Typology and
Grammaticalization
Heiko Narrog
6.1 Introduction
‘The goal ofthis chapter is to give an overview of the relationship between
language typology and grammaticalization as it is represented in the
research literature, structured from a specific perspective. This perspec.
tive is provided by the question of how language typology may influence
grammaticalization on the one hand, and how grammaticalization may
influence typological features of languages on the other hand.
Any understanding of this question is necessarily dependent on how we
define language typology and grammaticalization. Therefore we will start
Out here with a short review of definitions of the main subjects of this
chapter. The problem is clearer with language typology than with gram-
‘maticalization, so we will start out with language typology. First, consider
the rather traditional and narrow definition in (1)
(1) ‘Language typology is thhe classification of languages according to
their structural features.’ (Trask and Stockwell 2007: 309)
Now compare this with two wider and more modern definitions, first one
by Whaley (1997) as in (2) and then one by Velupillai (2012) in
(2) ‘Typology is the classification of languages or components of lan-
‘guages based on shared formal characteristcs."(Whaley 1997: 7-11)
(3) “To put it very simply, linguistic typology concerns itself with the
study of structural differences and similarities between languages
[It] is the study and interpretation of linguistic or language types.
More specifically, itis the study and interpretation of types of linguis-
tic systems.’ (Velupillai 2012: 15)
Its easy to see that the scope of language typology widens considerably
from (1) through (2) to (3) In (2) ‘components of languages’ are added to182 HRC NARKOG.
tse
“tanguages’ and in (3), the study of structural differences and similarities
petween languages is taken as the legitimate subject of language typology.
If we adhere to a rather narrow definition of language typology as the
classification of languages as in (1), the interaction between language
typology and grammaticalization will be rather limited. The more we
widen our scope. the more potential interaction is there to be observed.
Obviously, it is in the interest of this chapter to hold a wider view of
Tanguage typology that allows for more interaction. We will therefore go
with definition (3), which also more faithfully represents the state-oFthe-
lt of the field. A look through the accumulated contents of Language
Typoiagy the flagship journal of the field, reveals that language typology
indeed deals with practically every aspect of grammar from word order to
categories like voice, to features like animacy, and to morphology on the
fone hand and complex sentence structures on the other. Also note that the
focus ofthis chapter will be on language typology more than on universals,
pecause here as well we find more meaningful interaction and more
literature discussing this interaction.
Now we turn to the definition of grammaticalization. (4) is a classic
definition that is frequently cited in the literature,
(4) ‘Grammaticalization consists in the increase of the range of
a morpheme advancing from a lexical to a grammatical or from
‘a less grammatical to a more grammatical status.” (Kuryiowicz
[1965] 1975: 69)
‘The definition in (5), which is currently also often cited, can be considered
as an ‘updated’ version of (4), taking into account more recent ideas about
contexts and constructions.
(5) “{Grammaticalization] is concerned with such questions as how
lexical items and constructions come in certain linguistic com
texts to serve grammatical functions or how grammatical items
Gevelop new grammatical functions.’ (Hopper and Traugott
2003: 1)
Lastly, Heine (2003: 575) has provided us with the definition of grammati-
calization thats the broadest of all, namely of grammaticalization as
(6) ‘the way grammatical forms arise and develop through space and
time’
For the purposes of this chapter, it doesn't really matter which of these
three definitions are chosen. They are all sufficiently broad. As in the case
of typology, the problem arises if the definition is narrowed down. 1 will
spare the discussion of such narrowed-down alternatives forthe part ofthe
‘chapter where the difference really matters (§6.3), and will for now stick to
‘2 traditional and broad understanding of grammaticalization, which
allows fora lot of interaction with language typology.Typology and Grammaticalization
153
Table 6.1 The parameters of grammaticalization (Lehmann
2002: 110)
Parameter Paradigmatic ‘Syntagmatic
Weight Integrity ‘Struural scope
Cohesion _Paradigmaticty Bondedness
Variability Paradigrnatc variability Syntagmatic variability
Almost as important as the definition of grammaticalization are criteria
for grammaticalization, As is often the case with terms that come into
vogue, there is a salient tendency to overstretch this term to refer to all
Kinds of phenomena, and thus render it almost vacuous. Determining the
concrete criteria according to which a certain phenomenon should be
actually considered as grammaticalization is an important step.
‘According to my knowledge, the following two sets of criteria by
Lehmann, and Heine and Kuteva, respectively, although based on work
dating from the 1980s, still constitute the most comprehensive available.”
‘The set of criteria by Lehmann (2002) in Table 6.1 is based on the concept
of grammaticalization as decrease in autonomy of a linguistic sign.
‘The concrete criteria are structural in nature and introduced deductively
from this overall idea. Presumptive cases of grammaticaljzation can be
measured against them.
‘The combination of three parameters of grammaticalization and the
syntagmatic and paradigmatic dimensions results in six processes of gram-
‘maticalization, namely (1) oss of integrity (weight), Le. attrition, in the case
of phonological substance erosion, in the case of meaning desemanticiza-
tion, and in the case of morphosyntax decategorialization, (2) increasing
paradigmaticity (cohesion), ie. paradigmaticization, (3) loss of paradigmatic
variability (variability), ie. obligatorification, (4) shrinking of the morpho-
logical scope of a sign (weight), Le. condensation, (5) increase in bondedness
(cokeston), i.e. coalescence (also “univerbation’) and (6) loss of syntagmatic
variability (variability), Le. fixation. In contrast, Heine and Kuteva (2002)
take a bottom-up approach and, generalizing over many cases of gramma-
ticalization, define a set of criteria that is recurrent in grammaticalization
as opposed to other types of change. The criteria are listed in (7.
(7) a, desemanticization (or ‘semantic bleaching’) - loss in meaning
content,
». extension (or context generalization) ~ use in new contexts,
. decategorialization — loss in morphosyntactic properties charac-
teristic of lexical or other less grammaticalized forms, and
4. erosion (or ‘phonetic reduction’ - loss in phonetic substance.
It goes without saying that the focus on the individual Linguistic sign that is
apparent in both sets of criteria, especially in Lehmann’s, does not neces-
sarily sit well with recent approaches to grammaticalization in terms of184
Hero Naso
constructions (cf. definition (5), Lehmann’s set of criteria have also been
criticized as biased towards morphological change, since they basically
refer to morphemes as the ‘linguistic sign’, and the formal change mor
phemes undergo (cf, e.g,, Traugott 2010: 273-4). Still, these two sets are
the best currently available.
‘Now, coming to the main topic, there are logically speaking four possi-
ble ways in which grammaticalization and language typology may inter
act. Namely,
i. grammaticalization motivates structural features that can be typolo-
gized but not vice versa;
ii, typological features influence aspects of grammaticalization but not
vice versa;
iii, there is both an influence of grammaticalization on structural features
of language that can be typologized, and influence of typological
features on aspects of grammaticalization;
iv, no interaction,
Ifwe choose the definitions that we did above, we can firmly conclude on
the basis of extant research that grammaticalization and typological fea-
tures influence each other, that is, (ii) is the case. This will be shown
below. Section 6.2 deals with the influence of grammaticalization on
typological features of languages, section 6.3 with the influence of typoto-
gical features on aspects of grammaticalization and section 6.4 will give
a short conclusion.
6.2 Grammaticalization Motivates Structural Features that
Can Be Typologized
This section consists of three subsections, based on type of interaction.
The first. $6.2.1, deals with grammaticalization as a partial explanation
for some implicational universals, the next, §6.2.2, with grammatical
zation as a (partial) explanation for the onder of affixed material and
finally §6.2.3 with grammaticalization as an explanation for cross-
Linguistic types of the expression of certain grammatical categories.
‘One can view the order of subsections as an ascending order in terms
of relevance for the current research landscape in linguistic typology,
starting out with material that is meanwhile historical. A short conclu
sion follows in §6.2.4.
6.2.1, Grammaticalization as Partial Explanation for Some
Implicational Universals
Lehmann (1986: 12) observed that the process of grammaticalization can be
a causal factor of some of Greenberg's implicational universals}] such as theTypology and Grammaticalization
135
following (the numbers refer to the numbering of the forty-five universals
proposed by Greenbeng in his seminal 1961 paper
2. In languages with prepositions, the genitive almost always follows the
governing noun, while in fanguages with postpositions it almost
always precedes.
3. Languages with dominant VSO order are always prepositional.
4. With overwhelmingly greater than chance frequency, languages with
normal SOV order are postpositional
16. In languages with dominant order VSO, an inflected auxiliary always
precedes the main verb. In languages with dominant order SOV, an
inflected auxiliary always follows the main verb.
27. Ifa language is exclusively suffixing, itis postpositional. If itis exclu
sively prefixing, itis propositional.
All of the above generalizations can be explained by the fact that when
Texical heads of complex constructions grammaticalize (eg. relational
noun to adposition: verb to auxiliary), they usually remain in their original
position, For example, relational nouns grammaticalize to adpositions (2..
3., 4), and verbs grammaticalize to auxiliaries (16.) and temain in their
original position. However, Lehmann (1986: 13) does not consider this to
be a full ‘explanation’. He cautions that
it should be clear that the recourse to grammaticalization is not in itself
a sufficient explanation of such implicational generalizations. AS | said
before, the specific function common to the syntactic relations of
a relational noun to its complement and of an adposition to its comple-
‘ment, and the specific flinction common to the syntactic relations of
a verb to its object and of an adposition to its complement, are concepts
not inherent in grammaticalization, but are, instead, a presupposition for
these constructions to be connected by grammaticalization
‘Thus, Lehmann seems to consider grammaticalization asa channel but not
as the ultimate cause or motivation for the order and relationship between
elements of the sentence. Relatedly, Greenberg himself (1995) saw a role
for grammaticalization in his four-part approach to diachronic typology.
which consists of the following elements:
i, Dynamicized state-process model. Synchronically, there are certain
states within a specific typology, or within a specific language; eg.
OV (vs. VO}; AN (vs. NA} etc, The typology describes the process by
‘which the language changes from one state to another. In order for
this model to lead to interesting generalizations, a ‘strong connection’
must hold, i. it must be possible to move from one state to another
within that typology.
ii, Dynamicized state-process model of an elaborate (or: sub) typology.
This model looks at intermediate states between the main states in186
Heo NeROG
a typology and tries to find generalizations about the diachronic rel
tionship between these states.
Intragenetic comparison: comparison within one genetic language
group.
iv. Intergenetic comparison: comparison between two or more genetic
Janguage groups.
Within this four-part approach, Greenberg (1995: 157, 159-60) primarily
locates the study of grammaticalization in intragenetic comparison,
where grammaticalization theory provides knowledge about the direc-
tionality of change, but also in intergenetic comparison, where gramma-
ticalization theory does the same on a larger scale. He was mainly
interested in how grammaticalization interacts with global constituent
order and word order changes to come up for detailed variations in word.
order within one language, and counterexamples to implicational univer
sals such as the above. The following (reconstructed) stages of word order
development in Ethiopian Semitic languages is an example (we are here in
the context of intragenetic comparison).
(8) PrING/NA -+ Pr/NG/AN ~» Pr/GN/AN — Pp/GNJAN (p. 155)*
‘The highlighted part of the chain constitutes a violation of implicational
‘universal (2). However, it can be explained through the interplay of prince
ples of grammaticalization and global constituent order and word order
‘change in this language. Among noun genitive, noun-adjective and adpost-
tionhhead, nounadjective order is the type of word ortler that is most
unstable and susceptible to change. It will change first, followed by noun-
genitive order, while grammaticalization from relational noun to adposition
will take the most time, and therefore adpositions will lag behind. This
‘szadual sequence of changes, then, in which grammaticalization plays a
role, leads to violations of universal in the ordering of meaningful elements.
Lehmann (1986: 15-17) also discussed the role of grammaticalization in
this type of historically conditioned intra‘anguage (and intragenetic) var-
iation. However, he argues (pp. 17-20) that not constituent order and word
order, but the area on the borderline between syntax and morphology
should be at the centre of the study of grammaticalization and linguistic
typology. According to him, high-level syntax like constituent order is not
grammatically determined enough but may be pragmatically determined.
Morphophonemic processes tend to be irregular, and thus difficult to
generalize upon, crosslinguistically. Pure meaning (semantics) and
expression (phonology) are also excluded because it is the combination
of meaning and form that is typical for language. The lexicon (lexical
typology) is not regular enough. Indeed, to my knowledge, the study of
the role of grammaticalization in the development of constituent order
and word order has remained at the fringe of typologically oriented
research in grammaticatization.Typology and Grammaticalization
187
62.2 Grammaticalization as (Partial) Explanation for the Order of
Affixed Material
Via the principles of grammaticalization, we expect morpheme order tobe
frozen word order. Originally independent words that grammaticalize
become increasingly prosodically dependent on adjacent lexemes or
words and end up as affixes in their original position. In typology, there
are at least two related topics, namely the suffixing preference and mor-
pheme order in complex words.
6.2.2.1 Suffixing Preference
‘There is an overall tendency in languages to prefer suffixes over prefixes,
which has been well known in language typology at least since the 1980s.
‘As Bybee, Pagliuica and Perkins (1990) showed, this tendency even holds in.
the majority of head-initial, especially SVO, languages. Table 6.2 shows the
overall preference for suffixation, irrespective of basic word order, accort-
ing to data by Dryer (2005).
‘A more detailed breakup of where we find suffixes and prefixes shows
that there are considerable differences by grammatical categories. Related
findings by Gysouw (2006) are displayed in Table 6.3.
As Table 6.3 shows, there are significant differences with respect to what
categories get suffixed vs. prefixed. These differences may be ascribed to
the different positions in which the morphemes grammaticalize, which in
turn may depend on mechanisms of discourse.
Perhaps the most detailed study on the extent of the suffixation prefer
ence and its causes is found in Bybee et al. (1990), and to my knowledge
there is no research that substantially contradicts the findings and the
conclusions in this study.
Bybee et al. (1990), after giving an overview of the suffixation prefer-
cence, in a language as a whole and also by word order type, investigate
a number of psycholinguistic (processing) and phonological factors possi
bly leading to the suffixing preference, and find that none of them can
account for it, They conchide that the ‘fossilized syntax hypothesis’
explains the suffixing preference best (p. 35). The ‘fossilized syntax
hypothesis” goes back to the idea by Givon (1971) that ‘yesterday's syntax
Table 6.2 Prefixing versus suffiing in inflectional
morphology (Dryer 2005: 110)
Litle or no iflecional morphology 122
Predominantly suffcng 302
Moderate peterence for sufing na
Approximately equal amounts of suing and prefaing 130,
Moderate pelerenc for prefining 2
Predominantly prefxing 34
Total138
HEIKO NARROG.
Table 6.3 Rate of suffixation (vs. prefixation) by grommatical category
(Cysouw 2006: 11)
100.
|
74.4 % overall
‘Sufization (ne)
«0
20
10
‘case plurality tense/aspect pronominal person person
Possession grams paradigms
is today's morphology’. It assumes that ‘the position of an affix is the same
as the position of the nonbound lexical or grammatical material from
which the affix developed’ (Bybee et al. 1990: 3). AS is well known, this
idea is the starting point for the revival of interest in grammaticalization
in modern mainstream linguistics,
However, strikingly, fossilized syntax does not necessarily reflect nor
mal word order, Pethaps the most conspicuous example is (subject) person
endings being suffixed even in SOV languages, where they aze expected to
precede the verbal root. The most likely reason is that it is unstressed
pronominal subject pronouns postposed to the vetb, rather than pronouns
in their normal position that get grammaticalized (cf. Bybee et al. 1990: 10;
based on Comrie 1980).
Further research has shown that within person paradigms, there is
2 prefixing preference for very small and very large paradigms, while
those in the middle are predominantly suffixing (Cysouw 2006).
‘As Cysouw concludes, “The big riddle of the suffixation preference thus
actually consists of various smaller-scale riddles concerning different
Kinds of affixation asymmetry’ (2006: 14).Typology and Grammaticalization
159
Se eta eee ee)
we 6 | pe [ram nev [suv |naDavayeay| 3”frewr wobec Sb
ui en fda | falar
ree] L
Digit tao Comprcipeines |
Figure 6.1, Navajo verb template (according to Young and Morgan 1987: 37-8)
Mithun (2003) argues that the ultimate answer to the suffixing prefer
ence must be sought in the history (ie, grammaticalization) of the indivi-
dual morphemes that as an aggregate make up the suffixing preference.
Generally speaking. itis reasonable to assume that the position of a bound
morpheme vis-vis its host reflects the order of the erstwhile independent
‘word vis-i-vis this host, unless one can make the wellfounded assumption
that a morpheme changed its position after grammaticalization. Harris
and Campbell (1995: 64-70) refer rather abstractly to cases where reana-
lysis in changes that we would identify as grammaticalization led to
a change in position. These cases should probably be considered as excep-
tional, since bound morphemes are by definition less mobile than inde-
pendent ones. However, a change of position may become more likely if
a morpheme goes through a clitic stage. Comrie (1980: 85-7; 1989: 217-18)
observed that the position of clitics, while less free than that of indepen-
dent words, is more free than that of other bound morphemes, and that for
prosodic reasons. clitics may follow rules that differ from those of inde-
pendent words. Ifa clitic stage is involved in the grammaticalization of
a morpheme, the possibility that morpheme order does not reflect erst-
while word order becomes greater. Comrie has argued that this is the case
for pronominal clities, which are the most prominent type of clitis in the
Indo-European languages, and it remains to be seen if this is also true of,
other grammatical categories.
Note, though, that grammaticalization is the mechanism through
which grammatical elements become frozen in a certain position, but
grammaticalization alone does not really explain why an element is in
a position before or after the verb when it grammaticalizes in the first
place.
6.2.2.2 Morpheme Order in Languages with Highly Complex Words
In languages with a complex morphology, morpheme order can also be
‘a highly complex issue. There are different approaches, and it is fair to say
that motivations for morpheme order are not yet resolved. But grammati-
alization may be one important motivation. Morpheme order in the
Navajo verb, as presented by Mithun (2011), is a case in point. Figure 6.1
shows the Navajo verb template with the various morphemes involved in
abbreviation, and (9) spells out the abbreviations.160 HEIKO NARROG
_—_ ee
9)
Position classes (according to Young 2000)
0 Object of a postposition [applied objects}
la Null postposition
Ib Postpositions {applicatives}, Adverbial-Thematic, Nominal
prefixes
Ie Reflexive
1d Reversionary: ‘returning back’
Je Semeliterative ‘once more"
TT iterative
I Distributive plurat
IV Object pronominals
V Subject pronominals: third person
VI ‘Thematic and adverbial prefixes [three slots}
Vil_ Mode modality. aspect}
VIM Subject pronominals; first and second persons
IX [Classifiers] (valency markers}
X Stem
‘The order represented in Figure 6.1 posits a number of difficulties for any
attempt at explanation. According to Mithun (2011: 179-80), the following
four points are especially challenging.
i. Languages with verb-final syntactic structure are expected to be suffix
wv.
ing. Navajo shows strong verb-final order in clauses, but itis exclu-
sively prefixing.
. Mutually dependent morphemes are often expected to be contiguous,
but many Navajo lexical entries consist of parts scattered throughout
the verb. Also, mode-aspect categories are expressed by combinations
of non-contiguous prefixes and stem shape.
Inflectional affixes are expected to occur outside of derivational
affixes, but in Navajo derivational and inflectional prefixes are
intercalated.
Paradigmatically related affixes, usually mutually exclusive, typically
all occur in the same position in a template. But they don’t do so in
Navajo.
It is impossible to go into detail here, but according to Mithun (2011), the
‘most popular explanations of this order in terms of syntax (the ‘mirror
principle’; Baker 1988), semantic scope (Rice 2000}, and combination of
syntactic and phonological principles (Hale 2001) cannot really explain the
non-contiguity of morphemes of the same category and the overall ‘patchi-
ness’ of morpheme order: In contrast, there is good evidence that the order
of morphemes can be explained as reflecting the order in which they
grammaticalized historically, Mithun (2011) describe
some detail how
prefixes increase in (1) phonological reduction, (2) generality and abstrac-
tion and (3) diffuse meaning from left to right. Thus, from leftto right, theyTypology and Grammaticalization
161
show an increasing degree of grammaticalization (see the parameters by
Lehmann (2002) and Heine and Kuteva (2002) as presented in §6.1)
In conclusion, ‘the positions of prefixes in the verb correlate with their
age: those closest to the stem are the oldest. and those furthest the young:
est’ (Mithun 2011: 184). Furthermore, ‘the Navajo prefixes whose sources
are still identifiable developed from words that still occur syntactically
before the verb’ (p. 189). Hence grammaticalization explains morpheme
order best.
‘Thus, Mithun (2011) presented a wellargued case for morpheme order
reflecting grammaticalization, rather than any other (syntactic or seman-
tic) motivation in the Navajo verb. Note that this case is different from that
presented in the discussion of the suffixing preference in section 6.2.2.1.
In the case of the suffixing preference, grammaticalization is seen as
mediating between the word order of independent words and of bound
morphemes. In the case of order between morphemes in the Navajo
template, the question is the reason for the relative position of mor-
hemes to each other. In this case, grammaticalization appears to be the
immediate cause and therefore a valid explanation,
6.2.3 Grammaticalization as Explanation for Cross-linguistic Types
of Expression of Certain Grammatical Categories
‘The past twenty or so years of typological and functional research related
to grammaticalization have brought to the forefront the observation that
‘many linguistic categories are crossJinguistically expressed by a limited
number of structural types. These structural types are the product of
‘grammaticalization, Furthermore, the genesis of these types can explain
at least some structural and semantic features that they have. This
approach to linking typological patterns with grammaticalization is parti-
ularly associated with the lines of research of Heine (1997; Heine and
Kuteva 2002) and Bybee (2006; Bybee, Perkins and Pagliuca 1994). In the
following, | present a number of prominent examples that have been
brought up in the literature:
Indefinite articles: According to Heine (2004; Heine and Kuteva 2006),
¢.89 per cent ofall indefinite articles cross-linguistically are derived from
the numeral ‘one’. This explains some positional tendencies of indefinite
articles, the fact that they are often confined to singulars, and beyond that,
the fact there is the following implicational hierarchy for their apptica-
tion: mass noun > plural noun > singular noun,
Possessive constructions: Heine (1997) identified eight source sche-
mas for possessive constructions cross-linguistically. They are displayed in
Table 6.4.
According to Heine (1997: 102-3), these source schemas can account for
some characteristics of specific possessive constructions, such as why they
often have non-verbal, or copularlike, predicates, or how the possessor is182
HERO NARROS:
Oe
Table 6.4 Source schemas for predicative
possessive constructions (Heine 1997: 91)
Source schema Label of schema
Xtakes¥ ‘Action
Yis located atx Location
Xiswih ¥ Companion
XS Yexists Genitive
Y existsfor/toX Goal
Y¥ ens trom x Source
As forX Yess Topic
Yisxs 0) Equation
movement
characteretic
ater
Figure 6.2 Paths forthe development of future mosphemes (Bybee 1988: 374)
encoded in a specific language, ie. as a comitative, locative, etc., or why
they frequently have locative morphology, etc.
Furthermore, these constructions often undergo a development, at the
end of which
1. the possessor precedes the possessee.
2. the possessor has properties of a subject, and the possessee has proper-
ties of clausal object.
3, the possessor is definite and the possessee is indefinite.
Depending on the stage of development, possessive constructions often
display ‘bybrid’ properties between source and target structures (Heine
1997: 98-9) (see also Comrie 1989: 219-25 fora somewhat detailed descrip-
tion of such a process in Maltese). In this manner, grammaticalization also
accounts for their properties as a process.
Future: Bybee (1988; and elsewhere) investigated cross-linguistic poly
emy of future morphemes and found that the polysemy can be explained
by reference to the diachronic evolution of these elements, which can be
represented as in Figure 6.2 above.
‘According to Bybee (1988: 374-5), these paths of development are expla-
natory because they explain (1) why itis difficult to find a single abstract
‘meaning for a polysemous morpheme (like many future morphemes}, (2)
the cross-linguistic similarities of grammatical meanings by similar paths
of development and principles of historical change and (3) differencesTypology and Grammaticalization
163
between morphemes in different languages with reference to different
lexical sources and different extent of change along the universal paths of
change. Also, (4) they predict possible combinations of meanings, and (5)
they allow reconstruction of the lexical sources of grams. Bybee’s line of
research continues to emphasize the explanatory potential that gramma-
ticatization has for crosslinguistically reoccurring patterns of expression
of grammatical categories, The quote in (10) from Bybee (2009) sum-
‘marizes this perspective on grammaticalization and typology.
(10) ‘... | Ihave identified several mechanisms of change, all of which
are driven by increased usage: phonetic reduction, automatization,
increasing autonomy, semantic bleaching, and pragmatic infer-
fence. These are the basic mechanisms of change that can act on
any grammaticalizing material and render it part of the grammar
Although these processes explain similarities across languages, they
also allow for and create differences .... Thus, grammaticalization
hhas great potential for explaining the similarities as well as the
differences among languages." (Bybee 2009: 30)
Passives: Givon (2008, 2009) identifies six common types of passive con-
structions. (A) the adjectival tative passive (e.g. English common passive) (B)
the reflexive passive (eg. English getpassive); (C) the serialverb adversive
passive (eg. Chinese); (D) the VPnominalization passive (eg. Ute}; (8) the left
dislocation-cum-impersonakpassive (eg. Kimbundu); (F) the zero-anaphora
passive (eg. Sherpa). Note that (AJ-{C) are promotional passives (ie. the
erstwhile object becomes the new subject), while (D}{F) are non:
Promotional. The concept of passives, on which the identification of the six
patterns is based, is a functional one and fairly broad. It should be cautioned
that not every concept of passives would include the same range of
constructions,
In any case, the source constructions reflect functional similarities.
‘The concrete structural properties of these passive structures in individual
languages can be explained by the degree to which they have grammati-
calized (and emancipated themselves) to full-fledged, more prototypical
passives from their source constructions. For example, non-promotional
passives may eventually become promotional as subject properties gradu-
ally shift to the object patient. Similarly, oblique agents may eventually be
added (in early stages of development of passives, agents are usually not
expressed). In this way, ctoss-linguistically common types of passives and
their concrete morphosyntactic features can be explained with reference
to their source construction and their degree of grammaticalization.
6.2.4 Summary
First, since grammaticalizing items usually grammaticalize in a position
‘that was occupied by their lexical predecessor, the order of grammatical164
HEIKO NARROG:
items with respect to other items can be explained by their position at the
time of their grammaticalization. This seems to hold for bound as well as
non-bound morphemes and is a partial explanation for some language
patterns of typological interest. Secondly, commonalities and differences
in the expression of grammatical categories across languages can be
explained to a large degree by reference to grammaticalization. With
respect to the expression of grammatical categories, it has even been
claimed that true universals are diachronic. It is the processes and paths
of grammaticalization that are shared more or less universally, rather than
their synchronic outcomes. Here I will cite Bybee again, who formulated
this idea succinctly as in (1),
(11) [The] remarkable similarity in grammaticization paths across unre-
lated languages strongly suggests that universals of diachronic
development be included in a theory of language universals.
J would further argue that the diachronic universals are much
stronger than any synchronic universals we can formulate concern-
ing the presence and meaning of tense and aspect markers in the
languages of the world ... Thus the very robust and very specific
paths of development shown above constitute much stronger cross-
linguistic statements than any statements we could devise about
synchronic states.’ (Bybee 2006: 186-7)
63 Typological Features Influence Grammaticalization
‘Compared to the scenarios discussed in the previous section, some of the
scenarios presented here are more hypothetical. Clearly, certain historical
processes lead to certain outcomes, but are there target structures of
‘change that influence the pathways of change? This is more difficult to
demonstrate. Before going into possible scenarios, I wish to highlight two
presuppositions for the following discussion. First. very generally, motiva-
tions for grammaticalization itself must be sought in cognition and syn-
chronic Janguage activity, such as online processing of language and
‘memory storage. This is a point readily acknowledged by scholars who
have been concerned with the relationship between synchrony and dia-
chrony. For example, Greenberg (1995: 164) suggested that ‘there is
a gradual process by which innovations originating in discourse ultimately
‘become established as part of the system of language’, that is, he sought
the origins of change in discourse. Haspelmath (2008: 214) suggests ‘econ-
omy’ as the most important motivation of change: [Diachronic| changes
are motivated by economy, ie., by the innovating speakers’ desire to speak
economically. This is not a synchronic grammatical constraint, or even
& more general cognitive constraint. It is simply a constraint on any
rational behavior.’ Hawkins (2011: 225) considers grammars as theTypology and Grammaticalzation
165
‘conventionalizations of the same processing mechanisms that psycholo-
gists find evidence for in experimental and corpus data’. He also views
thems ‘adaptive systems’, and change as driven by efficiency and ease of,
processing, Thus, for Hawkins, efficiency and ease of processing are the
causes of language change and grammaticalization. However, factors such
as cognition, discourse or efficiency are not typological: they operate as
constraints on language activity in general, irvespective of the typological
features of the language.
Secondly, the extent to which typological features influence grammati-
‘alization, or the question if this is the case at all, strongly depends on the
definition of grammaticalization. The more abstract the definition of
grammaticalization, that is, the mote itis reduced to a universal essence,
the smaller the possibility for variation. While the common definitions of
grammaticalization (section 6.1) allow for wide interaction between gram-
‘maticalization and typological features, this interaction is considerably
reduced if the definition is narrowed down. For example, Himmelmann
(2005: 82) proposes that
4 grammaticisation process can be defined asa process of context expansion
on all three levels, as summarised in the following formula:
(Xn) ADB | Kn(Xn4-x) An+xb| Kn+x
where A and B represent full lexical items, ba grammaticised element and
the following three types of contextual changes occur:
{a} host class formationjexpansion: An + Andx (e.g. common nouns —
common and proper nouns)
(b} syntactic context expansion: Xn — Xn4x (e.g. core argument position
— core and peripheral argument positions}
(c) semantic-pragmatic context expansion: Kn -> Knéx(e-g. anaphoricuse
— anaphoric and associative anaphoric use)
(@timmelmann 2005: 82}
Himmelmann’s definition seeks to reduce grammaticalization toa process
of ‘expansion’ on several levels. Importantly for our purposes here, one
goal of his definition is to exclude typological factors from grammaticali-
zation, since they are viewed as epiphenomenal. Himmelmann (2005: 83)
is explicit about this as stated in (12).
(12) "The above definition of grammaticisation ... differs from previous
efinitions of grammaticisation in two ways. Firs, itis selective,
singling out context expansion in general and semantic-pragmatic
context expansion in particular as the major defining feature of
grammaticisation, All the other phenomena which are often observed
in grammaticisation processes and which are considered crterial in
other definitions (such as paradigm formation, erosion, obligatori-
nessjredundancy, etc) do not occur in this definition ... {Erosion166
Heiko NARROG
and fusion here are considered epiphenomena. Their occurrence
depends on at least two factors. For one, they depend on the overall
typological profile of a given language (eg. in isolating languages
the potential for fusion is generally very limited). For another, the
construction type also appears to play a major role.”
In agenerative view of grammaticalization as well, it should be possible to
reduce grammaticalization to a core or essence that is universal and not
amenable to typological influences. In generative grammar, fundamen-
tally, “grammaticalization” is regarded as the diachronic development of
lexical heads into functional heads’ (Roberts and Roussou 1999: 1011}
Similarly. van Gelderen (2004) views grammaticalization as change from
lower head to higher head (Late Merge’) or from Spec to Head. There might
bbe some typological variation in the heads that are the sources and the
targets of change, but the essence would remain unaffected. However, in
this chapter, we are interested in the traditional ‘full package’ of gramma-
ticalization, including phonological and morphological changes, and this
full package is the subject of the discussion in the remainder of this
section. Lastly, note that in the following it will also become clear that
typological features influencing grammaticalization can be shared across
a language area. This is a topic of great potential interest but cannot be
pursued here systematically.
6.3.1 Synchronic Universals Constrain Grammaticalization
‘Trivially, absolute universals will constrain grammaticalization, in that
grammaticalization cannot result in a structure that violates an absolute
universal. However, there are few true universals, especially in grammar,
‘and it is difficult to come up with an actual example of this type of
constraint.
It is potentially more fruitful to ask whether typological tendencies
(relative universals) constrain grammaticalization, that is, is
2 grammaticalization process influenced by the existence or non-
‘existence of a target structure that conforms to a relative universal? This
‘would be the converse of grammaticalization leading to typological ten-
dencies, and so the question is do paths of grammaticalization lead to
typological universals. as claimed in 2.1, or do typological universals con-
strain paths of grammaticalization? The latter would presuppose that
there is some teleology in change, which is an unpopular position in
‘current historical linguistics in general (see Luraghi 2010), and in gram-
maticalization studies in. particular (see Hopper and Traugott 2003: 74).
Haspelmath (p.c) suggested to me that in the case of ditransitive construc-
tions (see Malchukov. Haspelmath and Comrie 2010) he believes that there
are cross-linguistic target structures that may constrain paths of gramma-
ticalization. However, given the lack of related discussions in theTypology and Grammaticalization
167
literature, I don’t see myself in the position to pursue this possibility
‘meaningfully within the framework of this chapter.
A more remotely related predecessor of this issue can be found in the
discussion about the extent to which typology can be used for reconstruc.
tion (e.g. Greenberg 1995: 145; Shields 2011). It is only remotely related
because here we deal with a question of possible former synchronic states
rather than processes. In Comrie (1989: 210-18), we find a sceptical dis-
cussion of this possibility, based on the observation that there are too
many exceptions, Nevertheless, applied with some caution, itis possible
to argue that reconstructions should take into account the typologically
plausible (see Comrie 1993).
63.2 Typological Language Structures Constrain
Grammaticalization Processes
6.3.2.1 Phonological and Morphological Aspects
There are several ways in which phonological and morphological typolo-
gical features may be considered as a constraint or a guiding factor for
grammaticalization processes.
First, im head-final languages, grammaticalization is more likely to lead
to bound morphemes than in heact-initial languages because of the suffix-
ing preference; ie. a stronger tendency for postposed than preposed mor-
phemes to become bound (cf. §6.2.1; Bybee et al. 1990).
Secondly, it has been observed that grammaticalization in tonal, isolat-
{ng languages like the:Sinitic languages does not accompany reduction of
syllables. Ansaldo and Lim (2004: 345) observe that ‘{sjtrongly isolating
languages typically do not allow yesterday's syntax to become today's
morphology ... syllable boundaries are discrete and phonotactic con-
straints rule out reduced syllables of the kind observed elsewhere, the
material available for reduction is not easily found at the morphological
level’ (cf. also Bisang 2004, 2010}. However, Ansaldo and Lim (2004) show
that a reduction in suprasegmental features takes place. The authors found
that erosion was primarily found in vowel quality and duration. Thus, even
in this type of language, some of the phonological and morphological
criteria of grammaticalization (see section 6.1) do apply.
Along similar lines, Schiering (2007, 2010) proposes a three-way rhythm
typology of languages of mora-based, syllable-based and stress-based lan-
guages. He claims that in grammaticalization in general, and in cliticiza-
tion in particular, phonological reduction will be greater in the order
mora-based > syllable based > stress-based. Schiering (2007, 2010) provides
evidence from twenty languages. In Figure 6.3 below, the left side of each
bar represents phonetic strength of stress, and the right side degree of
segmental effect.Ccorroetes Ml Eifect
ruba, nna = 1X66, den = Slave Kt
ne Jasgue, fn = Fresh, ckt= Chukchi, ude = Udibe,
see Gana, aey = Amele,iae ~ Tariana, yd = Kayarih, 36a = Koyra
aun mao\< Maori, c= Mander Chinese, cat = Catalan, Pom = Ng
rian Pidgin, mit = Maltese)
Figure 63 Segretal fic of sess (Schiring 2007; 342; 2010: 88)
Itis puzaling to see that for Mandarin Chinese, classified as mixed stress]
tone, a high degree of erosion is predicted, while research in Chinese
Linguistics stresses how litte reduction takes place in this type of lane
sage, This may be due to the way in which segmental effect Is measured,
Samely through phonetic effects of stress as described in the grammars of
tach language, rather than through empirical or experimental data (cE
Schiering 2007: 342-3).
Possibly, the classification into mora-based, syllable based and stress
‘based languages can explain differences in phonological reduction only
partially. For example, there isa striking contrast in the degree of erosion
in grammaticalization between Korean (cf. Rhee 2011) and Japanese
{Nartog and Ohori 2011), which are both often classified as mors:based
In Korean, phonological and morphological erosion is rampant, while itis
moderate in Japanese. Furthermore, in both languages, contrary to
Schiering’s prediction, we find plenty of phonological and morphological
‘erosion in grammaticalization, unlike in Chinese. Perhaps, Schiering’s
{2010) findings should be understood as being valid for cliticization, Put
cannot be generalized beyond clticization. In any case, Schiering’s (2010)
hypothesis is a promising starting point for further research.
63.2.2 Semantic and Syntactic Aspects
Bisang, in a series of publications (eg. 2004, 2008, 2011) claims a number
of constraints on grammaticalization in Sinitic languages. Besides the
small extent of changes in phonology (lack of cocvolution of meaning,
and form’) that was already discussed above, these are
«+ lack of obligatoriness of grammatical categories; e. lack of grammatr
cal paradigms in general;
> « Jack of clearly determined semantic domains,Typology and Grammaticalization
169
‘+ existence of rigid syntactic patterns (word order patterns) within which
lexical items grammaticalize, and pervasiveness of inference. which
enables language users to encode and decode the function of a specific
item in a specific semantic context:
‘+ no grammaticalization chains (i, grammaticalization of one category
leading to the next category}
Thus, according to Bisang, none of Lehmann’s six parameters of gramma-
ticalization (Table 6.1) applies except syntagmatic variability, nor do other
traditional models of grammaticalization. Besides Sinitic, the same char-
acteristics are claimed to hold for Tai, Mon-Khmer and Hmong-Mien lan-
guages (cf. Bisang 2008: 25-31). That is, the same constraints on
grammaticalization hold for an areal language type
6.5.3 Extant Target Structures Influence Processes of
Grammaticalization
Very generally, it may be assumed that, cetera paribus, the typological
morphological tendencies of a language (fusional. agglutinative, isolating)
may induce the same morphological types as target structures
languages with agglutinative morphology, grammaticalization is more
likely to lead to affixation than in isolating languages. However, the
synchronic morphological structures are not necessarily the direct causes
for different outcome of grammaticalization. There may be indirect, or
deeper, causes for the historical continuation of morphological types. such
as prosodic structure (cf. §6.3.2.1). Furthermore, since the morphological
structure of languages can change significantly over the course of history,
it is also clear that morphological structures are not always replicated
through cycles of grammaticatization.
Replication of existing structures through grammaticalization may
apply not only to morphology (and syntax) but also to grammatical
categories. Ideally, every language would have one exponent for each
crosslinguistically available grammatical category. In reality, indivi-
dual languages, figuratively speaking, often cannot seem to get enough
of specific categories, while they don't care at all about grammaticaliz-
ing other categories. These categories may be expressed indirectly
through other categories, lexically or not at all (depending on the
analysis and theoretical point of view). It appears, then, that
a preferred set of categories is grammaticalized over and over in
waves or cycles (or spirals). The categories whose grammaticalization
is preferred may further impact the mode of grammaticalization of
other categories,
For example, in the area of tense, aspect, mood (TAM), Bhat (1999)
described how languages may be either tense-, aspect, or mood promi-
nent, and the not overtly expressed categories may be indirectly expressed170
eKONARROG
through the well-grammaticalized ones. In terms of grammaticalization,
‘we can surmise that languages tend to maintain this typological profile by
renewing their preferred category rather than grammaticalizing the
“neglected” ones.
Bhat (1999: 182} suggests different paths of grammaticalization:
“[Languages that give greater prominence to aspect than to tense develop
a perfective form from an earlier perfect construction and an imperfective
form from an earlier progressive construction, whereas languages that
give greater prominence to tense than to aspect develop past and present
forms directly from their perfect and progressive constructions respec-
tively.’ As Bhat further maintains (pp. 182-3):
(13) ‘aspect:prominent and mood-prominent languages show distinct
tendencies of change when they develop temporal distinctions.
In the case of aspect-prominent languages, we generally find a two-
way pastjnon-past distinction or a three-way past-presentfuture
distinction developing from an earlier perfective-imperfective dis-
tinction, which correlates with the fact that from an aspectual point
of view present and future appear as imperfective and past as per-
fective. For example, the Semitic languages had originally an aspec-
tual system, which distinguished between perfective and
imperfective events, but in some of the modem languages this has
changed into a temporal system in which the distinction is between
past and present ... In the case of mood-prominent languages, on
the other hand, the general tendency is to develop primarily
a future/non-future distinction.
(hat 19997 182-3)
While Bhat (1999) saw different degrees of ‘prominence’ of certain cate-
sgories in languages, Chafe (2000) coined the term ‘florescence’. He sug-
gests that, ‘like forests, languages may develop toward a climax stage
‘where particular combinations of features, like plant communities, may
flourish to define a particular language type. I think itis useful to think in
terms of the florescence of linguistic features in this sense - the flowering of
features that come to dominate the form a language takes’ (p. 39). Some
features of Troquoian languages serve as examples for this florescence.
First, oquoian languages (especially Northern Iroquoian) have extremely
elaborate pronominal prefixes that may lead to an inventory of 60~70
prefixes, Besides singular, plural and dual number distinctions in the
first and second persons of both agents and patients (which are also
differentiated), there is even an inclusive-exclusive distinction in
Cherokee, and the addition of gender distinctions with thitd persons (see
Chafe 2000: 41). Decisively, this elaborate pronominal prefix system is
described (reconstructed) by Chafe as the result of successive grammatica-
lizations. Secondly, Iroquoian languages are characterized by nounTypology and Grammaticalization
wm
incorporation of a wide range of concepts, e.g, animals, foods and body
parts. Even whole events can be incorporated as verb roots are incorpo-
rated into nouns, Again, interlanguage comparison suggests that these
incorporations developed not at once but over time, newer incorporations
following older ones. They also range from productive to fixatedjidiomatic.
‘Thinking of parallels in other, more easily accessible, languages, the
relatively large variety of modal and semi-modal verbs in English or
German may be a good example. South Asian languages also have many
(see Narrog 2010), but other languages have none at all. Historically speak-
ing, we have a wave of emergence of the modals from Old to Middle
English, and a wave of emergence of the semi-modals from Middle to
‘Modern English. The prior emergence of one or two auxiliary verbs first
seems to have drawn others along the same path thereafter, and the
existence of the modals may have induced the later development of the
semi-modals
‘Krug (2000) tries to account for the development of the semi-modals in
terms of a ‘gravitational model’, which operates on the principle that
‘larger masses (in our case highly frequent emerging auxiliaries) attract,
smaller masses (in our case less common constructions’ (p. 226). Based on
frequency and similarity (a qualitative measure), he calculates to which
extent specific semi-modals are influencing others (p. 232). Changes in
a field of related items can be easiest observed in phonology (e.g. want to>
‘wanna; go to > gonna), but actually ‘extend to all basic linguistic levels i.e
phonology, morphology, syntax and semantics’ (p. 227).
More generally and more traditionally, it may be possible to describe
such changes in terms‘of ‘analogy’. Grammatical items and constructions
‘with low frequency change in analogy to high-frequency items and con-
structions. Type frequency determines the influence ofa group of items on
other items outside that group, as already observed very early by Paul
(1891: 100) (‘But all that part of language which lacks the support of an
environing group. or which enjoys it only in a limited measure, proves,
unless impressed by repeated usage intensely upon the memory, not
strong enough to withstand the power of the larger groups”, or later by
Bybee and Thompson (1997: 384) ("high type frequency ensures that
a construction will be used frequently, which will strengthen its represen
tational schema, making it more accessible for further use, possibly with.
new items’), In contrast, token frequency may determine the item or
construction with the biggest influence within 2 group. In this manner,
the dominant structural and functional types of a language may have
a strong influence on new grammaticalizations.
‘An almost complementary concept to “florescence’ is DeLancey’s (2001)
“functional sink’, which he defines as ‘a function which is important
enough, crosslinguistically, that in language which does not formally
express it with dedicated grammatical machinery. any construction or
lexical means which expresses a related function is a likely candidate for1m HEKONARROS
pee
‘grammaticalization’ (p. 15). For example, adjectives are based on the func
tion of noun modification, which is universal, even if adjectives are not.
Certain nouns or verbs may be drawn into this functional sink, eventually
leading to the development of the new category adjective. The same can
‘occur with clearly grammatical categories. For example, Thomnes (2013)
claims that stch a functional sink is at work in the grammaticalization of
‘causative constructions in Norther Paiute. Causative is a grammatical
function with a high communicative need that is usually available even
in the absence of grammatical means and, because of the frequent need of
expression, attracts lexical materials to grammaticalize.
63.4 Summary
‘We have seen a number of cases where synchronic typological structures
of a specific language may either constrain or guide the direction of
grammaticalization, especially phonological and morphological aspects,
and the categories being grammaticalized. These aspects of grammatical
zation might be considered as ‘epiphenomenal’ in a stripped-down or
highly abstract concept of grammaticalization, but they are nevertheless
of typological interest. As for synchronic universals constraining gramma-
ticalization, we have found no good examples.
6.4 Conclusion
‘Paths of grammaticalization lead to cross-linguistically recurrent language
patterns. Deeper motivations for these paths must be sought in general
Principles of communication and cognition, Also, morphological aspects
of synchronic language structure tend to reflect properties of their gram-
‘maticalization. Reversely, synchronic language structures may constrain
and guide some aspects of grammaticalization. Here, the impact of extant
phonological and morphological structures, and the impact of extant
‘grammatical categories on further grammaticalizations may be of parti
cular interest for further study. This chapter is organized in accordance
with the question how grammaticalization may impact typological fea-
tures of languages, and how reversely, typological features of languages
‘may impact paths of grammaticalization. Under a different perspective,