad, showing how similes are developed over extended texts and in dierent genres. Allegory and parable are covered, described as extended texts from the source domain that


Book reviews Cognitive Linguistics 212 (2010)

are dicult to identify as metaphorical. This raises the issue of implicit versus explicit metaphor use; dierent analysts take dierent positions on the issue, but if we only look for metaphorically used language then we lose these and some other cases. Steen advances a variant on the pragglejaz procedure designed to deal with such cases. This chapter might appear less interesting for some researchers, if a decision has been made not to study implicit metaphor, but it contains some useful insights and tackles important aspects of simile. Chapter 12 turns to cognitive processes and products in usage. To date, most research on metaphor processing has been concerned with linguistic structure, not usage. Steen points out that conceptual metaphor theory does not set up a model of processing, so many psycholinguists are sceptical of it. An important exception is Gibbs, who works from the theory to devise deductive, experimental work. Steen uses Gibbs denitions of dierent kinds of understanding to tease out four aspects of processing: recognition, comprehension, interpretation and appreciation. Comprehension is essential, the other three are optional, and their nature depends on the language use and event. This raises the question whether entailments of cross-domain mapping are found in automatic comprehension, or only in conscious understanding, a question especially relevant for education and text design. Gioras work on salience (for example, 2003) leads Steen to argue that metaphorical entailments are not always mapped automatically, online; if the metaphorical sense of a word is the most salient, then no literal sense is activated. Steen also notes that psycholinguists have largely ignored genre, though all language use is genre-regulated (p. 353). The question of when automatic comprehension stops and conscious understanding starts takes a dierent perspective depending on which of the four language skills we are considering. In terms of data, Steen argues that it is impossible to get good introspective evidence for cognitive processes in usage, because we cannot introspect about our online production and comprehension, though we can consider the products of comprehension. This leads him to critique the use of introspective evidence in the standard pragmatic model of comprehension, and in conceptual metaphor theory and blending theory, arguing that all three hypotheses are founded purely in the researchers intuitions. Observation techniques are little better: conversation and discourse analysts can get close to mental processes through the detailed analysis of spoken data, but they cannot give denitive answers. Steen especially cautions against the analysis of published texts as an insight into the composing process, because of the amount of editing and reection that will have

Book reviews Cognitive Linguistics 212 (2010)


gone into them which means they cannot be seen as the product of online production. Manipulation is therefore the only technique that can hope to provide useful and reliable information about metaphor processing in usage. Chapter 13 concludes the book. It begins with an overview of the attempt to establish methodological guidelines for empirical research in metaphor, drawing together and comparing dierent approaches and different but related kinds of questions. Steen comments that it would be useful for researchers to dierentiate the stages of their research into the recognised categories used in science, for studies to be comparable. One of his conclusions is that the dierent models of metaphor have a reasonable amount in common, in that they all consider two domains. Steen again criticises many metaphor studies for not dening their categories tightly enough, and reiterates the need for caution and for careful distinctions, for instance between introspection as a source of data and as a tool for analysing data, and text as prompts for eliciting data from informants or as data themselves. In the latter part of the conclusion, Steen returns to the eight questions that were generated in Chapter 1, by the intersection of three distinctions to be made by metaphor researchers: metaphor in grammar versus usage, metaphor in language versus metaphor in thought, and metaphor as symbol or as behaviour. These pages usefully summarise the methodological points developed through the book and some of the principal procedures that have been described and developed. Gaps and possibilities for future research are pinpointed here, making for a thought-provoking research agenda for scholars of real world metaphor use. Many studies of metaphor have extrapolated from language data to claim to have discovered new conceptual metaphors, with no discussion as to how the linguistic metaphors were identied, or how it was decided that they represent any underlying patterns of thought. This book is a much-needed examination of all the issues involved. In a sense, it is not a book of answers; it is a book about how to tackle the questions, and it contains many thorough and perceptive critiques of other researchers attempts to do so. It is a huge undertaking, both reviewing and developing ideas, and a review such as this one can only hope to select a few of the many topics covered, inevitably in a biased and partial manner. For me, the complexity of the prose, both at the level of sentence and of argumentation, meant that the book is not an easy read- perhaps this is inevitable given the range of topics covered and the ambitious depth. However, this is a minor criticism of what is a signicant contribution to the eld, one that will form core reading for current and future researchers of metaphor.


Book reviews Cognitive Linguistics 212 (2010)

Barcelona, A. (2002) Clarifying and applying the notions of metaphor and metonymy within cognitive linguistics: An update. In R. Dirven and R. Porings (eds.) Metaphor and Meton ymy in Comparison and Contrast. (pp. 207277) Berlin: Mouton de Gruyter. Chilton, P. (1996) Security Metaphors: Cold War Discourse from Containment to Common House. New York: Peter Lang. Gibbs, R. W. (2006) Embodiment and Cognitive Science. New York: Cambridge University Press. Giora, R. (2003) On our Mind: Salience, Context and Figurative Language. New York: Oxford University Press. Grady, J. (1997) THEORIES ARE BUILDINGS revisited Cognitive Linguistics, 8, 267 290. Johnson, C. (1999) Metaphor vs conation in the acquisition of polysemy: The case of SEE. In M. K. Hiraga, C. Sinha and S. Wilcox (Eds.) Cultural, Typological and Psychological Issues in Cognitive Linguistics, pp. 155169. Amsterdam: John Benjamins. Koller, V. (2004) Metaphor and Gender in Business Media Discourse: A Critical Cognitive Study. Basingstoke and New York: Palgrave Macmillan. Kovecses, Z. (2000) Metaphor and Emotion: Language, Culture and Body in Human Feeling. Cambridge: Cambridge University press. Lako, G. and Johnson, M. (1980) Metaphors We Live By. Chicago: University of Chicago Press.

L. David Ritchie, Context and Connection in Metaphor. How Simple Ideas Shape Human Experience. Houndmills: Palgrave Macmillan, 2006, 248 pp., ISBN-13: 978-1-4039-9766. Hardcover $ 74.95. Reviewed by Ksenya L. Filatova, Ekatherinburg, Russia, Ural State University. Email 3ksenya.latova@gmail.com4 The book under review merges existing theories of metaphor into a highly consistent whole, drawing on current research in cognitive studies of the brain. Starting with illuminating and constructive criticisms, the author delivers a cogent synthesis of selected concepts from diering theories, to the general benet of his Context-Limited Simulators (CLS) theory. Compositionally, Context and Connection in Metaphor consists of nine chapters, including an introduction. Its index system (by name, by metaphor, by subject) is comprehensive and reader-friendly. The rst chapterintroduction gives a vivid overview of the authors stancemetaphors are claimed to be simultaneously cognitive, in a thoroughly biological sense, and social (4). The introduction sets up the books polemic tone with a debunking of meta-metaphors in which metaphor theorists are getting trapped (8), for example: mind as a machine,

Book reviews Cognitive Linguistics 212 (2010)


language as a code. Lako and Johnsons denition of metaphor is accepted as a working one and is elaborated upon later. Chapter 2 deals with attributional and relational models of metaphor, the two cornerstones of classical metaphorology. After careful examination of its subtle variants, Ritchie points out that both property attribution and superordinate category formation imply the existence of common salient features that have already been metaphorically interpreted, and thus, both theories are fundamentally circular (25). Chapter 3 presents a thorough overview of conceptual metaphor theory, from the original insights (G. Lako and M. Johnson) to some of the recent extensions (J. Grady, S. Narayanan) and principal criticisms (J. Vervaeke and J.M. Kennedy) concerning the theory. Ritchie then proposes a detailed analysis of the classic ARGUMENT IS WAR metaphor, and shows that the concept of WAR is far from being directly grounded in physical and social experience, and that people come to understand it in the rst place by metaphorical elaboration of their own embodied experiences of interpersonal conict, athletic competitions, and role-playing (45). Considering one by one the numerous problems raised by the conceptual metaphor theory (multiple meanings versus metaphorical meanings, xity of metaphorical meanings, suciency of metaphor in representing abstract, experienced-based concepts, and groundings of complex metaphors, inuence of metaphors on thought, lexicalizing conceptual metaphors), the author comes up with the main inconsistency between Lako and Johnsons theory and the avalanche of research that followed. Ironically, the revolutionary denial of literal meaning and all the referential logic in their seminal 1980 work was rather neglected by their followers who established one-to-one correspondences between metaphorical expressions and underlying root metaphors (54). According to Ritchie, the fundamental insight of Lako and Johnsons conception is that the co-occurrence of physical experiences . . . with each other and with more abstract emotional experiences . . . strengthens neurological connections that form the basis for subsequent understanding of concepts (55). He also makes a pertinent remark concerning languagedriven versus embodied knowledge. He argues that the role of the latter has been exaggerated, to the detriment of admitting the importance of communication and language in the development of conceptual metaphors (57). In Chapter 4 the author deals with G. Fauconnier and M. Turners conceptual blending theory, criticisms of that theory and responses to these criticisms. His major objection lies in the unnecessary complexity of the theory: faithful to his iconoclastic demeanour, Ritchie shows that many elaborations derive from meta-metaphors (space, blending),


Book reviews Cognitive Linguistics 212 (2010)

and without contesting the conceptual integration as a phenomenon, he promises a more coherent explanation thereof. Chapter 5 suggests an overview of some recent ideas about context and common ground. Beginning with relevance theory (D. Sperber and D. Wilson), Ritchie teases apart notions of context, cognitive environment and mutual cognitive environment, and accuses the theory of the same circularity vice: the decision whether to process a message in a particular context relies on the eects achieved by processing the message (83). Besides, he notes as unresolved the issues of innite recursion and of biological grounding. Moving to the conversational model of H. H. Clark, the author dwells on the social and interactive dimension it oers: the analysis of the communicative situation itself is considered to be useful for describing the coordination of multiple contexts and their inuence on the common ground. In Chapter 6, the author gives a detailed synopsis of L. Barsalous perception-based theory of cognition which he characterizes as an explicitly embodied approach (97) especially consistent with the limited processing capacity of the human neural system (123), and explains how it should be applied to language. Stating that, due to working memory limitations, simulations of concepts and schemas could never be complete, and that the previously activated contents of working memory must play a role in determining which features are simulated and which are omitted (114), Ritchie raises the question of how contextboth limited and extendedcan be incorporated into such a theory of cognition. That gives him an opportunity to re-analyze a number of theoretical constructs (experienced present, relevance, cognitive environment, social schemas, distributed cognition, cultural schemas, common ground ) and to show that they t into the previously described model. Though numerous gaps in the perceptual simulators theory are identied (123), Barsalous conception is proclaimed to be optimal for elaborating a new theory of metaphor. Chapter 7, the very heart of the book, presents a new theory of metaphor use and interpretationCLS simulators theory. The underlying ideas being already obvious to the reader, Ritchie takes his time to weave all the loose threads into smooth theoretical tissue. Though being compatible with conceptual metaphor theory, CLS accords more importance to language as a direct source of concept development (132). Ritchie then revisits the main assumptions of conceptual met