You are on page 1of 529

Meaning in Mind and Society

Cognitive Linguistics Research


Dirk Geeraerts
John R. Taylor
Honorary editors
Rene Dirven
Ronald W. Langacker

De Gruyter Mouton
Meaning in Mind
and Society
A Functional Contribution to the Social Turn
in Cognitive Linguistics
Peter Harder

De Gruyter Mouton
ISBN 978-3-11-020510-7
e-ISBN 978-3-11-021605-9
ISSN 1861-4132

Library of Congress Cataloging-in-Publication Data

Harder, Peter.
Meaning in mind and society : a functional contribution to the
social turn in cognitive linguistics / by Peter Harder.
p. cm. (Cognitive linguistics research ; 41)
Includes bibliographical references and index.
ISBN 978-3-11-020510-7 (alk. paper)
1. Cognitive grammar. 2. Sociolinguistics. I. Title.
P165.H37 2010

Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data are available in the Internet at

2010 Walter de Gruyter GmbH & Co. KG, Berlin/New York

Printing: Hubert & Co. GmbH & Co. KG, Gttingen
Printed on acid-free paper
Printed in Germany

I have been trying to write this book for about seven years. As usual, it
would never have happened without the encouragement and help of a
number of people. I am grateful to Anke Beck from Mouton de Gruyter
for starting the discussion that led to the book, and to her and Birgit Siev-
ers for remaining patient with me. Dirk Geeraerts on various occasions
discussed key ideas and offered a penetrating critique of a previous ver-
sion of chapter 6. I am grateful to Eve Sweetser for making it possible to
be a visitor at Berkeley at a formative phase of the project.
Without the pragmatics circle, an informal reading group that has
existed since 1977, I would never have discovered many of the perspec-
tives that made it possible to think of the possibility of writing a book of
this kind.
It has been extremely rewarding, as well as a great personal pleasure,
to be able to draw on the expertise of my sons Christoffer (in biology) and
Jonathan (in political science) in addressing the interdisciplinary issues
raised in the book.
Chris Butler, Troels Engberg-Pedersen, Hans Fink, Lars Heltoft, Lis-
beth Falster Jakobsen, Ronald Langacker, Chris Sinha, Ole Togeby, Ib
Ulbk, Peter Widell, Niels Erik Wille and Jordan Zlatev read various
parts of the book and gave helpful suggestions. Niels Davidsen-Nielsen
generously offered to proofread the book in the final phase. I am grateful
to my wife, Birthe Louise Bugge, both for remaining patient and for the
passages she read and discussed with me.
For all I owe to these and unnamed others, the faults remain my own.

1. What this book tries to do  1
2. A summary of the argument  5
2.1. There is no such thing as conceptual frames
(But theres a whole social-cognitive world)  5
2.2. On-line vs. off-line features  8
2.3. Social cognitive linguistics vs. analysis in terms
of discourses  9
2.4. Functional relations and adaptation  9
3. The progression of the book  10

Chapter 1. The heartland of Cognitive Linguistics

1. Introduction  14
2. Conceptualization and concepts  16
3. Frames, domains, and idealized cognitive models  22
4. Embodiment and image schemas. From conceptual
to neural patterns  29
5. Figurative meaning  35
6. Linguistic meaning: Polysemy, ambiguity, and abstraction  39
7. Mental spaces  42
8. Cognitive linguistics and cognitive grammar  47
9. Final remarks  55

Chapter 2. From conceptual representations to social processes:

aspects of the ongoing social turn
1. Introduction  58
2. Cognitivism and conceptualization in the sociocultural
sphere  58
3. Variation, lexical semantics and corpus linguistics  64
4. The developmental perspective: epigenesis, joint attention
and cultural learning  70
5. Extended grounding: situational, intersubjective
and cultural aspects  79
VIII Contents

6. Language as a population of utterances: an evolutionary

synthesis  88
7. Meaning construction  95
8. Final Remarks  101

Chapter 3: Social constructions and discourses

1. Introduction  103
2. The social construction of reality  105
3. Power, habitus, marginalization and discourse:
the French poststructuralists  108
4. The analytic practice: discourse(s) analysis  114
5. Discursive psychology  123
6. Systemic-Functional Linguistics  126
7. Socially based theories of meaning: overview and issues  136

Chapter 4. The foundations of a socio-cognitive synthesis:

social reality as the context of cognition
1. Introduction  138
2. Social facts: objective and subjective, intrinsic
and observer-relative properties  139
3. Niche construction  146
4. Individuals, collectives and the invisible hand  154
5. Functional relations  160
6. Mind in society: causal patterns and the individual  163
7. Summary: the socio-cognitive world  172

Chapter 5. Meaning and flow: the relation

between usage and competency
1. Introduction  179
2. Meaning as process input  182
3. Presupposition and the directional nature
of linguistic meaning  193
4. The procedural nature of competencies  197
5. Usage, competency and meaning construction  209
6. Conceptual categories and the flow. Messy
and precise semantic territories  212
7. Summary  219
Introduction IX

Chapter 6. Structure, function and variation

1. Introduction: the social foundations of structure  222
2. Usage, structure and component units  228
3. Function-based structure  236
3.1. Structured division of labour and arbitrariness
as a functionally motivated property  236
3.2. Slots and constructions: coercion
as construction-internal functional pressure  245
3.3. Functional upgrading: the dynamic dimension
of syntax  251
3.4. The interdependence of the top-down
and bottom-up perspectives  259
4. Norms and variation  269
4.1. Introduction: the interdependence of structure
and variation  269
4.2. Variation and the linguistic system  271
4.3. The role of norms  277
4.4. Norms and individual competency  282
4.5. The social dynamics of linguistic variation  287
4.6. Usage, structure and variation in an evolutionary
framework: a discussion with Croft  290
5. Summary: function-based structure and langue
in a social cognitive linguistics  298

Chapter 7. Meaning and social reality

1. Introduction  303
2. The growth and structure of social constructions  307
2.1. The bottom-up trajectory. From construals
to social constructions  307
2.2. Acceptance and efficacy: the evolutionary dynamics
of social construction  310
2.3. Habitus vs. conceptual models:
do we really need mental representations?  314
2.4. The Platonic projection: concepts as part
of the niche  318
3. The role of acceptance  326
3.1. Beliefs as social constructions: causal power
and grounding  326
X Contents

3.2. The interface between niche and flow in social

construction(s)  332
3.3. Bullshit: function and factual grounding  339
3.4 Re-conceptualization and social reality  346
4. Discourses analysis and social cognitive linguistics  354
4.1. What precisely are Foucault-style discourses?  354
4.2. How can discourses be understood in terms
of a social cognitive linguistics?  358
4.3. Where Foucault-style analysis belongs
and where it is inadequate  364
5. Meaning in hard social science:
the case of International Relations  369
5.1. The Copenhagen school of international
relations (CIR)  369
5.2. The need to distinguish between the niche
and the flow  372
5.3. Niche construction and the grounding
of layered identity structure  374
5.4. CIR, agency and factual grounding  379
6. Individual conceptualization and the social constructor  383
7. Cognitively based critical analysis
a comparative perspectivization  390
8. Summary: Conceptualization in society  403

Chapter 8. Multi-ethnic societies:

discourses vs. social cognitive linguistics
1. Introduction  408
2. The ethnic other: an ultra-brief historical overview  409
3. Social construction and universal humanity  412
4. A social cognitive analysis of the distinction
between us and them  415
5. Where discourses fail: the decline and fall
of the anti-racist discourse in Denmark  421
6. Strategies based on collaborative agency:
platform building and niche (re)construction  429
7. Asymmetric communication and the pathological they:
the niche for discourses  436
8. Final remarks: why critical analysis needs functional relations
and collaborative agency  438
Introduction XI

Chapter 9: Summary and perspectives  443

References  463

Index  505
Meaning in Mind and Society.
A Functional Contribution to the Social Turn
in Cognitive Linguistics


1. What this book tries to do

This book was undertaken with two purposes in mind, one academic and
one civic.
The academic purpose is to describe and contribute to the process
whereby cognitive linguistics is expanding to include the social side of
language and meaning. This development is one aspect of an even broader
intellectual challenge for the 21st century: cognitive science successfully
integrated a number of disciplines, including linguistics, in an umbrella
discipline to study the human mind but the very success of that endeav-
our has now carried it from a beginning where cognition was viewed as an
autonomous domain (the brain in the vat) into the study of cognitive
processes in society. Since there is no umbrella social science no soc-
sci that expanding cog-sci can team up with there is no easy blueprint
for how to take this step. No matter what ones preferred approach may
be, however, it will have to take the role of language into consideration
which makes it an exciting challenge for a language person.
Even more important than the academic motivation, however, is the
civic purpose. Recently, a prominent spokesperson for critical Muslims in
Denmark opened a debate on immigration by saying everything begins
with language and went on to argue that divisive ways of speaking were
at the root of the problems. Cognitive linguists would tend to disagree
with this statement, pointing instead to cognitive models in the mind.
Most ordinary people (and politicians) would ignore both and point to
social realities as they see them. The academic community can offer no
obvious way of making these different perspectives cohere. The book tries
to achieve its academic purpose in such a way that it can address this gap.
The specific focus of this book is expressed in the word functional. As
I use the term, cf. also Harder (1996), it refers to relations between a
dynamic object of description and the context more specifically the type
2 Meaning in Mind and Society

of pattern in which feedback from the environment helps to shape, pro-

mote or undermine their continuing role. I try to show that a full under-
standing of meanings must include an account of such functional relations.
This includes feedback from all relevant factors, including intersubjective
understanding and non-mental aspects of the way the world works. In
order to understand what for instance security means, ongoing feedback
across the whole spectrum, from the individual experience of being under
threat up to international relations, needs to be part of the framework.
The lengthy subtitle expresses the trajectory that the account follows:
the subject is Cognitive Linguistics (= CL); it describes the ongoing devel-
opment in CL that I describe as the social turn and it suggests that a
functional approach can add an essential dimension to it.
The topic of meaning in society has become increasingly focal for many
interconnected reasons in linguistics, in cultural studies, in organization
theory and management studies, and last but not least in politics, where
professional operators in the form of spin doctors have proliferated in the
last decade. It is becoming more and more important that those of us whose
fields involve meaning equip students as well as we can to understand and
engage with social processes of meaning creation and proliferation.
The problem is not that nothing is being done: there is a plethora of
different approaches on offer, and Cognitive Linguistics has done its
share e. g. Chilton (1996, 2004); Lakoff (2004, 2006, 2008); Kristiansen
and Dirven (2008). However, the book is based on the conviction that
both the academic and the civic issue have a shared and unsolved problem
in the present intellectual landscape, in that there is no clear answer to the
question: how does the analyst manage to get a full and integrated picture
of cognitive and social aspects of the topic of meaning in society?
The words full and integrated are crucial in relation to the academic
analysts civic obligations: if you go solely for those specific aspects that
constitute your special interests (personally as well as professionally) and
leave the rest to others, the field will be a prey to competing factions. This
may be okay or inevitable from an academic point of view; active research
never deals with more than a small part of the truth anyway. But the war-
ring half truths on the academic side tend to team up with warring inter-
ests in society leaving the field as a free-for-all.
This prevents the academic community from serving civil society as
well as it should as illustrated by the ethnic issue discussed in chapter 8.
An important factor is the influential academic approach that takes
discourses as the fundamental object of description. In this approach, the
free-for-all has the final word: the world consists of a cauldron of ongoing
processes of meaning creation that are caused by, and simultaneously caus-
What this book tries to do 3

ing, other social processes of the same kind. Since the distinguishing fea-
ture of this heterogeneous collection of approaches is that it operates with
the plural form discourses, I use the plural attributively to avoid confusion
with the uncontroversial non-count singular: a discourses approach is
something much more specific and problematic than a discourse approach.
From an individual perspective, the discourses perspective has the
attraction that it allows you to be the founder and sole proprietor of your
own local processes of meaning creation. Significantly, however, it is also
attractive to power holders who want to make sure that such processes
work so as to promote their interests. The claim that meaning is detached
from all foundations beyond the immediate process was originally put for-
ward by critical intellectuals, who wanted to tear the mask from estab-
lished interests parading as ultimate reality. Now it has become common
property, which means that those who have more power use it more effec-
tively. The civic position does not stand much of a chance if the winner
takes it all, also when it comes to meaning in society. That makes it worth
while looking for a different approach.
The book tries to show that a social cognitive linguistics can serve the
civic purposes I have described, and make the academic and the civic
agenda go hand in hand. Two basic features of the CL approach are essen-
tial arguments for thinking so.
First of all, as pointed out by Geeraerts (2003b; 2007), CL is inherently
oriented towards recontextualization. What created and continues to
unite the whole CL enterprise is the movement of going behind language
to set it in a wider cognitive context. What I call the social turn can be
understood as a new operation of the same kind: language-and-conceptu-
alization needs to be set in the wider context of meaning-in-society1.
Secondly, in the context of CL, the issue of foundations has been recast
in terms of grounding, which is an attractive way of thinking about the
relation between a focal object of description and the context in which it
belongs. In classic CL, the central form of grounding is bodily grounding
(cf. Johnson 1992). This dimension retains its crucial role, but in a social
cognitive linguistics, grounding also includes the anchoring of meaning in
feedback from the environment, outside the individuals body. This is in
keeping with the broader agenda in CL of experiential grounding.
Grounding contrasts on the one hand with dogmatic foundationalism,
where everything is rigidly determined at some basic level, and on the
other with the deconstructionist detachment of meaning from all founda-

1 This is a continuation of the agenda of Harder (1999) and Sinha (1999).

4 Meaning in Mind and Society

tional moorings. Where exactly between these extremes the description

ends up is an empirical matter. In this, it reflects the same point as the
concept of partial autonomy: certain facts stand on the shoulders of other
facts, which means that they depend on them without being reducible to
The scope of the book calls for a comment. It may appear that I am
trying to tell everybody in the social sciences and the humanities how to
do their jobs. But the academic purpose is actually quite specific: to show,
from the point of view of cognitive linguistics, how meaning as a feature of
individual minds is woven into the larger fabric of meaning in society.
Unfortunately I cannot address this issue without having to make a con-
siderable number of assumptions about meaning, language, human minds
and societies. While it does not quite amount to the universe and other
related matters, it may be a little too close for comfort.
One thing I clearly owe the reader is a definition, or an account (if
definition sounds a little too Aristotelian!), of what I understand by mean-
ing, so that it becomes clear how that entity can be both in mind and soci-
ety. As I understand it, meaning presupposes conscious experience, but
only experience understood as associated with a vehicle counts as mean-
ing. The most basic vehicle (cf. Sinha & Rodriguez 2008: 36468) is a mate-
rial object, such as a cup or chair which are meaningful entities to mem-
bers of communities in which they are associated with certain types of
experience. In this simplest case, meaning is a side effect of the objects
role in a form of life.
With the rise of signs, meanings acquire independent status in relation
to their vehicles. In human languages, this status gets its most sophisti-
cated manifestation. With the development of languages, the human form
of life becomes dependent on collective recognition of meaning, and that
in turn brings about a proliferation of objects whose causal powers depend
on what meanings they have (thus superimposing a new level of complex-
ity on the issue of meaning).
Of special importance in this book is the functional perspective (cf.
ch. 4). In the trajectory of meaning, it begins with the functions of objects
and continues via the functions of linguistic expressions to functional
relations between meanings and social structures (cf. also Zlatev 2001).
But the perspective extends beyond linguistic meaning to the general
issue of how the functional dimension interacts with the cognitive dimen-
sion in understanding what goes on in institutional, social and political
processes. Functions work at all levels, not only those that involve mean-
ing and the specific role of meaning needs to be understood in this
larger perspective.
A summary of the argument 5

The functional dimension of meaning-in-society constitutes the basis

of the specific contribution this book has to offer. Thus the book does not
pretend to give an equal and full account of all aspects of ongoing socially
oriented work in the cognitive tradition (which would also take more than
one book). It aims to describe the types of development that together
constitute the social turn, and show what a functional approach has to
contribute to it.
To sum up: this book tries to show how cognitive linguistics is expand-
ing from the classic version predicated on conceptualization towards a
social cognitive linguistics that grounds conceptualization in its social con-
text, and to show how a functional approach can provide the extended
foundation that this development requires. In doing so, it tries to show
that this will provide an approach to meaning in society that can do justice
simultaneously to the embodied experience of the individual and to social
For obvious reasons, the book does not try to give all the answers that
one might want such an approach to provide. What it does try is to show
that the approach addresses the relevant questions. Hopefully, it gives
readers a glimpse of what is missing in some of the partial answers, and
prevent these from turning into reductive half-truths.

2. A summary of the argument

The overall question is: how should Cognitive Linguistics (= CL) expand
in order to be an adequate framework for describing meaning as part of
social reality?
Some of the central parts of the answer are introduced in list form

2.1. There is no such thing as conceptual frames (But theres a whole

social-cognitive world)

A very brief summary of this book is to say that what a cognitivist sees as
conceptual frames is the tip of an iceberg that constitutes the whole social
universe, and a frame-based theory of meaning needs to broaden out to
encompass this perspective. In order to understand cognition-in-action, a
social cognitive linguistics needs an account of the social grounding of
meaning including relations between cognitive and non-cognitive
dimensions. This in turn involves the following issues:
6 Meaning in Mind and Society

(a) First of all, it requires a format for describing social facts as distinct
from cognitive facts: the book therefore presents and argues for such a

(b) A central claim about that format is that it involves various forms of
interaction between meaning in the individual mind and meaning in the
This is less trivial than it might appear at first glance. To say that there
is meaning in the environment, just as there is meaning inside the mind, is
one of those ideas that are sort of obvious but have not been given a clear
and consensual descriptive format. Cognitive Linguistics is basically pred-
icated on putting meaning inside the head (cf. Grdenfors 1998: 21), and
two of the authors I build on in moving into social territory also locate the
essential mental elements inside the mind of the individual. Searles (1995:
26) definition of social facts is based on a we-intention inside an indi-
vidual mind; and Croft (2000: 111) defines meaning as something that
occurs in the interlocutors heads.
In contrast, I place the individual agent in an interface position between
meaning emerging from within the body and meaning impinging from
outside. The approach that enables me to do so is entirely unmysterious
and implies no assumptions about a collective mind existing independ-
ently of individuals. Basically, it is a matter of levels-of-analysis. I do not
dispute what Searle and Croft are claiming the individual level just does
not capture all there is to say (as they are the first to point out in other
The rationale has two steps. The first can be illustrated with the proper-
ties of a traffic jam as opposed the properties of a car. Even if a traffic jam
consists entirely of cars, it has not just additional but apparently contradic-
tory properties; thus the location of a traffic jam that starts if two lanes out
of three are suddenly closed will quickly get extended backwards on a
congested freeway, although all the cars in it separately move forward.
And if we identify a traffic jam with the individual cars it consists of, how
can it be that the traffic jam persists while individual cars escape?
This issue reflects a basic ontological point that is related to Russells
theory of types (the implications of which were central to Bateson 1980).
At the basic level where the issue is the existence of assemblies with prop-
erties that differ from those of the individual instances, meaning-in-soci-
ety as understood in this book includes the traffic jams of meaning in the
human environment, an example being a rumour that arises and prolifer-
ates accidentally: it may have to be taken seriously while it lasts, but may
also disappear without a trace. But although I will occasionally refer to
A summary of the argument 7

this level of collective meaning, the focus is on more specific and struc-
tured forms of collective, meaning-imbued phenomena. These depend on
the uniquely human ability to engage in joint attention (cf. Tomasello
Joint attention (as part of joint activity) means that mental as opposed
to merely behavioural properties of others acquire the status as parts of
the world I live in. The most fundamental meaningful collective entity in
the human world is we understood as a group of which human indi-
viduals are members, and which is not reducible to the sum of its parts. In
other words, it is not the case that the we inherits all its mental properties
from properties of the individuals there are also mental, cognitive prop-
erties that the individuals have because they are members of a we. Evo-
lutionarily speaking, human beings are adapted to a world where it is
meaningful to be members of such collective wes and all other forms of
collective meaning are differentiated and specialized forms of the basic
we-type of experience. At this more advanced level, meaning enters into
structured and constitutive relations with the social world (cf. especially
ch. 7).
A special methodological difficulty is that when individuals relate to
meaning in the social and cultural environment, they do so by means of
representations in their individual minds there is no other form of access.
This means that much of the time it is difficult to know when you are deal-
ing with the content of your own mind and when you are talking about
meaning in society. But this problem has to be addressed rather than
denied. Cognitive linguistics can only successfully expand from the mind
into society, if it can tell the difference between what is going on in an
individual mind and what is going on in the social world.

(c) An account of meaning in society involves the causal interplay between

mental and non-mental factors. Not everything can be captured in terms of
mental facts. In the social environment, part of the way meaning operates
depends on causal patterns at the collective level.
An example is the proliferation of concepts across a community: the
spreading awareness of this seasons fashionable colours occurs outside
the individual and is interwoven with the proliferation of non-mental
objects such as this years fashionable clothes. Also inside the individual,
there is interaction between non-mental facts (including neural wiring)
and mental representations (cf. colour-blindness). The two sides interact
because impact from the environment affects both mental and non-men-
tal aspects of the individual (cf. ch 5).
8 Meaning in Mind and Society

2.2. On-line vs. off-line features

An adequate account of meaning in a social perspective needs to provide

a theory of the relation between the process (online) side and the product
(offline) side. This calls for a difficult balancing act, based on the following
two assumptions:

(a) The process or flow dimension is the basic and fundamental aspect
of social reality (as widely recognized after Wittgenstein, e. g. Toma-
sello (2008: 34243), citing Searle (1995: 36).

(b) There is more to social reality than facts about online processes
most obviously because there are constraints on the flow which are
due to factors outside it.

In relation to point (b), it is significant that interest in the product side of

social structure has gone out of fashion since the Marxist 1970s. As part of
resetting the political agenda, Margaret Thatcher once claimed that there
is no such thing as society and she is not the only one who thinks so.
Believers in free enterprise have always emphasized the sovereign indi-
vidual, and left-wing interest has by and large shifted to the issue of differ-
ent discourses.
The difficulty of maintaining the balance between (a) and (b) is partly
due to the fact that if you pursue what you think is the most interesting
dimension of the topic, you can very easily slip into thinking (or saying
things that presuppose) that this is the only dimension that needs to be
pursued. This happened when structure was discovered, and structuralist
pioneers went on to say that structure was the only thing that existed. At
present, the same risk is observable in the movement towards the flow
dimension. In linguistics, it takes the form of what I call usage fundamen-
talism, cf. the discussion in ch. 6. Elsewhere, it takes the form of the bot-
tomless semiosis, cf. the discussion of Derrida in ch. 3.
The book makes a point of understanding offline features in the flow
perspective, rather than in the traditional Platonic perspective where only
eternal features are real and flow is ephemeral. In chs. 5 and 6 it shows
what implications this has for understanding both linguistic meaning and
linguistic structure but in so doing, it also makes a point of showing that
offline features still need to be taken into account.
A summary of the argument 9

2.3. Social cognitive linguistics vs. analysis in terms of discourses

The book argues that a social cognitive linguistics is more adequate

approach to meaning-in-society than the discourses approach. This claim
can be broken down into three subcomponents:

(a) Non-agentive and conflictive aspects of meaning-in-society cannot

be properly understood as an autonomous domain. Although such
aspects exist and need to be addressed, approaches based on conflict-
ing discourses are at risk of producing dangerous half-truths about
their objects of description.

(b) A social cognitive linguistics (that integrates the process and the
product dimension) provides a framework that includes the founda-
tional role of agency and cooperation as part of the necessary context
for understanding impersonal and conflictive processes

(c) An analysis that combines cooperation with impersonal causal

pressures enables an ecologically valid approach to the normative
aspects of discourse conflicts. This will also enhance the critical poten-
tial of the analysis in comparison with the discourses approach, where
the norms are only in the eyes of the beholder.

2.4. Functional relations and adaptation

The book argues that functional relations are an essential part of the off-
line causality that structures meaning-in-society (including its linguistic
encoding). This claim builds on an extension of evolutionary dynamics
from the biological to the sociocultural time scale, whose basic features
are taken over from Crofts version (2000; 2006). In comparison to Croft,
there are two main differences: first, this book places greater emphasis on
the (off-line) concept of function, while Croft focuses more on the online
flow of usage; and secondly, it follows Andersen (2006: 59) in arguing that
there are significant differences between the transmission of genotypes
and of cultural traditions (the main difference is captured under the slo-
gan the visible hand). On both points, this book therefore assigns a
greater role to the functional dimension.
Functionality operates on both the individual and the collective level:
10 Meaning in Mind and Society

(a) Functional patterning constitutes an extra descriptive dimension

superimposed on the conceptual system. The linguistic as well as the
general conceptual competency in the individual (cf. chs. 4 and 5) are
partly shaped by functional pressures.

(b) Functional patterns resulting from adaptation in the individual can

only be understood in relation to the environment to which they are
adapted. This implies a successor concept to Saussurean langue
understood as language-in-society. In social cognitive linguistics,
langue is constituted by affordances in the sociocultural community
for using linguistic expressions to convey conventional meanings (cf.
chs. 4 and 6).

Affordance is a term borrowed from Gibson (1979) to cover factors that

are available in the environment for individuals to use (or not use). Lan-
guage as an affordance exists whether the individual taps it or not; if other
people in the community speak a particular language this constitutes a
potential for making contact with them, which an individual may or may
not make use of. The split between competency and langue is reflected in
a double dissociation between them: the individual may (accidentally or
on purpose) adopt a phrase for which there is no affordance in the com-
munity; conversely, there can be a conventional expression in the commu-
nity to which an individual has failed to adapt: no individual competency
is likely to tap the entire langue (which includes the whole variational
range of resources available in the community).

3. The progression of the book

Part One (chapters 14) lays the foundations, while Part Two (chapters
58) presents the main argument and the conclusions:

Part One:
Chapter 1 describes what I call classic CL: language as conceptualization.
Chapter 2 describes aspects of the ongoing social turn to which this
book aims to contribute. The main ongoing developments can be summed
up in the words variation and intersubjectivity. Among the elements I
build on are Sinhas theory of epigenesis, Tomasellos account of joint
attention and the evolution of communication, Verhagens inclusion of
the addressee in the grounding scene, Clarks account of joint action,
Geeraerts approach to language variation, Zlatevs exploration of evolu-
The progression of the book 11

tionary stages in sign use and Crofts application of evolutionary mecha-

nisms to language as a panchronic object in social space.
Chapter 3 describes the salient competitors to an extended CL as a
framework for describing the social dimension of language, focusing on
the discourses approach.
Chapter 4 discusses the nature of social facts, concluding with the ver-
sion that forms the basis for integrating the conceptual and the social
dimension. A central element is the trinity of flow, competency and langue
as the new, complex object of description for social cognitive linguistics.

Part Two:
Chapter 5 begins with the nature of meaning in the new perspective,
stressing its dynamic nature as a contribution to interaction (rather than
something residing in an underlying, Platonic landscape). It argues, how-
ever, that it is necessary to take account of both offline and online dimen-
sions of linguistic meaning, also in the new perspective. The offline dimen-
sion (viewed as an aspect of language ability or competency) is
functionally shaped so as to constitute potential-for-action, while the
online dimension takes the form of meaning construction in context.
Chapter 6 is about language structure, and argues for an integrated
account of (a) functional and cognitive dimensions, and (b) variation and
(a) Structure has two interdependent aspects: the bottom-up unit-ori-
ented conceptual dimension (reflecting a competency to fit an inventory
of units into whole utterances), and the top-down, recipe-for-action func-
tional dimension (reflecting a competency to handle formats for whole
utterances, saliently including clausally structured utterances).
(b) From the langue perspective, linguistic affordances in society reflect
a number of variational dimensions to which individual competencies are
differently adapted but all variants constitute potential sources of selec-
tion pressure for individuals who are exposed to them, and all are there-
fore equally part of a post-Saussurean, social cognitive langue. In contrast
to more radically variationist approaches, the main point is that varia-
tional linguistics presupposes structural identification of units. Variational
description is stage two in terms of descriptive adequacy, one level higher
than structural description, and not to be confused with a level zero
account simply in differences: saying that [t] and [tT] are variants in a
speech community is much more interesting than merely saying that they
are different.
Chapter 7 is about meaning-in-society. A major point is that while
meaning feeds into the construction of social reality, the results of the
12 Meaning in Mind and Society

construction process also involve other factors than meaning including

aggregate-level causal patterns.
Social constructions are thus part of social reality, not just figments of
imagination, and give rise to adaptive pressures on members of the com-
munity, involving both mental and non-mental aspects: to understand how
the law works gives selective fitness, but the law applies whether you
understand it or not. Sometimes mental aspects predominate (acceptance-
heavy social constructions, such as world knowledge), sometimes hard
facts (efficacy-heavy social constructions, such as the military).
Awareness of the different sides of social constructions is essential in
order to be able to capture not only the collaboration between causal and
conceptual dimensions but also the divergences. Among examples of col-
laboration is the Platonic projection, where concepts are endowed with
causal power in the community; among examples of divergence is bull-
shit (which prompts the formulation of a concept of factual grounding).
Functional relations once again have a partially structuring role, giving
rise to differential replication of social constructions depending on feed-
back from social (and material) reality, which determine whether social
constructions are stabilized or eroded over time.
An integrated account of social and mental facts makes possible a cru-
cial distinction between two senses of the word concept: competency
concept vs. niche concept. Competency concepts are internal to individ-
ual minds and enable individuals to grasp what is around them. Niche
concepts are socially entrenched ways of categorizing things, i. e. of sub-
suming aspects of social reality under particular conceptualizations
which determine how instantiations are treated in the community. The
two sets of concepts do not necessarily correspond: we live in social worlds
that put things in different containers, and each of us also develops an
individual competency to put things in containers but only in a mechan-
ical automaton would there be complete one-to-one correspondence
between niche concepts and competency concepts.2
Chapter 8 discusses the multi-ethnic society, arguably the most impor-
tant challenge for understanding meaning-in-society, and argues that the

2 In terms of the philosophical dilemma of internalism vs. externalism, this posi-

tion reflects a basically externalist position inspired by Burge (cf. Burge 1989:
181) and Putnams pragmatism (cf. Putnam 2001:1920). The main aim with
the distinction (rather than the more philosophical aim of conceptual clarifica-
tion), is descriptive: to make room for concepts both as possessions of indi-
vidual minds and as part of the world they adapt to.
The progression of the book 13

social cognitive framework proposed avoids the potentially damaging

limitations of analyses based on the discourses approach.
Chapter 9 briefly sums up the argument and illustrates the interplay
between classic-cognitive-linguistic and social aspects of the analytic
Chapter 1. The heartland of Cognitive Linguistics

1. Introduction

This first chapter is an overview of classic key positions and results in Cog-
nitive Linguistics (= CL). I hope to give a sense of both the unity and the
richness of the CL achievement: CL has been able at the same time to pro-
vide something new and distinctive and to introduce a broad and varied
range of phenomena into linguistics.
To understand the broad impact of CL, it is important to keep in mind
its rise as an alternative to formal generative grammar. Where generative
grammar is based on a rigid distinction between core linguistic properties
and everything else, CL is in its basic orientation a recontextualizing move-
ment, cf. Geeraerts (2003b, 2007). The central purpose that unites cognitive
linguists of different persuasions is to describe language as reflecting
human experience. Although this would not have surprised traditional
grammarians, it was a new start for linguistics as a science: according to
well-entrenched dogma, human understanding was too vague and uncon-
trollable to be admitted into scientific description hence the need for at
rigorously formal descriptive procedure, and for an anchoring in objective,
mind-independent facts.
The most prestigious manifestation in science of this approach was the
role of the formal structures of mathematics as the language of physics.
Formal linguistics hoped to achieve a similar success by distilling the basic
formal core out of the human language ability. Once the basic formal struc-
ture was found, so the reasoning went, actual language events would be
just as trivial from a linguistic point of view as the actual event of an apple
falling downwards would be for physics after Newton had distilled the
basic law out of it.
A centrepiece of the hard-nosed scientific position is that when lan-
guage is used in a controlled manner to talk about the real world, the concep-
tual level can be eliminated as epiphenomenal, i. e. as the unnecessary mid-
dle man. Formal grammar adopted a similar attitude to human understanding:
linguistic structure naturally had to map on to objective content in some
way, otherwise language would be irrelevant but worrying about how
actual speakers made the connection would only be a source of confusion.
The idea of allowing the full richness of human conceptualization in at
the front door is therefore truly a new step. It puts into well-deserved focus
Introduction 15

the domain in which language most immediately belongs, and also intro-
duces a way of thinking about language according to which it is neither
distinct from mental content nor forced to be in lockstep with it. Instead,
language is seen as a way of accessing mental, cognitive processes and
representations a window on the mind. In order to describe language in
that light, it was necessary for the new cognitive paradigm to put on the
map the whole range of cognitive resources that language could draw on.
Finding out about what the mind could do, as a prerequisite to finding
out how language used what the mind could do, was an exciting opening,
but at the same time a suitably constrained task. It also constituted a new
stage in the larger enterprise of cognitive science. Instead of moving in a
cautious and hamstrung manner, controlled more by ideas of what was
scientifically kosher than by insights into the object of description, cogni-
tive linguistics for the first time threw the door really wide open: let us
look at everything the mind can do, and stop looking over our shoulders
worrying about the science police.
When the tables were turned by CL, the whole idea of disembodied
objective understanding was rejected and the conceptual dimension was
reoriented away from reflecting objective reality towards the roots of con-
cepts in human life. Instead of being a weakness, the experiential and bod-
ily basis of understanding became a new foundation. Whatever human
understanding might turn out to be, it had to be based on the stuff that
goes on in a human body. Therefore embodiment or bodily grounding
arose as a key dimension in the landscape of conceptualization as under-
stood in CL.
This constituted an even more radical new departure. If the human
mind used to be a potential source of error, so much the greater was the
fallibility factor associated with the weaknesses of the flesh. This difficulty
illustrates a faultline in the role of the human mind in hard-nosed posi-
tivist thinking. On the one hand, the mind is the home of reason and
knowledge; on the other hand the mind is an object of empirical investiga-
tion. Behaviourists had declared the mind out of bounds to scientific
investigation because they believed nothing scientific could be said about
it. In order for that to make sense, however, they had to presuppose that
one could cordon off a mental sanctum of unpolluted truly scientific
knowledge. Yet if you do not recognize the existence of reliable properties
of the mind, it is hard to understand how the mind can host the inner sanc-
tum of knowledge.
First-generation cognitive science, including formal linguistics, revolted
against behaviourism by showing that strict formal simulation allowed for
investigation of the mind without polluting scientific knowledge with the
16 Chapter 1. The heartland of Cognitive Linguistics

kind of mush that behaviourists were afraid of. When Searle (1980) showed
that the strictly formal-computational approach to meaning was untena-
ble, his outraged opponents in the discussion talked about the threat to
the cathedral of science: if the human mind is cut loose of its science-
imposed shackles, who knows where that will lead to? The historical mis-
sion of CL was to look for the language-related answers to that question.

2. Conceptualization and concepts

CL understands meaning as conceptualization, understood broadly as the

stuff that mental processing is made of (cf. Langacker 1987: 5). In the gen-
erous spirit that replaced the hard-and-fast distinctions of formal linguis-
tics, conceptual representations were seen as part of a continuum all the
way down to the most basic mental level, which is generally assumed to be
felt subjective experience.3 From the point of view of language, however,
the kinds of mental processes and structures that have representational
content remain central and in this book the term conceptualization will
be understood as having representational content.4
For linguists, the special interest in representational content has to do
with the constitutive role of symbolic content in language; but representa-
tion is not relevant solely for language. It involves the basic relation
between mental content and experience of the world: taking perceptual
representations as an example, the mental image of a tree has an inten-
tional relation to the tree of which it is an image (cf. Searle 1983).5

3 The point was made by Nagel (1974) in an article entitled What does it feel
like to be a bat?. If we take bats to have a mind, the most fundamental
assumption we make is to assume that a bat has a subjective point of view, i. e.
that there is a felt quality of being a bat.
4 The reason for this is that non-representational mental states such as pain
have no direct relation to linguistically articulated expressions: the word pain
does not evoke actual pain. Conceptual representations of food or water, in
contrast, have a privileged relationship with linguistic expressions such as food
and water because the ability to evoke representational mental content is what
makes these expressions meaningful.
5 But the direction can also go the other way: once we have stored such a per-
ceptual representation, we can turn it around and use it as a basis for evaluat-
ing actual events: the remembered forest may trigger a reaction against envi-
ronmental destruction. The essential property is the element of duplication,
combined with the intentional relation between the representation and what
Conceptualization and concepts 17

As we have seen, in the logical tradition there was no room either for
subjective experience or for conceptualization understood as a mental
process only for concepts understood as a way for the mind to map pre-
cisely onto categories of the objective world. Classical, Aristotelian con-
cepts were therefore understood in terms of checklists of features (cf.
Fillmore 1975) that translated directly into truth conditions. For instance
the concept of bachelor is defined as an (1) unmarried, (2) male, (3) adult
(4) human, and conceptual classification of external objects (such as
potentially eligible males) can proceed by ticking off these four criteria. If
they are fulfilled, we have a bachelor, otherwise we do not: tertium non
datur there is no third option.
An important step towards the human perspective was the discovery
of differences between good and less good instances of categories, as cap-
tured by the term prototype, cf. a series of seminal papers by Rosch (1973,
1975), Rosch and Mervis (1975), Rosch et al. (1976), etc. Among the cases
investigated are colour categories. Rosch asked people to place a range of
colour chips in familiar categories such as (for English) red, yellow and
blue. It turned out that there was greater agreement on how good exam-
ples they were, i. e. whether they were more or less close to focal red
(roughly = the colour of blood), than there was about whether they were
inside or outside the category red.
Two consequences follow from this: first, there is a gradual transition
between red and non-red. Thus the Aristotelian point-blank division into
plus and minus is simplistic human categorization is subtler than that.
But secondly, because judgements of gradience agree better than judg-
ments about the cutoff point, prototypicality does not necessarily mean
lower levels of precision. Unclear boundaries and grey zones may appear
messy, compared with the tradition but the point-blank alternative
actually gives a less precise picture.
The difference is linked up with the true/false bias in the tradition: it
was implicitly taken for granted that the right categories would not only
serve the purposes of the ordinary conceptualizer, but also and more
importantly the purposes of the philosopher or scientist, whose goal was
to strive for the whole truth (and nothing but). For ordinary people, the
priorities are different: the question is mostly how to deal with the con-
crete matter at hand. For that purpose, focal points are more handy than
precise checklists. This element of focality-cum-gradualness was therefore

it represents. In combination they constitute the representational capacity

without which language would not be possible.
18 Chapter 1. The heartland of Cognitive Linguistics

a central point in struggling free of the rigidity of the logical view of mean-
ing that had occupied the scene for almost 2500 years.
Roschs findings also undermined the structuralist claim that it is only
the language that defines the content of a concept, while pre-linguistic
meaning is totally vague (cp. Hjelmslev 1943: 69, Lakoff 1987: 267): it
turned out that (after some initial training) the New Guinea Dani, who had
only two basic colour terms, one for dark and one for light, gave responses
to colour tasks that agreed to a surprising extent with Anglophone
informants e. g., about degrees of similarity with focal red. The conclusion
was that the level of human experience in this case colour experience
was more fundamental than the level of linguistic distinctions.
CL generalized this finding into a theory that the central concrete
foundation of experience-based categorization is sensory experience, i. e.
perception (cf. Evans and Green 2006: 7). One of the pioneers of CL,
Leonard Talmy (2000: 139), coined the term ception in order to stress the
continuity between perception and conceptualization. On a philosophical
level, this reflects an empiricist orientation in CL, in opposition to Chom-
skys rationalist orientation; there is a faint echo of John Locke saying
that there is nothing in the mind that was not first in the senses.
Colour concepts are also good examples of the bodily basis of concep-
tualization: we see blood as red and grass as green because this is the
experience of blood and grass that our kind of nervous system generates.
How it works is also scientifically well understood, and common knowl-
edge, because some people are genetically wired to see the world differ-
ently; they are, as we say, colour-blind. By the same token, it shows that
conceptualization is not a direct reflection of the objective properties of
the world out there. We only see colours the way we do because we have
a human body with certain properties, not because the world is inherently
red or blue. Moreover, we see things differently depending on the context:
the colour orange is midway between yellow and red, and so an orange is
more red than a lemon, while being more yellow than a (ripe) tomato. The
fact that it can be (truly) described by both linguistic predicates simulta-
neously, while it would still be true to say that it is neither red nor yellow,
shows that application of concepts to referents is not rigidly controlled by
the inherent nature of the object. Nor is this a rhetorical trick: the inde-
pendent existence of concepts as part of the way the world works implies
exactly this, that objects can be grouped in different ways, depending on
which concepts humans find most adequate for grasping the situation at
One way of capturing this perspectival element in conceptualization is
to say that human categorization reflects what is salient to people, cf.
Conceptualization and concepts 19

Geeraerts (1989). As pointed out by Geeraerts, prototypes are not all

alike, and they are certainly not as simple as this account may suggest.
While colour concepts are difficult to bring under an account in terms of
features, other prototype concepts can be captured by a list that shows not
all-and-only, but mark out the types of instantiations that are central.
The category bird is the classic example. Typical birds have feathers, wings
and tails, can fly and sing, build nests and feed their young, and sit around
in trees. This is why robins and sparrows come out as prototypical, while
chickens, ostriches and penguins are more marginal members of the cat-
As argued in Lakoff (1987), however, prototypicality turned out not to
be in itself an adequate theory of mental structure of conceptual represen-
tations. There are many reasons why the concept of prototype is not
enough to characterize what a mental category is like. The simplest idea,
that a prototype was something that had all relevant properties, does not
explain why peas and carrots come out as the prototype vegetable, cf.
Aitchison (1994: 68; Evans & Green 2006: 26869). Following Lakoff, the
term prototype effects has been used about the two central new and flex-
ible features, graded membership and core instantiations. Both are due
to the way the mind works rather than to fixed mental representations; we
may speak of (proto-)typification as a general feature of conceptualiza-
tion, which reflects its role as part of everyday coping, like the notion of
rule of thumb.
The precise nature of the actual conceptual constructs that we use is
thus something that requires further investigation. One such form of
investigation was already part of Roschs pioneering studies. Setting up a
scale of specificity, she established a hierarchy of three levels, with a cen-
tral level dubbed basic level categories, flanked by more general super-
ordinate categories and more specific subcategories. Basic level categories
encompass objects which play a significant role in the lives of people who
have them, like chair, table, car, in the modern world. Above these in terms

6 As pointed out by Geeraerts (1992; 1997), some of the features of the proto-
type-based approach reflect familiar and traditional assumptions of pre-struc-
tural lexical semantics. Dictionary makers have always known that words have
more central and more marginal senses, and that no single set of criteria would
suffice for each word. As he also stresses, however, the tradition had not pro-
vided a concept that would do the work of the concept of prototype, with its
new conception of conceptual grouping as based on a focal area of meaning
rather than on a bounded category.
20 Chapter 1. The heartland of Cognitive Linguistics

of generality are abstract superordinate categories like furniture or vehi-

cle, and below them subcategories like armchair and sports car.
Basic level concepts were empirically found to be associated with
rich and experientially well grounded conceptual representations. In con-
trast, superordinate categories elicited only a few properties (there is not
much to say about what a vehicle is); and subcategories only a few addi-
tional features beyond those already characterizing the basic level (a
sports car is a car with a few extra properties). Similarly, informants
were found to have motor routines corresponding to basic level objects
like car and chair, but no routines specific to vehicle or sports car. Gener-
alization over shapes, created by using a computer to average out visual
representations, also produced recognizable results at the basic level (for
example in the case of cars), but not at the level of superordinate catego-
ries (if you average out the shapes of a bus, a plane and a bicycle, the
compromise result is not the shape of a vehicle). Basic level categories
thus come out as rich and robust mental constructs, with more or less
built-in prototypicality.
This robustness is due to experiential grounding: people have experi-
ence with handling and recognizing chairs and cars from their everyday
lives otherwise specific motor routines could not arise. Precisely because
of the experiential and empirical basis of such concepts, however, one
mans basic level concept is sometimes another mans superordinate term.
If your form of life requires that you have detailed knowledge of many
different types of tree, then ash and willow may be your basic level con-
cepts (with assorted prototypes), while tree is a superordinate concept like
vehicle; if you are a city dweller, tree may be as far down in the taxonomy
as your everyday experience goes.
The basic level is thus a fairly solid new point of departure for under-
standing the kind of mental representations that real people construct:
down-to-earth, no more precise than required for everyday life, capable of
accommodating a broad spectrum of different cases, associated with prac-
tical as well as conceptual skills. In short, they reflect both properties of
conceptualization as a human skill and properties of its basis in experi-
ence. Basic level concepts are shaped by an economy factor: they end up
at a level of generalization and specificity that balance out costs and ben-
efits of cognitive efforts. In relation to human experience, they also reflect
the patterns of co-occurrence in the phenomena that constitute the input
to conceptualization. In the case of a car, there is co-occurrence of wheels,
seats, headlights, accelerator, brakes, hooting, etc. In the case of birds, the
same applies to wings, beaks, feathers, and chirping. Combinations across
the divide, in contrast, do not abound in everyday life; cars are not gener-
Conceptualization and concepts 21

ally found together with feathers, for instance (cf. Rosch et al. 1976 on
cue validity).
In general a concept, as a countable, individuated construct, represents
the outcome of segmenting the uncountable process of conceptualization
in a particular way. This general feature of dividing up the basically con-
tinuous world of conceptualization applies whether a concept is basic,
subordinate or superordinate, and whether it is understood in the classic
checklist fashion or as involving various forms of gradience. Boundaries
between conceptual categories may take different forms, but their exist-
ence is not in question; Croft and Cruse (2004: 89) rightly suggest that a
boundary is arguably the most basic of all properties of a category. If
concepts are containers, as suggested by Lakoff (1987: 283), we need to be
able to tell the inside from the outside.
The revolution in understanding concepts was also important in rela-
tion to historical change. As pointed out by Sweetser (1990), it had been
traditionally assumed that one could find ancestor concepts by abstracting
over extant present-day concepts. This assumption led linguists to postu-
late very general proto-concepts, like sharp object, as a source for
present-day diverse meanings. However, if the most solidly entrenched
concepts are those that reflect concrete everyday routines, this theory
loses its rationale.
Although concepts thus reflect recurrent features of experience, there
is no determinism involved: grounding involves motivation, not dogmatic
foundationalism. The element of choice is encoded in the term construal,
which is associated with the notion of alternative construals (cf. Lan-
gacker 1987:13841): the mental construct changes while the object of
conceptualization remains the same (as when the human subject shifts her
focus of attention). The relation between a conceptualizing mind and the
object of conceptualization is therefore called the construal relationship
(Langacker 1987: 128).
All linguistic distinctions can be understood as reflecting such alterna-
tive construals. Croft and Cruse (2004, ch. 3) discuss a range of grammati-
cal distinctions as reflecting pervasive differences of construal: leaves vs.
foliage (plural vs. non-count); something moved in the grass vs. there was
movement in the grass (agency vs occurrence); Joe killed Bill vs. Bill was
killed by Joe (from what perspective is the event conceived), etc. The same
applies to distinctions between lexical concepts. The classic examples of
lexical construal include the distinction between coast and shore: both
designate the border between land and sea, but coast views it from the
inland perspective, while shore views it from the sea. From coast to coast
is therefore a trajectory on dry land, while from shore to shore moves from
22 Chapter 1. The heartland of Cognitive Linguistics

one side of the sea to the other (cf. Fillmore 1985). In all cases, alternative
construals are simultaneously available as options and can be encoded in
two different lexical choices, allowing the speaker to impose his own per-
spective on a scene.7
Construal, in short, subsumes all modifications, online or convention-
ally encoded, that are superimposed upon a given input to conceptualiza-
tion, and pinpoints the ability of human minds to tinker in myriad ways
with the stuff they mentally represent to themselves. Construal therefore
reflects the seminal change described in the introduction: the realization
that the human mind has a role of its own, instead of being a source of

3. Frames, domains, and idealized cognitive models

When classic checklists were phased out, CL changed the direction in

which concepts are approached, beginning with larger contexts rather
than minimal features. The Aristotelian and logical tradition was to under-
stand meaning as constituted by component features, as in the analysis of
girl as analysable into the properties [human], [female] and [not adult].
CL instead starts by looking at where a concept belongs in the bigger pic-
ture. This change of perspective is reflected in different ways in the con-
cepts of frame, domain and idealized cognitive model. Frames were intro-
duced and developed in a series of papers by Charles Fillmore (1975, 1982,
1985); domains were introduced by Ronald Langacker (1987); and ideal-
ized cognitive models by George Lakoff (1987).
Frames were first on the scene and have remained central because the
concept of frame semantics is central in demonstrating the basic flaw in
truth-conditional and formal approaches to the study of meaning: mean-
ings depend on presupposed cultural and conceptual foundations in a way
that cannot be captured by analysing a concept into features. In this, the

7 Construal also includes the operations of conceptual coercion and conver-

sion as in there was cat all over the driveway, where the individuated cat is
conceptualized as a mass noun (coercion, because it is done without overt
signalling), or in a loaf of bread, where the unit loaf is explicitly imposed on
the non-count substance noun bread (conversion); it also covers modula-
tion (Croft & Cruse 2004: 128), whereby senses of words are enriched so as to
fit into online understanding without any semantic modification (as when
friend is interpreted as a male friend in the context A friend of Joe married
his sister).
Frames, domains, and idealized cognitive models 23

term reflects the same property that enters into the basic framing prob-
lem in artificial intelligence, i. e. the problem that because computers can-
not by themselves link up with the problems they are used to solve, the
user always has to provide the connection. If you analyse computer opera-
tions without taking into consideration what human beings use them for,
you end up with a string of zeros and ones.8 Two examples will illustrate
how the frame type of context-dependence works in understanding lin-
guistic meaning, cp. Fillmore (1985).
The word Sunday denotes a non-mysterious everyday entity that would
appear to offer no inordinate challenge to semantic description. However,
attempts to specify a list of features that would allow you to recognize a
given day as an instance of the category Sunday fairly obviously miss the
point. In order to explain what Sunday means, you have to take your point
of departure in a wider cultural and conceptual frame within which the
word belongs: the traditional Christian calendar, including the seven-day
week. Without that frame, the word itself could not have the meaning it
A slightly more complex example is the interpretation of what it means
to be a first-class hotel. You might try to understand it by seeing it in the
context of a whole set of other linguistic labels including second, third, and
possibly economy class, but that would not solve the problem. In order to
really understand what first-class means, you have to find out how the
categorization system works in relation to the available range of hotel
experiences, including everything from noise levels to language compe-
tencies of the staff. Once that background is in place, the linguistic context
would also be significant, including the question of how many classes are
higher on the list (luxury class, world class .). Here, too, there is no way
of understanding what first class means without understanding the frame

8 Framing can of course be helped along by specifying links and adding them to
the program. For each step of such a framing operation, you can add an extra
computational operation. But the ultimate framing operation will always
evade the program, receding one step each time: no matter how many specifi-
cations and framing instructions are included, the human operator has to pro-
vide the actual link with the world on her own.
9 The same problem would recur regardless of terminology and definitions.
Teachers will be familiar with marking systems and their descriptions of what
is required for students to obtain each mark and they (we!) know that how-
ever carefully you work out these descriptions, knowing the institutional con-
24 Chapter 1. The heartland of Cognitive Linguistics

Framing is not primarily concerned with the nature of the universe and
other related matters, but with how to fit a given, to-be-interpreted, unit of
meaning into the context. This can be illustrated with reference to a lin-
guistic application that depends on structural slots. Fillmore has used the
concept specifically in relation to one grammatically central form of con-
text-dependence: the ongoing FrameNet project (cp. Ruppenhofer et al.
2006) explores the whole inventory of contextual and grammatical slots
that are relevant to English verbs. In Fillmores theoretical development
this was a crucial step beyond the purely linguistic case frames for which
he became famous, cf. Fillmore (1968): the difficulty in constructing a
coherent and exhaustive theory based purely on syntactic case roles led
Fillmore to expand the foundation to the larger experiential and concep-
tual frames that were evoked by the relevant verbs.10
In Ruppenhofer et al. (2006: 6) the term frame is understood in a wide
sense according to which core nouns also evoke frames. However, it is also
clear that these are less central and can be ignored for the purposes of
annotation; hat or tower does not impose frames on surrounding elements
in the same way as verb meanings do (cf. chapter 6, p. 257). That is because
verbs (unlike nouns) have a core function in defining argument slots. A
verb like sentence, as in they could have been sentenced to five years in
prison, puts the patient in a criminal_process frame (cf. Ruppenhofer et
al. 2006: 126): because it designates a process that takes place as part of the
activities of a criminal court, the argument slot frames the filler as taking
part in a criminal process. The noun prison belongs in the same domain,
but does not in itself trigger a framing operation in the way that verbs such
as accuse, charge, convict and sentence do for the arguments they apply to.
In contrast, when it comes to the more general processes of contextualiza-
tion, there is no difference between nouns and verbs.
Although more specific than contextual understanding, the term
frame can nevertheless be individuated in different ways, depending on
what you are interested in. In the FrameNet project, the purpose is defined
in terms of the characterization of lexical entries: the frames they establish
are chosen so as to allow generalizations about the properties of the verbs.

text (frame) in which they are actually used will still be crucial for evaluating
what a given mark really means.
10 The key problem with case grammar was that the problem of identifying a
language-internal watertight system of deep cases was intractable and this
turned out to be connected with the more general insight on which CL is
based: all structural categories have to be understood in a larger context,
which needs to be included as the foundation of the analysis.
Frames, domains, and idealized cognitive models 25

Ruppenhofer et al (2006: 15) give an example of what counts as the same

frame for that purpose: verbs such as crawl, flit, slither and walk belong in
the same self-motion frame. But, if you wanted to focus on where a type
of motion belongs in the animal kingdom, these words would not neces-
sarily be seen as involving the same frame (slither may suggest a snake
frame, for instance). It all depends on what the relevant slots are for the
unit to fit into. For that reason, frames involve a possibility of misunder-
standing: if you do not infer the right slot, you may frame the phenomenon
in the wrong way; thus Joes girl may denote different persons depending
on whether you invoke the parent or the lover frame.
The concept of domain also involves going behind a concept to its
background. Langackers definition, however (1987: 147f), concentrates
on a somewhat different type of background-dependence than framing as
discussed above. His key example is that of the human body. In order to
understand what a finger is, you cannot define it by looking purely at its
component elements (knuckles, nails). Rather, you need to go to the whole
domain of which a finger constitutes an integral part. Langacker illus-
trates the idea by moving from finger to hand, from hand to arm, and
further on to the whole of the human body. The general idea is that con-
cepts are not atoms: each concept naturally belongs in a larger conceptual
whole. The body example might also be understood as involving simply a
meronymic (part-whole) relationship, but part-whole relationships in
general are occasion-specific (cf. Croft & Cruse 2004: 160): a lake may be
part of a park, but it is not built into the concept of lake that it should
have a park around it. Langackers point, in contrast, is precisely that
there is an inherent relation between specific concepts and the conceptual
areas from which they are carved.11
As you move through the operation of going behind a concept to the
domain out of which it is carved, you gradually get to larger and larger
domains. Some of these can be viewed as basic (cf. Langacker 1987: 148),
in that they link up with directly embodied experience. Because moving
around in space is one such type of embodied experience, space is regarded

11 Part of the same idea is the pair profile and base as understood by Lang-
acker (1987:183f). The notion of arc can only be understood against the back-
ground constituted by the concept circle, and you can describe what an arc is
by drawing first a full circle, and the draw a subpart of it in bold. The circle is
therefore, cf. the definition above, part of the domain within which it belongs,
but because it is the necessary immediate surroundings of the particular con-
cept arc, it is also the base. Other geometrical figures are part of the same
domain, but they do not constitute the base for the concept arc.
26 Chapter 1. The heartland of Cognitive Linguistics

as a basic domain. Langacker discusses, without attempting to settle the

matter, a number of basic domains, some related to specific senses (like
auditory pitch). The human body as the domain for body parts is regarded
as a more abstract domain (p. 150) because it is not grounded directly in
elementary experience but the distinction, like most other distinctions in
CL, is regarded as a matter of degree. One subtype of domain is the loca-
tional type (e. g. temperature and colour), which is constituted so that
each conceptual property can be described by its precise location within
the domain.12
Some concepts belong in more than one domain at the same time, like
book, which is a physical object and also a unit of mental content and
these of course must be described in terms of both domains. One domain
involves things like printing costs, while the other involves topics and
opinions. A set of domains associated with a single concept is called a
domain matrix in the case of book the domain of mental content and
the domain of physical objects are thus linked up in a matrix. Matrixes can
in turn be linked up in even larger structures, in a movement that ends up
facing the issue of the overall organization of human knowledge (Croft &
Cruse 2004: 24). Langackers target in pursuing the concept of domain, as
suggested by Evans and Green (2006: 231), is thus ultimately a conceptual
ontology that includes everything.13
A conceptual ontology is an overall system of resources for under-
standing and categorizing the world, not a precise theory. Part of what a
conceptual ontology contains is a folk ontology, an inventory of catego-
ries that people rely on in their culturally embedded everyday lives. In our
conceptual ontologies, such categories as the soul (cp. Croft and Cruse
2004: 27) are alive and well: you can search your soul, or you can suspect
that some people have no soul, etc. Everyday ontologies are not scientific

12 A mathematical modelling can be set up whereby concepts can be described

geometrically as points in multidimensional space, cf. Grdenfors (2000: 134).
13 When you place each concept where it belongs in the larger conceptual land-
scape, and gradually link up the different landscapes, ultimately you end up
with the totality of the human conceptual world. At the level of the whole
ontology, it may be asked what the difference is between conceptual and
objectivist semantics since an objectivist semantics would also aim at link-
ing up all concepts in a system that constituted an ontology, a theory of what
there is in the world. The difference is basically the same as the one that was
given above in the case of individual concepts. The point is that concepts exist
in their own right rather than passively reflecting the nature of the real world
(whatever that is).
Frames, domains, and idealized cognitive models 27

theories, but they function meaningfully and usefully in categorizing the

business of everyday life, and the philosophical questions of dualism ver-
sus monism, or materialism vs. idealism do not really arise for those pur-
Yet it would also be a mistake to set up an iron curtain between every-
day and scientific ontology. Scientific ontologies are a subcategory of con-
ceptual ontologies in general they are just rather special cases. If you are
trying to make human knowledge cohere at a very general level, the ques-
tion of whether human beings have souls as distinct from their bodies
raises itself for just the same kinds of reason that make you wonder how
to make everyday phenomena meaningful. There is a clear difference,
however, between operating with conceptual domains as used relative to
human experience, and the science-policed use of ontology in positivist
philosophy and first-generation cognitive science.
Frames and domains are essentially contexts for conceptual represen-
tations; they do not focus on the internal anatomy of conceptual represen-
tations. The difference between them is that frames focus on the operation
of linking an element with the contextual slot in which it belongs from
the syntactico-semantic frames of the FrameNet project to situation-spe-
cific interactive frames (cf. Lakoffs use of framing, p. 397). In contrast,
domains focus on the place of a concept in a generic matrix of relation-
ships that is independent of the situational context. This is central to
understanding what makes the term domain useful as a separate term.
Croft & Cruse (2004: 15) regard frame and domain as synonyms, which
makes sense if you do not want to emphasize the distinction between
generic and context-specific meaning. However, it is part of the project of
this book to enable CL to highlight this contextual vs. generic distinction
when it makes a difference, as it does when the social context and mental
categories interact. If you are unfamiliar with the domain, there is some-
thing about the meaning of a term that you simply do not understand, and
your conceptual ontology needs to be expanded. This is why the term
domain does not occur as a verb: domains are conceived as stable entities
with timeless properties, while frames may shift from one moment to the
next in online communication, involving re-framing operations that
speakers need to move along with in order to stay tuned.
After prototypes turned out not to be a generally applicable format (cf.
above), Lakoff (1987) introduced the notion of idealized cognitive model
(ICM) as a format for the conceptual content of the mind. ICMs carry over
from prototypes the element of idealization: they represent the result of
averaging out and prioritizing and generally knocking a model into a par-
ticular shape. ICMs, however, differ from prototypes in that they are not
28 Chapter 1. The heartland of Cognitive Linguistics

tied exclusively to the specific idea (the prototype prototype, as it were)

of a focal area with a corona of more and more marginal instances around
Closest to the prototype idea is the notion of a radial model, with
mother as the crucial example. The hub of a radial model corresponds
more or less to a classic prototype; in the case of mother it is one who is
a mother genetically, legally and in terms of actual nurturing, of a child
still alive, and who is married to the childs father, with motherly feelings
for the child, etc, etc. However, instead of just a gradual movement towards
a periphery, as pluses are changed into minuses for the feature list, a radial
model includes fully specified combinations such as single mothers, step-
mothers, adoptive mothers, surrogate mothers, etc, which all have their
own place in the picture. A radial model can be metaphorically character-
ized as a configuration in the landscape with a central conceptual summit
surrounded by lesser peaks and a gradual slope down to the non-mother
Two different, but related examples of idealized conceptual models are
social stereotypes and ideals. Illustrating the difference, Lakoff suggests
(1987: 87) that the ideal husband is a good provider, faithful, strong,
respected, attractive, while the stereotypical husband is bumbling, dull,
pot-bellied. Both are examples of the more general phenomenon of pro-
totypification introduced above, allowing for graded similarity judgments
and associated prototype effects, but neither constitutes the hub of a
radially structured conceptual complex of husbands shading off into non-
husbands on all sides.
As we have seen, it is sometime difficult to tell some of the basic
notions apart, and this is one point where the functional dimension that is
highlighted in this book may serve a useful role. Idealized conceptual
models (= ICMs) share with domains and frames the property of being
complex conceptual configurations, and may therefore function as both
domains and frames. Understanding the concept surrogate mother
depends on the domain of motherhood (including genetic, biological and
social dimensions). When viewed as a generic domain, this model provides
a background against which one may profile each separate subtype, such
as a single mother close to, but distinct from a mother who gives away a
child for adoption, for instance. This is different from the framing func-
tion, which also occurs: in actual social contexts, a single mother may well
be seen as framed by the prototype mother, in terms of which single moth-
ers constitute a (possibly shameful) deviation.
In terms of their role in the theory, ICMs are closer to concepts than to
frames, because they pick out a chunk of conceptual substance, and in
Embodiment and image schemas.From conceptual to neural patterns. 29

highlighting it detach it from its background (whereas frames constitute

the background for a concept). As the more general term, concept sub-
sumes the domain of idealized cognitive models: all ICMs can function as
conceptual categories. In the case of, e. g., Lakoffs model of the strict
father family, you may invoke instances of it just as you would invoke
instances of a concept such as car. However, ICMs capture the fact that
it is part of human cognitive practices to extract idealized representations
of parts of the world for purposes of everyday orientation. The canonical
function of conceptual models is therefore to serve as a partial world pic-
ture or Weltanschauung. The notion of a disaster caused by a large
meteor, for example, is a conceptual model that can be invoked to account
for dinosaur extinction or as a scenario for a thriller movie. Similarly, the
billiard-ball model of relations between things (cf. below p. 51) embodies
the knowledge that things can causally interact in ways that influence
their future direction. ICMs with a temporal dimension are known as
scripts: The restaurant script (cf. Shank and Abelson 1977) embodies
cultural knowledge of what happens when you dine out, linking constitu-
ent sub-events in ways that may enable text understanding, as demon-
strated in the AI literature, and thus also constitutes an ICM.
Functionally, ICMs cover an intermediate level of knowledge about
the world, between individual conceptual categories that enable us to put
all individual phenomena into separate containers and the whole concep-
tual ontology. An ICM is a familiar and coherent configuration in the con-
ceptual landscape that provides default knowledge allowing bridging
inferences and other forms of routine enrichment of understanding. Espe-
cially salient, and linked up with the term idealized, is the role of ICMs
as the carrier of generalized expectations of how the world works frag-
ments of the conceptual autopilot system that we depend on for most of
our everyday activities. The basic format, common to all conceptual rela-
tions, is the network: rather than a neatly ordered structure with every-
thing in its well-defined slot, all conceptual entries are connected to each
other via multiple relations of different types, of which Aristotelian fea-
tures and categories are only one subtype.

4. Embodiment and image schemas. From conceptual to neural


Just as frames and domains reverse the direction of approach to concepts,

embodiment reverses the priority between abstract reason and concrete
experience in linguistic description. The generative pattern of thinking
30 Chapter 1. The heartland of Cognitive Linguistics

based on abstract structures understands itself as reflecting Cartesian

thinking, based on the ghost in the machine, the rational mind within the
mechanical body. CL subjects abstract reason to the same reinterpretation
as conceptualization in general: instead of assuming that reasoning is
something quite distinct from ordinary human understanding, CL argues
that meaningful categories have to emerge out of human experience as
generated by the human body. The concept of bodily grounding expresses
the view that higher-order cognitive processes should be understood by
looking at their underpinning in the human body, not as the result of rising
above bodily limitations.14
A classic example is the concept of force dynamics (cf. Talmy 1985a,
2000). The ability to understand the force of logical argumentation is a
mental achievement, but in order to grasp the concept as it operates in
human understanding, we also have to understand it as an extension of the
felt experience of physical force. The understanding of figurative language,
especially metaphor, in Metaphors we live by (Lakoff and Johnson 1980),
is full of examples of the role of direct embodied experience as a source of
understanding abstract complexities, and the whole directionality of
explanation in cognitive semantics is predicated on this approach.
Perhaps the most central notion in CL for the embodiment dimension
is the notion of image schema. It played a central role in the foundational
account of embodiment in Johnson (1987), and it has also been used in
empirical developmental psychology, cf. Mandler (1992). The concept of
image schema was developed to capture a type of phenomena that was
first brought to the notice of linguists by Talmy (1975). In his broad-rang-
ing overall project of describing the kinds of meaning that characterize
closed classes Talmy outlines a range of meanings which differ from ordi-
nary lexical meanings in being more abstract and schematic. One such type
of meaning he terms topological, including cases like point, partition,
linear extent, adjacency (Talmy [1988b] 2000: 28). Grasping meanings of
this kind could be plausibly linked with experience of the type that is gen-
erally assumed to precede verbally and conceptually articulate thought,
Piagets sensorimotor stage. Before the child is ready to handle complex

14 One can understand the change without having to take sides: to some extent it
is a question of getting hold of the other end of the stick. From the point of
view of secure knowledge the foundation of mental content is what is objec-
tively (out) there regardless of the human observer. From the point of view of
human understanding, the foundation must be mental processes in human
Embodiment and image schemas.From conceptual to neural patterns. 31

content, she must be able to handle physical space and get her basic bear-
ings there.
Image schemas were launched in parallel by Johnson (1987) and Lakoff
(1987). Although (cf. Hampe 2005) no consensus has been reached on the
precise status of image schemas in CL, the concept is predicated on link-
ing up the two dimensions specified above: the structuring of (primarily)
space, and the privileged relation with felt experience. The example that
has perhaps received the greatest interest is the container schema (cf.
e. g., Lakoff 1987: 267). This schema illustrates both the structural simplic-
ity and the plausible relation with basic experience. It involves the con-
tainer itself, which has an inside and an outside, an element (which is
either contained or not), and two potential directions, in and out and
that is about it. But the abstract structure is of a kind that can be related
directly to perceptual systems.
As shown by Mandler (1992, 2005), empirical evidence suggests that
this kind of knowledge is indeed part of the very early cognitive develop-
ment. However, Jean Mandler points out that the plausible source of
image schemas in basic perceptual experience should not be confused
with the status of fully acquired image schemas in child cognition. These
are different levels, and may diverge (cf. the general discussion p. 383 f.):
image schemas as cognitive constructs are independent of sensory modal-
ity and can be used for inferencing already at the pre-linguistic stage, and
they are thus instances of abstract conceptualization rather than raw per-
ception. The word schema, in other words, captures an essential part of
their mode of being. The two sides of the idea, structural abstraction and
direct experiential basis, thus do not automatically cohere, as also pointed
out (with some regret) by Johnson (2005). Mandler also provides argu-
ments to show that the experience of spatial impact in the form of force
dynamics appears not to be crucial to image-schematic understanding:
spatial relations are sufficient in themselves. Thus the understanding of
containment is not necessarily bound up with the experience of being con-
tained (in the crib or pram, for instance). We will come back to this issue
below p. 79 f.
Schemas are one bid for integrating experience and conceptual struc-
ture. Another is the emphasis on emotional experience as a foundation for
thinking. This idea received scientific underpinning with the discovery by
Damasio (1994) of the dependence of rational thought on emotional
grounding. In the book (entitled Descartes error) he showed, based on
detailed clinical evidence, that rationality, rather than having to keep clear
of the mush of human emotions, actually became inoperative when the
link with emotions was severed because of brain damage. Without know-
32 Chapter 1. The heartland of Cognitive Linguistics

ing what feels right, the human subject can no longer take rational deci-
sions. Relating to the conceived world therefore depends on linking up
the conceptual and the emotional dimension.
The theory of embodiment thus involves very concrete relations in the
brain between concepts, schemas and emotional qualities, down to con-
crete neural implementation. The most radical manifestation of the project
of integrating concrete bodily phenomena with the description of mental
processes is the neural theory of language, which aims to provide a con-
crete specification of how to get from molecule to metaphor, with the
title of the recent book (Feldman 2006). With the experiential grounding
of someone who has made the whole journey from one pattern of thinking
to the other Feldman started out as a computer scientist and only later
began to take an interest in neural processes he explains in biographical
as well as theoretical detail why a purely computational approach based
on information processing is misguided. Using the word epiphany (Feld-
man 2006: 63) about his sudden realization that there was no distinct level
of physical symbols in the mind or brain (which in the formal paradigm
was necessary to provide formal symbols with causal relevance, cf. Newell
and Simon 1976: 116), he describes how the actual neurobiological feat of
deriving information from the environment must be carried out directly
by the neural wiring otherwise it would simply not be possible.
The core ability in Feldmans perspective is to respond appropriately
to information from the environment. Survival depends on being able to
come up with the best match between input from the environment and
the response from the organism. Beginning with the amoeba and its abil-
ity to follow a gradient of increasing concentrations of nutrition and away
from irritants, Feldman goes on to illustrate how the more complex neural
systems of frogs (citing Lettvin et al. 1959) are similar in allowing the
organism to respond to relevant features such as moving bugs and
approaching shadows. On the way to full human complexity we encounter
mirror neurons (more on this below p. 82 f.) which trigger relevant pat-
terns of action when animals perceive conspecifics coping with the envi-
ronment in particular ways (cf. Rizzolatti et al. 1996).
This is particularly interesting in relation to the process of linking up
embodied experience with relations between individuals: if the individual
is capable, by neural wiring, to share the experiences of others, a neural
theory does not entail a purely inward-looking approach to meaning, but
has a plausible link with social experience. By virtue of that, it also links
up with the basis of normativity and morality: human beings can act as
moral agents because they have access to experience about what pro-
motes or hinders the well-being of others.
Embodiment and image schemas.From conceptual to neural patterns. 33

In using this theory about human minds, a key element is the causal
link between mental connections and active neural connections (Feldman
2006: 91). This brings us to the level of priming effects, as driven by the
mechanism of spreading activation in a network. However, in order to get
to the level where mental connections are linkable with language, it is not
feasible to go all the way with the neurons alone (Feldman 2006: 151). Not
enough is known about the complex patterns of neuron circuitry to be
specific about how we actually understand the cat is on the mat in neural
terms. Feldman, in other words, cannot close the gap that everybody else
has also failed to eliminate, between the physical and the mental level of
description of the mind/brain: the slash remains.
To his credit, Feldman is quite explicit about this. He sees this as an
instance of the fact that sciences often need bridging theories: astrono-
mers need them in order to make facts about stars cohere with basic phys-
ics, and biologists need them to make facts about protein structure cohere
with theories of how sequences of amino acids fold into three-dimensional
shapes as specified by DNA. Rather than giving up the attempt to build a
neurally based theory of conceptualization, Feldman uses the level of
computational modelling to mediate between neuron circuitry and actual
linguistic performance.
This puts him in the company of classical computational modelling,
with the same basic framing problem: only the human operator can pro-
vide the essential link between simulation and simulated reality. The dis-
tinctive feature of the neural theory is the explicit commitment to an
adequacy criterion based on converging evidence, including especially a
criterion of neural plausibility and learnability. Based on these premises,
Regier (1996) constructed a computational model that was able to learn
spatial relation terms. One of the ways to achieve this feat was to provide
the system with a simple model of the visual system. In contrast with com-
putational models built on the blank slate (Feldman 2006: 154), which
could not learn these terms, a model based on what is neurally plausible
could do the trick.15

15 The same strategy was pursued by Bailey (1997) for action words, where the
crucial addition to the system was motor programs. The model was extended
from physical events to metaphorical interpretations by Narayanan (1997). He
showed how news stories about economics including statements such as
Japan continues its long, painful slide into recession could be understood as
transferring event structure (including aspectual phases from ingressive to
perfective) from the basic motor domain to the abstract target domain. This
34 Chapter 1. The heartland of Cognitive Linguistics

The final step to language is made through the apparatus of embodied

construction grammar. The basic idea in this approach builds on the
notion of image schemas, construed to include also event or action sche-
mas. Each word evokes a schema, or a network of schemas and these
then have to be combined according to a best match that repeats, at a
vastly more subtle level, the basic coping strategy of the organism in com-
ing up with the best possible response to input from the environment.
This approach is an essential part of the whole enterprise of working
out a theory meaning based on embodied experience. Thus the neural
theory of language lives up to a key commitment of CL in working out an
ever richer and more detailed theory about how explicit logical thinking
can be grounded in processes that reach all the way down into the purely
biological level of neural connections. Anticipating a discussion below (cf.
p. 200), however, let me mention some issues that are not resolved in this
approach. One is the question of precise relations between levels, already
discussed above. In addition to the neural and computational levels pre-
sented above, there are two others, including the topmost level of lan-
guage and thought (cf. Feldman 2006: 139). The project is not complete
before all levels are spelled out in ways that link up everything from bot-
tom to top. Until then, however committed the theory is to the bottom
level, the project is essentially in the same boat as other theories operating
in terms of separate levels with different vocabularies and properties.
To summarize the argument in this section: as part of the commitment
to meaning as a human-style rather than objective-scientific or formal-
logical phenomenon, CL is based on the approach to meaning as arising
out of processes in the human body. In this, it views meaning as uniting
emotional, motor and representational aspects in a bottom-up path from
baby to adult, and from biological to conceptual functions. Image schemas
have a central role in the way meaning is understood at the pre-concep-
tual level because they capture features that are relevant also for the
development of sensorimotor skills: paths, borders, beginnings and ends,
objects and obstacles. Force dynamics illustrates the involvement of all
levels from physical impact via neural and motor response to schematic
relations between agent and antagonist, including metaphorical exten-
sions of force to the force of logical arguments.

carried over to inferences appropriate to the abstract domain, such as eco-

nomic development, based on the elementary structure of motion events.
Figurative meaning 35

5. Figurative meaning16

Lakoff and Johnson (1980), Metaphors we live by, was the first powerful
illustration of how figurative meaning could enhance understanding of
language as part of a general project of understanding the human mind.
Putting figurative meaning at the centre of attention also effectively high-
lighted the narrowness of the science-imbued conception of meaning in
language. From the objectivist perspective, figurative meaning was mar-
ginal in two interconnected ways: ornamental rather than essential to the
message, and basically involving a misrepresentation: the statement you
are the cream in my coffee is not literally addressed to the cream in the
speakers coffee. There is an effect, rhetorical as well as humorous, of
putting it like that, but underneath is a real literal meaning which has to
be figured out. In terms of the standard position usually attributed to
Aristotle, metaphor was a figure of speech, rather than a feature of real
conveyed meaning.
The cognitive linguistic position on metaphor as stated by Lakoff and
Johnson is a continuation of the dimensions discussed above: meaning is
based on experience, on presupposed cognitive domains, and on pre-con-
ceptual embodied schemata. Figurative meaning adds an extra step to the
process of conceptualization, in the form of a mapping from one location
(the source) in the conceptual territory to another location (the target). In
the case of metaphor, the mapping goes from one domain to another, a
salient case being space as a source domain and time as a target domain,
as in you have your whole future in front of you. In the case of metonymy,
the mapping works by direct association, sometimes staying within the
same domain; in reading something with fresh eyes, the eyes are a part of
the whole that also includes other parts of the neurocognitive system;
sometimes it includes cross-domain connections, as in having ones eyes on
someone, with metonymic mappings that go from eye to vision to atten-
tion to desire (cp. Hilpert 2007: 87).
This makes figurative meaning an integrated part of the object of
description: taking an extra step inside a cognitive landscape is radically

16 For a long time, figurative meaning was the only area of CL that many people
had heard about, and it remains the best-known area of the approach. It has
also given rise to a plethora of different developments each of which would
require book-length accounts. For that reason, the various manifestations of
figurative meaning will not be in focus in this book. Only some of the most
pervasive features will be taken up as necessary dimensions of the social
extension of the framework.
36 Chapter 1. The heartland of Cognitive Linguistics

different from adding an external adornment to a cognitive essence. The

point, however, only becomes clear when it is understood how the effects
of figurative language can be better understood in terms of the whole
cognitive linguistic picture. The central element is the notion of under-
standing one thing in terms of another, where the detachment of meaning
from the objective nature of the referent again comes to play an essential
role. In objective terms, it would be a category mistake to understand one
thing in terms of a different thing, i. e. understand nature as a whole as an
animated being (as in you can drive nature out with a pitchfork, but she
always returns). In cognitive-linguistic terms, however, what happens is
something totally different: you use the same conceptual structure in more
places than one, when it is motivated in terms of human experience.
Rather than being a category mistake (as a hard-nosed objectivist would
say), it is the only sensible, and sometimes indeed the only possible thing
to do: you use your conceptual constructs wherever they seem to work.
The difference, again, is understandable in terms of the knowledge
bias in the tradition, as opposed to the desire to understand conceptuali-
zation as a domain in its own right. In the notation of Lakoff and Johnson,
the contrast is starkly profiled. When, for instance, you map the properties
of machines (source) on to human beings (target), as in Joe is due for a
maintenance check, this is expressed in the formula PEOPLE ARE
MACHINES but obviously this does not entail that Lakoff and Johnson
believe that people are machines from a knowledge point of view (oth-
erwise cognitive linguistics would be wrong, and formal semantics right!).
What the formula does suggest, however, is that there is a form of sta-
ble cognitive identification between the two domains, and that it is main-
tained by cognitive mappings from one domain to the other. These map-
pings therefore constitute part of the way the cognitive system works. The
identity is reflected in the regularity whereby you can invoke one domain
by invoking the other. This is the basis for downplaying the distinction
between dead and living metaphors: no matter whether we may have
forgotten that the expression the foot of the mountain recruits the human
body as a source domain, the mapping whereby human beings project
their own body on to the world in understanding it, is alive and well.
This has brought about a distinction between different types of meta-
phorical mappings. Some are intuitively more basic and entrenched than
others. Lakoff and Turner (1989) distinguished between conventional and
novel metaphors, and discussed how metaphors could be creatively
extended (thus avoiding the need to include cases such as Wallace Ste-
vens the emperor of ice cream as reflexes of basic and shared experiential
mappings in the mind). More recently, Grady (1997) has proposed a cat-
Figurative meaning 37

egory of primary metaphors consisting of those that have a basic experi-

ential status in the human form of life. A primary metaphor is motivated
by a primary scene, which is an activity or situation that recurs as part of
recurrent elementary experience.
One example is the kind of correlation that occurs between sensory
experience and subjective feeling, as in IMPORTANCE IS SIZE (cf.
Evans and Green 2006: 305); when you are faced with something that is
physically big, it goes naturally with a subjective feeling that you had bet-
ter treat is as important, as reflected in the collocation overpowering size.
A similar example is the metaphor SEEING IS KNOWING, reflecting
the fact the vision is the primary human source of knowledge as
reflected in countless historical (cf. e. g. Sweetser 1990) and synchronic
links between words for seeing and words for knowing. Co-occurrence in
primary scenes is a fairly clearcut criterion that legitimizes the special sta-
tus of a number of intuitively basic metaphorical mappings, including also
the links between more and up, change and motion, between end points
and results, etc.
What happens according to Grady is then that other metaphorical
mappings can be superimposed on those associated with primary meta-
phors. One of his examples is the theories are buildings metaphor, which
is exemplified by his ideas have shaky foundations or his theory was demol-
ished. There is a primary element in it, in the form of the metaphor that
links standing up with being in good working condition which is a recur-
rent scene for both people (up and about vs bedridden) and artefacts
(the windmill is not up yet). The choice of a building as the specific struc-
ture that can be either up or down, in contrast, is not similarly inherent in
experience, as evinced by the fact that theories do not have windows or
Embodiment therefore works in terms of a process whereby meta-
phors can be unpacked: the poetic extensions and the nonbasic aspects
can be traced back into the primary metaphors that they build on. Were
spinning our wheels as used about a crisis in a love relationship does not
mean that the lovers are understood in terms of lovers are cars; you have
to understand it at the primary level where the car is the vehicle in which
the lovers are travelling and the experience of being in a vehicle that is
not moving has the primary character that is necessary to understand the
embodied sense of frustration (cf. Lakoff 2008:256).
Primary metaphors are good candidates for the permanent mappings
that Lakoff and Johnson take as the prototype foundation for metaphor.
It would run counter to the general gradualist tendencies in CL, however,
if it were not assumed that there is a cline from the stable and entrenched
38 Chapter 1. The heartland of Cognitive Linguistics

and to the nonce and ephemeral end. In present-day Denmark we can still
use horse terms for processes associated with setting things in motion: in
addition to putting the cart before the horse we may also belong to the
kind of people who do not ride on the day they saddle their horse (i. e.
slow starters). The primary scene of horse-powered locomotion, however,
is not around any more, and we may expect an ongoing process of gradual
de-entrenchment of the mapping as a result.
An essential feature of metaphor theory is the basic asymmetry
between source and target. It reflects the same basic bottom-up direction-
ality that is associated with embodiment: we use what is basic and already
available to get at what is more sophisticated and intangible. Events in
time are less graspable than physical objects in space and therefore we
recruit spatial concepts in getting a grip on time. Physical processes are
more familiar and manageable than emotional processes, and therefore
we resort to physical source domains in trying to grasp what happens in
love relationships, including the physical journey as an experience of mov-
ing into a new location via various intermediate places and events. The
process could be seen as reflecting a strategy of metaphorical bootstrap-
ping, whereby we seize upon a familiar domain and use it to take a leap
into the unknown, retaining the familiar structure as a handle on the world
while we move on.
When we impose source structure on a target domain, however, we
cannot know in advance what conceptualization is most appropriate in
actually grasping the new domain. Lakoff (1993: 216) therefore intro-
duced a target domain override principle, explaining why not all infer-
ences from the source domain were valid. She gave him a kiss imposes a
transfer-of-goods structure on the act of kissing, but the inference that
when Joe gives me a book, I have the book afterwards does not extend
to the kiss, because the nature of the act does not leave a retainable prod-
uct (unlike he gave me a black eye). As has been pointed out, if the target
domain can selectively resist imposition of source structure, the formula X
IS Y is clearly simplistic it only holds for the current purposes of the
metaphor, if we express it with a Gricean twist (cf. below, p. 191 f.).
For the purposes of classical conceptual metaphor theory, metonymy
was something of a poor country cousin. It shares some of the essential
features, especially the asymmetry that in the case of metonymy is usually
expressed as going between vehicle and target, and the experiential basis,
including recurrent scenes cf. all hands on deck, for instance. In the per-
spective of permanent mappings there would have to be a stable link in
order to enable metonymy, while there would be nothing obviously cor-
responding to the bootstrapping process involved in going into a differ-
Linguistic meaning: Polysemy, ambiguity, and abstraction 39

ent domain and achieving new understanding. Therefore metonymies

were generally viewed in terms of motivating relationships that could
explain familiar extensions such as pars pro toto, cause for effect, rather
than as a source of conceptual creativity. This perspective has later
changed, cp. below p. 99.

6. Linguistic meaning: Polysemy, ambiguity, and abstraction

As a result of all this, the discipline of semantics as understood by CL

becomes much richer than previously understood but it also becomes
more difficult to pin down. From the point of view of language, a very
basic question is: what precisely is the meaning of a linguistic expression?
CL has not spent a great deal of time worrying about the question,
probably because that was something truth-conditional semanticists did.
The most generally accepted position is that of Langacker (1987: 161f):
while an expression evokes the whole domain, it only specifically desig-
nates the profiled subpart. The word daughter evokes the family domain,
but only designates the female offspring and therefore the female off-
spring is the point of access to the domain. Thus an individual linguistic
concept may be thought of as a point-of-access to something that is nec-
essarily bigger than the concept itself.
The access metaphor is useful because it stresses the overall point that
was crucial in the frame idea of semantics, namely that neat little sepa-
rate nuggets are not the stuff that human meaning is made of. The process
of understanding words necessarily invokes human experience, and the
idea of access allows that process to be active in a not rigidly bounded way,
while avoiding the undesirable consequence that each word means the
whole network of experience that it evokes (but cf. the discussion of word
meaning in ch. 5).
However, when you look at the linguistic perspective, there is one
question that is not exhaustively answered by this strategy. This concerns
the question of the individuation of conceptual structure vis-a-vis the indi-
viduation of linguistic meaning. This question raises itself whenever one
has to answer the question of how many different conceptual categories
can be designated by the same expression: the ambiguity-vagueness issue.
The truth-conditional, classical approach in its strict form presupposes a
view of meaning whereby each set of criteria constitutes one meaning
and any deviation means that you are into a different meaning. Depend-
ing on the view, bachelor is four different lexical items or one item with
multiple meanings.
40 Chapter 1. The heartland of Cognitive Linguistics

There is an element of notational choice involved here, but underneath

the notational choice is the real issue, which the discussion above has not
clarified: when we understand linguistic expressions as symbolic entities,
what are the criteria for individuating whatever we propose as the content
side, or the semantic pole of a linguistic expression?
With the overall emphasis on the dense network of mappings between
different locations in conceptual space, an important part of the early dis-
cussion in CL centred around the concept of polysemy. The idea is that the
nature of conceptual space allowed linguistic expressions to profile a plu-
rality of different but conceptually related regions. This highlighted a key
property of CL as opposed to the strict either-or mentality of classical con-
cepts: either it is one meaning, and no variation is required, or the two
meanings are different, and the item is then homonymous. The classical
example of this situation is (commercial) bank vs. (river) bank, in which
case lexical semanticists split the homonym into two (monosemous) items.
With the richly characterized conceptual universe that emerged from the
cognitive approach, this is clearly a much too primitive option, and the
option of describing a series of related but different meanings constitutes a
more revealing alternative. Wittgensteins concept (1953) of family resem-
blance is the seminal version of the idea, which is put forward in rejecting
an account in terms of all-or-nothing definitions (and in that respect similar
in spirit to CL), with game as the example of a concept where no single
characteristic applies to all instances (such as more than one participant,
a goal state, winning and losing, etc) had to be given up when confronted
with games like patience or throwing a ball back and forth.
The epitome of this approach in CL is the description of prepositional
meaning, where over (Brugman 1981, Lakoff 1987) is the prototype exam-
ple. It has been discussed many times in the literature (cf. e. g. Evans &
Green 2006: 329339), and there is no need to go deeply into it here. There
are more central variants, such as the plane flew over the city, involving
movement across and above. The full network includes a large number
of other senses, some closely related like above (hovering over me) and
control (power over someone), some less close, including covering (a
board over the hole) and excess (overextension). As a natural element in
the emphasis on the conceptual universe as an entity per se (cp the discus-
sion of metaphor above), all variants are claimed to be stored, even if a
mechanism could be constructed for generating them (cp. Pustejovsky
1995), and Lakoffs version is therefore dubbed the full-specification
approach (Evans & Green 2006: 336).
The full specification approach is the most sprawling way of looking at
word meaning, with senses proliferating in all directions. As described by
Linguistic meaning: Polysemy, ambiguity, and abstraction 41

Tyler and Evans (2003), it is possible to set up criteria for distinguishing

between polysemy and vagueness which can reduce the number. One such
criterion is inferrability: a sense is only an instance of polysemy rather
than vagueness if it occurs in a context where it can be taken as intended
without the support of a context that makes the sense inferrable. Thus the
spatial configuration obtaining in the tablecloth is over the table does not
constitute a distinct sense in relation to the above sense, because it can
be inferred from encyclopaedic knowledge about how we position table-
cloths with respect to tables, while John nailed a board over the hole in the
ceiling does constitute a distinct sense because it cannot be inferred from
the basic spatial configuration (Evans & Green 2006: 34344).
This criterion is related to Quines (1960: 129) criterion, according to
which a word with two senses can make a statement true and false at the
same time; thus if he put a board over the hole in the ceiling is true in the
above sense, it may be false in the covering sense, and vice versa. How-
ever, there is no hard and fast way of getting at a unitary maximal reduc-
tion, cf. Geeraerts ([1993] 2006: 101); depending on various factors, what in
one case may seem to be vagueness may in a different context come out
as polysemy. The CL approach is characterized by its general willingness
to allow for several simultaneous potential accounts; to insist that only
one description out of a set of competing options can be true is what Lan-
gacker (1987: 28) calls the exclusionary fallacy. Thus different readings
can be understood both as the product of a general mechanism and simul-
taneously as a list of distinct individual senses without any necessary con-
As a specific reflection of this non-exclusionary view of meaning, it is
also possible to approach the characterization of meanings in several
equally appropriate ways. The radically sprawling approach to meaning
discussed above stands in contrast to the radically restrictive definitions in
the Aristotelian picture but under the name of schema, the abstraction
associated with generalized overall meanings is also an essential part of
cognitive semantics. This plays a role both in the relation between lexical
meanings (male is more schematic than bull, for instance), but it is central
to the understanding of the (gradual) difference between grammatical
and lexical meanings. As described by Langacker (1987: 18990), the cat-
egory of noun may be captured by a semantic description based on the
most abstract sense of the noun thing (as in something) and this dimen-
sion is perfectly compatible with a simultaneous characterization in terms
of the prototype physical object. The schematic characterization, how-
ever, is useful in cases where there is no direct or obvious link between the
prototype and the extension: how does one get from cup to exacerba-
42 Chapter 1. The heartland of Cognitive Linguistics

tion?, Langacker asks, and indeed the path of extension would not be
easy to trace if there was never any resort to schematic abstraction. The
general sense of schema subsumes the more specific use in image schema,
cf. above, which is an abstraction from concrete sensory images. As usual,
it is not clear exactly where the distinction should be drawn; Langackers
schemas for capturing grammatical meaning often recur also in lists of
image schemas, including e. g. the distinctions between mass and count.
The existence of schemas does not license a general return to classical
semantics, however. Similarly, the notion of radial models does not justify
a strategy whereby you limit the meaning of an expression to only a pro-
totypical core and leave the fuzziness to vagaries of usage. The extent to
which there is a clear prototype or a clear schematic formula is variable
from one expression to the next.
For the notion of a meaning associated with a linguistic expression, I
suggest it might be practical to think of it terms of the territory, the area of
potential application within which the expression is sanctioned, to use
Langackers term. Territory differs from profile in being more general,
and naturally can be thought of as including a variational dimension, so
that a new usage extends the territory. (Thus when the practice whereby
internet con artists try to get people to mail their passwords is dubbed
phishing, it extends the territory of the spoken word fishing). The term
territory covers the full range of variant meanings for over, in other words
the range of potential use that may occur without infringing linguistic con-
ventions in the community. It is usefully primitive in that it abstracts from
category-internal semantic structure including ambiguity, prototypes, par-
tial family resemblances, metaphoric extensions and schematic charac-
The term thus ignores the details but the job it does is central to the
idea of a specifically linguistic as opposed to a general conceptual descrip-
tion. If you ask what a given expression means, the full answer requires a
specification of the whole territory whatever else it might be useful to
say about it.

7. Mental spaces

The last of the traditional centrepieces of CL to be covered in this chapter

is the notion of mental spaces, cf. Fauconnier (1985/1994). The notion has
become part of a larger discussion whose central concept is blending or
conceptual integration (cf. Fauconnier and Turner 2002), which will be
discussed in chapter 2.
Mental spaces 43

The concept of mental space was originally proposed in relation to

cases of referential opacity, as exemplified by classical examples like The
Morning Star is the Evening Star or Nancy wants to marry a Norwegian.
Mythical beasts have also frequently featured (especially unicorns),
because in relation to examples like John was looking for a unicorn the
fact that they may not occur outside the opaque (looking-for) context is
immediately obvious. The salient point is that referents may have an indi-
rect rather than a direct relation with reality: even if it is true that I am
looking for a unicorn, and the term unicorn thus correctly refers to the
object of my search, there is no guarantee that there exists a unicorn for
me to look for. Because it offers an account of logically troublesome cases,
the concept belongs among the foundational achievements of CL, tran-
scending the limitations of a purely referential, objective-realistic
approach to semantics.
The idea of a mental space captures the fact that in linguistic commu-
nication, we do not start out from the world as objectively given. Because
it is the human world that matters, we operate with local mental universes
such as my plans (I plan to be a home owner in 5 years), how I feel (I
long for cool evenings in the shade), or what I believe (I think ecological
accountability is the only way forward). Entities we talk about are charac-
terized by the way they are located in relation to such personal universes,
rather than by their place in an objective description of the world. I want
to find a nice present for Sarah sets up a want universe in which there
exists a present for Sarah, and the interlocutors can discuss this present
avidly because of the interest they take in the speakers wants and in
Sarahs upcoming birthday. The fact that the present for Sarah does not
exist anywhere else than in the mental space constituted by what the
speaker wants to do is entirely beside the point, in spite of the fact that
this issue mesmerized philosophers for 2500 years,
One of the significant properties of mental spaces is that they are only
partially specified. In this, they constitute a more flexible alternative to the
solution devised in referential semantics, the notion of possible worlds
(cf. Lewis 1973). Possible worlds are total specifications of alternative
worlds, in continuation of the assumption in the logical tradition that the
whole referential universe is taken as the basis. On the one hand this
would pose an unrealistic burden on mental processing of input, and on
the other hand it would strain credulity when it comes to a realist ontol-
ogy although some authors wanted to claim referential reality also for
alternative worlds, the idea did not really catch on.
Mental spaces are practical, because they carry less baggage. They are
bubbles devised for purposes of conceptual understanding and communi-
44 Chapter 1. The heartland of Cognitive Linguistics

cation and may be closed down as soon as they are no longer needed. They
are not defined by explicit referential specifications: they are underdeter-
mined (Fauconnier 1985: 2), allowing complex configurations of mental
spaces based on relatively simple grammatical input. They are triggered
by expressions serving as space builders, including conditionals (If she
accepts ), modals (They may refuse) and prepositional phrases (In your
situation, ).
The ultimate source of all such space-building is the ability of the
human mind to form representations, cf. the discussion in relation to con-
ceptualization above p. 16. In relation to opacity of reference, the prop-
erty of intentionality cf. Searle (1983) is central. Intentional mental struc-
tures are understood as representing the world, or more generally a
world. When we go to pick up a familiar person at the airport, we check
the remembered, mind-internal features against the faces in the crowd.
And while we are waiting, we operate with a mental space with specifica-
tions that we want to materialize in the base space of physical experience.
When the person actually turns up, we can close down the want space
All mental spaces owe their properties to the primordial mental space,
i. e. the human mind: because the mind can represent things to itself, it can
also create a split between a referent and its mental representation. It is
especially striking, however, in cases when the relation between the world
and our representation of the world is explicitly put on the agenda: condi-
tionals, modals, and thought experiments in general are manifestations of
this ability to imagine worlds, modulated by the relations of desire, belief
and imaginative projections of various kinds.
For reasons that will be discussed below, I suggest the term alternativ-
ity borrowed from modal logic (cf McCawley 1981: 276) for relations
between such spaces. Mental spaces, I suggest, arise when we imagine an
alternative scenario, and that means there is always an implicit compari-
son involved. The space that constitutes the basis of comparison is defined
by the current discourse but the mother of all these is the so-called
base space: the actual situation in which the conversation occurs. There
can be several such alternative spaces in play at the same time; thus in
Joe told me that Jill expected Jack to think of something, the something
is in Jacks mental space, which is situated inside Jills mental space, which
again is inside Joes mental space, which is part of the mental space
evoked by the speakers utterance but it all ends up in the actual dis-
course world.
The status of a mental space as an alternative licenses a very basic rela-
tionship in Fauconniers account, which is also associated with referential
Mental spaces 45

opacity, the counterpart relation.17 When an entity is construed as part of

an alternative to the actual discourse situation, it acquires two different
identities, which may have properties that are mutually contradictory. For
instance, he thinks he is Napoleon establishes a counterpart relation
between the main clause subject he and the delusion-world Napoleon.
And there may be several counterpart relations. With James McCawleys
example, I dreamed I was Brigitte Bardot and that I kissed me endows the
dreamer with two counterparts, linked by different identity projections.
If we take the concept alternative as basic, ontology does not run wild.
The base space of actual discourse is solidly located in time and space,
even when it is populated with the possibly hallucinated imagination of
the participants. Moreover, mental events are real events. Daughter spaces
are proper parts of the base space, even if they may have weird stuff inside
them. If someone believes he is a high-class professional assassin, that
belief is part of the discourse world, with potential causal consequences
for it. If you meet such a person, you would be well advised to get out of
the way and that means in reality, not in the belief space. This ontological
solidity continues no matter how many recursive embeddings there are. In
I think he believes she wants him to prove that he is a real man, a possible
continuation is:..which may explain why he is coming up to us with a leop-
ard on a leash. Beliefs, thoughts and intentions are hard facts about the
real world, while they give rise to mental spaces that contain not equally
hard facts.
These hard properties are also behind the ease with which we move
across space barriers. The alternativity relation means that the whole
point of having mental spaces is that they should share elements with their
parents and daughters. The sharing may be more or less extensive, but a
mental space with entities that are totally unprojectable on base reality
space would not sustain our interest for very long. In mental space under-
standing, entities are central in that you access spaces via counterpart
identity relations between entities, as defined by connector relations.
Such relations are variations on the basic representational theme, but
include (cf. Fauconnier 1985) value-role relations such as the one holding
between the French president and Sarkozy at the moment of writing. In

17 The counterpart relation was also part of the discussion in possible worlds
theory; at one time the preferred example was what Socrates could or could
not have for a counterpart (an alligator was one of the examples!). For obvious
reasons, the discussion becomes less tortuous if you do not have an objectivist
semantics, but instead view it as a matter of alternative construals of the same
situation in actual communication.
46 Chapter 1. The heartland of Cognitive Linguistics

this case the representational content (president) is additionally the con-

tent of a status function, so that it is superimposed upon the person who
carries the function. Another relation of the same kind is the actor-char-
acter relation, where the actor takes on the role of the character in ques-
tion for the duration, as it were.
Generally speaking, mental spaces constitute the central concept for
reinterpreting reference in a CL perspective. In the objectivist under-
standing that is built into the tradition, reference requires a solution to the
problem of the constitution of the universe; in the cognitive perspective,
human speakers are free to populate the universe of discourse with refer-
ents based on the human ability to construe the world in different ways.
Again, I suggest that highlighting the functional dimension will make it
clearer to establish a specific contribution for mental spaces, as opposed
to frames and ICMs. All three are little packages of mental content, which
may be connected in different ways. But in the normal case, the connec-
tor relations need not require separate mental spaces: thus, as long as
Sarkozy is president, there is no alternativity involved in the choice
between introducing him as the president and as Sarkozy he is one
and the same entity that can just be labelled in different ways. It is only
when the two identities are dissociated but at the same time kept in play
together that we need to put them in separate spaces (e. g. when compar-
ing Sarkozy with other French presidents). Similarly, the string of actors
who have played James Bond, from Sean Connery to Daniel Craig, are
ordinary co-constituents of the base space as long as they are viewed one
at a time as consecutive choices of movie directors for continuing a string
of box-office hits. You only need to put them in separate spaces when they
are evaluated as simultaneous and alternative James Bonds (because
Sean Connery, Roger Moore, Pierce Brosnan and Daniel Craig cannot
occupy the James Bond role at the same time).
Mental spaces, on the reading suggested above, are therefore neatly
complementary to ICMs: where ICMs provide partial configurations of
conceptual content, mental spaces provide partial locations where you
may put conceptual content. In this, they are similar to frames but they
have a special niche in their built-in alternativity: mental spaces always
come at least two at a time, only one of which is (viewed as being) real.
The functional role of mental spaces is that we want to have room for
alternatives to the way things are now, whether or not we attempt to
change the world to fit the alternative. In the classic case of Nancy wants
to marry a Norwegian, the Norwegian can be put either in the want space
or the base space. Depending on which space we choose, we may construe
the term differently: in the want space, there is more free scope for a
Cognitive linguistics and cognitive grammar 47

stereotypical construal, invoking idealized models of mountain-dwelling,

skiing (and possibly oil-affluent) Norwegians, than there is if there is an
already existing live Norwegian whom she wants to marry (because we
then want to fill out the picture by the real properties of the man she is
marrying (is he a nice person?). In either case, the roles of mental space
and conceptual content are fairly clearly separable.

8. Cognitive linguistics and cognitive grammar

Above I have tried to describe a number of ways in which CL has enriched

the general understanding of language and conceptualization. The descrip-
tion reflects the role of CL as a re-contextualizing force in linguistics. A
recurrent configuration in the sections above is that a particular cognitive
factor, which used to be considered as something that belonged outside
the proper domain of language, turns out to be able to throw new light on
the way language works. Essentially, the headlines of the sections above
are made up of terms that correspond to this pattern. The fortress of lan-
guage description was opened up, and language turned out to draw on a
variety of fascinating conceptual structures and mechanisms whose role in
language had been marginalized until then.
This approach, however, does not include what the second cognitive
revolution implied specifically for the specifically linguistic core area.
Although the thrust of the movement is clearly in the opening up and the
new vistas, the core area is essential, since without it CL might be cogni-
tive, but it would not be linguistics.
Since the overall focus was not on structure, but on moving beyond
structure, it would be misleading to try to set up a construct and dub it the
CL view of linguistic structure. Nevertheless, it follows from the overall
aims that a new, CL-oriented view of grammar needs to reinvent linguis-
tics in such a way that what used to be special and immanent features of
language can be viewed as a natural aspect of human cognition as a
whole without losing its ability to capture what is indeed specific to
human language.
The central figure in this endeavour is Ronald Langacker. It is no acci-
dent that while CL is a generic label shared by many and quite diverse
authors, cognitive grammar has remained a brand name for Langackers
theory: it is in his works that the implications of CL in general for the cen-
tral descriptivist agenda of the professional linguist (cf. Langacker 1999)
have been most systematically analysed, especially Foundations of Cogni-
tive Grammar (1987 and 1991); for a recent summary cf. Langacker
48 Chapter 1. The heartland of Cognitive Linguistics

(2008a). The first thing to point out about the framework, also in the con-
text of this book, is its commitment to the foundational relationship
between grammar and meaning.
That linguistic structure is symbolic is one of the points on which there
is a direct confrontation with the understanding of structure in generative
grammar. Generative grammar is predicated on the existence of a purely
syntactic structure, such that meaning is outside of that structure but may
correspond more or less to elements in it. Cognitive grammar takes its
point of departure in the symbolic relation between grammar and under-
standing-in-use; and symbolic structures have two poles, a phonological
and a semantic pole. Thus there are only three kinds of linguistic elements:
phonological, semantic or symbolic. Since sounds only count as linguistic
elements when they are used as part of symbolizing structures, all linguis-
tic elements are part of symbolic structures (the content requirement,
Langacker 1987: 53), and structure as such does not occur on its own.
As Langacker frequently emphasizes, this is by no means a free ride, or
a mere declaration of faith. While the distinction between competence and
performance initially set Chomsky free to postulate meaning-independent
structures in a way that was not easily controllable, Langackers commit-
ment to assigning all structure a role in a symbolic relationship means that
he has to provide an explanation for cases that are often assumed to be
obvious examples of non-correspondence. As a case in point, Langacker
(1999) offers an account of so-called formal subjects (like the pronoun it
in it is raining) in terms of the overall generalization that also applies to
contentful subjects. The schematic generalization he suggests is the role
of the subject as initial focus for the understanding of clause meaning. In
the case of prototypical subjects, the initial focus is assigned to an agentive
nominal, but the presentational frame that is associated with so-called
formal subjects can plausibly be regarded as a type of expression where
the initial focus is assigned not to an entity but to an abstract setting
(which may subsequently have an entity or a process assigned to it).
Whether or not one agrees with the description, clearly this symbolic com-
mitment has implications that the analysis must make good.
The symbolic view of grammar naturally leads to a bottom-up approach
to syntactic structure. The generative approach to structure starts out with
a question that is bound up with the innateness agenda: what abstract
relationships are characteristic of sentences in human languages? 18A sym-

18 This pattern of thinking was historically based on theories of formal automata:

structures are assigned a form of being that is independent of and comes
Cognitive linguistics and cognitive grammar 49

bolically based theory of linguistic structure, however, has to take its point
of departure in meaningful linguistic items in use. The existence of com-
plex utterances can be understood (Langacker 1987: 279) to arise from the
fact that if you want to encode a complex conceptualization, there will not
always be a single word that captures what you want to express which
means you have to combine your way to an expressive option that will do
the job. The point of departure for that process, therefore, is the list of
available items in the language.
At this point we need to distinguish between two separate applications
of the distinction between top-down and bottom-up descriptive strate-
gies.19 The application introduced above reflects the choice between start-
ing with general rules (top-down) and starting with actual instantiations
(bottom-up). It is on this point that there is a clearcut distinction between
generative grammar and CL but you can also apply the distinction to the
direction in which you approach a structure. Either you can begin with the
whole configuration (top-down) or you can begin with the single elements
(bottom-up). On this point neither generative grammar nor CL can be
placed unambiguously on one side of the divide.
Early generative grammar was clearly top-down (beginning with the
symbol S and moving downwards), but the minimalist framework has a
strong bottom-up element based on the operation merge. Cognitive
Grammar in its classic version is predominantly oriented bottom-up, being
understood as a Structured Inventory of Conventional Linguistic Units
(Langacker 1987: 73), which could be combined according to a composi-
tional path that moves from elements towards complex combinations. At
the same time, it is stressed that each conventional combination has prop-
erties of its own; this holistic aspect is one of the points on which Cogni-
tive Grammar differs from the strictly compositional approach of genera-
tive grammar. Hierarchy is always partial and only one possible dimension
of organization. An assemblys coherence has to be assessed holistically,
i. e. it is not inherently or exclusively bottom-up, top-down, or left-to-right.
In accordance with this, it is an open issue how structures constituting the
assembly are accessed in actual processing.
This has implications for the understanding of complex syntactic rela-
tions. To the extent they are compositional, i. e. if they are formed in
accordance with general principles (as assumed by generative syntacti-

before the concrete instantiations.

19 I am indebted to Ronald Langacker for pointing this out in relation to a previ-
ous version.
50 Chapter 1. The heartland of Cognitive Linguistics

cians), these general principles are not free-floating or autonomous but

must be understood as based in the semantic properties of the items that
are combined, and there is a cline with more idiosyncratic properties. In
combination with the content requirement, this has further implications
for the descriptions you can offer. Thus the kind of arbitrariness that in
generative descriptions is associated with syntactic (distributional) classes,
for instance with respect to what determines the membership of word
classes, is not compatible with this approach. It must be assumed that ele-
ments in them have a potential for being combined into complex syntactic
One illustration of this principle is that membership of the class of
nouns goes with a semantic description (symbolizing a schematic thing).
This semantic property at the same time qualifies the noun (or the nomi-
nal expression of which the noun is head) to combine with a verb, because
a verb designates a process that has an open slot for things that take part
in the process (an elaboration site). Because of this correspondence
between their symbolic contents, nominal and verbal expressions can
combine into complex expressions denoting process-cum-participant(s).
A corollary of this descriptive path is the claim that grammar is in itself
non-constructive it is the speaker who constructs complex utterances,
not the grammar that assigns structure to sentences. Structure is a prop-
erty of the inventory in that some units are part of other, larger units: the
unit [d] is part of dog, which is part of dogs, which is part of dogs are pro-
hibited, etc. and speakers can construct their own complex expressions
because combinability is not arbitrary, but motivated in terms of proper-
ties of the units themselves.
Following this descriptive path, grammar can be seen as (Langacker
1987: 97) the successive combination of symbolic structures to form pro-
gressively larger symbolic expressions. It is not an equally natural option
to begin with a complex structural relation and work downwards from
there. (But there is more to be said about that, cf. ch. 6).
The syntagmatic dimension that is involved in a compositional path on
the paradigmatic side by the third basic relation proposed by Langacker,
which he calls categorization (1987: 74), and whose core feature is schema-
tization (cf. above). Thus if a category subsumes another category, it is
abstract or schematic in relation to it, as fruit is abstract/schematic in
relation to apple. This relation is the successor concept in CL to the link
between Aristotelian categories based on assigning substances to their
superordinate kind, and may thus be used to form hierarchies reminis-
cent of the taxonomies that are traditionally used to illustrate conceptual
hierarchies (animal > mammal, reptile (etc); reptile > lizard, snake ).
Cognitive linguistics and cognitive grammar 51

However, because the relation is based on human acts of conceptuali-

zation, it does not carry the same ontological and universal implications
(Langacker 1987: 74n). It may depend on the situation how we choose to
set up the categorical relations between the relevant items. Apart from the
role of categorization in establishing relations between meanings, it is also
central to linguistic structure in that the formation of categories in lan-
guage itself depends on the formation of abstractions. Saliently, the kind
of meanings associated with grammatical items would not be feasible
without the formation of very abstract schematic categories: the sense
of thing in which all nouns are things is very abstract indeed, cf. above
p. 41.
The bottom-up approach applies also to this dimension. Abstractions
in the cognitive view do not arise out of thin air but from acts of categori-
zation applied to concrete items just as combinatory relations depend
on the concrete units that we combine. The bottom-up principle is also
associated with the general view of what language is. If you take units to
be basic, and combinations to be a subsequent and later stage, then the
only sensible way to approach syntactic combination is bottom-up.20
Langacker fleshed out this general picture with a systematic reinter-
pretation of the whole area of basic grammar in terms of symbolic struc-
tures. In all cases, he provides interpretations linking up linguistic struc-
tures with meaningful and generalizable cognitive models. What he called
the billiard-ball model may serve as a key example of this.
The billiard-ball model is posited as the cognitive underpinning of the
relation between nominal and verbal elements. It may be said to consti-
tute an idealized cognitive model of the relation between things and the
processes in which they take part. In this, it stands as a cornerstone in our

20 The same core contrast surfaces in relation to language acquisition: the bot-
tom-up approach entails that children learn units first, and then later get the
hang of how to build combinations, thus gradually mastering more complex
structures the structure-first approach entails an acquisitional path where
the underlying structural mechanisms gradually surface in the childs emerg-
ing language. And most general of all, the contrast is associated with a basic
understanding of the role of theory in relation to actual findings: Chomsky
sees the role of linguistic theory as analogous to the role of very general prin-
ciples of physical theory in relation to actual physical events stressing local
and bottom-up regularities is simply equivalent to focusing on the periphery
instead of the centre. Cognitive linguists, in contrast, reject the analogy with
physics and assume only the existence of those abstract categorizations that
actual human speakers and conceptualizers establish, which means that the
local perspective is basic.
52 Chapter 1. The heartland of Cognitive Linguistics

basic understanding of the world. In order to understand how it works, it

is necessary to maintain at the same time the basic analogy and the essen-
tial element of schematic abstraction. Nouns denote things in a very
abstract sense, which can only be defined precisely with respect to mental
operations but we understand thinghood also by reference to the cat-
egory prototype, which is a physical object located in space.
At the concrete level of the prototype, nouns are thus analogous to bil-
liard balls, which exist in space and are stable in time: billiard balls exist
also when no one is playing, in which case they just remain where they are.
In contrast, verbs are analogous to energetic interactions, which (1) exist
only in time and (2) depend for their existence on things serving as par-
ticipants in those processes. The event of hitting thus occurs in time and
requires for its occurrence two balls, which go through a process of hitting
that is extended both in space and time.
The abstractness is essential in order to understand why we can have
verbs like sleep, agree or resemble which have no direct analogy to the
energetic interaction of hitting. Some noun-verb relations are understood
not directly by reference to the prototype but rather with the schematic
level of understanding, which involves only the conceptual processes:
nouns are construed atemporally, while verbs are construed as unfolding
in time (Langacker 1987: 247248).
In addition to providing a semantic framework for the relation between
nominal and verbal meaning, the model also establishes a relation of
dependence between the two types of meaning. Verbal meaning is depend-
ent on nominal meaning in the same way that processes depend on things.
You can conceptualize billiard balls without events of hitting, but not vice
versa. Thus things are conceptually autonomous in relation to temporal
processes in which they take part. Temporal processes, in contrast, are
conceptually dependent on things. Like other relations, this dependency
relation has a symbolic basis: the meanings of verbs have reserved spaces
for participants (Langacker calls these reserved spaces elaboration sites,
i. e. sites reserved for being filled out by nominal elements, 1987: 304).
This conceptual account does a number of important things at the
same time: it provides a cognitive description of a very basic grammatical
relationship; it assigns meaning to grammatical categories that used to be
prime candidates for purely syntactic status; and it links up the nucleus of
a clause with basic elements of the way we think about the world, includ-
ing space, time, entity (thing) and process.
Other purportedly purely grammatical relations can be accounted for
by extensions of the same pattern of reasoning. The two nominal elements
in transitive predications do not have equal status in relation to the verb
Cognitive linguistics and cognitive grammar 53

they combine. Building on Talmy (1978), Langacker adopts the idea of a

basic asymmetry that is associated with the figure-ground asymmetry in
perception21 (which was pioneered by the Danish psychologist Edgar
Rubin, cf. Rubin 1915). A predicate with two participants thus privileges
one as the primary figure, the main character: Joe visited Jack is about
Joe, while Jack received Joe is about Jack. Langacker captures the relation-
ship by means of the distinction between trajector and landmark. The
trajector is the element with the privileged status, the subject candidate.
The landmark also has figure-like properties, because it is a participant
rather than part of the setting, but it is secondary in relation to the trajec-
tor (Langacker has later said that if had thought of it, he would just have
called them one and two, respectively, saving cognitive linguists a great
deal of tongue-twisting).
This dimension links up with the basic account of verbal meaning: tra-
jector-landmark alignment is part of the account of the elaboration sites.
In choosing the predicate, you therefore also choose a way of conceptual-
izing the participants: when you choose the preposition below you choose
a trajector that is vertically lower, while above has the trajector vertically
higher than the landmark. Thus although there is truth-conditional iden-
tity between the painting is above the fish tank and the fish tank is below
the painting, the two sentences do not mean the same thing, since one sen-
tence is about locating the painting, the other is about locating the fish
In general, when symbolic elements combine, the outcome can take
different forms it is not just a question of one meaning being dumped on
top of the other, with the combined meaning as an unstructured heap. The
concept of profiling acquires an additional function here. At the item
level, it highlights the carving out of the designated semantic area from its
base domain as a finger is profiled against the background of a hand (cf.
above p. 25). When it comes to understanding syntactic combination, it
can provide a symbolically grounded account of the grammatical relations
of complementation and modification.
The basic reason for this is that the profile indicates what the core con-
tent of a linguistic expression is. Profiles of contributing elements must be
involved when they combine, because there needs to be some consistent
overall profiling in the result if profiles stayed the same, meanings would

21 The force of the perceptual analogy in relation to a possible alternative source

in physical manipulation will be discussed in ch. 6. Both are of course instances
of bodily grounding.
54 Chapter 1. The heartland of Cognitive Linguistics

just lie around next to each other without collaborating. Modification is

elegantly captured by saying that a modifier-head construction takes over
the profile of the autonomous element, the head rather than the modifier.
Thus while the adjective dangerous on its own profiles a property, the
combined modifier-head construction dangerous road profiles the nomi-
nal thing (road) that has the property.
This may appear to be the natural and fully predictable solution: the
dependent element gives up its profile in favour of the element on which
it depends. However, complementation, which is the other main type of
grammatical combination, works the opposite way: the verb marry denotes
a process, and if we add an object, as in marry a Norwegian, it still denotes
a process. In other words, the autonomous thing does not impose its pro-
file on the combination instead, the dependent element, the verb,
imposes its process profile on the whole construction. Similarly for prep-
ositions: above denotes a spatial relation, and above the fish tank still
denotes a spatial relation, although a relation that has been comple-
mented at one end.
Viewing the issue of syntagmatic combination from a semantic and
bottom-up manner has other implications for the way in which properties
of combinations are related to properties of elements. Instead of purely
distributional laws as expressed in phrase structure rules like VP o V 
NP or PP o P  NP, we get a combination that also inherits semantic sub-
stance from its elements. Since the substance is brought into a new rela-
tional position, it is logical that new properties will arise as part of the
process: as a result of accommodation between the combined units, the
combination has semantic properties not found in the units themselves.
Thus the semantic substance gets pushed around in conceptual space
depending on what it teams up with, thereby extending the network of
meanings that an element can designate (cf. Langacker 1987: 76).
This also has consequences for the understanding of compositionality,
a cornerstone of formal semantics ever since Frege. Because of the proc-
ess of accommodation, the meaning of a complex expression is typically
not equal to the sum of the parts plus their manner of combination there
is almost always an extra element. Although it may be argued that accom-
modation is fundamentally a situational act rather than a feature of the
conventional meaning, typically the extra element will start to become
conventionalized whenever the combination is re-used. A bad loser is not
someone who is both a bad person and a loser; nor would it even be accu-
rate to describe him as someone who is bad at handling the event of los-
ing it is someone who is bad at it in a specific way (i. e., who cannot stand
losing and makes a nuisance of himself). Thus a feature of cognitive gram-
Final remarks 55

mar is partial compositionality (Langacker 1987: 449), reflecting again

that cognitive content rather than formal immaculate symbolization is at
the heart of language.
If we view language as a window on cognition, it is not always neces-
sary to specify precisely where the dividing-line between general cogni-
tion and specifically linguistic categories is: meaning is encyclopaedic in
the sense that there is no separate compartment for linguistic meaning as
opposed to the rest of the conceptual world. However, in combination
with profiling the three essential relations described above symboliza-
tion, categorization and syntagmatic combination provide the ground-
work for explicating the relation between linguistic units and the sur-
rounding encyclopaedic landscape. The region that can be designated by a
linguistic expression is wholly inside conceptual space, but cannot be
inferred from the non-linguistic conceptual region in itself.
In his concern to keep up the central descriptive agenda of linguistics
(cf. Langacker 1999), Langacker has to some extent ploughed a lone fur-
row in the CL landscape. In addition to showing how general cognitive
mechanisms and principles could be integrated, he has systematically
been rethinking the whole home territory of linguistics, shifting it from a
formal to a semantic foundation and showing how apparently arbitrary
features could be assigned a place in a theory based on meaning and moti-
vation. Precisely because of that, however, he has received growing recog-
nition also among authors who did not have the same commitment to the
descriptivist agenda. He has done a job that was essential to the creden-
tials of CL as a theory of language.

9. Final remarks

In this chapter, I have tried to give a condensed survey of the key ele-
ments in the rejuvenation of linguistics that CL brought about. I have
deliberately tried to create an idealized cognitive model of what may be
termed classic cognitive linguistics, and therefore backgrounded prob-
lems as well as issues which are central to the development that this book
focuses on; both kinds of issues will be taken up later on. While I recognize
the lacunae left by omissions due to condensation and idealization, of
course I hope to have given an impression of a view of language and cog-
nition that is both valid and innovative in a way that is not diminished by
subsequent qualifications.
All available truths are partial, and classic cognitive linguistics is no
exception. However, the tearing down of the wall between the general
56 Chapter 1. The heartland of Cognitive Linguistics

domain of human understanding and the understanding of language is a

lasting achievement, and I also regard the concepts discussed above as
lasting contributions to the analysis of the wider conceptual world that
opened up. All the concepts discussed above play an important role in
understanding meaning, the central property of language in CL.
In my discussion above, I have sometimes suggested how construals of
the central concepts that highlight their functional dimension may clarify
the division of labour between them. As we have seen, Croft & Cruse
(2004: 15) equate the terms frame and domain, and Barcelona (2007:
53) defines what he calls functional domains as being equivalent to Fill-
mores frames and Lakoffs idealized cognitive models. This is natural
because in purely conceptual terms they will often overlap: in such cases
domain, frame, ICM, mental space and concept may come down to the
same thing. But the distinction in terms of the functional roles of the con-
cepts should be maintained, so that they can be kept distinct when needed.
Let me end up this introduction by offering an example where they come
down to the same thing, and one where they need to be kept distinct.
In the right context, the dreamy-eyed utterance If I were a billion-
aire may evoke a concept, domain, frame, model, and mental space at
the same time. The concept captures the category of people who own ten-
digit fortunes; the domain is the generic hierarchy of affluence with bil-
lionaires at the top; the frame might be a magazine story about the super-
rich; the ICM invoked is the world of sixty-foot yachts and champagne
parties; the mental space is the alternative reality in which the speaker
imagines himself. Although the functional difference between the five
ways in which boundless wealth is evoked is just discernible, there is so
much duplication across them that simply talking about the millionaire
frame is a useful shorthand.
However, the more or less invisible functional differentiation may in
other cases be crucial to capture what is going on. To take a case in point,
the Nigerian oil emails that large numbers of people have received also
evoke the chance to become a millionaire, but here the overall picture is
rather more complicated. By asking the recipients help in getting an
enormous fortune out of the reach of an oppressive government, the email
raises the possibility of becoming a millionaire, but without evoking the
concept or the domain explicitly. The prospect of coming into possession
of a very large sum of money, however, can hardly fail to invoke the ideal-
ized cognitive model of being a millionaire.
The frame analysis has (at least) two levels: at the syntactico-semantic
level of the FrameNet, it evokes the money transfer frame, since the large
amount is mentioned in connection with the need to find an account to
Final remarks 57

which the money can be transferred so as to get out of the reach of the
oppressive powers-that-be: this is often a subframe of the commercial
transaction frame, but in these letters it tends to occur on its own (with
only three central frame elements, i. e. the money and its source and target
At the level of the whole utterance, there are two frame analyses pos-
sible: there is an explicitly invoked frame of political abuse of power,
which defines a slot for the message as a request for help (such that com-
pliance would coincidentally put some money in the recipients posses-
sion). The real frame is that of a scam to drain the accounts into which the
Nigerian money was supposed to flow, and in that frame the letter consti-
tutes the bait. Since there is an alternativity relation between the two
frames, we need to distinguish between two mental spaces, one for each of
the two frames which are linked by counterpart relations such that the
sender is a victim of persecution in one space and a crook in the other, and
the recipient is a potential millionaire in one and a sucker in the other.
Although this construal of the basic concepts in CL on some points
contrasts with proposals from other authors (some of which are cited
above), the point is not to criticize these alternatives, which are well-moti-
vated in terms of the purposes they serve in their respective contexts.
These narrowing-down construals are suggested in the hope that they may
serve to bring out the full and differentiated potential of these concepts,
reflecting an overall strategy of differentiation-within-continuity, where
CL in my view has tended towards emphasizing continuity at the risk of
making precise differentiation difficult. I think the need for continuity has
now been emphasized strongly enough to make it safe to move in the
other direction and stress differentiation: if all the theory offered were the
millionaire frame/space/domain/model with no clear differentiation, the
analysis would be somewhat less precise.
In the next chapter, the subject is the ongoing developments in CL that
constitute the social turn, the expansion from the individual mind to soci-
ety; this will affect especially the understanding of grounding, prototypes
and image schemas. I return to some core concepts of classic CL in chap-
ter 5, with a view to suggesting a reprofiling of their role in view of the
extended framework proposed in chapters 14.
Chapter 2. From conceptual representations to
social processes: aspects of the ongoing social turn

1. Introduction

This chapter is about the aspects of CL that go beyond the individual

mind. Some of them have always been there but are expanding, others
are new developments. In combination, they are shifting the focus of CL.
The key change is the recognition of factors outside the individual mind
as explanatory dimensions. These developments can be seen as a process
of living up to Langackers (1988) commitment to a usage based linguis-
tics which has gained momentum since the turn of the millennium (cf.
the title Usage Based Models of Language, Barlow & Kemmer 2000).
Another pervasive motif is the role of fellow subjects in human concep-
tualization. Two keywords that reflect these dimensions are variation and
I begin with conceptualization as a feature of the social and cultural
sphere (section 2). Section 3 introduces variational description (see also
ch. 6). Section 4 is on development and acquisition, with intersubjective
dimensions in focus. In section 5, I discuss the implications of these dimen-
sions for the central notions of embodiment and grounding. Section 6 dis-
cusses the evolutionary framework as an overall model for understanding
cognition in a wider context. Section 7 takes up a new focus that is directly
linked to the usage orientation for the study of meaning: the shift from
conceptual structures in the individual mind to the interactive process of
meaning construction.

2. Cognitivism and conceptualization in the sociocultural sphere

There has always been a social dimension in cognitive linguistics, but

social factors have been at one remove from centre-stage position in CL.
They come into the picture by the same route as physical factors: via cog-
nitive representations in the mind. Focusing on the mind-internal dimen-
sion goes naturally with the conceptualization of linguistics as part of cog-
nitive science, whose job it is to model the powers of the human mind as
part of the general description of how a human being works.
Cognitivism and conceptualization in the sociocultural sphere 59

In this way the social dimension can easily be accommodated within a

mind-internal perspective; it is hard to tell the difference between social
facts and cognitive representations of social facts. Frames (cf. p. 23) illus-
trate the natural conflation: calendrical systems are parts of the social
order and also mind-internal cognitive models. Theories of the individual
mind can naturally be extended to the analysis of cultural phenomena, as
described by Talmy in the article entitled The Cognitive Culture System
(Talmy 2000. vol. II, p. 373):
Cognitivism indicates that cultural patterns exist primarily because of the cog-
nitive organisation in each of the individuals collectively making up a society.
This analysis arrives at particular positions on the issues of what is universal
across cultures and what varies, of what is innate and what is learned, and of
how the individual and the group are related. This cognitivist view of culture
disputes several other theoretical positions, such as the position that culture
has mainly or solely an autonomous existence beyond the cognition of individ-
ual humans.
Adopting a mind-internal source of explanation was a pervasive trend
during what may be called the cognitive era from 1960 onwards.1 Cogni-
tivism does not entail a focus on mind-internal objects of description
only on mind-internal sources of explanation. The interest in cultural and
political issues within CL is manifested most flamboyantly in work by
George Lakoff (1987, 1996, 2004, 2006, 2008) demonstrating how idealized
cognitive models and metaphorical mappings can be used to understand
American politics.
His analyses illustrate some of the basic properties of classical CL, the
grounding of abstract thought in more concrete domains, and the role of
embodied experience in motivating conceptual mappings. In a series of
publications, beginning with Moral Politics (1996), Lakoff has shown how
the ideological divide between the liberal and the conservative positions in
the US can be understood in terms of two different idealized cognitive
models of the family: the strict father family and the nurturant parent
family, and has suggested what implications may follow from this link
between politics and cognitive structures. Lakoff argues that the conserva-
tives have been much better at motivating their supporters because they
have grounded their policies in terms of something everybody could under-
stand and relate to. If the liberals are to be effective in fighting back, they
must reconquer the lost social territory by finding an equally powerful way
of grounding their policies in structures that resonate in individual minds.

1 Cf. also Hutchins (1995) as quoted below p. 84 on cognitive anthropology.

60 Chapter 2. From conceptual representations to social processes

Arguing for the superiority of a nurturant family model, Lakoff points

out how conceptual models based in bodily experience have implications
that generalize into the sociopolitical domain: The nurturing family can
serve as a model for a society based on mutual support rather than on
unquestioning support of the president as chief executive. The family is
where human beings first meet the practice of government, cf. Lakoff
2008, and so the response to government on the larger scene draws on
those formative, embodied experiences (cf. the discussion of primary met-
aphor above p. 37).
Lakoff has analysed other concepts, including the key concept of free-
dom (Lakoff 2006) on the same essential premises. Framing is a key instru-
ment in political campaigning: if you can get people to mobilize a frame
that motivates your positions, the battle is half won. Lakoff has helped to
found a think tank (The Rockridge Foundation) based on the idea of
grounding liberal political argumentation in cognitive models that are
immediately understandable in terms of the everyday embodied experi-
ence of individual voters.
Lakoffs work on cultural and political issues is predicated on essen-
tially the cognitivist position expressed by Talmy. His view of the relation
between the mental and the social sphere comes out in an interview by
Pires de Oliveira (2000). Entitled Language and Ideology, the interview
challenges Lakoff from a point of view that stresses the potential constitu-
tive role of socially entrenched value systems for cognition and for cogni-
tive linguistics. Pires de Oliveira, while sympathetic to Lakoffs views,
probes the question of how facts belonging in the social dimension may
have implications for cognitive facts. Lakoffs position, however, is clearly
that the directionality is from mind-internal conceptual facts to social
facts. In analysing socially relevant cognitive structures, including ideolo-
gies, the place where cognitive linguistics can really make a difference is in
analysing unconscious frames and metaphors lying behind conscious
beliefs (p. 37).
If we temporarily abstract from the issue of the relation between con-
scious and unconscious phenomena (but cf. the discussion in ch. 5), there
is nothing wrong with doing research from the cognitivist perspective, i. e.
to look for the impact of mind-internal cognitive structures on social phe-
nomena. From the point of view of this book, it covers only one half of the
issue but you can only pursue one explanatory direction at a time. The
problem lies in ruling out the direction that begins with social processes
and looks at the mind in that perspective and there is little room for that
in Lakoffs thinking. His view is clearly that language is only indirectly
social: Since language reflects our conceptual systems, it will reflect the
Cognitivism and conceptualization in the sociocultural sphere 61

social aspects of our conceptual systems (p. 37). When Pires de Oliveira,
pressing her point, suggests that socially based assumptions might to some
extent shape Lakoffs views on language and ideology, she gets rebuffed in
no uncertain terms2.
This position did not remain unchallenged. The sense of stepping over
a new threshold can be exemplified with reference to two programmatic
articles: Taking Metaphor out of our Heads and Putting it into the Cultural
World (Gibbs 1999) and The Social Dimension of a Cognitive Grammar
(Hawkins 1997).
As a psychologist engaged in CL, Ray Gibbs has devoted much of his
attention to research into the mental reality of the phenomena that are
central to CL, especially metaphor (cf. Gibbs 1994, Gibbs & Colston 1995).
One strand of this research had taken the form of experiments showing
empirically that activating conceptual metaphors has effects for cognitive
processing that cannot be accounted for by purely linguistic similarity or
generalizations based upon them.3 But looking back on the evidence,
Gibbs points out that there is another step that needs to be taken (the
continuing path of recontextualization, cf. p. 3). The classical CL assump-
tion is the following (cf. Gibbs 1999: 146):
Most cognitive scientists supportive of the conceptual view of metaphor tacitly,
and sometimes explicitly, assume that conventional metaphorical mappings
must be internally represented in the individual minds of language users.

Instead, Gibbs suggests that

cognitive linguists and cognitive psychologists, like myself, should think
about metaphor and its relation to thought as cognitive webs that extend
beyond individual minds and are spread out into the cultural world.

2 Lakoff summed up the discussion as follows (Pires de Oliveira 2000: 34): You
asked, Isnt it based upon our own biases (that of a white, male, Anglo-Saxon
Protestant)? Putting aside the racism and sexism of the question, the answer
is no. It is an empirical issue.
3 Thus the conceptual mapping underlying the metaphor ANGER IS HEATED
FLUID IN A CONTAINER, once activated, will connect up with expressions
such as make ones blood boil, blow ones stack and hit the ceiling, although
there is no literal similarity (or literally based abstractions) linking these
expressions. This work was crucial to the empirical underpinning of the basic
assumption of CL, namely that linguistic meanings are not specific to language
as a self-contained domain but draw on general conceptual mechanisms and
62 Chapter 2. From conceptual representations to social processes

Taking himself to task (1999: 155) for not explicitly acknowledging this
dimension in previous work, Gibbs points out the essential link between
the social evaluation of an (annoying) event such as being kicked in the
leg and the question of whether it gives rise to the embodied experience
of anger as a fluid with yourself as the container. Depending on whether
the kick is perceived (socioculturally) as a deliberate unmotivated attack,
an accident, or a well-deserved revenge, the embodied response may dif-
fer considerably.
As an extra dimension, he points out the role of external symbols for
internal experience (including pictorial representations of people with
steam blowing out of their ears, p. 158), as a significant locus for meta-
phorical mappings. With the therapeutic practice of Erickson as an exam-
ple, Gibbs goes on to show how overt social practices such as sharing a
meal can serve as scaffolding for approaching emotional relations.4 For
the purposes of this book, however, it is significant that there is still a
sense that the object of description remains basically the same: in his con-
clusion, Gibbs suggests that there is much less of a difference between
what is cognitive and what is cultural than perhaps many of us have been
traditionally led to believe (1999: 162).
From a different perspective, Hawkins (1997, 2000) expresses similar
views, pointing out both that there is a field outside the individual mind
that needs to be explored, and that cognitive grammar is well equipped to
handle meaning in the social sphere (1997: 22). Hawkins announces his
agenda in the first lines of his article: The purpose of this paper is to dem-
onstrate that a cognitive grammar can and should attend to the socio-polit-
ical aspects of language use. Hawkins takes up the direction from social to
individual meaning, focusing on the role of ideology (with reference to
Althusser (1971) and Hodge & Kress (1993). In his approach he defines a
field of interest constituted by strongly emotional (positive or negative)
ways of referring to socially salient entities5. Presenting two obituaries of
Richard Nixon, he demonstrates how one tends towards the extreme pos-
itive pole (one of the great statesmen of this century) while the other is
located at the extreme negative pole (the only president forced to resign
the nations highest trust).

4 From his perspective, he is on to the same mechanism that is discussed later in

relation to material anchors (cf. p. 85): if overt practices are adjusted first,
emotional relations may piggyback on them, reducing the burden imposed on
human subjects.
5 Hawkins calls it iconography (in a different sense from the one associated
with Panofsky).
Cognitivism and conceptualization in the sociocultural sphere 63

In spite of the possibly alien flavour,6 Hawkins suggestion is fully in

harmony with Langackers (1988) view of cognitive linguistics as part of a
wider project of usage-based linguistics: the processes that create socially
defined positive icons and negative caricatures are cognitively imbued
forms of usage. In England, there has been more contact between critical
linguistics and cognitive linguistics, with Paul Chilton as a key figure; in ch.
3 below, I discuss approaches to meaning that start from the social rather
than the cognitive end, including the French anti-humanists; in ch. 7, I
present my own view of how a social cognitive linguistics can be set up to
handle the legitimate issues that they address, and I postpone the discus-
sion of the cognitive/critical interface until then (p. 394).
The political agenda is fully in harmony with Lakoff-style analyses
based on cognitivist premises, as described above. But forces operating in
the political sphere call for analyses that go beyond cognitivism. The dif-
ference is between seeing the mind as the sole explanatory factor and
essential object of description and opening the door to mind-external
facts as a source of explanation, cf. Geeraertss warning ([1988] 2006: 47)
against a tendency that is a natural characteristic of Cognitive Seman-
tics: the tendency, in fact, to look for purely cognitive or conceptual expla-
One example of CL-based analysis following this trajectory into the
sociopolitical sphere is the work of Brigitte Nerlich, which has broadened
from traditional linguistic topic areas in ways reflected by the designation
of her present chair as Professor of Science, Language, and Society. The
three topic areas are combined in Nerlich, Elliott and Larson (2009),
which deals with the social understanding and communication of science.
Among the striking results of this interdisciplinarity is the use of CL to
characterize the changing institutional pressures on science and the media
practices that affect the institution in terms of conceptual models, frames
and metaphors that are used in communicating about science. An exam-
ple of how these concerns address the same concerns as this book is their
discussion of increasing role of hype in science communication (cf. Ner-
lich, Elliott and Larson 2009: 3) as a result of sharpened competition for
funding and public support cf. the discussion of bullshit in ch.7.

6 The position he presents is one in which ideologies as socially entrenched

ways of thinking are at the same time part of the CL project and not quite
recognized as such. This view of what CL included was not universally
accepted: in Hawkins (2000:19), Hawkins explicitly reports the dismissive
response that questions of ideology drew from a CL audience when he pre-
sented his 1997 paper.
64 Chapter 2. From conceptual representations to social processes

The volume reaches beyond academia, including also accounts of insti-

tutional practices: the director of an institution (cf. Fox 2009) set up to
improve science communication in the media gives an (optimistic) account
of the bridge-building that has been accomplished, and an experienced
journalist (Strauss 2009) describes his professional efforts to provide the
ways of (metaphorically) conceptualizing science in a way that will enable
optimal understanding across the divide between experts and the public.
A further part of the perspective is the analysis of the potential impact on
science as a social institution (cf. Nerlich 2009) and its impact on and rela-
tions with the public. This illustrates how the object of investigation in CL
is widening to include the role of cognition in a larger social context.

3. Variation, lexical semantics and corpus linguistics

With respect to the usage based commitment, Dirk Geeraerts has occu-
pied a unique position in the development of cognitive linguistics, both
because of his background in historical lexical semantics and his dual
commitment to theory and to practical lexicography. Across his multifari-
ous research topics, the social life of words and meanings, as opposed to
the purely mind-internal dimension, stands out as the focus of interest.
Especially from the lexical point of view, his approach has provided a
broad sociocultural enrichment of cognitive semantics.
The concept of prototype is one example where Geeraerts (e. g. 1989)
reorients the significance of the concept from conceptual structure per se
to what prototypicality reveals about the relation between incoming expe-
rience and the mental organization of information.7 The key feature is
that prototypicality is essentially a result of the cognitive effort of manag-
ing experience in a flexible way.
This reflects the onomasiological perspective on meaning, i. e. the
approach that begins not with a word or concept but with the designated
phenomenon. Starting with the communicative task, onomasiology asks:
how can we name this object by linguistic means? This contrasts with the
traditional semasiological perspective that starts with the word and asks
what kinds of phenomenon it can denote. To illustrate the significance of
this perspective: Dutch offers two alternatives for destroy, cf. Geeraerts

7 This very general point is reflected in a broad range of disciplines (including

for example Kuhns theory of science, with its concept of scientific paradigms,
cf. p. 105 below).
Variation, lexical semantics and corpus linguistics 65

(1988), vernielen and vernietigen, which can be used about virtually the
same range of conceivable cases. Working from the words themselves, it is
hard to establish a clear difference. However, starting out with a usage
survey, looking at the kinds of things the words are used about, Geeraerts
can establish a pattern: vernielen (whose root is to do with tearing down)
is mainly used about concrete physical objects, and the result can typically
be viewed as constituting damage. In contrast, vernietigen (whose root is
to do with negation) is typically used about abstract objects, and results in
absence of the object. This also enables a greater degree of precision in
the area of conceptualization: atypical uses do not eliminate the core
senses, but impose atypical conceptualizations, like demolishing ideas or
negating physical objects.
The historical trajectory of meaning and naming is the subject of a
large-scale empirical investigation reported in Diachronic prototype
semantics (Geeraerts 1997). The introduction of legging as a new type of
clothing is studied in minute empirical detail, with mail order catalogues
and womens magazines as the corpus material. Only sources with picto-
rial representations are used, enabling researchers (1) to provide a feature
representation of instantiations named by the terms (including length,
width, material, etc.), (2) to cover also potential referents that are actually
named by a different term (including superordinate terms like trousers).
The development is followed in the years from 1988 to 1992, during which
the category undergoes a rapid development (cf. Geeraerts 1997: 3640).
In all years there was a core instantiation type (below calf length, maxi-
mally tight-fitting, without crease, smooth and finely woven/knitted,
upper rather than underwear, for female use) but the clothing article
became a success and the linguistic labels associated with it were carried
along by this success. One of the results that illustrate the process of social
expansion is that the increased incidence of core instantiations went hand
in hand with a gradual expansion in the range of variation: from the pro-
totypical centre, the successful new concept spread out into the surround-
ing terrain, establishing a rich new radial model in this lexical area.8

8 The significance of the onomasiological approach can be illustrated with the

measure they develop for degree of entrenchment of the linguistic terms:
because also cases that were nameable but (in concrete cases) not actually
named by the new terms could be captured (by means of the pictorial repre-
sentations), figures could be given for the development whereby the new word
acquired a gradually increasing salience, since it was used to name more and
more of the potential cases.
66 Chapter 2. From conceptual representations to social processes

In the legging case, the social process involved a new physical object.
This makes it easy to see the existence of both directions: from the social
phenomenon to the word, and from the word to the social phenomenon.
Social processes, however, also have their own role to play when the phys-
ical world remains the same, but the sociocultural processes cause them to
be understood differently. This creates a situation where it is more diffi-
cult to tease the cognitive and the social dimensions apart. This can be
exemplified with the case of the humours (etymologically = liquid)
model of emotions (cf. Geeraerts and Grondelaers 1995), which looks at
social processes that involve the history of high culture, including the his-
tory of ideas.
This case presents a direct contrast between a usage based and social
approach and a traditional cognitivist approach. Arguing against Kvec-
ses as a representative of the classical CL position of analysis in terms of
bodily based cognitive mappings (cf. Kvecses 1986), Grondelaers and
Geeraerts argue that the physiological basis (whose existence they do not
deny) is overlaid by a tradition from antiquity of analysing emotions in
terms of the four humours: black bile for melancholy, yellow bile for
anger, phlegm for sluggishness, and blood for light-heartedness The analy-
sis of the anger metaphor, understood in purely physiological terms, does
not explain the salience of the fluid element (as in ANGER IS THE HEAT OF A
FLUID IN THE CONTAINER) over the heat and fire elements. Also, the wealth
of expressions of other terms of fluids and temperatures show that there
is a historical pattern that cannot be predicted from universals of physiol-
ogy (cf. also Gevaert 2005). Among other things, this raises the question,
underplayed by Lakoff, of the significance of the distinction between liv-
ing and dead metaphors: if a way of thinking slowly becomes obsolete,
retrieving it as a source becomes gradually less natural. The understand-
ing of choleric and phlegmatic emotional responses in the 21st century is
likely to evoke bodily humours in a diminishing segment of language
A major priority for Geeraerts is to expand the methodological base of
cognitive linguistics. One of the really hard core consequences of taking
the sociocultural dimension seriously is the need to go beyond introspec-
tion: since usage is not accessible by introspection, a usage-based model
cannot be based solely on it; empirical, quantitative data are needed. This
calls for a number of different enrichments of traditional descriptive prac-
tices. A central factor is the rapidly growing availability of electronic cor-
pora and associated statistical descriptive tools, which make possible a far
greater breadth and depth of empirical coverage than previously availa-
ble. A descriptive aim facilitated by this tool is the possibility of account-
Variation, lexical semantics and corpus linguistics 67

ing for variational patterns, i. e. the type of fact that used to be the prov-
ince of sociolinguistics (including dialect geography), cf. Geeraerts (2005).
Grondelaers and Geeraerts (2003) demonstrate that lexicology is incom-
plete without an integration of the sociolinguistic dimension: the key
object of description, lexical choice, simply cannot be described without
an integrated sociolinguistic dimension. Just as the choice of phonetic
variants is determined by the social pressures (including the Labovian
parameters of style and social class), so is lexical choice.
One of the factors in prototypicality that become highlighted in the
usage-oriented perspective is the role of salience itself: what factors are
most important depends on speakers and the situation. This expands a
classic point made by CL against objectivist semantics: rather than being
an unfortunate misrepresentation of the real world, this is a reflection of
the fact that words inherently serve purposes for language users; thus con-
cepts need to be understood in relation to both naming and linguistic con-
ceptualization practices, including pragmatic variability (cf. Geeraerts
As an example of the interfacing between meaning relations tradition-
ally viewed as intra-linguistic and social factors that are traditionally
viewed as external, Grondelaers and Geeraerts investigate lexical choice
in names of cancerous diseases. The internal dimension they focus on is
lexical specificity, i. e. the choice between superordinate terms like illness
or disease, and more precise diagnoses such as cancer, or breast cancer.
Based on a corpus divided into medical and non-medical texts and
between personalized and generic contexts it is found (as predicted) that
there is a significantly higher frequency of the less threatening and more
general terms illness and disease in personal contexts (as compared with
generic and professional contexts), reflecting an avoidance strategy when
speaking of people close to you. Clearly, it would be arbitrary and unre-
vealing to separate this description into two sub-descriptions, one of which
dealt with level of semantic specificity and the other dealt with the impact
of social contexts.
The kind of issues that are thus brought into CL can be illustrated with
a long-standing research process involving the use or non-use of er, the
Dutch equivalent to existential there, in construction types with initial
locative (cf Grondelaers et al. 2002, Grondelaers, Speelman and Geer-
aerts 2008, 2008b). The illustration case is In the ashtray (there) was a
hailstone. Introspection about conceptual content does not provide
immediately striking ideas about how to account for the choice, if you fol-
low the standard structuralist commutation procedure of looking at the
same example with or without er. It is hard to verify any ideas that may
68 Chapter 2. From conceptual representations to social processes

come into your head, first of all because the difference is not very great
(and does to pertain to truth conditions or other more tangible criteria).
As always in such cases, looking for too long at the examples may con-
found any intuitions you might spontaneously have had.
Corpus investigations therefore have a distinctive explanatory poten-
tial, and they have made it possible to show that er serves as an inacces-
sibility marker, i. e. its occurrence correlates with the extent to which the
object that comes after er is unpredictable. Thus after the concert came a
reception would be without er, while after the concert there was an
earthquake would tend to have er. In addition, there is regional variation.
Belgian speakers are more prone to use er than speakers of Netherlandic
Dutch. By logistical regression, it can be shown that you can account for a
very large part of the variation (85 %) that way.
In one investigation, Grondelaers and Speelman (2008) pursued the
methodological agenda one step further, in an attempt to get at the
remaining 15 % of the variation that eluded logistical regression based on
the features that had previously been found to make difference. Although
questionnaire investigations have obvious limitations as guides to actual
usage, since they tap error-prone introspective judgments, they might be
one way of going beyond what corpus data could yield. The study did
reveal some regularities but on closer examination confirmed the reserva-
tions. Among other things, a retest revealed that there was an instability
factor in the judgements. Then again, it was interesting in that the instabil-
ity primarily affected cases without er, rather than the er examples: intro-
spection-based judgments changed significantly for the er-less examples.
While this does not settle the issue with respect to remaining factors, it
illustrates the difficulty of getting reliable data by asking people to intro-
spect. The research group at the University of Leuven to which Geeraerts
belongs (Quantitative Lexicology and Variational Linguistics, QLVL) has
since 2000 produced a steady stream of work that combines the social and
conceptual dimensions, thus providing an institutional focus for this kind
of work within cognitive linguistics.
The variationist and the history of ideas perspective are combined in
the analysis of culturally salient alternative approaches to linguistic varia-
tion in different national traditions (Geeraerts 2003c). There are two
influential models: the rationalist and the romantic model. The rationalist
or enlightenment model focuses on the virtues of a standard language as
a crucial means to enable all people to take up the ideal role of citizen
(citoyen, in the terminology of the French Revolution). This role (as
opposed to the stereotype bourgeois) involves emancipation from out-
moded constraints, and shared and equal participation in the political
Variation, lexical semantics and corpus linguistics 69

processes of the Republic. The Romantic model focuses on language as a

means to express your personal, traditional, Herderian Weltanschauung,
and sees standardization as a Procrustean bed that violates local and
authentic identity. Emphasis on the role of dialects goes with a romantic
view of language9 while standardization in language, including a shared
form of spelling, goes with the establishment of a coherent and universal
school system.10
While emphasizing the need for a solid empirical and statistical foun-
dation for lexical research, Geeraerts simultaneously points out that the
type of investigation that follows meaning through historical and cultural
changes inevitably also brings back a hermeneutic dimension into linguis-
tics. Lexical semantics is a Geisteswissenschaft in Diltheys sense (cf. Geer-
aerts 1992; Geeraerts 1997: 175), belonging inside the purview of human
interpretation. There is no contradiction here: empirical solidity is neces-
sary to ensure that interpretive processes are on firm ground (Geeraerts
1997: 185). There is an important distinction between introspection and
intuition: any kind of linguistic description depends on sound intuitive
judgement (cf. Itkonen 2008: 291 and below p. 156 on participant access);

9 In Denmark, for instance, the influence of the religious, educational and cul-
tural reformer Grundtvig gave rise to dialectology as the study of the authen-
tic voice of the people.
10 Geeraerts follows the two models through two series of transformations. In
the age of nationalism in the 19th century, the enlightenment model becomes
associated with the nation state as the basis of a liberal democracy, while the
Romantic model assumes the guise of the medium of national identity; in the
present postmodern age, rationalism underpins the movement towards global
English as a medium of international opportunity and co-operation, while the
Romantic model becomes recruited to defend multilingualism, whether of
ethnic minorities or of nation states under the pressures of increasingly anglo-
phone international collaboration. Quantitative measures of uniformity in
different strata in different time periods of Flemish and Netherlandic Dutch
are subsequently used to discuss the development in the light of the twin pres-
sures of ethnic identity (romantic) and standardization (rationalist). Three
possibilities are raised, with no attempt at prediction but the dual contextu-
alization of lexical choice in the context of regional and diachronic variation
as well as in the cultural pressures of standardization and ethnic factionaliza-
tion means that the question of whether a speaker of Dutch decides to use
(over)hemd vs. shirt (for a shirt) or caleon vs. legging(s) becomes fraught with
somewhat greater significance than a pre-variationist glance in a dictionary
would lead one to believe.
70 Chapter 2. From conceptual representations to social processes

but using introspection as the only source of data is a severe limitation on

what your intuitions have to work with.
It would be inaccurate to say that this is a new turn, since it has been
around from the beginnings of CL. But it has gathered momentum in
recent years, and its concerns are becoming generally shared, as exempli-
fied by the recent volume on Cognitive Sociolinguistics (Kristiansen and
Dirven 2008) which gathers the cultural, lectal, social and political aspect
under one theoretical aegis.

4. The developmental perspective: epigenesis, joint attention

and cultural learning

For an approach to language based on experiential grounding, it is neces-

sary to have an account of what happens on the path from the womb to
full membership of the speech community. The process of establishing the
ontogenetic basis for CL can be described in relation to two main figures,
Sinha and Tomasello. They both emphasize the intersubjective and cul-
tural dimensions as foundational aspects of the process.
As a developmental psychologist, Sinha has a different perspective on
cognition and language from that of cognitive psychologists. Cognitive
psychology in its modern version started with the computer metaphor,
because that was what made mental content scientifically respectable.
Developmental psychology had a broader base: for clinical reasons alone,
no one questioned the need to understand typical paths of mental devel-
It is from this broader perspective that Sinha approached the develop-
mental issues that are central to the agenda of CL. Language and Repre-
sentation (Sinha 1988) is an attempt to provide a new grounding for the
developmental study of human representational capacity with linguistic
representation in a focal position. Taking his point of departure in a criti-
cal stance towards the computational modern synthesis which domi-
nates contemporary cognitive psychology (1988: xiii), Sinha sets out to
establish what he calls a socio-naturalistic approach, i. e. one that inte-
grates natural and social sciences both in the conception of the object of
description and in methodology. A centrepiece of this foundation is the
notion of the materiality of representation, i. e. one in which the repre-
sentative powers of the mind basically unfold themselves not in an imma-
terial inner space but in a universe constituted by interaction between the
subject and the environment. This does not entail a reductionist scepti-
cism towards the mind as such, but an orientation towards describing the
The developmental perspective 71

mind as an integrated part of the overall practical and material world in

which human beings live.
At the time of the publication of the book, it was necessary to do a
great deal of clearing before such descriptive foundations could be marked
off effectively from its vociferous contemporary competitors, including
not only logical positivism and Chomskyan formalism but also Marxism,
Freudianism, and critical theory. The key difficulty is perhaps in carving
out a position from which you can keep at bay both the abstract and inna-
tist, computational approach to cognition and the sociologically oriented
belief in relativism, going back to Marxs belief that human essence is
nothing other than the ensemble of social relations, cf. Sinha (1988: 169).
This is why a central concept for the position that avoids both pitfalls is
the notion of epigenesis (borrowed from Waddington 1975, 1977 (cf. Sinha
2006; Sinha 2007: 1277). It designates a type of development that is super-
imposed upon rather than triggered by genetic determination but
depends on interaction between organism and environment rather than
simply being imposed by environmental factors.
This position is in harmony with classical assumptions about child
development. Although Piaget is frequently misunderstood as describing
an innate, maturational developmental trajectory, his approach to learn-
ing in terms of adaptation goes beyond innateness: adapting old schemas
to new operations, as in accommodation, is clearly an instance of superim-
posing experience-based learning on innate skills. The Piagetian trajec-
tory that begins with pre-representational modes of understanding also
avoids the extreme of Chomskyan nativism, in which learning really does
not exist. An even more significant figure in the picture outlined by Sinha
is Vygotsky, who represents the directionality that is underplayed in Piaget
(cf. Sinha 1988: 92f): the internalization of sociocultural practices, and its
importance for addressing the issue of integrating the study of social and
biological factors. The existence of patterns outside the individual, in cul-
tural and interactive practices, can thereby enter into the picture without
rendering the cognitive processes of the subject irrelevant. Another
important foundational figure (cf. Sinha 2007) is Karl Bhler, whose most
familiar contribution to the psychology of language is the organon model.
This model views the utterance as constitutively embedded in a field of
three interlocking functions: one reflecting its human source, the sender of
the message (the expressive function), the other reflecting its orientation
towards the addressee (the appeal function), and the third its role as rep-
resenting the external world, cf. Bhler (1934).
The central object of description for Sinha is therefore the human sub-
ject in its interaction with fellow subjects and the world, rather than the
72 Chapter 2. From conceptual representations to social processes

human cognitive system per se. The development of cognitive capacities

must be understood as part of the overall development of abilities to cope
with the environment. A central part of Sinhas position is that children
are born to be fellow subjects: intersubjective embedding is not something
that is subsequent to the development of the subject as such. Rather,
human infancy has evolved so as to prepare the emerging new subject for
intersubjectivity as its medium of existence (Sinha 1988: 107). On this
point, Sinha therefore moves beyond Vygotsky whose materialist
approach to some extent presupposes the divide between subject and
object that his approach subsequently seeks to bridge from the social side,
cf. Sinha (1988: 103).
These foundational views also have implications for language learning.
In Sinha et al. (1994), this approach is applied to an empirical study of the
acquisition of spatial semantics. Findings from Danish, English and Japa-
nese confirm the central assumption of this approach, that the rise of spa-
tial categories must be understood as a process of gradually extending an
ability that is based on (but not predictable from) universal cognitive cat-
egories in the spatial domain. Although containment is a well-supported
pre-linguistic cognitive category (Sinha et al 1994: 28283), actual facts
about the acquisition of markers of containment can only be explained by
taking into account features of the language that is being learnt: there is
more to it than simple labelling.
The fact that learning is conservative is also in harmony with a pattern
of gradually extended coping: each successfully mastered point in the tar-
get area is used as a vantage point for further progress. Among early
acquired expressions are those which are basic both from a cognitive
point of view and from the linguistic point of view: simple expressions for
simple spatial relations. This gives rise to a feature that reflects a core CL
concept, that of radial structure: Both in acquiring the whole field and in
acquiring the individual item, learners start out with central elements
(central spatial concepts and central senses of individual spatial concepts,
such as the canonical containment relation for in). Subsequently they ven-
ture into more marginal positions, gradually establishing a richer network
while preserving central points of orientation (cf. also the pattern found
for diachronic development of lexical meaning by Geeraerts above p. 65).
More radically, the potential of the epigenetic approach for causing a
revision of classic assumptions of CL is evidenced by a joint article (Sinha
et al. in press). It gives an account of Amondawa, an Amazonian language,
in which there are (almost) none of the familiar linguistic and cultural
resources for organizing temporal reference: no calendars, no verbal
tense, no word for time, and reference to temporal points and periods is
The developmental perspective 73

possible only via events occurring at those times (beyond simple deictic
temporal reference based on time of utterance). There is an extended
grammatical system of nominal aspect, so the language is by no means
generally impoverished, but it has no cultural or linguistic evidence of
either segmentation of time, metaphorical division of time via spatial con-
cepts, or any other obvious way of documenting an awareness of time as
an independent and separable dimension of the human world. (The argu-
ment is rather more powerful than Whorfs famous one for Hopi, cf.
Whorf 1956).
Among their provisional conclusions is that Time as Such is not a
Cognitive Universal, but a historical construction based in social practice,
semiotically mediated by symbolic and cultural-cognitive artefacts and
entrenched in lexico-grammar. The classical CL trajectory whereby
human beings gradually develop concepts based on universals of primary
bodily experience, mapping previous and more concrete domains such as
space onto later and more intangible domains such as time, thus needs to
be supplemented with a trajectory that starts from the sociocultural
dimension. We return to this discussion in chapter 7 (cf. niche concepts,
p. 318).
Michael Tomasellos research interests cover several issues central to
the social agenda in CL: language acquisition, the pragmatic dimension of
cognitive development, and the evolutionary path of language and cogni-
tion. In relation to language acquisition, he took up a unique position in
addressing the central interest of generative linguistics, the acquisition of
syntax, from a CL-oriented perspective. Because of the polarization in
relation to Chomsky, syntax has to some extent been a contaminated area
within CL, and as a result there was little attempt to offer alternative solu-
tions for the kind of problems that occupied the attention in Chomsky-
inspired psycholinguistics. For a long time, psychologists and psycholin-
guists interested in language acquisition, cf. Tomasello (1998), therefore
tended to take the generative theory of language to be the only bid for a
linguistic starting point for their research.11

11 Of course this is not the whole story: research by a number of functionalists

including e. g., Elizabeth Bates and Brian MacWhinney (with the competition
model as possibly the most well-known element, cf. Bates & MacWhinney
1987), Schlesinger and others established a counterposition to the generative
mainstream. Many of the elements of a usage based approach had been pro-
posed in the functionalist literature, including the role of item-based patterns
and frequency effects, but they tended to be seen not as constituting a strong
74 Chapter 2. From conceptual representations to social processes

Understanding language acquisition as usage based was nothing new,

since all opponents of Chomskyan innatism had to base their theory of
acquisition on actual usage. But much of the discussion could be framed
as a question of whether everything could be explained in behaviourist
terms leaving the generative position in possession of the cognitive high
ground: the connectionist approach based on pattern-recognition proc-
esses was vulnerable on the question of whether there is anything else
going on in the world than probabilistic assessment of feature complexes,
cf. Pinker and Prince (1988).12 Tomasello was able to reset this agenda by

alternative position but as a question of being for or against Chomsky there

was a strong hegemony effect in the pragmatics of the acquisition debate.
12 Generative linguists believed they had the only bid for higher-level cognitive
factors, while the opposition was relegated to blank slate reduction of human
beings to rat-like forms of learning. The discussion was to some extent predi-
cated on the argument based on the dichotomy between two forms of compu-
tational simulation, the classical algorithmic form associated with generative
grammar, and the connectionist or neural net approach, cf. Rumelhart &
McClelland (1986). Although this discussion brought out a number of signifi-
cant issues, it also took the discussion away from the core concern of CL, in
that the neural net approach was even further away from the semantic side of
language, earning it the epithet of neo-behaviourism. What appealed to CL in
the neural net approach was the absence of any reliance on pre-given struc-
ture, and the fact that machines could be trained to impose distinctions based
on any empirical input whatever, thus getting round the poverty of the stimu-
lus argument. In a neural net, everything is featurized, and categorization is
probabilistic rather than based on all-or-nothing Aristotelian categorization.
Neural nets can therefore be used to mechanically reject defective specimens
in a sorting process, simply by starting out randomly and then gradually being
adjusted, without any explicit description of what a defective specimen must
be like.The classical algorithmic approach based on higher-level structure
was also known as the symbolic approach because it presupposed the exist-
ence of symbols in the sense defined by the physical symbols hypothesis of
Newell and Simon (1976). However, there was a pervasive confusion in the
discussion between physical symbols as a hypothesis about the workings of the
brain and the discussion of lexical symbols as part of a human language. Con-
nectionists therefore sometimes felt they had to deny that there was such a
thing as a word constituting a unitary lexical category: the argument for reject-
ing symbols in the classical Artificial Intelligence sense (formal meaningless
syntactic elements) also carried over to rejecting meaningful symbols which
is against the basic CL position. Therefore most CL practitioners did not go
too deeply into the connectionist issue an exception being Lakoff, whose
position reflects this dilemma (cf. the discussion on embodiment p. 80). Out-
The developmental perspective 75

combining three interlocking dimensions in his impressive research pro-

gramme: phylogenetic (evolutionary) development from hominid ances-
tors to humans, human ontogenetic development, and detailed corpus
documentation of the path into language (that generativists claimed
would be impossible without innate knowledge). These three elements
will now be described in that order.
Based on extensive experimental evidence, Tomasello has argued that
the key difference between humans and other primates can be captured in
terms of the concept of shared intentionality, as manifested most tangibly
in joint attention (cf. Tomasello 2008: 4). Human beings are unique in that
they can relate to each other by (1) attending to the same thing and (2)
attending to it not because of its inherent interest, but because they are
doing it together (for instance because they are playing with it). Alarm
calls in animal groups are used to create attention to the same thing, such
as a prowling leopard but the interest in attending to it is not that this is
the game we are playing together at the moment: each group member has
a purely individual interest in calling out (Tomasello 2008: 18). Human
beings, in contrast, will look at things merely because fellow subjects look
at it, and thus assign an importance to the attended object that is due
solely to the fact that they are (co-)engaged in an act of paying joint atten-
tion, and pursue joint activities that have no apparent goal (Tomasello
2008: 177178). The engagement in being in this together also goes with
a species-specific form of altruistic orientation, cf. Warneken and Toma-
sello (2006), where available affordances are to a certain extent under-
stood as belonging inherently to all those present, rather than being up for
Joint attention and action is subtler than (but presupposes) intentional
action which again is subtler than (but presupposes) general coping
skills. If we look specifically at communication, Tomasello argues that
non-human vocal communication is non-intentional, and intentional com-
munication is very rare (but is found in the gestural communication of
apes, Tomasello 2008: 14).13 The crucial extra step into human communica-
tion is the ability to read the other persons intention in order to share that

side the specific CL context, in relation to learning and cognitive architecture,

it was a huge issue, however, cf. e. g. Elman et al. (1996).
13 Unlike vocalizations, which are directly associated with emotional states and
are emitted without regard to addressees, gestures are directed at addressees
and are sensitive to monitoring their attentional states. Awareness of the dis-
tinction between intentional and non-intentional action is demonstrated by
the fact that apes get annoyed when they perceive that the human caregiver is
76 Chapter 2. From conceptual representations to social processes

intention. As Tomasello points out (2003: 23; 2008: 91), this is what is
required for Gricean non-natural meaning the kind that goes beyond
depending on information available in the environment. While smoke
means fire by a mechanism that does not depend on human intentions, hi
is a greeting only because I attribute to the sender the intention of greet-
ing me and this requires fellow subjects who attribute intentions to each
other as part of their form of life.
In addition to communication, this ability also has implication for
learning and for the kind of social environment that human beings inhabit.
What Tomasello calls cultural learning is special and different from basic
trial-and-error learning, because the target goes beyond an individual
goal. The aim is to do what the other person is doing, and because of the
role of intentions this also depends on what the other person has in mind.
In adopting cultural patterns you cannot expect an immediately tangible
reward for successful acquisition to be part of what the others are doing
must be felt as a reward in itself, otherwise the process will not work. Play-
ing football may be fun, but it is not really fun until you know how what
makes the initial humiliations worth while is that there is a point in being
part of the action. Therefore the only game in town is always worth play-
ing, whatever it is. This enables human children to learn things of no osten-
sible value, simply because it makes them part of the community. Lan-
guage change over time thus trades on what Tomasello calls the the
ratchet effect of cultural learning: speakers do not have to start from
scratch, but can stand on the shoulders of previous generations, starting at
the point to which the community around them has arrived (because they
enjoy standing on peoples shoulders).14

withholding food intentionally as opposed to bungling the job (Tomasello

2008: 45).
14 In contrast, the evidence suggests that other animals can only learn new prac-
tices if they visibly contribute to their own individual purposes, such as using
tools to get at the food. This ability is consistent with cultural differences in
that different groups of animals may have figured out different solutions
(which they pass on) to recurrent problems in the life of a primate. A cele-
brated example is the Japanese snow macaques where some groups have
learnt to wash their food. However, it does not include the kind of accumula-
tion that goes with the ratchet effect. The food-getting strategies may be
transmitted from generation to generation, but only if they have a concrete
anchoring in the form of a tangible purpose (such as getting at the nut by
cracking the shell with a rock, or avoiding getting pebbles in your mouth by
rinsing the food), which remains the point of departure for all successive gen-
erations. Humans, by detaching themselves from the necessity of knowing
The developmental perspective 77

The specifically human experience of attending jointly to objects in the

world goes naturally with another difference in relation to chimpanzees.
Chimpanzees do not share food or show other spontaneous altruistic ten-
dencies in relation to conspecifics15. That is not to say that they are loners
without social relations but grooming and other such activities are
strictly on a quid pro quo basis: you scratch my back, and Ill scratch yours.
The activities are shared, but not joint, or (more abstractly) dyadic: there
is no common object of attention that takes priority over separate and
individual purposes.
The phylogenetic difference has an ontogenetic correlate. A human
child is not born fully equipped to enter into joint attention and purposes.
A key event in the development is when, somewhere between nine and
twelve months (cf. Tomasello 2003:2126), human children develop the
capacity for joint attention. Instead of having a merely two-way relation-
ship with the external world (acting on the world, including other people,
and getting reactions in return), children begin to tune in on what others
are attending to.
This is what makes language learning possible the real missing link
between mechanical input and language in the mind, in contrast to the
innate grammar postulated by Chomsky. In a series of experiments, Toma-
sello (2001) has shown that children can learn names for absent objects
(as opposed to visually present objects) when they have good reason to
think that the absent object is what other people have in mind. Tomasello
argues that this attunement can solve the gavagai problem: children
learn what words mean because they become skilled at figuring out what
others intend to name. Even if there may be occasions where the gavagai
problem is not tractable, intention reading points to where the problem
lies in such cases: the lack of a jointly understood situation. Expressions
only get meaning through joint practices. Language learning takes place
in contexts where people know how to do the same things and therefore
also know what each other are doing. Language acquisition is driven by

what precisely there is in it for me as an individual, can take over cultural

practices merely for the purpose of doing what we are doing. Obviously, there
are risks as well as advantages in this, cf. also below p. 106.
15 Zlatev (personal communication) reports findings in current work suggesting
that chimpanzees and bonobos (the Pan species) allow for a greater degree of
passive sharing (mothers allowing offspring to feed off their food) than in
other primates.
78 Chapter 2. From conceptual representations to social processes

the inherent motivation that is associated with being on the same page as
As a last dimension, this process also opens a pathway into a normative
universe thus establishing an foothold within empirical psychology for
an aspect of the human world that is more familiar from philosophy and
sociology (cf p. 277 on Itkonen, p. 141 on Durkheim; p. 364 on Habermas).
Joint attention and joint activity have causal underpinnings just like eve-
rything else in the universe, but the fact that it can only come off if speak-
ers are actually on the same page means that one can be looking at the
wrong object. The attentional stance of the other person is the criterion
for whether I am getting it right. In this primal scene, there is no norm
beyond the here-and-now but once cultural learning gets under way,
learning the norms inherent in cultural practices is part of it. You cannot
learn to play soccer without learning that scoring a goal is good and that it
is bad if the other team scores a goal, cf. Rakoczy, Warneken and Toma-
sello (2008) on the acquisition of norms in relation to games. Language
works the same way: water is called water, and if you ask for it by using the
word zunk, you are doing it wrong simply because you are not hitting
the target in joint social space.
Once this mechanism is there, acquisition does not have to depend on
an innate crib. You can take the steps one at a time, as part of figuring out
what the other speakers are on about. The joint attention hypothesis thus
feeds into Tomasellos long-standing research into language acquisition,
including the precise path into complex syntax. The first major step was
the discovery of the asymmetry between nouns and verbs in early acquisi-
tion, known as the verb island hypothesis (Tomasello 1992). While chil-
dren learned to insert nouns in slots in a productive manner fairly early,
generalizing overall noun properties across instances, they tended to use
verbs only in ways that they had heard, i. e. with complementation pat-
terns that were actually attested in usage. Thus the idea that there was a
general category verb at the outset of the acquisitional path a core

16 The absorbing interest which children invest in those acts of pointing, gaze-
following, mutually co-ordinated movement etc, where shared attention and
understanding are at stake, can be understood as an instinctual basis for lan-
guage learning but obviously this is a different kind of instinct from Pinkers
Language Instinct (1994). In relation to language learning, a key difference
between humans and animals is therefore that human children can learn the
language without being rewarded with nuts at intervals. Being able to do what
others can do and being with the others and knowing what they know are
experienced as inherently valuable.
Extended grounding 79

assumption for innatist theories did not appear to be plausible in the

light of the data. Rather, a very cautious path of generalization based on
actual usage seemed to be not only possible but also more realistic.
This idea was followed by a more general strategy of tracing actual
language acquisition by means of dense corpora of young children in the
process acquiring language. Because of the density of the corpora Toma-
sello was able to follow the emergence of new syntactic forms in greater
detail than previous research and show that rather than innate patterns
suddenly being evoked in full-fledged form by confrontation with data,
children generally followed a very cautious and conservative trajectory
through the space of possibilities. The overall pattern was that children
took only one step at a time away from known forms. The celebrated crea-
tivity that was the key argument for innatism, where wholly new sentences
were produced on the basis of very general rules, was very hard to find
examples of, while tentative forays into new territory at the margins of
known usage patterns were everywhere.
Just like the variational dimension, the developmental dimension car-
ries a methodological requirement. In addition to the corpus methods
described above, rigorous experimental investigation is necessary in order
to get at the relevant differences between age groups, as well as between
humans and apes. Here again, the kind of issues you want to include has
implications for the methods you need to employ.

5. Extended grounding: situational, intersubjective

and cultural aspects

The previous section brought out one central social factor, which is equally
essential to the development of language, culture and personality: human
subjects are inherently disposed towards establishing a community of
understanding with other human subjects. This has profound conse-
quences for some of the basic assumptions in CL.
These consequences affect the understanding of grounding. Ground-
ing is the key foundational difference between first-generation cognitive
science and CL (cf. the discussion p. 3 above). CLs platform was and is
that language and meaning are not autonomous and self-contained enti-
ties but grounded in human cognition as a whole, which in turn is grounded
in the human body as a whole. The title From molecule to metaphor sums
up this path of understanding.
In order to accommodate the social dimension in the picture, this
notion of grounding is no longer sufficient. Rohrer (2005) has pointed out
80 Chapter 2. From conceptual representations to social processes

the dangers of the lack of clarity in the concept of embodiment, suggesting

that it would mean a form of reductionism that is actually not too distant
from behaviourist attempts to keep mind and consciousness out of the
scientific world picture altogether. As also argued by Zlatev (2007),
attempts to ground meaning directly in a neural level ultimately end up
with no way to distinguish mental from non-mental phenomena. Lakoff
and Johnsons position (1999: 17) is illustrative here: according to them,
the amoeba categorizes objects by moving either towards them or away
from them and there is no such thing as inner representation (Johnson
and Lakoff 2002: 24950). If grounding is understood in terms of this pic-
ture, it means more or less the same as reduction. 17
Rather than trying to split up the concept in different subcomponents,
Zlatev relocates the whole grounding process into the social sphere. The
title Situated embodiment (Zlatev 1997) sums up the dual focus on social
and biological aspects of grounding. His position takes over from Witt-
genstein (1953) the idea that meaning emerges from language games, i. e.
from interactive practice, but stresses the contribution of the human sub-
ject to this interaction. This proposed synthesis is illustrated with the
emergence of spatial meaning; based on a child language corpus he argues
that neither an approach in terms of direct embodied experience alone,
nor an approach in terms of learning downloaded from the social process
alone can explain the data.
The task is then to show how the duality works in practice. Zlatev
(2003) has proposed a theory that accounts for the broad concept of
meaning in terms of four stages that constitutes a scale of nature, with
relevance to both the evolutionary and the epigenetic time scale. The rock
bottom point of departure for understanding meaning is the emergence of
living creatures: meaning can only exist in relation to life. In harmony with
the epigenetic perspective, meaning is seen as essentially a property of the
relation between organism and environment. What distinguishes meaning
from other such relations is the concept of value, which (following v.
Uexkll and Cisek) Zlatev anchors in the basic contrast between desira-
ble and undesirable (or favourable vs. unfavourable) input from the envi-
The idea of value therefore ties in with Gibsons (1979) idea of
affordances and also with the central role of emotions (cf. Damasio 1994)

17 The issue of embodiment in a cultural perspective is discussed in a recent

interview with Tim Rohrer and Mark Johnson, cf. Pires de Oliveira and Bit-
tencourt (2008).
Extended grounding 81

for cognition: before anything else, there must be a criterion for what is
good or bad.
On that basis Zlatev suggests four types of meaning systems: cue-
based, associational, mimetic and symbolic creatures. Cue-based beings,
the simplest type, work solely through responses to a fixed and innate set
of environmental cues, i. e. event types. The standard minimal example is
the coli bacillus which propels itself towards food and away from toxic
substances. Habituation is the only form of learning, entailing adaptations
of the response system to ignore stimuli below a certain intensity. Above
these, we find associational creatures (comprising most higher animals),
which are capable of modifying their response system by (re)forming
associations between environmental input and actions. Evaluation in
terms of intrinsic desirability remains basic: responses are modified based
on whether they give good or bad results. For higher animals, the limbic
system, as described by Damasio 1994, has a crucial role in generating
primary, hard-wired emotions which constitute the criterion for evaluat-
ing success or failure.
Since the third stage is in important respects transitional, it is easier to
take the fourth stage, symbolic creatures, first. The crucial properties
include abstraction and conventionalization, i. e. symbolic meanings that
are detached from direct relations with the stream of experience, and soci-
oculturally shared. This also means that they are intentionally controlled,
i. e. they constitute non-natural meaning: you understand an expression by
making assumptions about what is meant by it, rather than by simply asso-
ciating it with other features of the environment. This stage is strongly
associated with a radically increased role for the cortex in the brain.
The third stage is transitional in that its key feature, taken over from
Merlin Donald, is that it is mimetic, i. e. capable of understanding and
using bodily responses for the ulterior purposes of representation, includ-
ing gestural communication. This entails a form of detachment from the
strictly associational link with the environment that is characteristic of
stage two, but falls short of the achievement of shared and abstract sys-
tems of meaning at stage four. Only the apes are candidates for belonging
at this stage. This placement is motivated by their ability to form culturally
distinct groups, which may plausibly arise by mimetic transmission of
behaviours, and by their ability to take over and iconically represent
human forms of communication. At the same time, they fail to make the
transition into the symbolic stage, and therefore they cannot undergo cul-
tural development because their cultural differences are still concretely
anchored in the shared context, rather than based on shared representa-
tions alone.
82 Chapter 2. From conceptual representations to social processes

Zlatev (2008) adds a differentiation of the mimetic stage into dyadic

and triadic mimesis, where apes are somewhat capable of the first (imi-
tation, imperative pointing, ritualized gestures), but without human encul-
turation, not really of the second (where declarative pointing and produc-
tive iconic gestures belong).
The discovery of mirror neurons, cf. above p. 32, has played a role in
this discussion. Their defining feature is that they fire equally when others
perform an action and when the subject does it herself. (Parents will be
familiar with a salient manifestation of this mirror phenomenon from the
experience of opening their own mouths when spoon-feeding toddlers!).
Mirror neurons are automatic and unconscious, and appear before the
mimetic stage but they are important in demonstrating that other peo-
ple matter to the individual also at a purely mechanical level, not only in
the subtler forms based on joint attention.
This scale is then used to outline a scale of ontogenetic development.
No evidence is suggested for a purely cue-based stage, but Zlatev traces
a development from an early associational stage, dominated by the acqui-
sition of sensorimotor schemas (before 9 months) via a mimetic stage,
with pointing as an example (before language acquisition) to the final
symbolic stage (or post-mimetic, cf. Zlatev 2008: 219), which involves a
socially shared symbolic system.
Regardless of the precise epigenetic path of intersubjectivity, the broad
picture outlined above suggests that there is a sense in which intersubjec-
tivity is present from birth, and also that its foundation includes a hard-
wired ability to respond directly to the feelings and actions of others
(Zlatev 2008: 223). Further, that the representational dimension is mini-
mal at the initial stage and develops through a dyadic stage (where the
point is to influence the actions of others) to full, triadic communication,
where the intersubjective relation includes a shared world. This can be
understood as building up to the full Bhler triangle of self-expression,
appeal and reference as the presupposed foundation of all linguistic com-
As also argued by Sinha (cf. p. 72 above), intersubjectivity is basic in
this picture, not something that is added at an advanced stage. Rather than
being essentially imprisoned in their individual minds, human beings are
inherently grounded in a world that includes aspects of the internal states
of other people; crucially, we have a natural quasi-perceptual access to
other peoples elementary qualia, cf. Zahavi (2007). Increasing cognitive
sophistication is part of a process that mediates in increasingly sophisti-
cated ways between organism and environment, including a social envi-
ronment that is constitutive from the beginning. The special cognitive
Extended grounding 83

sophistication of human beings depends on their embedding in a cultural

context of shared meanings that are not directly embedded in the stream
of experience, but in membership of a community. Meaning in the fully
human sense thus arises in a space that is both cognitive and sociocultural,
and therefore marks out a new ontological territory.
A crucial concept in the epigenetic approach is a further development
in the general notion of a schema (cf. Sinha 2007, Zlatev 2005), beyond
the classic image schema. At the most general level it subsumes all types
of skills that depend on extracting something generalizable from previ-
ous experience. The concept is therefore general enough to be used across
the board from neurobiology via cognitive representations and all the
way to the level of conventional sociocultural schemas, cf. Sinhas discus-
sion of the history of the concept (2007: 1275). Thus, when the child learns
to grasp a ball, increasing success can be viewed as the result of a success-
fully entrenched grasping schema just as the successful learning of a
linguistic pattern is an instance of schematic learning. Authors as diverse
as Bartlett, Piaget and Bourdieu have found the word useful in that
However, further development of this basic insight depends on differ-
entiation. Zlatev has proposed (Zlatev 2005) the concept of mimetic
schema as an alternative, or at least a complement to the not entirely
clear role of image schemas (cf. above p. 31), which hovered between
abstraction and concrete perception. Zlatevs concept offers a way of
relating the two types of schema without conflating them. While neural or
sensorimotor schemas are embodied in a very concrete sense, they are
strictly the property of an individual organism and thus cannot underpin
a system of public meanings. On the other hand, if schemas are under-
stood as abstractions over sets of more concrete meanings, they lose the
foundational status that they have been assumed to carry in the view of
embodiment that is central to cognitive linguistics. The developmental
perspective suggested by Zlatev locates mimetic schemas at the transition
point, where bodily experience acquires a dual identity as shared abstrac-
tion. Mimetic schemata as a form of internalized imitation are at the same
time features of bodily action and public representations of the relevant
bodily experience. They can thus have a feel associated with them (Zlatev
2005: 322), which is naturally understood as part of communal experience
at a pre-linguistic level but which may be recruited in understanding
linguistic utterances.
Zlatevs theory is primarily a way of extending bodily grounding to the
situated body; and mimesis is an embodied skill. His theory is based on the
Wittgensteinian perspective of grounding in shared practice (cf also. Sinha
84 Chapter 2. From conceptual representations to social processes

(1999) and Harder (1999) on discursive grounding). In fleshing out this

approach, an anthropological dimension is necessary.
Anthropology was affected by the same narrowing of the perspective
that went with the cognitive revolution. One of the anthropologists who
have served as an inspiration for CL expresses this in the following way:
As part of the cognitive revolution, cognitive anthropology made two crucial
steps. First, it tuned away from society by looking inward to the knowledge an
individual had to have to function as a member of the culture. The question
became What does a person have to know? The locus of knowledge was
assumed to be inside the individual. The methods of research then available
encouraged the analysis of language. But knowledge expressed or expressible
in language tends to be declarative knowledge. It is what people can say about
what they know. Skill went out the window of the white room. The second
turn was away from practice. In the quest to learn what people know, anthro-
pologists lost track both of how people go about knowing what they know and
of the contribution of the environments in which the knowing is accomplished.
(Hutchins 1995: xii)

The project in which the limitations of a cognitivist approach became evi-

dent to Hutchins was a study of navigation as a cognitive process. From
the first moment, one thing was obvious: in navigating a large navy ship,
success depends on more than one mind. To the extent the successful out-
come of any navigational task can be considered the outcome of a cogni-
tive process, that cognitive process is distributed across a collective group.
The knowledge required is therefore a truly social fact in the sense pro-
claimed by Durkheim (more on this in ch. 3), cf. Hutchins (1995: 176).
The fact that different people have to know different things about how
to carry out a cultural practice such as navigation raises several issues.
From the managerial point of view, the answer to the problem is simply a
question of division of labour. But from the point of view of human under-
standing, it raises the question of the link between what I know and what
other people know as we collaborate on reaching a common goal. The
kind of cognition that goes on is inherently cultural in the sense that it
goes with the status of being a member of a community of understanding,
rather than working through the individual mind alone. Moreover, there
is always going to be a social organization that mediates between individ-
ual-internal structures and individual-external structures and that social
organization is an entity of a kind that cannot be viewed as mind-internal
(Hutchins 1995: 262). Organizational knowledge is present in any complex
organization as distributed knowledge, for example about how to produce
buses or how to govern a city. And organizational knowledge is a special
case of cultural knowledge, i. e. the kind of knowledge that allows a human
Extended grounding 85

society to go about its collective business. In both cases, the larger group
constantly faces tasks that are beyond the capabilities of any individual
member (ibid.).
One of the things that make this possible is another type of extension
beyond the body of cognition-imbued activity: the use of material
anchors. The case he discusses in detail is that of the astrolabe (Hutchins
1995: 9699); a simpler example is the compass. If we look at the collective
task of navigating on the one hand and the individual cognitive systems on
the other, it is obvious that the success of the enterprise depends on a reli-
able way of mediating between on the one hand the demands of the task
and the cognitive systems, and on the other hand between the different
cognitive systems between which there is a division of labour. The role of
material anchors in that context can be illustrated with the apparent plau-
sibility of behaviourist reasoning: there is something mysterious about the
idea that invisible processes in the mind can control the material world,
and it feels more scientific to attribute effects to overt material causes. The
type of operations that are necessary to keep a ship on its course may well
appear mysterious to rookie navigators for the same reason and if the
necessary cognitive operations are mediated by an object where its ele-
ments are represented in material form, it will literally make the task
more tangible. Learning to handle the compass and the astrolabe is an
intermediary step to learning how to handle the whole navigation of the
ship. Similarly, the abacus makes calculation visible, and thus more man-
ageable in the initial stages.
In order to understand the significance of this operation, however, it is
necessary to understand it as involving a kind of back-formation, a top-
down modification of a relationship whose basic mode is bottom-up. As
pointed out by Sinha & Rodriguez (2008), the primal scene of intersubjec-
tivity is not going from individual knowledge to common knowledge as a
special and refined form of knowledge. Instead, intersubjectivity is based
in the experience of participation in shared material practices. Physical
objects such as chairs enter into this process: they acquire shared signifi-
cance by entering into shared practices. An object, socially speaking,
counts as a chair if it enters into those practices by serving the function
of chair. As an example, bean bags at one point in the sixties began to
count as chairs, without acquiring new physical properties simply
because they entered into shared practices as things you would sit on.
Material anchors are therefore a special case of the relation between
material practices and cognitive representation. The basic way in which
objects acquire significance is that their symbolic significance emerges
from their place in (non-representational) shared practices material
86 Chapter 2. From conceptual representations to social processes

anchors, in contrast, are designed to represent something that would oth-

erwise be more abstract and intangible. An abacus helps a child to get into
calculation, because it provides a concretely accessible path towards the
abstract learning target
Nevertheless, material anchors may have a constitutive role in ena-
bling forms of cognition that would not otherwise be possible. Conceptu-
alization is easier if it is supported by immediately available input from
the environment, cf. Brown (1994); thus remarks on how to carry out a
practical task are easier to understand if they are made in the context of
actually carrying out that task. A material anchor such as an abacus may
not only provide a short cut, but be a condition on making abstract numer-
osity culturally sustainable. (More on this issue p. 323).
Yet material anchoring is not an account of the way cognition is inher-
ently and basically grounded. In order to approach that issue properly, we
need to follow the bottom-up path and see how sophisticated phenomena
emerge from the simpler ones on which they depend, and recognize both
the dependence on and the difference from the simpler phenomena (cf.
p. 145 below).
In relation specifically to language, Verhagen has developed a theory
of intersubjective grounding that involves an additional dimension of the
expanding CL world view described above. He takes his point of depar-
ture in Langackers theory of subjective grounding. In the standard dia-
gram reflecting Langackers canonical viewing arrangement, encoded
meanings are put on stage. For example, the prototype billiard ball
model, as in the red ball hits the white ball, puts on stage a scene where the
red ball is the figure, the white ball is the landmark, and the hitting consti-
tutes the profiled event. In Langackers classic version (but cf. Langacker
2008a), the onstage event is grounded in the mind of the conceptualizer or
speaker. When subjective grounding plays a special role (as in processes
of grammaticalization, cf. Langacker 1990), what happens according to
Langacker is an evocation of the conceptualizer, rather than the objective
scene of conceptualization. However, grounding predications, including
tense, are not part of the onstage event they evoke the position of the
conceptualizer as part of the speech event. The here-and-now is not on
stage but part of the off-stage ground.
What Verhagen (2005) argues persuasively is that instead of the lone
conceptualizer, we have to posit a speaker-addressee relationship as the
grounding scene. Rather than relating to the mind of the speaker alone,
the event is being located in relation to a speaker-hearer axis. Referring to
Bhler and Tomasello, Verhagen (2005: 2 and 6) sees the basic ground for
meaning as being a matter of cognitive co-ordination rather than intra-
Extended grounding 87

mental grounding: other minds are presupposed from the beginning.

Aligning himself with the argumentative semantics of Ducrot, he suggests
that regulating and assessing others (Verhagen 2005: 9) is as central as
understanding the informational content on stage. In the book, he goes
through a number of cases showing how this regulatory intention can
change the perspective. One example is negation: negation evokes two
conceptualizations at the same time and performs an act of regulating,
more specifically correcting, an informational state.
The kind of context in which cognitive linguistics locates elementary
meaning, and in which embodiment takes place, in other words, has clearly
been moved from the inner recesses of the individual mind-cum-neural
system to embodied interaction with a world including other minds. Sinha
and Jensen de Lopez (2000) give a striking illustration of what they call
extended embodiment. They compare the canonical image schema of
containment as encoded in the relevant preposition in two communities,
the European (represented by English and Danish) and the Zapotec.
Rather than containment being directly grounded in universal bodily
experience, it turns out to reflect differences both in linguistic encoding
and sociocultural practices. Specifically the canonicality effect associated
with the basic image of a container described by Lakoff and found in
European investigations, was not found in the Zapotec community
which could be easily correlated with the ways in which woven baskets
were habitually used for containment by the Zapotecs.
Such an extended view of embodiment entails a rejection of the more
radical version of Lakoffs neural approach to embodiment: although the
neural element is important, it is not constitutive of meaning in the emerg-
ing intersubjective recontextualization of cognition and embodiment.
This also has consequences for the basic understanding of the con-
strual operation that is the heart of cognitive semantics, cf. Verhagen
(2007). In the diagram that sums up the basic scene for the construal oper-
ation, cf. Langacker (1987: 487488) with a subject and an object of con-
ceptualization, Verhagen shows that it is necessary to replace the solitary
conceptualizer with a dyad of speaker and hearer. Instead of a subjective
ground, an intersubjective ground is necessary which in turn means a
common ground (Verhagen 2007: 60).
All the dimensions discussed above have an obvious affinity, which
constitutes the motivation for bringing them together in this chapter. Per-
haps the best way to summarize it is in terms of the significance of the
shared material world. Epigenesis, intersubjectivity, material anchoring,
and cultural development occur in that shared world. Cognitive represen-
tations, processes and mappings depend on objects, people and practices
88 Chapter 2. From conceptual representations to social processes

with properties that do not only impinge upon cognition but constitute
the presupposed context for it and therefore have to be included in a full
account of grounding. From Bhler via Sinha and Tomasello to Verhagen,
the triadic relationship between the first, second and third person roles
recurs in different variants as the basic scene for understanding language,
replacing individual cognitive representations as the object of description.

6. Language as a population of utterances: an evolutionary synthesis

Embodiment implies a commitment to a form of biological realism. First-

generation cognitive science has been accused of physics envy in its com-
mitment to mathematically precise models, which from a CL perspective
is clearly a source of distortion. CL, however, is moving closer to the terri-
tory of the second-most successful scientific theory, evolutionary biology.
The most obvious place to explore neighbourly relations with evolution-
ary theory over the fence is of course historical linguistics. When Bybee,
Perkins and Pagliuca (1994) published a synthesizing volume on gram-
maticalization, they called it The Evolution of Grammar but it was
essentially an analogy only, rather than a case of theoretical inspiration.18
A step towards a more theoretically ambitious integration of theoreti-
cal thinking from evolutionary biology was taken by Croft with his two
volumes on historical and synchronic linguistics, Explaining Language
Change (2000) and Radical Construction Grammar (2001). In the follow-
ing, I will discuss the historical part first, as the one with the most direct
analogy with biology, and the synchronic part afterwards.
In order to arrive at a theoretical synthesis with evolutionary biology,
it was necessary to go a roundabout way. Languages are not biological
organisms and do not physically interact with the environment the way
organisms do. Too direct analogies are therefore likely to unravel on close
inspection. It is necessary to extract the structure of the causal mechanism

18 At that point the analogy was centred on the general attitude to the subject
and the priority of diachronic patterns and pathways over synchronic rules.
The focus was on showing that the source of understanding why languages
look the way they do was in the types of development that could be found,
rather than in the relations between different parts of languages with other
parts of languages which is analogous to the contrast between morphological
theories of organisms as opposed to theories based on trajectories of evolu-
tionary development.
Language as a population of utterances: an evolutionary synthesis 89

that makes a system undergo evolution-type developments. Hulls (1988)

evolutionary theory of the development of scientific ideas had both the
necessary distance and the necessary precision to make explicit what
needs to be present in order for the dynamics to work.
The basic architecture of the model can be described as follows. There
is a basic distinction between two levels at which changes happen, the
individual organism and the whole population. Change happens first at
the level of a single change in an individual, and may then go on to the
level of overall, aggregate change in the population. In biology the indi-
vidual change is a mutation, a random change in the genetic constitution
of a single individual; the aggregate change is the change in a species that
results from the new gene becoming propagated across the whole popula-
tion. This distinction in turn depends on the existence of reproduction, or
in Hulls terminology (Hull 1988: 408), replication. Replication is the
central mechanism that makes evolutionary change possible at all: an
individual must be able to pass on its own structure (more or less intact)
to a new individual in the next generation otherwise there is no pathway
from individual to aggregate change. That is why only reproductive sys-
tems allow evolution. Entities that persist as individuals, such as atoms
and oceans, cannot evolve, although they can undergo individual change.
The four basic elements that Hull (1988: 40809, cf. Croft 2000: 22)
outlines as basic to his theory of evolution are replicators, interactors,
selection and lineages. Replicators are the elements that get copied from
one generation to the next, in biology the genes. Interactors have the role
of either being selected or not; in biology the prototype interactors are the
organisms. Organisms play an independent role because it is their survival
and reproductive success that determine whether the genes are passed on
or not. There is a link, of course: part of reproductive success is due to the
genes, whose proliferation depends on how well they serve their bearers,
the organisms, in interacting successfully with the environment, including
members of the opposite sex (this is taken up in relation with the defini-
tion of function, p. 160 below).
The difference in the reproductive success of organisms is what gives
rise to selection, causing some genes to spread in the next generation and
others to become rarer. Comparison across generations is possible because
the replication mechanism gives rise to lines of descent, or lineages: when
an entity is copied from one generation to the next, the result is a space-
time worm (Hull 1988: 410) that wriggles on as long as reproduction takes
place. Lineages can arise at several levels: genes, organisms and species all
form lineages across generations. We can follow (1) what happens to a
gene from one end of a lineage to another, (2) what happens to the organ-
90 Chapter 2. From conceptual representations to social processes

isms from one end of a lineage to another, and (3) what happens to the
whole population. The changes are linked, but not identical.
When all these elements are present, they allow the mechanisms
whereby evolution becomes possible. The essential two-step nature of the
process of change is ensured because of the distinction between replica-
tors (genes) and interactors (organisms): the small thing that may change
in a way that is passed on to the next generation (by replication) depends
for its success on how it contributes to the success of a bigger thing, namely
the interactor. Although the standard way of thinking is anchored in the
gene as the replicator and the organism as the interactor, the point of
Hulls application of this idea to scientific development is that it is the
causal structure that is important, not its particular application in par-
ticular the co-existence and mutual dependence between reproduction (at
the individual level) and selection/proliferation (at the population level).
Science may evolve in an evolutionary manner because the same causal
structure is involved, regardless of how (dis)similar the history of science
is in other respects from biological evolution: ideas are produced and
reproduced through the work of individuals but they are selected and
proliferated as a result of selection mechanisms that work over the indi-
viduals head (through the reception history in periodicals, at conferences,
etc). This applies to language change, too (more on this p. 157 below). Evo-
lutionary theory is therefore something totally distinct from embodiment.
What matters there is the causal structure of evolutionary dynamics.19
There is a direct link in Crofts theory with the doctrine of usage based
linguistics. Evolutionary dynamics involves causal processes that affect
the relevant entities through actual usage events such as matings, births
and deaths. A necessary property of all elements belonging in such a sys-
tem is therefore that they constitute spatiotemporal particulars, i. e. physi-
cal entities bounded in space and time. This is a condition on having the
relevant causal properties. Thus species of animals, in order to enter into
evolutionary processes, must be understood as populations rather than
essences (cf. Croft 2000: 13, 2006: 95). Even if we assume for the sake of
the argument that the definition homo sapiens encapsulates the essential
property of the human species, thinking animals might in principle exist

19 The word evolution carries so many overtones that it is essential to distin-

guish between biological evolution proper and systems that have the same
type of causal dynamics. In places where there is a danger that the use of the
word would lead to unwanted inferences, I try to use terns like evolutionary
dynamics or selection-adaptation mechanisms about the abstract similarity
in causal structure.
Language as a population of utterances: an evolutionary synthesis 91

elsewhere that do not belong to the human species as a historical entity.

They would therefore not be affected by the same causal factors that
shape the human species across generations.
This has other interesting consequences. The changes that are the
whole point of having a theory of evolution in the first place can in princi-
ple affect all an organisms properties, and so would-be essences may be
lost in time20. A species is defined in terms of interbreeding, since this is
what causes replication and enables selection among its members to cause
the proliferation of changes. Certain properties of individuals may be
essential in order for that to be possible but the essence depends on its
role in interbreeding, rather than the other way round.
Crofts version of the theory takes the form of a theory of utterance
selection. The most basic item is the replicator. In Crofts theory
The replicators themselves parallel to genes are embodied linguistic struc-
tures, anything from a phoneme to a morpheme to a word to a syntactic con-
struction, and also their conventional semantic /discourse-functional (informa-
tion-structural) values. The replicator is the particular linguistic structure as
embodied in a specific utterance. (Croft 2000: 28)
A crucial question to which we shall return later (p. 292), turns on the way
we understood the relation between the two elements in the definition of
the replicator, their status as being embodied and their status as struc-
Since utterances are the elements in which structural units (linguemes)
are replicated, it is the population of utterances that constitutes a language.
Although perhaps not intuitively obvious, the analogy is sustained by the
parallel with the causal structure that is necessary for evolution: a popula-
tion that constitutes a species is made up of the units (organisms) that in
them have the replicators (the genes) and pass them on. The population
that constitutes a language is the population of units (the utterances) that
contain the replicators (and pass them on).
The view of language as a population of utterances is the basis of the
theory of synchronic structure that constitutes the natural complement of

20 This does not, of course, mean that it is forbidden to investigate and enumer-
ate the properties that characterize members of a population that constitutes
a species. What is forbidden is if these properties are taken out of their his-
torical context and reified as constituting the general object of description.
Even a perfectly adequate descriptive generalization (e. g. that human beings
have a language ability) is only a partial truth about the species because it
belongs in a wider context of causal pressures and changes.
92 Chapter 2. From conceptual representations to social processes

an evolutionary approach to language change. Radical Construction

Grammar (Croft 2001) is radical in the sense that it is the syntactic the-
ory to end all syntactic theories (2001: 4) because it eliminates the idea
that there are any scientifically justifiable basic syntactic categories. Good-
bye to nouns and verbs, clauses and phrases as shared basic concepts in
All other theories of syntactic structure, formalist or functionalist, have
set up a descriptive apparatus and argued that this apparatus is the one
that captures best what really goes on in structural combinations of lin-
guistic elements. Such theories are therefore somewhat like eternal
essences in biology, attempts to define units in Platonic or Aristotelian
terms. From a really radical construction grammar perspective, they are
more or less plausible and attractive mirages whose fundamental mistake
is to try to extract something basic and eternal from the only basic reality
there is, the sprawling and fluctuating multiplicity of actual utterances.
Construction grammar will be discussed in ch. 6. The crucially social
idea that is pursued so consistently in Crofts radical version is that it all
comes down to a population of utterances. One of the consequences is
that variation, diachronic as well as typological, is the basic fact that lin-
guists have to cope with. Not only are all grammatical structures essen-
tially language-specific; even within the individual language, linguistic
generalizations are merely inductive generalizations based on actual spec-
imens found in usage (as the population-genetic biologists generaliza-
tions are based on specimens collected in different sub-populations).
The question arises, therefore, of how the linguist can get a terminol-
ogy at all? Where can we get a frame of reference that can deliver con-
cepts in terms of which the variation can be described, if we cannot with-
out distortion use terms such as noun and verb? The answer, in harmony
with the basic assumptions in CL, is that language depends on pre-existing
conceptual resources. These resources can be understood as constituting a
conceptual space also known as mental map, cognitive map, semantic
map, or semantic space, cf. Croft (2001: 92). In that space we find things
like reference, modification and predication along one axis and objects,
properties and actions along the other. Linguistic categories in different
languages can therefore be described in terms of the conceptual territory
they cover.21

21 This idea is of course widely used across linguistic traditions; it is in harmony

with structural diagrams of dividing-lines between colour categories in differ-
ent languages, cf. Hjelmslev (1943: 49), and also the modern tradition of
semantic maps from Anderson (1982).
Language as a population of utterances: an evolutionary synthesis 93

This idea is a new twist of a familiar idea: to base the understanding of

language on a presupposed underlying order. From the point of view of
European structuralism, it would correspond to the idea that content sub-
stance or purport is universal (cf. Hjelmslev 1943: 46): it consists of con-
ceptual categories which, in contrast to linguistic categories, are assumed
to be cross-linguistic: all languages are used for reference and predication
and encode objects, properties and actions. Croft emphasizes (2001: 93)
that conceptual space also represents conventional pragmatic or dis-
course-functional or information-structural or even stylistic or social
dimensions of the use of a grammatical form or construction. Although
these presumably belong at the less universal end of the section, this claim
is in harmony with the CL tradition of assuming that there is a total uni-
verse of conceptualization that can subsume all relevant properties (rep-
resentational or social) for the purpose of understanding language22.
Croft (2009) discusses the crucial changes that must take place in the
classical apparatus of CL in order for it to take the full step towards inte-
grating the social dimension. He sums up these in terms of expansions of
four basic principles of CL.
First, the interpretation of processes in language as instances of gen-
eral cognitive abilities must be extended to social cognitive abilities of
which the most important are joint action, coordination and convention.
Joint action is action of the kind that reflects the human capacity for joint
attention and joint action, cf. above; co-ordination enters into the practical
execution of joint action in ways described by Clark (1996); and conven-
tion is one way of achieving co-ordination.
Secondly, to the symbolic commitment (which in Langackers classical
form pairs off the phonological and the semantic pole) Croft adds a third

22 The evolutionary analogy is also found in other versions, cf. Frank (2008), who
focuses on the way metaphors arise, spread and disappear as part of the socio-
cultural process. In her approach, the memetic dimension plays a more sig-
nificant role. A meme, following Dawkins (1989:192), is a unit of cultural
information, such as a tunes, catchphrases, beliefs, clothes fashions, which
are like linguistic expressions in that they may proliferate across populations.
Frank points out the problems with this notion, recognizing the need to flesh
it out in order to set up a fully fledged theory. The point of the article, however,
is to trace a way of thinking about language, with the complex adaptive sys-
tem as the most fully developed version, as the wider frame in which the life
history of discourse metaphors must be understood. The idea of extending
the evolutionary analogy to sociocultural formations will be pursued in chap-
ter 7 below.
94 Chapter 2. From conceptual representations to social processes

corner anchored in the community. The point is in harmony with the

revised theory of intersubjective grounding proposed by Verhagen, but
emphasizes the population level rather than the I/you relation, emphasiz-
ing that meaning is variable according to what members of the community
you are talking to (cf. Clark 1998).23
Closely related to this move is the third extension, from encyclopaedic
to shared meaning: it is not enough to say that meaning is recruited from
the general store of knowledge about the world, we also need to say that
meaning is recruited from the common ground between speaker and
addressee). In defining what this means, Croft stresses the role of com-
munities of practice (following Wenger 1998) over and above shared
expertise (Clark 1996). For Wenger, a shared history of learning is essen-
tial, thus making more stringent requirements than would apply to all
members of a speech community (beyond a face-to-face, hunter-gatherer-
type community). Croft argues that there can be a mediation such that
immediate and direct communities of Wengers type can form the basis
for indirect communities (via the rise of historical lineages, cf. above), so
that the foundational role of shared activity can be preserved also if com-
mon ground can extend beyond direct experience (this discussion is taken
up in relation to social construction below p. 331). The requirement of
shared practices brings the sense of common ground close to the directly
experienced form of common ground that is the focus of attention for
both Sinha and Verhagen.
The fourth and last extension involves construal and takes the step
towards construal for communication, reflecting a move towards some-
thing like the argumentative semantics proposed by Verhagen. The cen-
tral idea is, as for Verhagen, that construal is not individual but construal
for a communicative purpose. This socially amended view of construal
goes with the basic role of variability (illustrated with Chafes pear sto-
ries): rather than assuming that each alternative formulation reflects a
precise conceptualization chosen by the speaker, we must understand the
encodings as the result of the inherent variability of encoding practices:
Alternative construals provided by alternative verbalizations cannot be
precise. We thus end up in the fundamental fact of usage: Language is
a fundamentally heterogeneous, indeterminate, variable, dynamically

23 Croft uses the term subject in its grammatical and psychological senses as an
example: part of knowing the language is to shift your understanding depend-
ing on the actual discourse situation so that you avoid linking grammatical
agreement with psychological subjects (and vice versa).
Meaning construction 95

unfolding phenomenon, just like the human society it constitutes a part

of. (We return to this issue in ch. 6).
The introduction of evolutionary dynamics opens up a larger theatre
of operations than other socially oriented developments. The central ele-
ment is the dynamic relations between the individual and the community
that unfold in a panchronic four-dimensional space. No individual intui-
tion and equally, no linguistic corpus can contain the whole of that

7. Meaning construction

With the emphasis on variability and indeterminacy, another social, online

process becomes elevated to a more significant role. Meaning construc-
tion is attracting increasing attention (cf. e. g. Radden et al. 2007) as the
process whereby complex meaning is created as part of communicative
interaction. Like other new developments, this has in principle been part
of the framework from the beginning, in the sense that language is only
part of the story of understanding (cp. Reddys (1979) seminal criticism of
the conduit metaphor). It just did not attract so much attention, because
the focus was on the cognitive models and mappings in the mind, rather
than on their role in situated meaning construction.
This changed with the rise of blending (or conceptual integration)
theory, cf. Fauconnier and Turner 2002. Blending is an outgrowth of the
theory of mental spaces, cf. Fauconnier 1985 (and above p. 42), and
describes what happens, simply put, when two distinct mental spaces are
brought together to create a third mental space with elements from both.
Among the staple examples is the philosopher who portrays himself as
having a debate with Immanuel Kant ending up by putting Kant in a
position where he has no answer. One input space contains Kants philo-
sophical writings and views as well as the contemporary philosopher and
his work with philosophical problems in the tradition that is heavily influ-
enced by Kant; another input space contains the academic symposium
scenario. When these spaces are blended, we can cast the philosophical
disagreements in terms of a debate with turns-at-talk, replies and refuta-

24 Kant the philosopher is still with us in his works as a result of cultural learning
(in one of its more demanding forms!), and in terms of the restrictive defini-
tion of mental spaces adopted on p. 46 above, philosophical discussion can
therefore be carried on by people inhabiting one undivided mental space that
96 Chapter 2. From conceptual representations to social processes

Blending has some similarities with metaphor, and sometimes incorpo-

rates metaphors as subcomponents. Like blending, metaphor depends on
mapping from one conceptual sphere to another, and the result has ele-
ments from both: when love is understood as a journey, it has both love-
properties and journey-properties. But where metaphor lends itself to an
understanding in terms of stable underlying mappings, blending is more
obviously an online dynamic event. It is plausible that our understanding
of space is in certain ways permanently mapped onto our understanding
of time, for instance when it comes to understanding distances in time.
However, we cannot assume a debate with Kant that is stored in the per-
manently available cognitive unconscious, and the blend therefore brings
about something new.
This is also reflected in what is the key difference between metaphor
and blending, namely the role of emergent effects. Metaphorical map-
pings in the classic Lakoff and Johnson sense are not emergent they are
permanent parts of the mental universe, as reflected in the (provocatively
ontological) format of metaphor description X IS Y, as in LOVE IS A
JOURNEY. The mappings can be activated by linguistically articulated
metaphorical descriptions, but the real metaphors are already there, in the
form of cognitive mappings. In contrast, when you blend two spaces online,
there is a whiff of mixing powder and embers: you may have something
explosive on your hands. However, here is no absolute dichotomy between
dynamic and static conceptual mappings.25 Once it has emerged, a blend
may stick around. The Grim Reaper, for instance, is a blend that has
become culturally entrenched.
The distinction between blending and metaphor brings out two impor-
tant differences. First, blending is broader than metaphor because it does

includes the results of the process of cultural tradition. But online debate
requires the integration of distinct mental spaces also in the restrictive sense,
because Kant the debater is no longer with us and cannot enter the space in
which the living philosopher puts forward his arguments.
25 A step towards more dynamic conceptualization of metaphor was taken by
Lakoff and Turner (1989), describing literary metaphors as extensions of exist-
ing conceptual mappings, with novel literary effects, which might equally
deserve the epithet of emergent effects. Conversely, blends may involve stable
mappings. In Turner (2001: 19f), cockfighting in Balinese culture as analyzed
by Geertz (1973) is described as involving a blend between man and cock that
endows the fighting with a significance which depends on the co-presence of
man elements and bird elements in the fight to the death. The blends status as
staple element is revealed in the fact that outsiders are likely to miss the point
if they drop accidentally past an actual cock fight.
Meaning construction 97

not have to involve two different domains like space and time or philo-
sophical tradition and campus debating arrangements any two concep-
tual spaces can be blended. Another standard example can be invoked
here: the riddle of the monk (adapted from Koestler, cf. Fauconnier and
Turner 2002: 39). Imagine a monk who walks from the foot of a mountain
to a sanctuary at the top, taking from eight in the morning to eight in the
evening to reach his goal. After staying the night at the shrine, he follows
the path in the opposite direction, again taking from eight in the morning
to eight in the evening to complete his journey. The riddle asks: is there
a place where the monk is at exactly the same time on the two consecu-
tive days? The answer can be arrived at by blending the two journeys,
imagining the monk setting out at eight oclock simultaneously from the
top and the foot of the mountain. The place that the riddle asks for is the
point where the monk meets himself. This blend is clearly distinct from a
metaphor, since there is no recruiting of structure from one domain to
The second distinct property of blending in its most interesting form is
that it is double-scope, i. e. it does not have the clear directionality that
metaphors have. The whole point of a metaphor is to recruit structure that
you know in order to impose it upon something that you are trying to
understand. If theories are buildings, you know that they take an effort to
construct, for instance. If argument is war, you know you have to counter
your opponents moves. Blends lack this directionality because the rela-
tionship is not from source to target; it goes from two source spaces to one
output space that did not exist before, and where anything can there-
fore happen.
This is where the distinctive creative potential of the blending opera-
tion comes in. It is no accident that blending theory gave rise to a book
called The Literary Mind (Turner 1996). Blending offers the unique
opportunity of mixing conceptual elements in ways that transcend the
familiar cultural environment. The figure of Satan in Miltons Paradise
Lost achieves his literary effect because his status as personification of
Evil is blended with certain properties of heroic figures in narratives
(including being the arch-rebel).26 Successful writers at some point dis-
cover that their creations are no longer willing to do their creators bid-
ding but start to act in ways that need to be discovered rather than deter-
mined and this testifies to the strength of emergent effects. If a blend is

26 There are also a great deal of other blending activities going on in the Satan
figure, cf. Fauconnier and Turner (2002: 16062).
98 Chapter 2. From conceptual representations to social processes

successful and has a life of its own, a new entity has arisen and the concep-
tual world is no longer the same.
This potential is obviously a source of meaning construction rather
than evocation or invocation of existing meanings. Deviant but fascinating
figures abound in great works. When many of us read a great many more
run-of-the mill crime novels than great works of contemporary fiction, it
is partly because blending may be a conceptual challenge, and sometimes
you may not feel comfortable with some of the newly emergent creatures
that start prowling around in your mind27.
One type of conceptual integration operation illustrates the reversal of
the perspective that is central to this chapter. The directionality that begins
with the human body has its counterpart in the goal that unites the princi-
ples of blending: achieve human scale! (Fauconnier and Turner 2002:
312). This principle is a constraint on the processes of meaning construc-
tion and thus affects the output end rather than the origins of the process.
The process of compression works so as to convert a multiplicity of inputs
in both spaces to a compact version in the blend, such as when the evolu-
tionary story of the population of pronghorns across millions of years
evolve to run away from predators is compressed into one pronghorn that
escapes which is open to human identification because it is a kind of
event that fits into ordinary experience. The human element is active at
both ends: in the body that provides the neural wetware and in the online
interactive process that shapes the outcome.
In Coulson (2006), the potential of blending for online construction of
meaning outside literary contexts is analysed in relation to thought, rheto-
ric and ideology. Stressing the temporary and online nature of the opera-
tion of blending, she traces the new interpretations imposed by blends
such as snowflake kid used by a Christian pro-life group about surplus
fertilized eggs resulting from in vitro fertilization. Speaking of these eggs
in their liquid nitrogen tanks as tiny humans located in frozen orphan-
ages triggers a process of meaning construction whereby it would be
unethical to use them for stem cell research (which is what the debate was
about). Note that the operation works by the active re-construction from
eggs to children, encoded with the endearing term kids, and by conferring
the affective value that is associated with baby humans on the eggs. If you
think of them in those terms, it would indeed be unethical to experiment

27 Sharing this low-brow propensity, I found it comforting to learn from Ray

Monks biography of Wittgenstein (whose uncompromising existential stand-
ards were daunting in other respects) that he was particularly fond of the
Black Mask stories and assiduously avoided any variations on the pattern!
Meaning construction 99

with them. (There is of course a step from running the blend in your
mind to adopting this as the way you think about them; more on that
below p. 307.)
When this powerful mechanism of generating new meanings has
become part of the landscape, it becomes logical to think also of less
pyrotechnic (the term is Turners) forms of conceptual processing in
terms of meaning construction rather than in terms of meaning evocation.
One unresolved issue is the division of labour between the creative build-
up of surprising new forms of meaning and the evocation of familiar types
of meaning in standard combinations. As pointed out in Bundgaard,
stergaard and Stjernfelt (2007), although blending analysis may be uni-
versally feasible, it will often be overkill because standard schemas are
more plausible and economical. Their illustration case is compounds (one
of the issues listed as examples of what blending analysis applies to in
Turner 2007). They are essentially asymmetrical because they are modifier
constructions with a head that defines the point of departure for under-
standing, and therefore you can understand modification in the simple
case as based on schemas associated with the head. Thus fingernail can
be understood without requiring blending, because nails have a schematic
relation with fingers. Similarly house rat can be understood without
blending while mall rat (a term for delinquent teenagers who hang out
in shopping malls) do indeed require blending: no schema associated with
rat will provide the right interpretive link with mall28.
One of the processes that have acquired new significance as a result of
the increased focus on meaning construction is metonymy (cf. Panther
and Thornburg 2007, Panther 2006, Radden et al. 2007). Basic so-called
metaphors like more is up, cf. Panther (2006), might with greater justifica-
tion be understood as metonymies because it is the experiential, indexi-

28 Embodied construction grammar (cf. Bergen and Chang 2005; Feldman 2006)
views the process of meaning construction in a way that is relevant to this dis-
cussion. In accordance with the central assumption in construction grammar,
it is assumed that constructions are entrenched as wholes rather than com-
posed online. The comprehension process therefore has a necessary first step
in the form of evoking the constructions which are involved in a given utter-
ance. For the classic example of she sneezed the napkin off the table you there-
fore need to evoke caused motion, sneezing, napkin, etc. However, once the
constructions are activated, down to sensorimotor level, the second phase con-
sists in running what is termed a simulation of the entirety of the construc-
tional subcomponents. The analogy is with the way a computer runs a pro-
gramme, and the way Fauconnier and Turner speak about running a blend.
100 Chapter 2. From conceptual representations to social processes

cal connection that motivates them. Generally, Panther argues based on

examples that metonymic links must be understood as the primary source
of contextual enrichment. With the example a Pearl Harbour must never
happen again, Panther (2006: 169f) shows that the metonymic interpreta-
tion place for event is no more than a prompt for further interpretation:
whatever the reader knows becomes available as a source of added mean-
ing (negative effects, surprise attack, loss of military capability ). The
function of metonymy as meaning enrichment in context is based on
indexical relations and means that you build up larger meanings instead
of merely substituting one (contiguous) element for another.
Barcelona (2007) explicitly links up metonymy with Gricean pragmat-
ics, using metonymic links as input to Gricean implicature. Thus a cause-
effect relation at the textual level, where introducing a salient new refer-
ent causes the expectation of more information, drives the textual
inferencing process that creates coherence between the elements. Faced
with the textual beginning If you have ever driven west on Interstate 70
from Denver to the Continental Divide, you have seen Mount Bethel, read-
ers infer that a description of Mount Bethel is now coming up. The aware-
ness of how things hang together is not just a passive conceptual construct
but an active force driving cognitive processes in understanding towards
richer and more coherent meanings29.
In general, the process of metonymically driven meaning construction
reflects the fact that there is an online process of grabbing chunks of
meaning going on, and Langackers idea of point of access works also in
the pragmatic dimension: once you have gained access to a semantic point,
you may accidentally or by design take in more or less of the surrounding
Croft and Cruse (2004) use the term dynamic construal rather than
meaning construction, but their orientation shares with blending and

29 Stressing the methodological dimension, in harmony with Geeraerts, Gibbs

(2007) discusses the question of what evidence there is for the actual causal
role of figurative language. The metonymy place for event (in addition to
Pearl Harbor, cf. above, one may cite Gibbss example he was shocked by Viet-
nam), for instance, was found in one study to be active only in 1 % of all coun-
try names (Markert and Nissim 2003, as quoted by Gibbs 2007:22), and it is
necessary to make ones claims precise in relation to such empirical figures.
One finding that supports the role of metonymy in actual text construction
rather than as an abstract principle is the fact that familiar relations have
stronger effects: you recruit understanding based on actual available links-in-
context (Gibbs 2007: 23).
Final Remarks 101

online metonymy the emphasis on situated processes rather than static

general constructs. They quote, with approval, Smith and Samuelson
(1997: 167), saying that These foundational ideas of stable categories and
stable concepts, have led to little progress (Croft & Cruse 2004: 92).
Instead of taking the classical approach of locating stable structure in the
lexicon and accounting for variability by means of pragmatic rules and
principles (cp. the process from meaning via metonymy to Gricean impli-
cature explored above), Croft and Cruse (2004: 97) take a different
approach whereby neither meanings nor structural relations are speci-
fied in the lexicon, but are construed on-line, in actual situations of use
(referring to Moore and Carling 1982 as previous proponents of this view).
This discussion will be taken up in ch. 5.
The rise of meaning construction as a major issue is thus another man-
ifestation of the general trend explored in this chapter: from the anchor-
ing points of language and meaning in the general properties of cognition
grounded by the human body to the online and fluctuating processes of
making language work in a social context.

8. Final Remarks

There is obviously more to say about the socially oriented work that is at
present developing within CL than described in this chapter. This chapter
has attempted only to single out the main directions of this work. Some
additional aspects, including the specifically political dimension, will be
taken up as part of the presentation of the overall social cognitive frame-
work in Part Two of the book, especially in chapter 7.
In the introduction to this chapter I mentioned variation and intersub-
jectivity as the two keywords. To sum up at sound bite level what the
essence of this chapter has been, the most striking manifestation of the
usage-based trend is variational description, while the most radical change
from the individual mind to minds in an interactive relationship is the shift
to intersubjectivity. One phenomenon is especially central to the whole
project of the book: joint attention and action. Because joint attention and
action create a third element in addition to what is in each individuals
mind, collective mental phenomena have more emergent properties than
traffic jams (cf. the introduction, p. 6). All the issues taken up in this chap-
ter must be understood in the light of the constructive powers of joint-
attention events: conceptualizations viewed as part of culture, the role of
the population level in variational and evolutionary patterns, the dynam-
ics of meaning construction in context.
In the next chapter, we are now going to have a look at the competi-
tion. The social arena is not virgin territory that advancing cognitive lin-
guists can claim as their own merely by planting the flag, and in order to
discuss the potential role of a social CL, it is necessary to assess the find-
ings of approaches who regard it as their home base.
Introduction 103

Chapter 3: Social constructions and discourses

1. Introduction

The social sciences do not traditionally take a primary interest in mind and
language. However, there is one type of approach, associated with the two
keywords in the title of the chapter, which has gained ground in the past
generation, and which has assigned a significant role to language and mean-
ing. Social constructionism1 is the belief that social processes have the
power to create a range of entities that used to be understood as facts (dis-
eases, scientific theories, genders, etc). Discourses (in the plural) are key
agents in that process: what precise entities are generated depends on what
discourse is at work. The social semiosis is one name for the whole overall
process in which we find discourses and the entities they give rise to.
Although this approach is radically opposed to that of CL and cogni-
tive science generally, there is another sense in which they start from the
same point of departure: the demise of objectivism. In order to under-
stand what the two approaches share and where they differ, it is necessary
to understand this historical point of departure.
From antiquity onwards it had been assumed that words had meaning
because solid reality lurked right behind them. The precise nature of that
seemingly obvious link, however, stubbornly resisted all attempts to nail it
down; and in the 20th century it was abandoned.2 The transition is gener-

1 Although the two terms social constructionism and social constructivism

are sometimes used synonymously, often a distinction is made in terms of
which the latter is used about learning, especially as associated with Vygot-
skys position, cf. this passage from Wikipedias entry on constructivism: ().
social constructionism focuses on the artifacts that are created through the social
interactions of a group, while social constructivism focuses on an individual's
learning that takes place because of their interactions in a group. Throughout
this book I use the term social constructionism as a cover term for both; in
terms of the framework I propose, the difference is captured by the distinction
between the niche and the competency dimension of social facts, cf. the conclu-
sion of ch. 4. (I am indebted to an anonymous reviewer for pointing this issue
out to me).
2 Putnam approached the problem of linking language and world with the
sophisticated intellectual technology of logical empiricism and showed that
it was logically insoluble. In Putnam (1980) he showed (by an argument that is
104 Chapter 3: Social constructions and discourses

ally associated with the change from early to late Wittgenstein. In Philo-
sophical Investigations (Wittgenstein 1953), meaning is described as some-
thing that emerges from language games, rather than being based in
mind-internal structures or as labels of objects. Placing the naming game
as merely one type of activity that you can use language for, Wittgenstein
in one fell swoop gets rid of both objective and mental categories as priv-
ileged sources of meaning. An alternative example of a language game is
one enacted by two builders, one of whom has the task of passing on build-
ing materials to the other, who does the actual building. An utterance like
slab gets its meaning from being part of that language game, and the other
builder understands it because he is part of the game. Looking for the
source of meaning in the mind would thus be mistaken: if we abstract
away the building activity (and imagine the two people lying on the beach
instead), the utterance slab would no longer make sense.
This was a truly monumental change. Suddenly it became clear that the
entire 2400-year-old approach to meaning had got it backwards. When the
objective facts of unified science went away, both mental and social facts
came out to play but they played in separate ball parks. Cognitive science,
including CL, pursued the mental side of the issue. The social side did not
join forces in a parallel umbrella discipline of social science, but across
disciplinary boundaries social constructionism had equally spectacular
consequences. If social processes are the key causal agent behind every-
thing from the scientific world picture to Tolstoys War and Peace, it under-
mines the foundations of the whole tradition of knowledge not just the
tradition in the humanities (which had been slowly undermined by the
progress of science anyway), but also in the hard sciences. In various forms
it has continued to exert corrosive influence on the authority of apparently
solid knowledge. We are still in the process of digesting this change.
This chapter begins by outlining the rationale of social construction-
ism, with a view to identifying the main true insight and the main fallacy
in the development. The next section describes the most influential mani-
festation, the French poststructuralists or anti-humanists, and then goes
on to sketch out the type of discourse analysis that is associated with the
plural form discourses. I then describe first a form of psychology and then
a form of linguistics in which the social semiosis is the basic point of depar-

related to Gdels famous proof of the incompleteness of mathematical sys-

tems) that any language, formal or everyday, can only be linked to reality by
virtue of an existing social consensus about the meanings of words and state-
The social construction of reality 105

2. The social construction of reality

In order to understand what this approach implies for language and mind,
it is necessary to begin by considering what it means for knowledge in
general. In relation to language, the causal power of social forces is most
familiar from Austins and Searles theories of speech acts. Acts such as
marrying and promising bring about things that were not there before
(marriages and promises). Social constructionism views the creation of
scientific knowledge from the same point of view: what are the processes
whereby a piece of information is assigned the status of fact?
From that perspective, facts are constructed through social processes.
The same goes for another centrepiece of the tradition, the concept of
knowledge itself: only if we assign the status of fact to a given piece of
information can we say that we know it. In addition to the iconoclastic
thrill, it created a new exciting project: investigating the processes that
contribute to the construction of facts. The first major work that thema-
tizes the role of social factors in determining what counts as scientific fact
was The Structure of Scientific Revolutions (Kuhn 1962), which put the
notion of a scientific paradigm on the map as a name for the collective
belief system of a school of scientists. This concept is set up expressly as
something distinct from scientific theories viewed as the best available
approximation to true knowledge. Paradigms are familiar facts of life for
linguists, since the field is a very obvious example of how scientific disci-
plines may contain competing sets of basic assumptions. When a linguist
has to present his investigation of a given linguistic body of observations,
the issue of how to adapt his conclusions to the paradigm is clearly a sepa-
rate and laborious issue that comes on top of the process of adapting his
conclusions to his object of description (previously understood as real-
ity). Scientific paradigms are cognitive constructs, and thus illustrate that
this development is part of the same historical process that gave rise to
cognitive linguistics. But like other authors discussed in this chapter, Kuhn
is less interested in cognition than in the workings of social power, in this
case the power of scientific communities.3
A second major work, which came to define the post-positivist take on
scientific knowledge also in a broader community perspective was The
social construction of reality (Berger and Luckmann 1966). They dis-
cussed science as a key example of how social processes determine what

3 Like this book, Kuhn understood such socially constructed paradigms as

developing in analogy with Darwinian evolution (Kuhn [1962] 1970: 172).
106 Chapter 3: Social constructions and discourses

is to count as reality. In their perspective, scientific knowledge does not

begin with the real world, then enter our minds and finally move into
social processes of discussion and teaching. Instead, knowledge arises
from social processes of collection, sorting and gaining acceptance, with
the link to the real world as a tenuous and recalcitrant issue that is never
finally settled.
There has been a considerable degree of confusion about exactly what
follows from the development outlined above. There is one insight that I
regard as both important and uncontroversial: the processes whereby
something acquires factual status in human minds are to some extent
social and knowledge can only exist in human minds. Therefore no one
can deny that facts as we know them are social constructs. Universities are
social institutions, and since they have a role in sorting and proliferating
facts, facts undergo social processing.
Adherents of objective knowledge may, however, be forgiven for feel-
ing that this is beside the point that social constructionists want to make.
But from this point it becomes difficult to get any further. It is hard to get
a grip on facts in their pristine, non-social mode of being. Only the type of
knowledge that consists in direct perceptual observation, as the case of
seeing the tree in front of the house, can be kept apart from social proc-
esses. Positivists therefore generally took that as the prototypical case of
knowledge creation. But not all knowledge is grounded solely in direct
perception, and new scientific knowledge never is. New findings have to
be systematized, discussed, replicated and evaluated before it is clear how
they should affect our knowledge state. These processes take place in the
scientific community. The processes whereby something gets the status of
new knowledge in the scientific community are indisputably social, and
they involve the same processes of spreading and evaluation as the proc-
ess whereby something comes to be accepted as knowledge in the social
community as a whole. Any lingering feeling that an individual direct
observation would be a better prototype of scientific certainty cannot, on
reflection, be maintained. A sole individual may be deluded or halluci-
nated. Even if somebody had seen the Loch Ness monster clear as day-
light, we would look for confirmation from other sources. If you try to
bypass social validation procedures, you are not in the inner sanctum of
science you are merely on your own.
This argument, which is persuasively represented by Rorty (1980), still
leaves something that causes scientists to go but ! The only source of
resistance, however, is the fundamental conviction that whether or not
something is actually true, and thus deserves the status of knowledge, does
not depend on social processes. Let me confess at once that I share that
The social construction of reality 107

conviction. Who would want to deny that I can be innocent of a crime

even if all available evidence points to me and everybody is firmly con-
vinced that I did it?
Rorty suggests that we give up the idea because we cannot do anything
with it, and argues that we should see knowledge as founded in tried and
tested social procedures alone because they are what make the differ-
ence. But this argument depends on assuming that only operational crite-
ria are definitive for adopting a belief: if nothing follows logically from a
belief, we might just as well throw it away. However, this is not the way
actual human beings work we stubbornly insist that there is a fact of the
matter although we cannot be sure when we have found it. This may be
described as an act of faith, but I suggest it is better described as a form of
adaptation (more on this ch. 5): we cannot help taking the existence of a
material world for granted, because that is the way we are wired up. This
is my personal favourite for an unconscious cognitive routine with an
innate foundation: if something happens to you, you assume that there is
something in the world that brought it about.
And that is why it makes sense to insist that there exists a fact of the
matter, also on the level of explicit beliefs: if this is the way we cant help
thinking anyway, we might as well bring our conscious beliefs in line. Only
hardened intellectuals feel free to ignore this commonsense perspective, at
their peril: with a quip usually ascribed to Wittgenstein, surely the people
who deny the reality of the material world would not wish to deny that
under my trousers I wear underpants. Hard-nosed science (cf. Dawkins
2009) and common sense thus join hands at this point: what should count
as knowledge about an object of investigation does not depend on socially
constructed representations of that object, but only on what the object is
really like.
As powerfully illustrated by the Sokal Hoax4, unconstrained social
constructionism thus raises two related issues. The first is the question of
validity: if fact status is something we can just assign to propositions, how

4 The difficulty in pinning down exactly how much was determined by social
processes, combined with the common sense acceptance of the distinction
between solid fact and illusion has created considerable leeway for disagree-
ment and extreme positions. One of the most celebrated clashes was the so-
called Sokal Hoax, where a physicist wrote an article which was a travesty of a
social-constructionist critique of belief in hard physical facts. The article in fact
succeeded in getting published in Social Text, a leading social-constructionist
periodical (cf. Sokal 1996).
108 Chapter 3: Social constructions and discourses

can we know whether they are trustworthy or not? The second is the ques-
tion of rationality: If the individual is causally bound to believe whatever
he downloads from the social group he hangs out with, what happens to
the process of sorting and evaluating incoming information? The latter is
associated with what Pinker (1994, 2002) calls the standard social science
model, in which human minds are blank slates, overwritten by whatever
social processes may inscribe on them. While the anything goes position
is not credible in science, we still need to know the precise nature of the
influence of social processes in determining what counts as good science.5
What such a precise account is going to look like is controversial. What
is uncontroversial is that the process by which we assign the status as sci-
ence to something is social, and reliable procedures for evaluating scien-
tific results thus depend on a good social framework for science. Ironically,
social constructionism about scientific knowledge therefore revolves
round one pivotal, objective, inescapable scientific fact: all the stuff that is
recognized as knowledge has undergone a process of social construction.
This means that whatever other factors enter into the picture, such as mol-
ecules, neurons and metaphors, an account of the role of social factors is
mandatory in order to understand the way we think.

3. Power, habitus, marginalization and discourse:

the French poststructuralists

The social constructivist revolution affected not only the theory of hard
science but had an even more pervasive impact on fields dealing with the
softer forms of knowledge in the areas of culture and society. In this
context, it allied itself with a broad agenda of anti-authoritarianism that
affected also the more general political and cultural agenda of the 1970s.
Revolts against established forms of knowledge went hand in hand with a
revolt against the set of assumptions that constituted the legitimization of
established political practices in the western world.
In the early phase of that development, Karl Marx was an important
influence, especially with respect to one key point: mental categories are
shaped by social processes ultimately driven by power relations. The state-
ment that the ruling thoughts are the thoughts of the rulers (Marx/Engels
1932: 35) can be regarded as the foundational statement in the left-wing

5 Cf. p. 89 above on the Darwinian theory of Hull (1988).

Power, habitus, marginalization and discourse 109

agenda of social constructionism and is echoed in countless different con-

texts, e. g., the Bourdieus critique of reproduction in education (Bourdieu
and Passeron 1990: xv;1011). On another point, however, Marxist think-
ing was at odds with social constructionism: Marx believed rather too fer-
vently in solid material reality6. What was generally accepted in the intel-
lectual climate of the time was the element of social determination,
combined with the suspicion against accepted patterns of thinking. The
status of the thoughts of the ruling classes in the community came to be
understood in terms of the concept of hegemony rather than ideology
in the classic sense of illusion. In the version that was widely adopted,
hegemony as defined by Gramsci (Gramsci 1971, 1991), cf. Wright (2004:
16768), was a form of domination that did not depend on raw power, but
on acceptance of beliefs and positions by non-dominant groups. In a state
of hegemony, people outside the dominant group pursue their own inter-
est in ways that reinforce the hegemony (following the maxim if you cant
beat them, join them).7
Among the most influential figures in the kind of social construction-
ism that set the agenda in the 1970s are a number of French intellectuals
known variously as post-structuralists and anti-humanists, with Bourdieu,
Foucault and Derrida as key figures. There are radical differences between
them, not least in the role they assign to language (more on this below).
The justification for lumping them together in this section is only that they
all take the force of social processes as their point of departure, and so

6 One way of summarizing the main thrust of his ideas is to say that what is real
is in fact not decided by social consensus. That was why all the socially
entrenched ideas that young and old Hegelians were haggling about were all
equally misguided. Instead, the bedrock reality of peoples lives is created by
their own interaction with the material world. This is where the term ideology
gets its sense of illusion from: there is an actual social reality which can be
factored out from all the talk. While the freshness of the indignation behind
Marxs attack on German idealist philosophy has survived, the project of fil-
tering out the realm of non-reality from material reality turned out to be dif-
ficult to bring to a successful conclusion without an element of reductionism
(vulgar Marxism). Reducing superstructure to being determined by the
base constituted by material production gradually lost ground; the remains
were swept away by social constructionism and with it the intellectual inter-
est in the strictly economic part of the agenda.
7 As emphasized by Wright (2004), this is a more specific situation than simply
one in which there is simply a superior position that dominates over less pow-
erful beliefs and practices.
110 Chapter 3: Social constructions and discourses

they see the understanding of the individual as subject to limitations due

to the processes of social determination. Further, instead of looking for
purely economic forms of determination in the Marxist tradition, they
look for the causal source of what counts as facts in broader types of social
processes and practices. Bourdieu expresses the last point as follows:
The only way to escape from the ethnocentric naiveties of economism () is
to carry out in full what economism does only partially, and to extend eco-
nomic calculation to all the goods, material and symbolic, without distinction,
that present themselves as rare and worthy of being sought after in a particu-
lar social formation which may be fair words or smiles, handshakes or
shrugs, compliments or attention, challenges or insults, powers or pleasures,
gossip or scientific information, distinction or distinctions, etc. (Bourdieu
1977: 178)
What Bourdieu describes as the driving force, equally in the traditional
Kabylian culture where he did his early field work and in the modern
world, are the social processes, including everyday as well as ritual prac-
tices, that assign symbolic value and meaning to actions and material
goods alike. For the scientist, the task as he sees it is to construct a theory
of practice, or, more precisely, the theory of the mode of generation of prac-
tices (Bourdieu 1977: 72). The symbolic values are of course mentally rep-
resentable, but that is not their central property:
We see yet again how erroneous it would be to consider only the cognitive or,
as Durkheim put it, speculative functions of mythico-ritual representations
(Bourdieu 1977: 165)

One of the reasons why mental representations are assumed to be insuf-

ficient is that the generation of practice is seen as involving crucially not
just the collective practices, but also (as a reflex of those practices) a set of
pre-conceptual dispositions inscribed directly in the body, as subsumed
under Bourdieus most famous concept, habitus.
The concept of habitus illustrates at the same time an affinity and a
difference between the cognitive-linguistic and Bourdieus social-interac-
tive approach to cultural meanings: on the one hand, it affirms the impor-
tance of bodily grounding of cultural meanings, and also places elements
of culture backstage (cf. p. 198 below), as inaccessible to conscious
inspection. On the other hand, it locates the explanatory factors else-
where, outside the embodied mind. Bourdieus interest is in pointing to
the existence of social processes that drive certain cultural patterns so
deep into the human subject that they are out of sight. Educational proc-
esses are among those patterns that, according to Bourdieu, have the cre-
ation of a habitus as its goal: the society reproduces itself by imprinting
Power, habitus, marginalization and discourse 111

itself directly into the bodily dispositions of the students (cf. Bourdieu &
Passeron 1990: 31), dispensing with the detour that goes via cognitive rep-
The key type of social practice, and the central explanatory factor in
critically oriented social constructionism, is what is called discourse in a
wide sense that includes all language-mediated social processes from edu-
cation, religion, and government to everyday communication. The con-
cept of discourse as understood by Foucault and especially his followers,
has gradually assumed centre-stage position as the term for what analysts
need to get their hands on in order to understand what shapes post-objec-
tivist forms of reality. The most striking renewal of the standard way of
talking about discourse is the introduction of a countable sense, a dis-
course, understood broadly speaking as a particular way of talking about
things (more on this in the following section).
Initially, Foucaults interest was in large-scale quasi-objective forms of
determination, with historical epochs in focus and shifts in discourse as
parts of a new form of history of ideas. Foucaults (1969) study of the
change in scientific discourses at various points in history points to a
mechanism related to the one made famous by Kuhn (1962) in the theory
of science9. At any given point in history there is a particular way of talk-
ing and also disagreeing about a field of study that is taken for granted.
When the participants in such historical circles of practice make their con-
tributions, they act according to unwritten rules to which there are no
alternatives available. By the same token, these practices also determine
what is beyond the pale, i. e. what cannot be articulated within a context
dominated by a given discourse. It is not that it is physically impossible it
is just that it does not arise, or if by a freak accident it does, it becomes
invisible because it does not qualify as a relevant way of talking about

8 Cultural and material practices are interwoven in this view of human practice,
since symbolic and financial capital are interchangeable. An illustration is that
in certain societies it is rational to spend all your money on a wedding since
that wedding will give more status in the sociocultural order than hoarded sav-
ings. Although Bourdieu explicitly does not see either language or cognition
as the central factors, the processes that he as well as the other French anti-
humanists are interested in are those that are accompanied by language and
other forms of symbolic mediation; Bourdieu speaks of lconomie des
changes linguistiques in the original title of the book that in English became
Language and symbolic power (Bourdieu 1991).
9 As pointed out by Piaget 1972, p. 112 (in the Danish translation).
112 Chapter 3: Social constructions and discourses

Later, Foucault turned to the interplay between social power and

human subjectivity, with the history of prisons and social control as a
major motif (cf. Foucault 1975) One focus is on the dividing practices
whereby external forces impose a split between the normal and the mar-
ginal: the mad and the sane, the sick and the healthy, the criminals and
the good boys (Foucault 1982: 208). This brings him into the territory
of what cognitive linguists would call idealized or stereotyped conceptual
models, although at this point Foucault views them as constituents in proc-
esses of determination in a macro-social perspective. In his third phase, he
continued the trajectory from historical formations into the human sub-
ject itself and the impact of the self-definitions on experienced identity,
especially in terms of sexuality. At this stage his focus moves beyond
socially distributed patterns of thinking to the issue of embodied under-
standing, albeit still from the point of view of social determination.
Following up on the studies of deviation and normality, a privileged focus
for Foucault and his followers has been the study of texts reflecting the
polarity in social categorization between us and them, or between self
and other; more on this in ch 8.10
It is part of this extremely influential way of thinking that discourses
always appear to represent reality, even as they subtly reorganize the way
we think. The approach thus opens a chink in the armour of ruling repre-
sentations, allowing criticism to expose the latent discrepancy between
what is presented as reality and what is behind it. Yet since social reality
only manifests itself in discourses, and as discourse never stands still, we
can never get at the ultimate reality that lies behind them if, indeed,
there is any reality behind them. The concept of floating signifier goes
back to Lvy-Strauss, but became central when structuralism turned into
poststructuralism, because of the constant re-articulation of meanings.
Taking its point of departure in the traditional assumption that there is a
fixed Platonic meaning underlying all signs, the notion of floating signifier
came to embody the role of signs as part of the semiotic flux: meanings are
generated as we go along and are carried along with the flow, with no ulti-
mate foundation.
In the theory of literature, the concept of intertextuality reflects the
same type of mechanism: rather than representing diverging forms of

10 This contrast has been adopted and used in vast numbers of different guises,
including gender studies (where heterosexual patterns are mainstream), class
structure (where the ruling classes are mainstream) and post-colonialism
(where established western doctrines are mainstream).
Power, habitus, marginalization and discourse 113

reality, texts take their cues from each other and reproduce as well as
reorganize particular organizations of meanings. In the most radical ver-
sion of this position, there is nothing but interpretations as far as the eye
can reach: Every decoding is another encoding, as succinctly expressed
by David Lodges fictitious professor Zapp. A key point of reference is a
passage from Derrida which is often rendered as there is nothing outside
the text (but which in the French original is il ny a pas de hors-texte, Der-
rida (1967: 227), which is a pun on hors doeuvre and means something
like there is no privileged domain that is totally extra-textual).
In Derridas own context, this much-discussed quote is not part of a
discussion of what really exists Derrida is playing a different language
game and is not really into ontology talk. This becomes clear in the dis-
cussion with Searle in Glyph, Derrida (1977a&b), where rather than
entering into the discussion, Derrida is being naughty according to the
standard rules of the game.11 Although the plausible intended message is
not to express the thoroughgoing ontological scepticism which detractors
use it to illustrate, but rather the text-like interpretation-driven features
of human existence as such (cf. Derrida 1981: xiv), there is a certain irony
in having to resort to the authors intended message in trying to defend
Derridas general position. However, this should not obscure the rele-
vance of the theme that Derrida is pursuing: that readers are in the grip of
forces they cannot hope to get behind, and we should therefore be scepti-
cal of the claims of those who pretend to have found a secure foothold
beyond the semiosis itself. The risk is only that this particular agenda is
taken to be the whole truth, cf. pp. 336 and 450.
With the alliance between discourse and the exertion of social power
of marginalization and oppression, there is a critical edge shared with
Marxism, even if there is no material base for it. The factors that shape
discourse in ways that speakers cannot defend themselves against still cry
out for being pointed out and examined. This gives rise to a pattern of
interpretation which, following Ricoeur (1970), has been called the
hermeneutics of suspicion: the task of meaning construction in under-
standing discourse is associated with an orientation towards dismantling
illusions and unmasking hidden power play in the text.

11 As pointed out in Siegumfeldts 2005 (counter-)obituary, the fact that his posi-
tion emphasized the unfinished nature of all interpretations and openness
towards the future did not mean that questions of sources and origins were
foreign or irrelevant to him.
114 Chapter 3: Social constructions and discourses

4. The analytic practice: discourse(s) analysis

Discourse analysis is an umbrella term for a large variety of analytic

practices whose only shared feature is that they analyse actual instances
of language use. The following should perhaps be printed as a government
warning on all textbooks in discourse analysis:
What gets addressed under the rubric discourse is so varied that the default
expectation should be the non-generalizability of what is said about some type
of discursive object of attention to others. (Schegloff 1999: 167)
The Foucault-inspired type of discourse analysis introduced in the section
above, however, is possibly the most influential variety. As a framework
for analysing ways of thinking in literary and political contexts, it is also
the approach that constitutes the most conspicuous alternative to CL
when it comes to understanding meaning in society and meaning con-
struction in texts. Bluntly speaking, if CL in its new socio-cognitive mani-
festation wants to stand as an attractive method of analysing conceptual
models in relation to social forces, it must be seen to have advantages that
post-Foucauldian discourse analysis lacks. I think it does have such advan-
tages, and also that post-Foucauldian analysis is potentially dangerous if it
is not handled with extreme care (which is far from always the case, cf.
Antaki et al. 2002).
For those reasons I will take some trouble in giving an account of core
features of this analytic practice (or collection of practices12), henceforth
called the discourses approach, or discourses analysis. In this chapter
I first try to place it in relation to other forms of discourse analysis. Sec-
ondly, I take up concrete examples of how it is understood and used in
analysis and try to show both what it typically covers and what it typically
leaves out. When I have presented my own approach to meaning in soci-
ety in ch. 7, I compare it with examples of the discourses analysis, and in
ch. 8 I illustrate why the differences in approach are not just important
academically, but also from a civic point of view.
This form of discourse analysis involves two key elements: (1) uncover-
ing a pattern of conceptual organization as reflected in a body of texts
(statements) and (2) showing what particular position this discourse

12 It follows from the heterogeneous nature of the enterprise that what I present
as core features will not fit all instances equally well, cf. also chapter 7, section
4. What I claim is merely that there is a position in the landscape that my
account fits well enough and that has enough impact to make it worth chal-
The analytic practice: discourse(s) analysis 115

takes up in social space, especially as a manifestation of power relations.

The conceptual dimension is thus understood in relation to its social posi-
tioning. Some key concepts are taken over from the Marxist tradition,
including ideology and hegemony. After the strict division between
truth and illusion faded out of the picture, ideology is typically understood
as a coherent set of culturally salient assumptions and values adopted by
a (sub-)community, with the capitalist ideology still in its position as a
historically central example. The concept of hegemony, as introduced
above, is used about the dominant position of an ideology in the relevant
body of texts, understood as reflecting and maintaining power relations in
the community by the collusion of people outside the dominant coalition.
The focus is on uncovering the economic and ideological interests associ-
ated with the position manifested in the discursive practices that are ana-
lysed (cf. Laclau & Mouffe 1985).
In this approach, discourses are the central objects of description in
accounting for the social role of mental representations, rather than eco-
nomic interests or pervasive practices. This emphasizes the force attrib-
uted to patterns of linguistic interaction and the beliefs that are expressed
in such interaction. It suggests that what our so-called knowledge derives
from is essentially the flow of language with changing patterns of mean-
ing, referring back to previous patterns while continuously reorganizing
In the linguistic tradition, there is a different form of discourse analysis
which arose as the result of crossing the sentence barrier, cf. Harris (1951).
From that perspective, discourse is the domain of relations between sen-
tences (understood as the maximal unit of grammatical analysis), and con-
stitutes a new linguistic domain, where patterns associated with grammar
gives way to other patterns. The Rubicon-like crossing of the sentence
boundary has remained defining of the field, cf. Schiffrins (2006) charac-
terization of discourse as structure operating above and beyond the sen-
This linguistic definition represents an adaptation to the fact that the
intra-sentential turf was already occupied by Chomskyan linguistics. One
notes the element of unconscious collusion that is characteristic of victims
of hegemony: in taking over the remainder, linguistic discourse analysts
give up the sentence-internal territory without a murmur as lying outside
discourse(!). Although defined by phenomena that go beyond the sen-
tence, discourse is actually more or less equal to language use: processes
like repair, topic nomination occur inside as well as outside sentences,
but are among the characteristic features of discourse organization. The
focus is on the online creation of meaningful relations between utterances
116 Chapter 3: Social constructions and discourses

and participants, with the emphasis strongly on local discursive events. In

Conversation Analysis (cf. Sacks, Jefferson & Schegloff 1974), this empha-
sis on the local, empirically given course of events has been at the most
marked, yielding substantial insights in the microstructure of oral com-
The linguistic tradition of analysing discourse is understood as part of
the umbrella construction of usage-based linguistics, cf. the discussion
above p. 58. The strength of this conception of linguistics can be described
as a joining of forces between CL, which was heavy on mental content and
light on interactive patterns, and the linguistic tradition of discourse anal-
ysis, which was heavy on interactive patterns and light on mental content.
Textbooks in discourse analysis, ignoring Schegloffs warning, tend to
treat the two traditions as varieties of the same overall enterprise. There
is, however, a valid point in seeing the two enterprises as belonging to the
same field of inquiry especially from the point of view that I am adopt-
ing in this book. Both usage-based linguistics and poststructural dis-
course analysis subscribe to the claim that discourse, i. e. linguistic interac-
tion, is the central channel through which individual mental representations
engage with the social world. The affinity extends to descriptive sociolin-
guistics, where the importance of socially constructed, power-laden proc-
esses has become a major issue.
Although sociolinguists focus on the factors that determine choice of
linguistic variants, this topic also involves power and social construction,
because the choice of particular variant forms does not just reflect, but
also in itself constitutes or constructs membership of social groups (more
on this in ch. 7). Thus by speaking in a particular way, choosing special
linguistic variants, you define your own position and identity (or let other
people define it!) in relation to the social world. The sociolinguistic vari-
ants you choose have a position in social space and contrast with variants
selected by rival communities, just as the core conceptual models you acti-
vate in discourse contrast with models favoured by rival groups.
The pattern is observable at both micro- and macro-social levels. On
the macro-political level, the issue of opting for a national standard lan-
guage in the tradition of the French enlightenment is directly tied to the
social construction of the nation (cf. Geeraerts 2003c as discussed p. 68).
On the individual level, sociolinguistic choices such as speaking immigrant
multi-ethnolects (cp. Quist 2005) or other down-market linguistic forms
also reflect conceptual models of who you are and who you belong with.
The distinction between jocks and preppies on the one hand, and burn-
outs on the other (cp. Eckert 2000) establishes a choice between socially
entrenched group identities in youth culture based on a contrast between
The analytic practice: discourse(s) analysis 117

mainstream and marginal with affinity to social class, which is indexed

by linguistic variants. Issues of the kind that reflect conceptualizations of
ethnic group identity involving issues of marginalization pervade sociolin-
guistics. Tannen (1994) similarly links up discursive choices and patterns
with cultural assumptions that may differ between genders and subcul-
tures, including media culture. There is no point in trying to enforce apart-
heid between the traditions. But when the post-Foucauldian style of anal-
ysis on a broad scale blends in with the linguistic tradition, this only makes
it more urgent to be explicit about its limitations.
As discussed above, the theoretical centrepiece is the (countable) con-
cept of a discourse, understood roughly as a group or type of statements
or texts sharing a common orientation. Originally (cf. Foucault 1969) this
was understood in relation to texts produced by particular intellectual
groupings such as the physiocrats, or of texts constituting a particular
academic discipline such as economics or biology in a specific period.
Foucault at one point defined a discourse as constituted by a particular set
of historically determined rules (Foucault [1969] 1989: 131). However,
these rules did not produce a clear-cut set of shared properties. Foucault
rejected the traditional assumption that the key common denominator
was to be sought in the subject matter, or in the shared body of thought.
In looking for the unity of such bodies of texts he looked instead (cf.
above, p. 111) for unity in the pattern of dispersion, i. e. in the structure of
their different positions in common space (Foucault 1989: 4041). The
key idea was that a body of statements could be understood as reflecting
a set of rules for positioning yourself within a pre-given field, such that the
players were unaware of the processes of determination that kept them
within the rules.
This idea caught on for many reasons. The basic reason was that it
pointed to a type of phenomena that had until then been overlooked:
when we learn to talk in a particular way about particular subjects, we are
more or less aware of the field within which we can choose to move and
the options we have within it, but we are not conscious of the borderline
between what is possible for us and what lies beyond our ken.13 An insidi-

13 This is true in more than one way. First of all, it applies to the borderline
between the field that constitutes our world and the untrodden domains
beyond and secondly it applies to the distinction between the types of things
we say and the types of things we habitually do not say within the everyday
cosmos we live in. The latter we only meet in the shape of embarrassing fail-
ures and transgressions, without being aware that what is a failure by one
standard may be an achievement by another.
118 Chapter 3: Social constructions and discourses

ous added attraction of the model, however, is that it situates the analyst
at a level of consciousness above the language user: the analyst is looking
for something the speaker herself cannot see. This privileged position for
the observer follows directly from the limitations of the position of the
subject: if the analyst was equally incapable of seeing what eludes the sub-
ject herself, there would be no analysis possible. There is no dialogic, col-
laborative relation between the sender and the analyst, who is not an
addressee but an observer.14
Foucault pioneered the approach of looking for the backgrounded,
hidden and invisible elements in the picture rather than the foregrounded
and explicitly highlighted. The distinction between the focal and the mar-
ginalized is related to the distinction between normality and the other.
His general approach is well suited for a description of the way in which
dominant patterns of talk are instruments of upholding a particular bal-
ance of power that keeps some people in and others out, while masquer-
ading as simply representing the way the world is.
I would therefore like to draw a distinction between on the one hand
the specific contribution of Foucault to understanding the historical
changes he describes, and on the other hand what Blommaert and Ver-
schueren (cf. chapter 8) call the Foucauldian lore. For instance, Foucaults
analysis of the development of the correctional dimension of institutions,
cf. Foucault 1975, brings together an impressive amount of knowledge,
linking up language, architecture, institutional history and many other
diverse fields. In doing so, Foucault shows how the detailed and compre-
hensive historical knowledge of a range of social processes allows you to

14 Foucault and his followers are essentially only interested in statements, ways
of representing the world, and thus deal mostly with texts that might to others
naively appear to represent reality. But in continuation of Foucaults investi-
gation of social processes of marginalization and exclusion, the interest in the
role of practices of discussion and communication in setting up the borderline
between what is inside vs. outside the field of accepted practice has played a
tremendous and well-deserved role because it brought something under the
microscope that had eluded the attention of investigators. If it is assumed that
language gets its meaning from reality itself, the role of talk in making parts of
reality invisible does not arise as an issue for the analyst and thus escapes
attention, just as the deft movements of the magician keep certain acts out of
sight of the audience. The processes of social determination that lead people
to talk about some things but not others, and talk about those things in certain
ways but not in other ways, would therefore be potentially successful also in
misleading the critical investigator.
The analytic practice: discourse(s) analysis 119

bring out the rise of interwoven patterns of thought and social control that
would be inaccessible if you stayed within traditional assumptions and
discipline boundaries. I thus follow Putnam (2001) when he expresses his
admiration of Foucaults analysis of specific institutional practices, while
remaining more sceptical about the more general features of his work
(those that are taken over by the Foucauldian lore in Blommaert and
Verschuerens terms).
One common mildly diluted variant is discourses analysis used with
texts as the object of description, i. e. without an independent analysis of
institutional practices. The following description of it is somewhat boring
and not very dense in information, but this unfortunate feature is moti-
vated by the sprawling nature of the phenomenon, its dispersion in social
space (as it were), combined with my wish to provide a reasonable amount
of documentation, since I would otherwise suspect myself of tending
towards unfairness in my description.
As described above, discourses analysis begins with a distinction
between different discourses, thereby allowing the analyst to assign a
piece of discourse (non-count) to a given discourse (countable), such as
the neoliberal discourse or the discourse of tolerance. The use of this
form of analysis is characteristic of approaches that use the adjective crit-
ical about themselves, such as critical theory, critical linguistics, and
especially critical discourse analysis (= CDA). Among the most influen-
tial representatives of Foucault-inspired CDA is Norman Fairclough (e. g.
1995, 2000, 2003), which makes it natural to use his work as a point of
To illustrate the vagueness as well as the central point in the concept of
discourse in this approach I begin with a selection of quotations:

Fairclough (2003: 123f) introduces the concept with reference to a discus-

sion in Foucault:
I believe I have in fact added to its [the word discourse, PH] meanings: treat-
ing it sometimes as the general domain of all statements, sometimes as an indi-
vidualized group of statements, and sometimes as the regulated practice that
accounts for a number of statements. (from Foucault 1984)

The analysis of discourse for Foucault is the analysis of the domain of state-
ments that is, of texts, and of utterances as constituent elements of texts. But
that does not mean a concern with detailed analysis of texts the concern is
more a matter of discerning the rules which govern bodies of texts and utter-
Then, from Fairclough (2003: 124125), on his own behalf:
120 Chapter 3: Social constructions and discourses

As in the case of genres, it makes sense to distinguish different levels of abstrac-

tion or generality in talking about discourses. For instance, there is a way of
representing people as primarily rational, separate and unitary individuals,
whose identity as social beings is secondary in that social relations are seen as
entered into by pre-existing individuals. There are various names we might give
to this discourse for instance, the individualist discourse of the self, or the
Cartesian discourse of the subject. It has a long history, it has at times been
common sense for most people, it is the basis of theories and philosophies and
can be traced through text and talk in many domains of social life, and its scale
is considerable it generates a vast range of representations. On a rather less
general, but still very general, level, we might identify in the domain of politics
a discourse of liberalism, and within the economic domain a Taylorist dis-
course of management. By contrast, in Fairclough (2000b) I discussed the polit-
ical discourse of the third way, i. e. the discourse of New Labour, which is a
discourse attached to a particular position within the political field at a particu-
lar point in time (the discourse is certainly less than a decade old).

In the same tradition, Riggins (1997: 2) intensifies the sceptical dimension:

In everyday language, a discourse traditionally has been understood as a state-
ment or an utterance longer than a sentence (Fiske 1987, p.14). But in the
humanities and social sciences in recent years, the term has come to have a
more elusive meaning that usually takes the work of Foucault (1972, 1984) as
its starting point. Foucault seems to have emphasized the structural nature of
statements, including those that are spontaneous, and the way in which all
statements are intertextual because they are interpreted against the backdrop
of other statements. The anthropologist William OBarr, following Foucaults
lead, provides a useful general definition of discourse as a flow of ideas that
are connected to one another (OBarr 1994, p.3). A more technical definition
might be to say that a discourse is a systematic, internally consistent body of
representations, the language used in representing a given social practice
from a particular point of view (Fairclough 1995b, p. 56). The practice of social
work, for example, could be conceptualized in terms of at least two discourses.
Social work can be seen as the provision of benevolent, professional care or as
a negative and repressive form of population control (Stenson, 1993). Dis-
courses do not faithfully reflect reality like mirrors (as journalists would have
us believe). Instead, they are artifacts of language through which the very real-
ity they purport to reflect is constructed.
The quotations illustrate both how elusive the concept is (cf. especially
Riggins), and also how sharply profiled its central function is: to pinpoint,
and typically challenge, a specific point of view that underlies a particular
representation of the world.
As suggested in the Fairclough quote above, there are generous crite-
ria for how discourses can be individuated; you can in fact define a par-
ticular discourse pretty much as you like. But since the point of the analy-
The analytic practice: discourse(s) analysis 121

sis is to pin down a particular point of view, a given discourse has to stay
within the defining point of view. This becomes clear when Fairclough
(2000: 24) analyses a Blair speech containing the following passage:
In the increasingly global economy of today, we cannot compete in the old way.
Capital is mobile, technology can migrate quickly and goods can be made in
low cost countries and shipped to developed markets ()

()The fourth (point, i. e., that technology can migrate quickly, PH) is perhaps
the most interesting: it is represented as an action, but technology is repre-
sented as itself an agent in a process, rather than something that is acted
upon (i. e. moved) by the multinationals. Notice the metaphor here technol-
ogy migrates, like birds in winter. The sentence might be differently worded,
for instance as : The multinational corporations can quickly move capital and
technology from place to place, and they can make goods in low countries and
ship them to developed markets. But this is not just a change in the wording; it
is a change of discourse: in the discourse of New Labour, the multinational cor-
porations are not agents responsible for what happens in the global economy.
What prevents the hypothetical reformulation from joining the existing
Blair discourse is its status as an ideological odd-man-out. Purely abstractly
there is no necessary contradiction between seeing multinationals as
agents and other elements in Blairs universe; but the ideology of agency
is different from the one Fairclough is after. It is not a matter of either
logical contradiction or strictly linguistic choices, but of ideological stance.
If Fairclough had chosen to define other ideological features as criterial or
if Blair had chosen as his overall strategy to profile the beneficial effects
of agentive corporations, this would have been part of the New Labour
The discourses analysis therefore and that is crucial in comparing it
with a cognitive-functional approach inherits one half of the classical
objectivist understanding of language: it assumes a truth-based semantics
The difference is, of course, that what is to count as reality is part of the
battle. Similarly, as pointed out by OHalloran (2003), it also takes over a
truth-conditionally based view of language processing, corresponding to
first-generation cognitive science: meanings provide a precise world
view although the reality claim and the world view are approached with
scepticism. There is no awareness that language could be linked to reality
in other ways, such as by its functions within a way of life.

15 One may observe that Foucaults own analysis of the way economic power
works would appear to coincide with Blairs rather than with an analysis in
terms of explicit agency.
122 Chapter 3: Social constructions and discourses

Most authors, Fairclough included, recognize the existence of a real

world behind the representations; but they simultaneously point to the
difficulties in getting at it. The following quotation can illustrate the orien-
tation towards characterizing the discourse itself rather than what it rep-
Do we perhaps, beyond its fables and follies, pretend to know what terrorism
is? No, indeed, we question the very possibility of defining, and thereby giving a
satisfactory account of, the facts categorized as terrorism. Our goal is not to
elaborate yet another typology, but rather to redirect the study of terrorism
into an examination of the very discourse in which it is couched. As is the case
with other discourses of the postmodern world we inhabit, the terrorist signifi-
ers are free-floating and their meanings derive from language itself. (Zuleika &
Douglass 1995)
Sometimes the causal role of discourses is upgraded to the point where
discourses themselves are seen as agents, leaving the speakers themselves
as puppets. This is in keeping with one of the points stressed by Foucault,
namely that the rules are hidden: speakers do not know the discursive
rules that they are following. Jrgensen & Phillips (1998: 24) cite Kvale
(1992: 36) for the claim that the self no longer uses language to express
itself, it is rather the language (or discourse) that speaks through the per-
The central analytic mechanism in poststructural discourse analysis
reflects the social constructionist pattern of thinking: extra-discursive
reality is out of sight; the human mind is a passive medium formed by the
social process; language-mediated social processes with their baggage of

16 How the discourse can be personified and abstracted out of the speakers can
also be illustrated by Hansen (2006), using discourse analysis in the political
science field of International Relations:
The construction of the Balkans as incapable of change and with the capac-
ity of entrapping the West functioned to legitimize a Western policy of inac-
tion. But as accounts of the warfare surfaced in Western Media, the Balkan
discourse had to engage in a debate on whether ethnic cleansing and geno-
cide warranted Western intervention. Here the Balkan discourse made a
double move. First, it homogenized the inside of the Balkans by constitut-
ing the subjectivity of anyone involved in the war in Bosnia as one of being
Balkan, more specifically as parties or warring factions.
We see how the social agent, the Balkan discourse finds a counter-ploy to a
move by an opposed agent by regrouping parts of the underlying Weltan-
schauung (more on the role of meaning and language in the theory of Inter-
national Relations in chapter 7 below).
Discursive psychology 123

representational content are the sole actors on the scene, and the analyst
(standing above the fray) can essentially choose or even construct the
object he wants to focus on.
Foucault (1969) argued that this view constituted a new phase in intel-
lectual history and provocatively claimed that the age of man was end-
ing that human being as a concept was becoming obsolete, replaced by
the impersonal forces of discourse(s). But instead of viewing this as a
world view in which human beings are mere pawns, it makes more sense
to see it as a type of analytic practice that chooses a different vantage
point compared to the individual-based perspective with the human agent
as the hero. Foucault-inspired discourse analyses such as Edward Saids
([1978] 2003) Orientalism have succeeded in bringing out patterns that are
indeed invisible to the individual speaker, precisely because they adopt a
perspective that is beyond the scope of single individuals: there is a cluster
of related but variable ways of speaking about the Orient which has
developed over centuries and which, for the individual speaker at any
given historical time, are already there before she even opens her mouth.
We can only get at that kind of pattern if we abstract from the individual
and therefore this type of analysis brings out elements that were over-
looked before Foucault, and in so doing puts a much-needed critical spot-
light on the power of social forces to determine (partly) what we think
and what we take to be real without being aware of it.

5. Discursive psychology

The discourse-based pattern of thinking gave rise to a new form of social

psychology that is presented in the following quotes from key members of
the group:
We argue that the researcher should bracket off the whole issue of the quality
of accounts as accurate or inaccurate descriptions of mental states. The problem
is being construed entirely at the wrong level. Our focus is entirely on discourse
itself: how it is constructed, its functions, and the consequences which arise
from different discursive organization. In this sense, discourse analysis is a radi-
cally non-cognitive form of social psychology. (Potter & Wetherell 1987: 178).

the self I portray in my experiment has no voluntary agency. Ones sense of

self, in this context, is determined by social feedback; I am simply the reposi-
tory of others attitudes toward me. (Gergen 1996)

Discursive psychology understands mental processes as driven by online

discursive activity. Instead of the encyclopaedic view of conceptual rep-
124 Chapter 3: Social constructions and discourses

resentations in CL, we may think of it as a print-on-demand model:

representations are produced when events require them. The ancestry
outlined in Potter and Wetherell (1987) is mostly non-psychological:
anti-mentalist philosophers including Wittgenstein and Ryle, the concept
of speech acts (especially in Austins version), ethnomethodological soci-
ology and Saussure-inspired thinking about language. The role of lan-
guage is central as the carrier of the discursively mediated content that is
the chosen focus of psychological investigation. Discursive psychology is
thus a good candidate for the purely blank slate approach to the human
mind reviled by Pinker (2002), and for the deconstructionist extreme in
Johnsons (1992) account of where CL is situated in the academic land-
scape. As if speaking directly to cognitive linguists, Gergen describes his
stance as follows:
Let us first deconstruct the traditional emotional terms concepts such as
anger, love, fear, joy, and the like. That is, let us view such terms as social con-
structions, and not as indexing differentiated properties of the mind or the cor-
tex. (Gergen 1996)

Discursive psychology also aligns itself with the conversation analysis that
grew out of ethnomethodology, including its commitment to close-knit
analysis of the situated unfolding of conversational practices. What the
discursive psychologists add to these foundations is the attention to the
psychological maneuvering involved in the process of online construction.
They use the work of Atkinson and Drew (1979) on courtroom interroga-
tion to illustrate how the discursive process of attribution of blame is the
key motor in the process, shaping questions and especially answers. They
note (Potter & Wetherell 1987: 89) that allocating blame takes a number
of turns, usually eight or nine, and point out that witness testimonies are
organized so as to best deflect the process. Understanding of what goes on
as well as the causal mechanisms driving the process needs to be based in
the process of blame attribution and the construction of events that goes
into that process, rather than in facts about the inner mentality of partici-
One of the analogies used is that of dancing: like people engaged in a
square dance, participants engaged in discursive interaction know what
expectations and role attributions are at work in the unfolding situation
and navigate as best they can so as to acquit themselves (sometimes liter-
ally so!) as best they can. Referring to Garfinkel (1972), discursive psy-
chologists invoke the existence of a dense network of everyday normative
mechanisms working to keep things within existing patterns, but also
shaping events when developments stray from those patterns.
Discursive psychology 125

Instead of the causality associated with agency or with determination

from outside the discursive process, discursive psychology concentrates
on the factors at work in the interaction itself. Specifically, the use of cat-
egories is seen as explained neither by the existence of mental, conceptual
models, nor as deriving from classification of things in the real world, but
for their causal power in discourse. In the court case, categories such as
girl, race, black sister and white man are invoked because they count
towards relevant inferences, and are variable (indeed inconsistent), in
terms of content within the same stretch of discourse: categories are
selected and formulated in such a way that their specific features help
accomplish certain goals (Potter & Wetherell 1987: 137).
Potter (2003), defending his approach, stresses the advantages of this
view as opposed to traditional methods which treat society and its actors
[] as structured sets of causal entities. (Potter 2003: 791). Arguing
against a critique that accuses the theory of having a thin view of the
human actor, he claims that It is a rather weak idea of what is thick or thin
that treats these intimate, consequential studies of psychology in practice
as somehow lacking in comparison with traditional models of agents with
inner motors. He also points out that discursive psychology does not
straightforwardly contradict the traditional (including cognitive)
approaches to psychology, only in that it develops an alternative under-
standing of language and its role in the machineries of psychological
research and assessment.
This defence is well suited as a point of departure for understanding the
role of causality in the picture. Causality is crucial also in the foundations I
suggest for a social cognitive linguistics (cf. ch 7). When Potter distances
himself from discourse-external causes, I believe it must be understood pri-
marily in contrast to a positivist, single-step form of determination by
objective forces (cf. the discussion of Wver, p. 370 below). In psychology
such a view would imply that human beings always acted out a certain set of
objectively fixed personal drives, regardless of the particular circumstances,
including unfolding discursive practice. Clearly it would be a simplistic and
reductionist enterprise to set up predictions about individual responses in
interaction solely based on inner, pre-defined properties (biological or
mental) without bothering to study how purposes and responses are modu-
lated by the force of unfolding reality, including discursive pressures.
However, the view espoused by Potter in the quotation above is also
fairly obviously only a thin slice of the full story. Behind the online causal-
ity of blame attribution, working from turn to turn in court interrogation,
there is a deeper pattern in which the institutions of law enforcement are
active (more on this in chapter 7 below), and there is also a set of concep-
126 Chapter 3: Social constructions and discourses

tual models and status functions in terms of which blame is attributed on

a principled basis. Thus when suspects know that certain sorts of motiva-
tions may be seen as more culpable than others (Potter & Wetherell 1987:
130), it is presumably due partly to causal forces associated with practices
of societal discipline and punishment (cf. Foucault 1975), and correlated,
pre-existing conceptual models of culpability. The process of witnesses
who try to minimize blame assignment thus cannot be understood without
reference to those discourse-external factors.
If taken literally, the position of discursive psychology reflects the turt-
les-all-the-way-down ontology of social constructionism. Since it is clear
that something has gone before the online discursive causation, it is part
of the pattern of thinking that what goes before is also the result of discur-
sive processes (rather than pre-given essences). Behind the social con-
structions there are other things, which are also socially constructed.
In relation to the interpretation of that particular type of linguistic
process that involves the understanding of texts, the pattern of argument
is reminiscent of the hermeneutic circle: understanding the whole presup-
poses the understanding of parts, and vice versa. In both cases there is a
valid point made by those statements. In order to capture that point with-
out ending up in a vicious circle, we have to see it as part of a bigger pic-
ture, as argued above: since we are not dealing with a static relation of
causal determination, but dynamic processes of meaning construction,
there is nothing mysterious in the claim that each act of interpretation has
consequences: if we understand a part of a text in a particular way, it has
subsequent implications for our understanding of the whole and when
we have then constructed a reading of the whole text, it has subsequent
implications for the way we understand its parts. Similarly, when we
impose a particular construction on the event we are testifying about in
court, it has implications for what further constructions we can make in
later answers (otherwise the construction might collapse on us). But this
in no way rules out causal impact from factors outside the discursive pro-
cess. A full account thus has to consider how the full story can be put
together, while welcoming the contribution from detailed investigation of
the strictly online part of the psychological process.

6. Systemic-Functional Linguistics

Also within linguistics, there is a long-standing group which is essentially

social in its orientation, viewing semantics as based in social semiotics,
namely Systemic-Functional Linguistics (= SFL) with Michael Halliday
Systemic-Functional Linguistics 127

as the central figure.17 SFL shares a number of basic beliefs with cognitive
linguistics, and Halliday has had most of those beliefs since before Chom-
sky took centre-stage position on the linguistic scene. For almost sixty
years, he has been pointing in essentially the direction that is associated
with this book, and in so doing been right from an early date about an
impressive number of important things. It is worth outlining some of
First of all, following the lead of his mentor, Firth, he points to the
foundational status of actual linguistic usage in linguistics: language must
be understood as ongoing practice. Secondly, meaning rather than form is
the name of the game: the social process that is the primary manifestation
of language is a process of creating, handling and exchanging meanings.
Thirdly, as opposed to the American tradition, the issue of structuring is
not primarily associated with the formal side, but also and essentially with
the side of meaning. Meaning, as in CL, is understood as encoded in a
continuum rather than a split between lexicon and grammar, as expressed
in the term lexicogrammar Fourthly, the experiential dimension is
embedded in a larger, functional conception of what semantics is that also
includes an explicitly recognized interactive, or interpersonal, dimension,
as well as a textual dimension.
Some of Hallidays theories about links between sentence structure
and interpersonal and textual features have gained very wide recognition
both inside and outside linguistics. For example, the reanalysis of standard
notions of subjecthood (Halliday 19671968), in which are into actor, sub-
ject and theme, with roles defined in terms of the different major subcom-
ponents of the semantic system, has remained a classic.18 Halliday has also
stressed the role of construal, including what he calls grammatical meta-
phor, such as the reification associated with nominalizations (the shooting
rather that they shot him). Combined with agent deletion, it can be used to
transform the police shot a demonstrator into the shooting, a kind of
transformation with obvious political implications.

17 It should be pointed out that there are different groups within the larger com-
munity of Systemic-Functional Linguistics, and the discussion below does not
apply equally to all of them. In particular, the Cardiff community with Robin
Fawcett as a central figure would not in general agree with the positions I
criticize below.
18 Cf. Butler (2003) on the changing relations between the subcomponents dur-
ing the development of SFL.
128 Chapter 3: Social constructions and discourses

These are among those SFL-based linguistic concepts that have had
considerable impact in discourses analysis as described above.
As a final point, Halliday (cf. Halliday 1975) has analysed the roots of
language in early embodied practice, including as a particularly important
point the role of intersubjectivity as the basis of language development,
not as something that arises when two cognitive individuals happens to
meet; his contribution there has been stressed by e. g. Trevarthen, cp.
Trevarthen and Reddy (2006).
All in all it is therefore interesting, also in the context of the history of
linguistics, to ask why (I believe) not only Cognitive Linguistics in general
but specifically this book has not been rendered superfluous by progress
in systemic-functional linguistics?
In answer to this I would like to suggest that the systemic-functional
endeavour as a whole has taken a different course, and as a consequence
Hallidays impact has perhaps been greater outside than inside linguistics.
The basic feature, which is a strength in some respects but which has lim-
ited its power to bring about the kind of overall clarification that is the
aim of this book, is its inherent user-orientation.
In a lecture in Copenhagen (2006), Halliday talked about appliable
linguistics (rather than applicable, or applied). By the term appliable he
understood linguistic concepts developed for or by the potential users
themselves, devised so as to be usable in the particular context in which
users were interested. In the preface to his collected works (Halliday 2002:
2), he talks about how
when it came to asking questions about language, I always found myself lin-
ing up with the outsiders I was interested in what other people wanted to
know about language.

This personal bent combined with his theoretical orientation towards

ongoing practice as the basic level has created a certain centrifugal
momentum in his overall approach. One result of that is the laudable but
also intimidating level of detail with which individual language events,
including texts, are in principle supposed to be analysed. In Dimensions
of Discourse Analysis: Grammar (Halliday 1985), he analyses a short
spoken dialogue in ten different ways, giving rise to charts in which the
dialogue is characterized in terms of each of the ten different sets of
descriptive categories, with minute attention to ongoing language activity
at all relevant points. From the point of view of the reader, the balance of
interest in the account is strongly oriented towards the full richness of the
actual specific event, while it takes a certain effort to extract the generaliz-
able features of the overall analysis.
Systemic-Functional Linguistics 129

This would be nothing but a wholesome counterweight to the masses

of over-abstract and convoluted products of the fertile brains of linguists
who never went near an actual utterance if they could help it if it were
not for a more principled issue that arises in continuation of this. The crux
is the status of the criteria by which the concrete analyses have been con-
ducted. What is the relation between criteria used in an actual concrete
analysis and the theory?
This is a very difficult issue but also essential, not only for SFL, but for
any approach that aims to capture the variable and context-specific nature
of social events. I am therefore going to discuss it in some detail and
approach it from several directions. Those who feel that they have got the
point may skip to the next section!
The basic problem is (as pointed out by Butler 2003) that although
there is in one sense a large and complex set of categories permanently
available in SFL, these categories are not always defined explicitly enough
to make instantiations of these categories generally recognizable, and dif-
ferent users may therefore interpret them in ways that reflect their differ-
ent purposes. The price of flexibility, therefore, is that the analyses do not
add up to a theory that constrains subsequent analyses.
This is not a question solely of the possibility of disputing why in con-
crete instances a certain description has been proposed this charge
could be made of all linguistic analyses. The problem has deeper roots in
SFL and goes with a cornerstone of the whole theoretical edifice, namely
that there is assumed to be a full semantic system at the bottom of all lan-
guage events, including texts. This semantic system is not to be confused
with the lexicogrammatical system it constitutes a large complex of
purely semantic choices that comes before the interface with actual lin-
guistic forms. Halliday envisages a future in which this whole set of
semantic choices can be enriched to such an extent that it can handle all
the meaning-assigning operations that are now accounted for in terms of
inference, knowledge of the universe, and the like What we have to do is
extend and enrich our semantics to the point where we can handle these things
as part of the system and process of language (2002: 11).

Every new insight that you get when analysing a text, every addition to
your knowledge of the universe, must be understood as reflecting a part
of your semantic system. This system would have to be pretty big; but the
complexity is perhaps not the main worry the world is after all a compli-
cated affair. The problem is that it is not clear that there is a well-defined,
existent object of investigation which warrants all this complexity in the
130 Chapter 3: Social constructions and discourses

semantics (viewed as part of the language system) as envisaged by

By taking this stance, Halliday becomes at the same time the most
devotedly data-oriented linguist around, and the one who insists most
strongly on the fundamental role of the language system. Speaking again
of the semantico-pragmatic rather than the lexicogrammatic part of the
enterprise, he says If you dont know the system, you dont understand
the text (Halliday 2002: 10). The entire meaning potential of the language
user, as it unfolds itself in any given text, therefore has to be prefigured in
the system. The project is reminiscent of producing a universal map to the
scale of 1:1 of language in action or perhaps that is even an underestima-
tion: as pointed out in Bache (2008), the systemic grammar includes a
number of forms which are entirely unattested, so the system prefigures
not only everything that is out there but a great deal more besides.20
The need to have everything inside the system is linked to another part
of Hallidays orientation, the priority of the top-down view (2002: 12). This
need therefore does not arise if you combine a usage-based view with a
bottom-up approach, as in CL because the categories that we choose on
the hoof may exist nowhere else than in a particular text at a particular
time. In other words, output categories do not have to be prefigured in the

19 There is a fundamental difference between complexity as a property of the

description of an actual empirical object and complexity as a property of the
descriptive system. Any empirical object can be described in infinitely many
ways, and with an arbitrary level of complexity it depends on the person
doing the description what categories to apply, what perspectives to view it
from, and how to carve up the object. How far such descriptive complexity
should be carried depends on the concrete purpose of the description and the
funds available, as in a police investigation of a crime scene. It is a different
matter when you are considering what semantic categories are inherent in the
system itself. Anyone who wants to operate with a language system (as this
book does, cf. chapter 6), instead of taking actual language in use as the only
real object of description, is faced with the risk of hypostatization. When you
postulate a virtual category of the system in order to account for the actual
empirical occurrence of an instance of the category, you have to find a way to
avoid circularity and it is hard to see how this circularity can be avoided if
you assume by fiat that every empirical choice must reflect pre-defined fea-
tures of the general system as Halliday does.
20 The need to prefigure the entire potential for meaning creation in the linguis-
tic system (rather than allow a role for non-linguistic practice) marks a clear
contrast to Bourdieu, cf. Halliday (2002: 11): Bourdieu is an expert in exploit-
ing the power of language to proclaim that language has no power.
Systemic-Functional Linguistics 131

semantic system in order for the text to be understandable (pace the quo-
tation above). This is where the understanding of natural language differs
from the understanding of a formal system.21 For that reason, there can be
no valid criteria for setting up the whole set of semantic choices relevant
for text understanding in advance setting up such a system is either
something you can do any way you like, or a circular process whereby you
smuggle the output categories into the premises of textual understanding.
In practice, the semantic categories therefore have an uncertain status,
which is perhaps best understood in terms of the concept of appliability,
cf. above: they are there because it seems a good way to analyse the text
you happen to be dealing with.22
In practice, it thus appears to be up to the individual systemic linguist
to posit any categories that he finds revealing. An illustrative early instance
of this is when Halliday (1973: 72), envisaging a sociological semantics,
sets up a network of semantic choices in the area of parental control over
children. Inspired by Bernstein, he sets up a socio-semantic network
that includes a choice between person-oriented and position-oriented
forms of control, as the first tier of semantic choices that mediates between

21 In natural language understanding, the input is not determined solely by the

way the system works; in addition, it is shaped by the mental content that
arises in response to other forms of input than language. In contrast, if you
want to understand a number such as 111753, everything there is to under-
stand is given by the number system, which is indeed presupposed without it,
you will be unable to grasp any individual number that manifests the system.
Input from outside the number system cannot interfere in the process, the way
experiential input can shape the understanding of linguistic utterances.
22 The same uncertainty adheres to general categories of the analysis. As an
example, we may take the concept of field (together with tenor and mode
constituting the context of situation which is the last of the ten sub-analyses
in Halliday 1985). Field is defined as what is going on: the nature of the
social-semiotic activity. In the actual example, the interlocutors discuss
whether you can use the word it of a baby and the field of this text is said
to be (Halliday 1985: 283): Field: a general, imaginary problem of verbal behav-
iour: how to refer to a baby whose sex is unknown, without offending against the
parents, the baby (later in life), or the language.There is no obvious way of
delimiting the number and kinds of fields that you might want to assign the
text to. The term field is very appliable, in that it points to the fundamental
pragmatic issue: if you do not understand what is going on, you will have little
chance of doing the rest of the analysis. But precisely for that reason, it cannot
very well be a feature of the pre-existing system: ongoing activity by its nature
does not exist except while it is actually going on.
132 Chapter 3: Social constructions and discourses

purely behavioural options (outside language) and the choices that are
closer to the lexicogrammar. It is fascinating to follow the path that Halli-
day outlines from the situation of confrontation between the childs and
the parents agenda to the actual, if unfortunate, choice of If you do that
again Ill smack you. It is also clear that the socio-semantic network is set
up based on both a particular sociological condition and a particular soci-
ological theory. Had the sociologist conceived of the situational options in
different ways, the semantic network would have looked different. It
would appear also to be in the spirit of this practice to import for instance
the grid of choices offered on company phones (press one for complaints,
two for bookings .) as a systemic network into the socio-semantic
domain of customer inquiries.23
This broad scope for the theorists free choice may arise also in connec-
tion with categories close to the lexico-grammar. Thus Halliday (1994:
6869) operates with four basic speech act types (seen as (discourse)-
semantic options), divided two by two: offer vs. request, and goods &
services vs. information. These can be matched to lexicogrammatic cate-
gories, in that declaratives and interrogatives are respectively offers and
requests of information, and imperatives are requests for goods and serv-
ices but what about offer of goods and services? Well, for that semantic
choice there is no congruent lexicogrammatical category. This is perfectly
true but again, it is not clear how the semantic category offer of goods
and services is validated, or whether or not the descriptive principles of
Systemic-Functional Linguistics actually entail that it needs any validation.
This is the essential issue in the critique I have presented. There can be
no argument against positing new relevant descriptive categories; there
can be no objection to users applying existing categories in a way that
suits their purposes best; there can be no objection to describing an empir-
ical object in arbitrarily fine detail. But at the end of the day there has to
be a process of evaluating and validating all descriptive categorizations, if
they are to have any claim to a status as part of the descriptive apparatus,
rather than disposable tools that have served their purpose.

23 There is of course nothing wrong in using descriptions of social reality devel-

oped within another discipline as background for understanding linguistic
choices. The problem lies in the duplication or circularity that is the result if
the sociological description is imported wholesale into the semantic descrip-
tion of the linguistic choices themselves. Duplication arises if we describe the
choices first as sociological and then as semantic options; circularity arises if
we then go on to explain the semantic choices as reflecting the (identical) soci-
ological choices.
Systemic-Functional Linguistics 133

I believe the grand project for a socially based semantics that Halliday
outlines is unrealizable for reasons of principle, both because of the open-
endedness of the universe of potentially relevant semantic choices and
because of the hypostatization that would be involved in attributing them
all not only to ongoing process but also to the system from which choices
are supposed to have been made. I think this explains why a number of
linguists who have at one point rightly been fascinated with the many
attractive features of the approach, do not in the end align themselves
with the overall endeavour of systemic-functional linguistics: it is not
designed to bring about clarification of issues about which one might be in
doubt it is designed for users, not scientists.
This is also reflected in the practices of the community. Logically
enough, since the issue is not really whether this is the best possible
description, systemic linguists do not concentrate on that point:
Perhaps the most common criticism of systemic practice is that argumenta-
tion is severely lacking, to the point of being non-existent. (McGregor 1997:

The result is that although much work is done, and much of it is not only
useful but also potentially valuable, there is not much sorting and selec-
tion going on:
since, as we have seen, systemicists place a very high value on applications
and their feedback into the theory, this tendency acts to reinforce the accept-
ance of hypothesis as received wisdom. (Butler 2003: 204)

Bache (2008: 1) points out the consequences of this intellectual conserv-

atism also for core areas of the grammar, in his case tense and aspect.
Systemic-functional linguistics is best understood essentially as a tool-
box for the process of dealing with segments of linguistic reality, with out-
comes whose value is determined by what use they can be put to in con-
crete cases rather than a systematic attempt to validate a particular version
of the theory. As pointed out by Butler (2003: 202), SFL on Hallidays own
admission tends to work by accretion rather than by critical reassessment
of earlier positions. There can be no objection to the emphasis on useful-
ness: give me a useful descriptive tool rather than a useless theory any day
of the week. But if positions are not reworked but just abandoned and
new ones invented based on usefulness for the purpose at hand, this must
be part of the directions for use: do not take systemic-functional linguis-
tics as a linguistic theory. You may use it for your own descriptive pur-
poses, just as you may use it for any other purpose you may have, but you
do so at your own responsibility. You cannot have it both ways
134 Chapter 3: Social constructions and discourses

This feature of the system acquires an additional dimension when the

linguistic categories are brought into use in Critical Discourse Analysis.
Widdowson (2004) subjects this alliance to a thoroughgoing critique,
whose central feature is the too great reliance on features of the language
itself and the linguistic concepts, with too little attempt to demonstrate the
validity of the descriptions offered in relation to actual discourse.24
Hallidays analysis must be understood in the light of the general SFL
approach to the relation between cognition and language. Where CL
views language as emerging from general cognitive processes, SFL is
predicated on the opposite directionality, cf. the title of Halliday and Mat-
thiessen (1999), Construing Experience through Meaning: A Language-
based Approach to Cognition. If it is assumed that language is the original
site of meaning, it is less odd than it appears from a CL perspective to
assume that the linguistic formulation can be taken as the point of depar-
ture for the analysis of cognitive effects. As in discursive psychology, cf.
above p. 125, this assumption involves a thin view of inner mental con-

24 The appeal of Hallidayan grammar that comes from the goal of basing gram-
mar in actual text thus turns out to have the drawback that once the gram-
matical description has been made, it freezes the understanding of text into a
grammatical mold. If there is no distinction between the two, there can be no
dynamic interplay between the grammar and the process of textual under-
standing in context, understood as an online process that mediates between
the general categories and the concrete act of understanding. In a number of
cases, Widdowson shows that grammatical metaphor and agent deletion does
not really hold up as an argument for manipulatory language use. One exam-
ple is newspaper headlines, where there are good reasons why full sentences
instead of nominalizations are not to be expected. Another is the critical anal-
ysis of literature, cf. Widdowson (2000), which he shows to be un-cooperative,
because literature invites readers to carry out a genre-specific language activ-
ity which the critical analyst refuses to perform.In a similar vein, OHalloran
(2003:79) points out how Hallidays assumption that grammar as a description
of the linguistic system can at the same time be a grammar of the text has the
consequence that critical analysts can use Hallidays system to analyse texts
under the assumption that clauses are mentally facsimilated in cognition. If
something is coded as a thing (as a result of nominalization), this reification
is directly copied into cognition. A sentence that does not explicitly encode
agency is copied into a cognitive representation where agency is missing (see
also the discussion in ch.7).
25 Another fairly striking failing analysed by OHalloran (2003) is the absence of
a consideration of the role of the reader: whether the text spreads a political
bias or not depends on the activities of the reader as much as on the text. As
Systemic-Functional Linguistics 135

Robert de Beaugrandes riposte to Widdowsons critique (On use-

fulness and validity in the theory and practice of linguistics, de Beau-
grande 1998) while arguing against a number of distinct claims by Wid-
dowson, does not really address the validity issue, except at the end where
he says that a successful application can very well be one reliable indica-
tor of the validity of a theory. One can agree that there is a connection
between appli(c)ability and validity, in the sense that a theory that one
cannot meaningfully apply to an object is also disqualified from giving a
valid description but being a necessary condition is different from being
a sufficient condition. By leaving the argument at this stage, de Beau-
grande effectively concedes the point.
Each separate piece of systemic-functional analysis therefore stands or
falls by its own merits. The most inspiring work done within systemic-func-
tional linguistics, it seems to me, is found in analyses of areas that are
inherently bounded by the nature of the particular topic selected. In such
cases, the looseness of the framework matters relatively little, and the
openness and the nuanced attention to the whole field of social and psy-
chological factors involved can come into its own. Thus, for instance, the
territory of the classic work on cohesion (Halliday and Hasan 1976) is
defined by a fairly specific range of phenomena and a clearcut substance
area: it is placed at one of the crucial interfaces between the structured
area of the sentence and the less structured area of textual properties, and
has given rise to a rich subsequent development of the area. Similarly,
Halliday (1967) broke new ground by exploring the interface between
grammar and intonation, cf. Bache & Kvistgaard Jakobsen (1980); Mat-
thiesen, collaborating with American functionalists, contributed to a the-
ory of intra-sentential rhetorical acts, etc. In relation to the user perspec-

OHalloran points out, if you read for gist rather than immerse yourself in the
text, you are not likely to be mentally molded by the text. Rather than stamp
a precise reading into the minds of unsuspecting readers, news texts are sub-
ject to experiential meaning construction which means that text-based ideol-
ogy does not translate directly into ideological mystification to the extent
presupposed by the alliance between SFL and CDA.To some extent, the prob-
lem is the same one that has been approached from a number of perspectives
above: the assumption that the text world reflects categories of a large seman-
tic system, which makes it superfluous to reserve a role for interaction with the
extralinguistic world. Since the actual reader is part of the world-text interac-
tion, it is taken for granted that he can be subsumed under the description that
characterizes the text as meaning potential. The world, in a sense, is treated as
a feature of the text.
136 Chapter 3: Social constructions and discourses

tive, it can be mentioned that at the University of Southern Denmark

there is an education in organizational communication based on Halliday-
inspired principles: in this relatively bounded domain of options, the open-
ended, practice-inspired analytic practice has proved very successful.
In conclusion, although Systemic-Functional Linguistics has champi-
oned a social approach to meaning, it has not provided an overall theory
of social semantics as such, or a distinctive central conception of the field.
What it has done is to put social semantics on the agenda in its user-
friendly way, and given rise to examination of a number of interesting
disparate pieces of the overall puzzle.

7. Socially based theories of meaning: overview and issues

The historical shift associated with the late Wittgenstein combined the fall
of positivism and the rise of interaction as the home ground of meaning.
In spite of the lack of a unified umbrella (soc-sci) for the social sciences,
the idea that social processes, some more powerful than others, are the
driving forces of meaning construction is widely shared within fields
including literature, psychology, anthropology, philosophy and political
science. The approaches discussed in this chapter are just a small selection
of those that directly bear on meaning.
The unity-in-diversity is reflected in criss-crossing links between these
approaches. The role of social determination and power, and the critical
attitude to it are common ground. Hallliday is explicitly committed to lin-
guistics being used as an ideologically committed form of social action (cf.
Halliday 1985: 2, as quoted in Butler 2003: 158), and the most influential
work in SFL-inspired social semantics is probably Faircloughs school of
Critical Discourse Analysis which also uses Foucault-inspired concepts.
Discursive psychology looks for the same kind of power-driven mecha-
nisms as Foucault and also reflects the hermeneutics of suspicion. Both
discursive psychology and Halliday adopt a minimalist approach to cogni-
tive content: Halliday and Matthiessen (1999) links up language in the
social domain directly with neuroscience, thus cutting out mental cogni-
tion as an unnecessary middle man, cf. Butler (2003: 158). All in all, the
foundational importance of social processes, viewed in a critical perspec-
tive, and the focus on understanding conceptual content as more or less
derivative of the social processes, makes this broad position a clear alter-
native to an emerging social cognitive position.
In the chapter I have tried to make clear both where I think there is
something in each of these approaches that a social cognitive linguistics
Systemic-Functional Linguistics 137

needs to address, and also why I think their account of it is not definitive.
The points of interest have to do with the social life of meaning above
individual level the traffic, as well as the traffic jams of meaning, to
broaden the metaphor I used in the introduction (p. 6). Of these, I take up
later two elements from Bourdieu: the idea that cognitive constructs enter
into the economic sphere as symbolic capital, and the notion of habitus
as embodiment imposed from outside. From Foucault I take up the con-
cept of a discourse understood as a collection of utterances that exerts
pressure on subsequent contributions, defining what it is (im)possible to
say. From Derrida I take up the idea that there is a powerful internal
dynamics to processes of meaning construction in society, which cannot be
grasped from the outside. From discursive psychology I take up the power
of specific language games such as the blaming game to get people to
adopt conceptual constructions on the fly, as dictated by the internal logic
of the game, rather than their own mental content, and finally Systemic-
Functional Linguistics offers an approach to language as a vast tool-box
for user-friendly meaning construction.
The reason why these approaches are not definitive have to do with
two things. The first is the part of the picture that they ignore (or actively
suppress), which in most cases includes the human mind. The second is the
vagueness in the understanding of the nature of the all-important social
process. The two are related in that the absence of a clear role for the mind
also makes it unclear how to understand the role of mental constructs in
the social process.
For the project of an integrated social cognitive account, this raises a
pervasive question of clarification in cases where what appears to be the
same issue can be addressed both as a social and cognitive phenomenon.
Thus where Fairclough (p. 120 above) speaks of the Cartesian discourse,
Lakoff and Johnson speak of the Cartesian metaphoric model (Lakoff and
Johnson 1999: 409). From a Foucauldian perspective, Lakoffs strict father
model would be an instance of the discourse of governmentality, cf.
Foucault (1975), etc. There is also an affinity between the variationist
dimension of the emerging socio-cognitive position and the Foucauldian
concept of dispersion: the fact that it is not one well-defined conceptual
framework that constitutes a discourse but a fluid and moving constella-
tion of related positions in mutual interaction reflects the same insight.
We are left with a question with a chicken-and-egg whiff about it: what is
the relation between conceptual models and discursive processes? In
order to address this issue, it is necessary to provide an account of how
mental and social facts interact. This is the subject of the next chapter.
Chapter 4. The foundations of a socio-cognitive
synthesis: social reality as the context of cognition

1. Introduction

The approaches discussed in the previous chapter left us with an unre-

solved problem: how can we get a picture of social facts that can form the
basis of an integrated account of social and mental facts?
The revolt against objectivity left the field open to relativity instead,
offering no account of how causal factors in the social domain interact
with the rest of the world. But we have to take social facts seriously for the
same reason that we have to take physical facts seriously: we could get
hurt if we ignore them. Instead of the ubiquitous but intangible miasma of
social determination, we need to get at the hard facts.
One key difficulty is that the classic distinction between subjective
and objective is insufficient to get at hard facts in the social domain. It is
necessary to disentangle the question of what the intrinsic features of
social reality are (cf. Searle 1995) from a reductive understanding of
objectivity. This is the topic of section 2. I begin with the Platonic and
positivist positions and argue that we can avoid both objectivist and rela-
tivist reductionism if we understand social reality as having a specific kind
of complexity. The rest of the chapter tries to describe what that intrinsic
complexity consists in, ending up with an answer to the question raised in
the beginning: how does social reality tie in with the mind?
In section 3, I invoke the process of niche construction as a mecha-
nism for anchoring social facts in the rest of the world. Niche construction
is a special twist of the evolutionary scenario described in ch. 2. It does two
essential things: it shows how nature and nurture become conflated in a
way that can be precisely described; and it shows how the community can
become a causal factor in its own right.
From the biological time scale, section 4 moves on to the sociocultural
time scale. Here, evolutionary dynamics takes the form of the invisible
hand that operates over the heads of individual members of the commu-
nity. A separate section is devoted to the key area of evolutionary causal-
ity that establishes functional relations between individual action and the
community (section 5). In section 6, I present a theory of the interplay
between social and cognitive dimensions of language based on the
Social facts 139

premises established in previous sections. In section 7, I conclude by dem-

onstrating what this entails for language as an object of description in an
integrated theory of the social and cognitive properties of language.
This chapter focuses on social facts rather than the social process. I
return to the flow dimension in ch 5.

2. Social facts: objective and subjective, intrinsic

and observer-relative properties

What kinds of things are social facts?

The social constructionism debate has raised this question forcefully in
the last generation, but the problematic status of social facts is not new.
From the days of Plato, the social domain has generally been viewed as
antithetical to the whole idea of looking for facts, because of the associa-
tion between surface appearances and social phenomena on the one hand
and deep underlying ideas and the mental domain on the other.
A distinction is necessary here between Plato and the Platonic tradi-
tion. Plato himself was deeply concerned with the social domain; the title
of his main work was The Republic, a projected social construction in the
sense proposed in ch. 7.1 True reality had to be located at an underlying
level only because the distinction between appearance and reality was too
easily blurred in the social process. The problem was the sophists, not
social reality as such. That problem is alive and well, and on that point the
present work is entirely on Platos side. However, in the tradition of the
humanities, the Platonic pattern of thinking came to be associated with a
principle that located reality in eternal factors that were in principle invis-
ible. Actual events were thereby demoted to derived and secondary status.
It is this pattern of thinking that is designated with the phrase the Platonic
tradition in this book.
Until recently, the Platonic tradition has dominated western thought.
As a consequence, there is something oxymoronic about the phrase social
reality because the social sphere has been understood as the domain of
opinion and prejudice. And the reversal that created social construction-
ism, while proclaiming a new winner of the argument, essentially pre-
served the dilemma: you might say that radical social constructionism
confirms the worst suspicions about the social domain in the Platonic tra-

1 I am indebted to Hans Fink for pointing this out to me.

140 Chapter 4. The foundations of a socio-cognitive synthesis

dition. The search for hard facts clearly needs to look elsewhere; and there
are a number of useful points of departure in the tradition.
Like mental facts, social facts crept out of the shadow of supposedly
objective facts in a very gradual process, whose effective starting point
was somewhere in the Enlightenment. Until that point the official assump-
tion had been that the social order had no existence of its own, but was a
reflection of the divine order (and disaffection with it was therefore a
Satanic revolt). Instead, it was gradually realized that human beings were
to some extent responsible for making their own history. The processes
that culminated in the French Revolution put social processes on the map
as something that required attention in its own right.
This development took place in many arenas simultaneously. From
being viewed as divine, the social order came to be viewed in terms of a
contract (cf. Rousseau and Burke) still binding, but now an agreement
among human beings in a social relationship. Hume put on the agenda the
more radical issue of whether our patterns of thinking were merely the
result of what was socially accepted. But perhaps most significantly, the
privilege of being the invisible hand operating behind the scenes while
engendering events beneficial to everybody, was transferred from Divine
Providence to economic, market-based mechanisms (more on this in sec-
tion 4 below).
The idea of a general science of social reality was proposed by Auguste
Comte, the founder of positivism, in a version that explicitly aimed to
throw out the old idealistic tradition and begin with solid scientific facts.
This pattern of thinking continued with Durkheim, who was the founding
father of sociology, in the sense of being the first to give substance to the
new discipline. He investigated a number of social issues based on the
conviction that social facts are just as objective as physical facts. In this, he
was right in one sense: if anything is a fact, in the intrinsic sense (cf. p. 142),
it is objective in the sense that it does not matter what an observer may
think about it. We return to this issue in ch. 7, p. 310.2

2 Durkheims insistence on objectivity is also due to two strategic factors, which

are recognizable from linguistics and from cognitive science: first of all, in
order to acquire respectability, a new discipline needs to look as much like
established and prestigious disciplines as possible (for the same reason that
the first cars looked like horse-drawn carriages). Physics, dealing with solid
material fact, is the most prestigious contender. Secondly, a new discipline
needs to carve out a domain of its own, with facts that belong specifically to its
own turf otherwise it will not be able to get to the stage of becoming a going
Social facts 141

Durkheim makes two points that are central also from a linguistic per-
spective. One is about the link between covert mind-internal and overt
mind-external facts; the other concerns the nature of autonomy. First, one
should use overt facts to get at covert facts about social reality. Durkhe-
ims position is positivist in resting on a belief in objective fact, but is in no
way reductionist or physicalist. He views moral, normative ties as the basic
fabric of society; but since norms as such are not directly accessible, for
methodological reasons the scientist has to look for external manifesta-
tions (such as the law, cf. Durkheim 1893: 28). In studying whether life is
felt to be (normatively) good, getting at quality-of-life directly may be
impossible, but we can study it indirectly via the prevalence of suicide
(1893: 226f) and use it to get at the less accessible issue, the felt experience
of the quality of life. Social facts thus include a subjective as well as an
objective dimension. Secondly, the kind of autonomy proposed by Durk-
heim(1898) for social facts is not absolute but partial.3 Partial autonomy
applies also to mental facts: Durkheim argues that although mental life is
grounded in cellular mechanisms, mental representations are not reduci-
ble to facts about cells.
Some of the issues that Durkheim took up have remained striking
examples of the causal power of social facts. The study of suicide (Durk-
heim 1897) is illustrative also from the point of view of the social turn of
cognitive linguistics. Suicide clearly involves subjective experience: unless
you conceive of your life as not worth living, suicide is not a relevant
option. Durkheims point was that this apparently most private and sub-
jective of all issues could be systematically related to features of the large-
scale social order. Coining the concept of anomie for a societal condition
in which established norms and patterns of behaviour are breaking down,
he suggested that this kind of social condition had direct causal impact on
the suicide rate. What manifests itself as a subjective sense of having
reached the end of the road is therefore linked up with causal factors in

3 Although socially entrenched representations (such as religious beliefs) are

anchored in facts about the way in which societies are built up, they are not
reducible to statistical facts about social structure. I think Durkheims point is
as valid today as it was then (cf. Harder 1999 for an analysis of partial auton-
omy in relation to language). The reason it has become submerged in the dis-
cussion is that the hedge partial tends to be forgotten when the identity of a
given field of inquiry has to be asserted. The need to put a scientifically respect-
able sociology on the map made Durkheims position appear more hard-line
objectivist than it was. Other disciplines, of course, have pursued similar strate-
gies, with or without hedges.
142 Chapter 4. The foundations of a socio-cognitive synthesis

the social domain. In modern times, the issue of suicide in indigenous

communities that are losing their cultural identity has made this relation-
ship topical again, cf. Chandler et al. (2003).
Durkheims views on the relation between subjective and objective
facts are determined by his basic assumptions about the nature of science,
which included an insistence on quasi-physical status for social laws. This
strategic choice represents one extreme in the social science landscape
that CL is moving into, where the other extreme is radical social construc-
tionism. These two otherwise antithetical views of social science agree on
one thing: you cannot trust subjective mental representations. Although
social and subjective facts are linked, the nature of the link therefore
remains unresolved.
The distinction between subjective and objective is often confused
with another important distinction, which also concerns the issue of solid
vs. intangible facts: the distinction between intrinsic properties and
observer-relative properties (cf. Searle 1995). In the case of physical
objects, the two distinctions coincide: the moon has certain (objective)
properties which do not depend on me and other (subjective) properties
which it only gets because I observe it (for instance, triggering a romantic
state of mind in me). But in the case of facts about human beings, they do
not coincide: being miserable is a subjective property, but it is an intrinsic
property of the person who is in that state you can be miserable without
requiring the presence of an observer.
One mistake is easy to make in applying the distinction: it may appear
that by saying a property is intrinsic I am claiming that it is necessarily
inviolate to observer interference and that is not the case. Although
being miserable is intrinsic to me at a given point, the feeling may very
well be strongly influenced by the presence of observers and their stance
to me (whether they alleviate the feeling or make it worse!). The point is
that both a mental fact such as being miserable and a social fact such as
the state of welfare institutions in Denmark are real constituents of the
world, and therefore they have real, intrinsic properties, which are distinct
from the properties that depend on what observers think or do.
Observer-relative properties include those expressed by human cate-
gorization. For physical objects, once again, it is easy to tell categorizations
from intrinsic properties: they can be categorized in many ways (in princi-
ple, infinitely many), without implications for their intrinsic nature (thus
Plutos intrinsic properties are unaffected by the fact that it has been
demoted from its status as a planet).
Mental and social facts can also be categorized in different ways but
in the case of social and mental facts, any actual act of categorization is
Social facts 143

likely to impinge on the object categorized. If a religious person catego-

rizes his mental state as sinful, or if a politicians conduct is categorized as
reprehensible, it may have consequences for the mental state and for the
politicians social status and career. Because mental and social facts are
liable to be influenced by the way we categorize them, we need to be extra
careful in making the distinction. The alternative would be to give up the
distinction between the object and the categorization. But if we do that,
we can no longer claim that categorization influences the object or
rather, the claim would become circular. Only if I can maintain an analytic
distinction between social object X (e. g., the social importance of reli-
gion) and my own categorization of social object X (what I think of the
importance of religion) does it make sense to claim that one influences
the other.
This is a crucial point for understanding the relation between solidity
and variability in social facts, and thus for the whole point of the book. At
any given time, there is a social reality out there with certain intrinsic
properties. As part of that social reality there is a multiplicity of observer
positions, all with their idiosyncratic features. These are just as real as
other aspects of reality but they may (1) change in ways that are difficult
to predict and (2) cause changes in the rest of social reality in similarly
unpredictable ways.
These two points are important for understanding exactly where radi-
cal social constructionism goes wrong. First, the fact that the observers
stance to social reality is part of and causally relevant to social reality does
not mean that it is the same thing as the whole of social reality. Secondly,
changeability is different from unreality: in these parts, the weather may
change very quickly but the fact that it may start pouring down in a
minute does not mean that the sunny spell is an illusion.
The radical social constructionist mistake is to say that because there
are several legitimate perspectives on an object, you can say anything you
like about the object. This is wrong on two counts: first of all, you may
categorize the object wrongly even from your own perspective; secondly,
there may be properties that are relevant for you which you cannot see
from the perspective you have adopted.
In the context of this book, the distinction between intrinsic and
observer-relative properties can be applied to the different approaches to
language and mind. Linguists can freely choose to adopt a particular
observer position, a vantage point from which to view linguistic facts. That
vantage point will determine what they see when they observe language,
just as your vantage point will determine whether you see the tail or the
trunk of an elephant. An essential part of the synthesis I propose is to take
144 Chapter 4. The foundations of a socio-cognitive synthesis

on board the moderate social constructionist insight that part of the total
reality of a social condition depends on the way people view it; that posi-
tion will be elaborated in the rest of the chapter and more fully in ch. 7.
What more radical social constructionists overlook is that since you can-
not see everything from a single perspective, it follows that there exist
things which happen to be out of sight. As long as you stay down in the
undergrowth, everything is observer relative and partial, but there is one
thing which is not subject to observer relativity, namely everything (cf.
Fink 1988: 152); the fact that no one can see it from her own perspective
does not prove that it does not exist.
On the basis of this understanding, we can now be more precise about
the relation between the three central positions in the discussion about
language and meaning: objectivism, cognitivism and social construction-
ism. They are best understood in terms of different observer positions.
Inevitably, however, each of them is imbued with some spillover from the
observer position that colours its views on intrinsic properties. They have
two shared and one distinguishing attitude to cognition, objective reality
and social process. Each regards one domain as basic and the two others
as either intangible or derivative. Objectivism takes observer-independ-
ent physical reality as its focal object and regards social practice and men-
tal representations as flaky and unreliable as sources of distortion. Cog-
nitivism focuses on individual mental representations and regards
objectivity as a mirage and social processes as generated by cognition.
Social constructionism regards social activity as basic and takes it to be
determinative of both cognitive representations and (what is mistakenly
taken to be) objective facts.
Each thus shares a minus with one of the others. Thus cognitivism
agrees with social constructionism in rejecting objectivity.4 Objectivism
agrees with social constructionism in seeing mental representations as
epiphenomenal. Objectivism and cognitivism, finally, both see social proc-
esses as derivative in relation to the nature of language and meaning.
To the extent these views reflect the stance that is adopted, they are
neither right nor wrong, just a matter of choice. But it follows from the
priority of intrinsic properties as asserted above that the matter does not
end there. There are inherent relations between physical, cognitive and
social facts that an adequate description has to capture, whatever the pre-

4 Cf. Janicki (2008), who makes a point of the rejection of essentialism in CL

and (rightly!) gets a raised editorial eyebrow in the introduction of the volume
for leaving meanings undefined (=floating), cf. Kristiansen and Dirven (2008:
Social facts 145

ferred vantage point is. Therefore a chosen observer position may be more
or less adequate, and may miss out on more or less of the intrinsic nature
of the object. In order to address that issue, one must have a picture of
what the whole rounded object of description is like. This is why it is nec-
essary to try to establish an overall view in order to correct warring half-
truths without inevitably making complementary mistakes, and that is my
excuse for offering what may appear to be a somewhat overextended
framework. The baseline position for that framework is the following:
I assume (cf. also Harder 1996), along with most of contemporary sci-
ence, that there is a hierarchical relation between different types of facts,
such that some types of fact presuppose others, and not vice versa.
Physical facts are ontologically basic. If there were no physical particles in
fields of force, there would not be anything else either (cf. also Searle
1995). But particles in fields of force enter into more complex levels of
organization. When higher levels are superimposed upon simpler levels,
new properties emerge which cannot be described in terms of objects at
the simpler levels. That is why reductionism is wrong (cf. Kppe 1990).
Attempts to describe language with reference to only physical objects
would miss almost everything of what there is to say about language,
because language only arises at higher levels of organization. Physics is
generally taken to be the most successful scientific discipline, and quan-
tum mechanics the most successful theory. Part of the reason is that it
deals with the most basic constituents of the universe and leaves the hair-
ier facts to others.
A major step upwards in the hierarchy is when physical objects enter
into the complex set of relationships that characterize the biological
world. Being a biological entity confers properties that cannot be pre-
dicted from physical components alone. These have to do with the ongo-
ing process of life. Biological properties are part of the physical world and
biological entities are also physical entities, but the difference is that bio-
logical properties exist only as long as they are alive when an organism
dies, physical mechanisms are again alone in the arena.
Within the biological world there are simpler and more complex life
forms. For social animals including ants and human beings, the group is
the most crucial part of the environment and thus adaptation to the group
is crucial for survival. The two examples illustrate that there is no one-to-
one link between sophisticated social organization and sophisticated cog-
nitive powers; adaptation to group life does not necessarily require men-
tal mediation.
Evolutionary biology is usually viewed as the second-most successful
science. Its domain is defined in terms of the mechanisms that shape the
146 Chapter 4. The foundations of a socio-cognitive synthesis

progression of the flow of life from first beginning to present-day human

At still higher levels of hierarchical organization we find the sociocog-
nitive universe in which human language operates. This is where it is less
easy to nominate universally acclaimed success stories and where this
book belongs. In accordance with the general logic of this levels-based
ontology, biological facts presuppose physical facts, and social facts pre-
suppose biological facts. But when we get to the level of human societies
things get less easy to disentangle. The next section shows why this is.
In suggesting what hard facts need to be integrated into the theory, I
begin with the level of evolutionary biology. The causal pattern I am get-
ting at here works in a time scale that antedates and is inaccessible to
individual human cognition, and forms the presupposed background for
phenomena that work in a more humanly accessible time scale.

3. Niche construction

The significance of niche construction for understanding the causal power

of social facts can perhaps best be appreciated in relation to the basic old
chestnut: the nature-nurture discussion. Is the source of explanation in the
nature of the individual, or in impact from the environment? Social con-
structionism and Chomskyan innatism are sophisticated modern develop-
ments of the nurture and nature-oriented positions (respectively). The
issue goes back much further than Darwin, but evolutionary biology is the
generally recognized context of the scientific approach to the question.
Genetic factors, including mutations, are causal agents inside the organ-
ism, while adaptive change is a mechanism anchored in the environment.
Some of the intricacies of the argument in concrete biological cases have
been made accessible to the public by Stephen J. Gould (e. g., The Pandas
Thumb, Gould 1980).
Niche construction comes in as the most radical example of why the
question nature or nurture? sometimes has no meaningful answer.
Strictly speaking, it never has a fully adequate answer, because the influ-
ence of the environment is always to some extent mediated by the genetic
properties of the organism. When niche construction is at work, however,
any attempt to answer the question will be wrong.
The concept of niche is a familiar metaphor in biology. It denotes an
environment that offers a specific set of affordances (cf. Gibson 1979)
that certain organisms are adapted to. For example, cacti have developed
the ability to conserve water in a dry environment, which gives them a fit-
Niche construction 147

ness edge in deserts and the desert therefore constitutes their niche. In
the standard case, a niche is thus a condition provided by the environ-
ment, into which species may find their way. The two are matched if the
niches have the right affordances for the species. Part of the dynamics of
evolution consists in mutations that create beings with coping skills that
fit into hitherto unoccupied niches as when life conquered dry land by
means of hard egg shells.
The simplest, classic relationship is one in which it is just a question of
competitive survival of organisms in a given environment (including the
social environment). This the scenario presupposed by the brute survival
of the fittest social Darwinism of Herbert Spencer. However, evolution-
ary thinking about individual and social processes has passed through suc-
cessive stages of sophistication since then. Sociobiology (cf. Wilson 1975)
has shown how altruism could drive evolution by enabling co-operative
behaviour that increases overall fitness. Also, relations between niches
and organisms may change. Cultural traits have been shown to impact the
gene pool (Laland et al. 1999, citing Feldman and Cavalli-Sforza 1989), as
when domestication of cattle led to the spread of lactose tolerance in the
population (since among cattle-keepers those who could digest milk had
better chances of survival).
Niche construction is a recent extra level in this increasingly subtle
understanding of evolutionary mechanisms (cf. Odling-Smee, Laland and
Feldman 2003).The central idea is that there are processes of linked feed-
back going both ways: the species changes its environment in certain ways;
the environmental changes influence the selection processes in the spe-
cies. As a result of that altered selection, in the next round the population
has a different impact on its environment than before, thus shaping the
environment to which it is concurrently adapting. Deacon (1997) illus-
trated this mechanism with the case of the beaver. Over the years, beavers
have adapted to living in an environment with ponds (cf. their flat tails, for
instance). Simultaneously, however, they have developed a behavioural
repertoire that includes building dams thus creating ponds. Beavers are
thus adapted to environments with dams built by beavers. This is where
the nature-nurture dichotomy becomes entirely inapplicable. If you ask
whether the explanation for the beavers lifestyle is in the beavers nature
or in influence from the environment, there is no sensible answer.
Niche construction has been suggested by Deacon (1997, 2003a, 2003b)
as a scenario for the process that gave rise to language, and the discussion
in the following is based on his views. The key idea is that human beings
have been selected so as to do well in a niche that includes language pro-
duced by human beings.
148 Chapter 4. The foundations of a socio-cognitive synthesis

The niche that is thus constructed is the speech community. Introduc-

ing the speech community as a factor in evolution changes the founda-
tions of the discussion between innatist and usage-based linguistics.5 In
the evolutionary scenario suggested by Deacon, the process that created
language started out with a change in mutual understanding in the com-
munity. Among pre-linguistic factors, Tomasello (2008) suggests that
intentional gesturing as found in non-human apes was probably a crucial
factor in the evolutionary process: while ape vocalisations are directly tied
to emotional states, gestures depend on monitoring the attentional states
of the addressee and thus create a relation between individuals that goes
beyond direct causal impact. For the development of human communica-
tion, the decisive extra step is when enhanced communicative ability starts
to provide a selective advantage for a hominid community. Once there
was enough of a selective advantage to give enhanced communicative
powers a critical edge, all changes which contributed to this critical edge
were positively selected for and they in turn changed the nature of the
community to which subsequent adaptation would take place6 a case of
100 % nature and 100 % nurture, as expressed by Deacon.
In addition to the general logic of niche construction, Deacon (cf. Dea-
con 2003a) makes a suggestion that turns the question of genetic change

5 One of the innateness issues is the question of continuity or discontinuity

between human and animal communication. The Chomskyan innateness
assumption goes naturally with an assumption of total discontinuity. In Chom-
skys classic view language arose by a genetic mutation serving as the magic
bullet that provided the African Eve with the language faculty, in the same
way that the first proto-reptile one day laid a hard-shelled egg. Any assump-
tion that there was a gradual evolution towards language, in contrast, would
suggest that language had functional properties built into it, thus providing a
rationale for explaining structure by means of function. Chomskys dislike of
evolutionary biology fits naturally into this picture.
6 This kind of dynamics is unique in being able to account for why a series of
physically unrelated changes could be part of the developmental pattern that
ensued: greater cortical control over the organs of speech, lowered larynx,
bipedalism etc. These changes are not credible as disparate constituents of a
genetic magic bullet, but they could all plausibly arise as standard mutational
accidents along the way. They would thus spread across the population because
they enabled better communication, driving an evolution towards ever greater
linguistic powers. If this is correct, we see again how the niche hypothesis
supersedes discussions about nature vs. nurture. It no longer makes sense to
ask how much is due to the genes inside the communicative hominid, and how
much is due to the external environment including the fellow communicators.
Niche construction 149

for language on its head. Genetic predispositions come and go in animal

populations, and loss may be as significant as gain. A crucial part of the
dynamics is the role of selection pressure in promoting the maintenance
of certain properties. When that selection pressure decreases for one rea-
son or another, tolerance of deviation increases. As a result, the genes may
take a random walk, as the phrase is, and may ultimately get lost alto-
gether. One type of process where decreasing selection pressure is central
is domestication.
Domestication means that an animal, for instance the dog, no longer
has to struggle with an environment that is uncompromisingly given;
instead it lives in an environment that has been modified so as to offer
better affordances. There may be food sources available that do not come
naturally, and there may be protection against predation. This means that
specimens that would previously have been incapable of fending for
themselves may now live to produce offspring. The ensuing proliferation
of genetic diversity, partly due to random walk and partly to subsequent
controlled breeding, is exemplified by comparing the morphological
diversity of domestic dogs with that of wolves, their genetic ancestors.
Pekinese or Chihuahua dogs would not stand much of a chance out there
in the wolf niche, but it does not matter in a human environment where
survival and reproductive success is determined by other factors.
The process whereby the speech community became a more and more
salient environmental factor for emerging humans has some of the fea-
tures of this process, even though there was no master species involved.
Rather than surviving by direct interaction with the physical environment,
hominid individuals (we may suppose) increasingly survived and pros-
pered by the success of the social group in creating conditions for group
survival. Good communicators, we may suppose, contributing to the suc-
cess of the group, were valued members of the community. Domestication
would then take place to the extent that the niche itself (= the speech
community) grew in importance as a factor for reproductive success. The
brute environment would then gradually lessen in causal relevance, while
the causal weight of the human environment would gradually increase in
weight. With Deacons phrase, humans may have become a self-domesti-
cated species. We created domestic humans along with domestic dogs.
Deacon suggests that as chances of reproduction became increasingly
attuned to success within the speech community, the previous system of
vocal responses that were attuned directly to input from the physical envi-
ronment began to diminish in importance. This caused a gradual loss of
calls reflexively triggered by environmental stimuli. Compared to the
inventory of e. g. monkey species, humans have very few environmentally
150 Chapter 4. The foundations of a socio-cognitive synthesis

triggered calls because they became less important as the information

state of the group became more and more important.
When the selection pressure to stay attuned to direct impulses from
the environment decreases, the scope increases for behaviour that is not
genetically determined. Deacons (2003a) illustration example is the
white-backed munia which has been bred in captivity for 300 years, losing
its genetically determined song in the process. While the genetic song is
necessary to attract mates in the wild, in the cage it was the breeder who
selected the mate, so the song genes took a random walk and instead of
having a genetically determined song, the captive munias have become
capable of mimicking sounds that they hear. When the fixed program dis-
appeared from the genes, environmental factors could get an enhanced
There are several morals of this story. One is that rather than a specific
innate genetic addition, one key development may have been that vocali-
zation as a behavioural field became increasingly free from genetic deter-
mination.7 A niche in the form of a community emerged to which indi-
viduals gradually adapted and that involved a gradually smaller degree
of direct dependence on bodily attunement to specific environmental
The rise of symbolic meaning which is the centrepiece of Deacons
theory, cf. The Symbolic Species (Deacon 1997), inherently involves a
redefinition of meaning from a direct relationship with the environment
to a relationship mediated by meaning in the mind. Symbolic meaning is
cognitively different from iconic and indexical meaning in that it requires
a higher-order organization in the brain. One way of describing the transi-
tion to symbolic meaning is that the cognitive higher-order level takes
over and marginalizes the more direct links between signs and sources of
meaning in the environment. Symbolic meanings can persist and be
evoked in the absence of direct links with the environment; adapting to a
world of symbolic meanings therefore goes naturally with loss of adapta-
tion to direct situational-pragmatic stimuli.
This element of detachment is essential in understanding the special
nature of human language. At the beginning of The Symbolic Species,
Deacon points to the absence of primitive languages as a significant fact
about the story of human languages. No other species have anything like
a language in the human sense. Symbolic meaning, and its neurocognitive

7 Other factors would obviously also be required, or munias would have

invented human speech as well.
Niche construction 151

underpinning in cortical development, is an evolutionarily distinctive

development in human beings.8
This general evolutionary logic is compatible with the key evolution-
ary step proposed by Tomasello. In theory, the capacity for joint attention
could be a genetic achievement based purely on random mutation and
thus an alternative to the Chomskyan language gene as a kind of magic
bullet. However, if we look at the pointing ability, there is experimental
evidence suggesting that the dynamics of domestication may have a role
to play. Pointing requires a degree of attunedness to other minds: in order
to read the extended finger as a message for you, you have to respond to
it as a message from someone with an intention to inform you. The exper-
imental setup involves the task of figuring out where a cache of food is
hidden. The subjects are brought into a room with hiding places, and the
experimenter then points to the one where the food is. In the simple setup,
chimpanzees fail to learn to respond to the pointing, while dogs can do it,
cf. Hare et al. (2002), Byrne (2003). In contrast, wolves (the ancestor spe-
cies) cannot. By analogy, the critical factor appears to be in the domestic
niche rather than in any specific evolutionary addition to the hominid
branch of primates.9

8 One can postulate a fairly direct link between this feature and the rise of syn-
tactic organization, cf. Deacon (2003b), Harder (1996:26869): while calls are
indexically linked to a feature in the situation, symbolic meaning does not
have a direct link with the situation and therefore it needs to be attached to
the situation in order to get a situational function. That job can be done by
indexical (= grounding) expressions which thus form a syntactic link with
purely symbolic (= conceptual) meanings. More generally, the existence of
ungrounded meanings is a prerequisite for the possibility of evoking and com-
bining meanings freely: call-type meaning by definition depends on a situa-
tional stimulus, and any combination would therefore signal a combination of
situational stimuli not a complex combination where the meaning is more
than the sum of its parts. The step from holophrastic languages, as found in
animal communication, to syntactically complex languages, is part of the same
development that enabled the speaker to operate with meanings that are not
directly tied to immediate stimuli.
9 Responding to pointing only makes evolutionary sense if there are individuals
out there who altruistically point out food and other items of interest which
by and large does not occur in the wild (cf. Tomasello 2008: 41). Learning to do
that thus appears to require losing some of the genetically fixed response
mechanisms that may be a matter of life and death in the wild, but are less
critical than the ability to respond to socially provided information in a domes-
tic environment rather than wait for a mutational fluke. However, the point-
152 Chapter 4. The foundations of a socio-cognitive synthesis

The central point for the purposes of this chapter is that niche con-
struction provides a scenario for how a social unit (the speech community)
can be a causal factor in its own right.
Its causal powers become apparent not by isolating it from the indi-
vidual, but by showing the causal interplay between the two factors. Point-
ing at the properties of an individual organism as the source of explana-
tion is insufficient: we can only understand the properties of the individual
by seeing her as a member of a particular kind of social community. Point-
ing at the community as the sole source of explanation is equally insuffi-
cient: there would be no community without individuals with the proper-
ties necessary for becoming members. The evolutionary spiral of niche
construction uniquely demonstrates how the mutual dependence can
arise in a panchronic perspective, avoiding the usual gratuitous assump-
tion of influence-in-both-directions.
This further reinforces the pivotal role of joint attention as described
by Tomasello. Joint attention is the crucial evolutionary novelty because it
brings a special intersubjective relation into being: a joint mental state by
definition cannot be reduced to the sum of mental states in individual
minds. When mum and dad pay joint attention to their child, this complex
mental state is not reducible to the sum of two simple mental states, mum
paying attention and dad paying attention. It follows that we is not reduc-
ible to the sum of two separate individuals, so a new and more powerful
distinction has come into being between individual and collective. Toma-
sello (2008: 6) shows how this provides an underpinning for Searles con-
cept collective intentionality, the building block of social reality. Joint
attention is thus on the one hand a constituent of the individual mind, an
ability that uniquely distinguishes human beings and at the same time it
is a building block of a new kind of community.
In both Tomasellos and Searles perspective this new and irreducible
we is viewed basically as a constituent of the individual mind: Tomasel-
los introductory chapter is entitled focus on infrastructure, and the dia-
gram in Searle (1995: 26) shows two individuals, each with a we inside the
head. Tomasello uses the term ratchet effect (cf. Tomasello, Kruger and
Ratner 1993) that is the result of joint attention allowing individuals to
learn what others know (instead of each having to learn from personal
experience). In understanding the nature of social reality, however, the
complementary external perspective must be emphasized equally. Niche

ing skill of domesticated animals still falls short of the triadic pointing, where
it is the pointers state of mind that is in focus rather than the food.
Niche construction 153

construction involves an external ratchet effect, since reality is rebuilt

based on the output of previous generations, not by each generation start-
ing from the same baseline. (We know the external ratchet effect under
the name of history).
Thanks to joint attention and the ratchet effect, a human baby is born
into an environment that already contains a human we, which exerts
selection pressure on the individual mind to take its place in the joint
world. Possessing a mind-internal we is an entry condition otherwise
you cannot join the we that is already out there. The ability to become a
well-adapted participant in the existing community, a member of the niche
we, depends on the ability to take part in the emerging speech communi-
ty.10 We may assume that there has been considerable selection pressure
to be good at language acquisition.
In the context of CL, this scenario has consequences for how the role
of bodily grounding must be understood in a human as opposed to a pre-
linguistic community. The speech community presupposes a weakening or
loss of the simplest and most direct form of bodily grounding, the form in
which a vocal response is fully pre-wired and tied to specific environmen-
tal situations. Accounts of embodiment, as we move along the evolution-
ary trajectory of niche construction, will therefore have to be increasingly
mediated by those factors that intervene between situational stimuli and
bodily responses, i. e. the features of the community to which human
beings have become attuned. The causal role of culture as a population-
level phenomenon, and the integration of the biological and the cultural
level of description that cultural evolution gives rise to has been empha-
sized by Richerson and Boyd (2005), who also point out the resistance
against this pattern of thinking that is due to the entrenched division of

10 For the hominid individuals who are born into an emerging speech commu-
nity, the properties of the speech community are a factor outside themselves,
to which they may be more or less well adapted, i. e. their individual bodies
may be more or less adequate in relation to the crucial features of the speech
community. At this initial point, the speech community is not in their minds
but their (embodied) minds are inside the speech community, understood as
the niche in which they will live or die. Language, by implication, is also out-
side the individuals mind at that point the tool of tools (in Deweys terms)
or supreme artefact (cf. A. Clark 1997) that members have to adapt to in
order to thrive, analogous to the ponds and dams that are a feature of the
beavers niche rather than of the individual beaver. The community, saliently
including the language, is part of the human niche, just as dams are part of the
beavers niche.
154 Chapter 4. The foundations of a socio-cognitive synthesis

labour according to which some academic communities believe only in

social determination while others think only in terms of hard facts. The
need to integrate the two sides will have implications for the understand-
ing of grounding, to which we will return in chapters 5 and 7.

4. Individuals, collectives and the invisible hand

With the speech community, we move from the biological to the sociocul-
tural time scale. Only its most basic features can be accounted for in terms
of biological change. Most of the features that play a role in actual speech
communities are due to changes that come on top of biological features
without necessarily demanding further genetic adaptation, i. e. changes in
community practices. Such changes also causally affect individuals in the
community, but they are epigenetic (cf. Sinha as discussed on p. 70) rather
than genetic: they bring about sociocultural adaptation that is superim-
posed on the effects of genetic dispositions. The full consequences of the
biological evolution described in the previous section therefore did not
arise immediately in the primeval speech community. They are the results
of a new round of selection-adaptation dynamics triggered by cultural
learning (cf. the discussion of Tomasello p. 76).11
This brings about a new kind of hard social facts. In the sociocultural
time scale, the mind acquires a special causal role (whereas biological evo-
lution was at work before there were human minds around). At the socio-
cultural level, two types of processes are at work: one where the social
level and the individual mind are mutually attuned, and one type where
they may diverge radically. Both types are ubiquitous and at work all the
time. The main point of this section is to give an account of the causal role
of the second, invisible-hand type of process, where the individual mind
has limited access to what is going on. But it is easier, especially from a CL
point of view, to profile its specific properties if we start with the kind
where there is a mutual relation between the aggregate level and the indi-
vidual mind.

11 Relations between individuals and collectives as we know them from the mod-
ern human scene are the product of both kinds of processes, intertwined in
ways that are hard to disentangle. In this section, however, the attention is on
the kind of causal pattern that involves cultural learning. The only relevance of
genes is to enable the relevant kind of epigenetic adaptation to community
Individuals, collectives and the invisible hand 155

This kind of collective phenomenon is what Clark (1996) calls joint

action.12 Joint action (1996: 3) is carried out by an ensemble of people
acting in co-ordination with each other, such as waltzing, paddling a
canoe, playing a piano duet or making love. Two properties are central:
such activities cannot be reduced to a combination of individual acts
thus they transcend the flock of birds case that can be decomposed into
the responses of the individual birds: birds do not play flock the way
musicians play a symphony. Secondly, joint actions are composed of two
sets of intentional actions: individual acts of participation (each inten-
tionally playing their separate scores) and the collective act that they par-
ticipate in, i. e. playing a symphony (which is the collective intention of
the orchestra).
This form of complexity is only possible by virtue of the existence of
common ground (Clark 1996: 92110). Joint projects can only succeed if
all participants share an awareness of the whole, without which they could
not know what their own role is in relation to the others. This may appear
to be a tall order, especially if you take the traditional investigative path
that begins with the isolated individual and wonder how to get from there
to collective, joint action. But with the help of the evolutionary account,
we can see why that approach would be inherently wrong. Just as increas-
ingly communicative hominids were born into a community with proper-
ties to which they had gradually become adapted, in the same way it takes
a process of sociocultural adaptation to become a member of an orches-
tra. Speech communities and symphony orchestras are there before the
individual, and have arisen out of a long process of niche construction
which works also in the sociocultural time scale.13 Via this path, the devel-
oping individual grows into different types of subcommunity activities,
from street games to football matches to orchestras to democratic elec-

12 Joint action in Clarks sense fits naturally into a theory based on joint atten-
tion, cf. Tomasello (2008: 83).
13 If you start out with an isolated individual, you cannot get him into the
charmed circle of joint activity and that is why normal human development
cannot start with such an isolated individual, cf. the discussion of Sinha p. 72.
Participation in joint activity grows out of the evolutionary gift of attunement
to the community via joint attention, plus the sociocultural gift of cultural
14 As always, the activity dimension is basic, rather than the product dimension.
The channel of initiation is the experience of taking part, not the completed
artefact (the whole football game or the whole symphony). Participant activ-
156 Chapter 4. The foundations of a socio-cognitive synthesis

Such collective processes are consciously accessible to the individual,

and would not work unless they were: the conceptual representations and
the collective social action go together in participant awareness: if I am a
second violin in a symphony orchestra, in the normal case I represent
myself mentally as playing second violin. In a state of seamless adaptation
there would be complete identity between conceptual models in the indi-
vidual mind and conceptual models as part of the sociocultural niche.
However, we now come to the second type of relation between indi-
vidual and collective facts, where there is no such continuity.15 In the
macro-social domain, this dimension of social reality is captured by the
concept of the invisible hand, so named by Adam Smith in the founda-
tional text of economics, The Wealth of Nations (1776). His canonical
example constitutes the central argument for economic liberalism. At the
individual level we have the entrepreneur who looks out primarily for
himself; he does not see his activity as contributing to a shared aggregate
goal that will either come off or fail. Nevertheless, there is an aggregate
outcome at the level of the economy or market. Just as in the case of
joint activities, the aggregate outcome has properties that go beyond the
sum of its parts; and in this case, the two levels may even have apparently
conflicting properties.
The logic of liberalism illustrates this. In spite of the fact that each
entrepreneur is trying to get his hands on as much of the loot as he can, the
aggregate effect may be that everybody is better off. In the assumed case,
looking out for your own economic interests entails driving down costs
and being as efficient as possible in using resources. If you are better than
your competitors at this, you make more money for yourself. This means
that you become richer than the others but it also means that the others,

ity, which is the individual contribution to joint activity, is learnt through the
experience of being part of the whole. The state of co-ordination is not some-
thing that is guaranteed in advance; it always depends on individual, inten-
tional performance. The basic entity could not very well be a collective inten-
tion, because that would be a rather mysterious entity. But if the pre-existing
social community is understood as a causal factor, individual intentions can be
analysed in a way that makes explicit the inherent link between them and their
presupposed social embedding (Clark 1996: 61).
15 Clark (1996: 22) distinguishes between anticipated and emergent products
of individual activities. Only when the collective outcome is anticipated do we
have joint action; however, individual activities also give rise to aggregate
effects that are not anticipated. This distinction matches the distinction
between the levels of the visible and invisible hand.
Individuals, collectives and the invisible hand 157

if they want to stay in business, have to emulate your efficiency and thus
get better at what they are doing. For your customers, this means that they
get products at lower prices because goods that are produced with fewer
resources can be sold cheaper. Within this foundational fairy tale of eco-
nomic liberalism, what at the individual level is driven by pure self-inter-
est, at the aggregate level works towards the common good16.
The invisible hand does not always bring such beneficial consequences
for everybody. Another foundational story in liberalism is that of the
tragedy of the commons (cf. Hardin 1968). Traditional villages, in addition
to individual plots of land, also had a shared, communal area, where the
poor and landless could graze a cow or two. This was an idyllic feature of
traditional village life which disappeared in the transition from feudal to
capitalist societies, widely bemourned in the literature (cf. Goldsmith, The
Deserted Village). But a discrepancy between causal patterns at the indi-
vidual and the collective level tended to undermine the idyll. The poor
individual peasant would gain something by putting an extra cow on the
commons as long as it survived and gave just a trickle of milk. But above
a certain number of cows, each new cow brings about a decrease in the
aggregate yield because as the cows are nearing starvation level, the
total milk production declines. At the end of the curve, all cows drop dead,
but even if it is in no ones interest to go on till that point, the logic of the
invisible hand tended towards a situation where there were too many
cows for the common good, with overgrazing and low yield as a result.
These examples illustrate that invisible hand effects are hard facts
just as evolutionary causality in general brings about hard facts. In order
to understand the position of the individual in a social context, it is not
sufficient to address what is intuitively accessible (visible) from the indi-
vidual perspective because part of the individuals situation is due to
factors that operate over the individuals head, at the aggregate level.
Invisible hand effects were brought to linguists attention by Keller
1990.17 The individual who chooses to use a certain word rather than

16 Although the invisible hand is older than Darwinian evolutionary theory, in

public consciousness it became more or less subsumed by the social Darwinist
notion of the survival of the fittest (coined by Herbert Spencer). My use of
the term captures the distinction between individual-level facts and popula-
tion-level facts within the social sphere, but does not assume either a social
Darwinist or an economic liberalist ontology.
17 As pointed out by Andersen (2006: 63), the need to distinguish between differ-
ent levels of analysis in understanding change was familiar from Coseriu
158 Chapter 4. The foundations of a socio-cognitive synthesis

another does not intend to change the language. But if enough people
make the same choice such as stop using the German word englisch in
the sense angelic, as happened in the 19th century the word used in that
sense may drop so much in frequency that it is not learnt by the next gen-
eration. At a time when the English became more and more visibly domi-
nant at world level, the possibility of being misunderstood as meaning
English when you really meant angelic was on the increase, and so we
may assume that people increasingly chose the alternative engelhaft, thus
unwittingly driving englisch = angelic out of business, as suggested by
Evolutionary causality as manifested by the invisible hand is a key fac-
tor in shaping sociocultural reality. To sum up: a social cognitive linguistics
will have to cover three different objects of description: processes in the
individual mind, collective processes based on individual awareness (the
visible hand, by analogy with the hand of the conductor of the symphony
orchestra), and collective processes operating over the individuals head
(the invisible hand). There are connections between them; but they are
not the same thing.
Although the main point here is the invisible hand as a feature of selec-
tion-adaptation dynamics, its role in relation to cultural phenomena
including language differs crucially from its role in relation to biology and
economics, precisely because of the foundational role of the visible hand
of joint action. Visible-hand phenomena constitute a crucial middle
ground between the individual and the whole population, and the mutual
coordination between the individual and aggregate levels means that soci-
ocultural facts differ from biological facts on precisely this point. In the
area of language change, this is relevant for the disagreement of principle
between Croft (2000; 2006) and Andersen (2006) with respect to the role
of intentions and cultural factors: meaningfulness operates at collective
levels, too, as manifested by the creativity associated with reanalysis and
adoption of new forms. The Danish word for car (bil) arose as a result of
a newspaper competition in 1902. This feat the newspaper has not man-
aged to repeat in spite of several attempts but it shows the potential role
of visible and intentional process at aggregate level.
This means that the analogy between biological and sociocultural evo-
lution is not as a complete as suggested by Croft: in sociocultural contexts
neither the selection criteria nor actual processes are necessarily invisible

([1958] 1974); the new element in Kellers analysis was the clear separation of
the two sets of causal patterns.
Individuals, collectives and the invisible hand 159

and meaningless at the aggregate level. People do not only innovate inten-
tionally but also to some extent propagate innovations intentionally for
instance because they want to do their bit (like returning empty bottles
and batteries), or because they feel they are part of a joint activity (such
as a democratic election) in a Clark-like fashion (Clark 1996). But the
invisible hand is always part of the game at the aggregate level. No one
votes for a hung parliament that cannot govern, and yet it may be the
aggregate result of individual intentional votes. And those who think that
putting forward an attractive conceptualization of the future is enough to
change the world are not likely to have much experience of the process.
Putting a real social construction on its feet is hard work and depends
crucially on finding a path through the Darwinian jungle of macro-social
selection pressures.
The rise of an aggregate level at which the invisible hand can operate
is a historical process. In economics, the process involves the rise of a mar-
ket, where the value of an object can be detached from individual embod-
ied experience. In the field of language, it involves the process whereby
the meanings of linguistic expressions first became detached from con-
crete situated acts. And because of the external ratchet effect, the proc-
ess goes on. With a familiar analogy, the rise of paper money and of writ-
ten communication makes possible a radically increased detachment of
value and meaning from situated joint experience. A paper note and a
text, with the now familiar dual nature, both represent and constitute
social statuses (as value and meaning) and because of their detachment,
they can both survive beyond the moment and also undergo changes due
to the invisible hand. Money may depreciate and words change their
meaning for reasons that have nothing to do with the embodied experi-
ence of local participants.
As a consequence of this form of complexity, we also get types of com-
munication that operate at levels distant from immediate personal experi-
ence (non-basic types of communication in Clarkss term, 1996: 9). Politi-
cal and corporate communication have different ecologies from everyday
conversation. The dimension of detachment from personal experience is
also essential for understanding the social dimension of the sociocognitive
world. It has consequences both for the kind of experience that is availa-
ble in the community and the kinds of talk. Capturing such indirect rela-
tions is the key to understanding the interplay between embodied experi-
ence from below and aggregate social pressures from above; this subject is
taken up in detail in ch. 7.
Socially oriented analysis of language tends to start at the aggregate
end, with written texts for collective audiences, while cognitively oriented
160 Chapter 4. The foundations of a socio-cognitive synthesis

approaches start with meanings that reflect personal embodied experi-

ence. The detachment of aggregate-level communication from basic
human interaction may again raise the spectre of the suspicious ontologi-
cal status that tends to cling to social entities. In conclusion it is worth
stressing, therefore, that invisible-hand processes are extremely respecta-
ble from the point of view of the science police.18 In a free market, prices
are the result of forces of supply and demand that are not intuitively
accessible but their reality is not in doubt. Similar mechanisms are at
work in relation to language, even if they do not manifest themselves in
properties that are as measurable as prices and wages.

5. Functional relations19

The foundation for social cognitive linguistics that I am proposing is at the

same time a functional-cognitive synthesis. Although this reflects my own
take on the subject, I believe that it is an inherently necessary dimension,
rather than an extra issue that is brought in: it is functional relations that
make the cognitive and the social dimension cohere. This relationship is
implicit in the account of the invisible hand given above, but depends on
an understanding of what function is.20

18 Ever since Adam Smith, they have been the centrepiece of economic theory
a discipline that has become increasingly formal and mathematical in the last
generation, to the exclusion of older-style economists that are disparagingly
called verbal. This is not to say that economic theory has a grasp of reality
that is necessarily on a par with its formal precision (a problem also found in
other disciplines). But it means that there is nothing inherently mushy in this
type of intuitively inaccessible facts.
19 I have discussed the issue of functional causation in greater detail elsewhere
(see Harder 1996 and 2003).
20 The point cannot be understood based on the commonsense meaning of the
word function, which can mean a range of different things including utterance
functions such as greeting, discourse functions like repair, grammatical
functions like subject, and social functions like legitimation. Moreover, the
term function is often associated exclusively with intended function (cf. Croft
2000), and functional explanations based on functions as intentions at the
aggregate level would be nave functionalism, rightly criticized by Croft
(1995): like infantile megalomania, nave functionalism assumes that the world
is tailored directly to the speakers needs and intentions. Formalist critiques of
functionalism generally take this version as their adversary, and too often it is
not clear what the alternative is, as pointed out by Tomlin (1997: 165).
Functional relations 161

Function as used in this book is an aggregate-level phenomenon, and

part of the causal structure of evolutionary dynamics (as described in ch.
2, p. 89). As such, it does not depend on individual intentions (cf. Allen,
Bekoff and Lauder 1998, Wright 1973). Functional relations obtain as a
property of the two-stage mechanism of replication and selection, causing
certain elements to persist as part of the whole evolving system. Without
functions, there would be no evolution, only change.
Function is a type of effect (in biology, typically of an organ). It differs
from knee-jerk effects in that it is defined in terms of what this effect
causes in the next round. When you look for the function of a biological
organ (wings, for instance), you look for effects (flying, for instance) that
contribute to the survival and persistence of the animal. Function thus
involves two cause-effect sequences: wings cause powers of flight and
powers of flight cause enhanced reproduction and selection rates. Func-
tion in that sense is part of the way the world works; there is no subjectiv-
ity involved.21
Three features must be present in order for functions to exist.
(1) Functional relations can only exist in entities that survive by repro-
duction (and are thus exposed to selection pressures). These include
organisms as well as human artefacts: a product that goes out of produc-
tion and an animal that becomes extinct do so because the causal chain
does not bring about sufficient (re)production. Inorganic objects, in con-
trast, may persist as individuals, like Mount Everest, but there is no selec-
tion process going on, hence no complex, feedback-driven causal cycle.
Elements of inorganic matter such as molecules therefore cannot have
(2) Functions also involve a part-whole element: effects have to favour
the survival of a larger whole in order to qualify as functions. Functional
relations are therefore a feature of complex wholes with independently
describable parts. Wings have functions for birds, and artefacts have func-
tions for human beings.
(3) Functional relations are parts of a dynamic panchronic system, not
a synchronic moment in time. They are thus dynamic in two ways: first,

21 In the case of wings, an experimental test would be to release pinioned birds

in the wild, along with birds with intact powers of flight. If the pinioned birds
never managed to breed while their otherwise identical conspecifics did, that
would be evidence that powers of flight are indeed the function of bird wings.
For some species, there might be no difference and that would tell us that
powers of flight had become non-functional for them. Endemic island species
(for instance water rails) occasionally lose powers of flight.
162 Chapter 4. The foundations of a socio-cognitive synthesis

they depend on causal cycles between parts and wholes and across gen-
erations. But they are dynamic also in the sense that the contribution-to-
persistence of an element may change over time. In the case of wings, a
challenge to functional explanation was how an animal ever made it from
no wing to a fully functional wing. On the powers-of-flight hypothesis, half
a wing would make no difference to survival chances (in which case wings
could arise only by evolutionary, mutational chance, rather than func-
tional selection). A suggested answer was that pre-wings might contribute
to heat regulation and only accidentally develop so as to provide powers
of flight in which case their function might subsequently shift.
The same applies to functions in the social domain, such as the buying
power of dollar bills and the meanings of words. The function of dollar
bills is to be exchanged for goods and services, and the function of words
is to convey their meaning. But in the economy, the value of money
changes with shifting market conditions, and in language, the meanings of
words change with shifting conditions of use (as corn came to mean
maize in the US). Functional relations are thus always defined by actual
replication and selection mechanisms, and have variational features (just
as wings may to variable degrees have a heat regulation function on the
side). Functions as hard facts about linguistics expressions depend on the
same panchronic scenario as the function of organs. As pointed out by
Croft (2000: 2324), the selection/replication scenario is at work also when
the language stays the same. A word that remains unchanged does so
because the aggregate invisible hand effect of individual choices repro-
duces it in identical form (just as a money can retain its value only when
invisible-hand effects cause the prices to stay where they are a far from
trivial outcome).
I began this section by saying that functions understood as individual
intended functions are not an adequate foundation for understanding
functions as part of the way the world works. There is a double dissocia-
tion between intended utterance functions and conventional functions of
words: the same situational purpose might be achieved in other ways, and
the same linguistic expressions might also be used to achieve a different
situational purpose. But there is an indirect relationship as well as an
equilibrium condition under which they match up. Individual utterances
are acts of reproduction, which feed into the aggregate process. When the
speaker means something that does not get conveyed, this influences
selection in the next round. A speaker who chooses the word engelhaft
because he intends to avoid the potential misunderstanding of englisch,
contributes his mite to the aggregate processes (like the individual entre-
preneur who brings down her production costs).
Mind in society: causal patterns and the individual 163

When conventional meaning is in a stable equilibrium, the two levels

coincide in a satisficing proportion of cases. The word hello has the actual
as well as conventional effect of conveying a greeting, thus bringing about
an everyday interactive relation of mutual recognition between speaker
and hearer, and that effect is simultaneously the cause that makes people
use it again on other occasions. But functional relations are only part of
the picture; selection implies the existence of other effects than those
which are functional. They are complex, changeable over time and many-
stranded, rather than simple, eternal and unambiguous. 22
Functional explanations have generally been unpopular in the social
sciences in the last generation, for two diametrically opposed reasons. On
the one hand, the spread of social constructionism has discouraged
attempts to constrain the free scope of random social forces; and on the
other side of the spectrum, more direct causation has appeared to be more
scientific, cf. Hull (1988: 354f). But as pointed out by Hull, the fact that
equilibrium-oriented functions and feedback mechanisms are always only
part of what is going on does not prevent them from having a crucial role
to play. If there are functional mechanisms that keep certain practices in
satisficing condition while leaving others to an uncertain fate (such as sci-
ences, systems of government, monetary systems, and linguistic meanings),
we need to know what they are. The fact that function cannot explain eve-
rything is a property it shares with all other interesting theoretical con-

6. Mind in society: causal patterns and the individual

We can now return to the question that social constructionist approaches,

as discussed in the introduction to this chapter, could not help us with.
How can mental and social facts be captured in an integrated picture?

22 Functional relations are also emergent in the sense that they emerge from
processes of interaction with the environment and may change with those
processes, as in the case of wings. That does not mean they are emergent in the
sense that they are never really there (cf. Dahl 2004: 2737 on the flip-flop use
of the word emergence to indicate either a generous or a reductionist ontol-
ogy). Once emerged, functional relations are, by the definition above, causally
relevant features of the objects involved. If functional relations break down (if
something happens so that wings no longer enable the animal to fly), the ani-
mal may not be able to persist in its present form.
164 Chapter 4. The foundations of a socio-cognitive synthesis

In previous sections, we have established a skeletal picture of the pre-

cise causal roles that hard (= observer-independent) social facts play for
cognitive processes. The account has two advantages compared with the
ubiquitous social determinism that was found in approaches described in
chapter 3. First of all, the causal impact is not direct knee-jerk determina-
tion that reduces individual cognition to a dependent variable. Secondly,
with more precise mechanisms to refer to, we can hope to mark off the
place of mental, cognitive facts with more precision as well.
Starting from the mental end, it is practical to distinguish between
three aspects of mental facts: internal, representational and (social-)con-
The internal aspect of mental facts is their basic status as a distinctive
part of the real world. Reality includes facts about minds alongside facts
about atoms, bacilli and coal mines. The cognitive revolution unleashed a
flurry of activities whose aim was to chart this scientifically unexplored
domain. Subjectivity and conceptualization are examples of the things
you find when you examine the mind, just as particles and spin are things
you find when you examine atoms. As part of this movement, CL contrib-
uted to the charting of this territory, cf. the overview in ch. 1. The internal
aspect, however, can be pursued quite happily without raising the issue of
relations with the rest of the world23, all the more so since fundamental
aspects of the link continue to defy scientific investigation, cf. the slash
between mind and brain as discussed p. 33. This internal aspect was the
undisputed focus of classic CL.
The representational aspect has been the focus in the philosophical
tradition, casting the mind in the role of a map of the world (at the risk of
forgetting that the mind was also part of the world). From the point of
view of causal relations including selective fitness, maps are useful in nav-
igating the world when you move beyond the territory where the auto-
matic pilot works. Since Popper (1972) it has been taken for granted that
representational powers have survival value because they enable the indi-
vidual to perform mental simulations before putting themselves at risk.
Mental creatures do not have to step in front of a truck to see what hap-

23 This does not entail that the internal aspect of mental phenomena is outside
the world of cause and effect. Brains must have the causal power to generate
what goes on in the mind, and mental phenomena have causal relevance (e. g.)
for learning abilities, cf. p. 204 n. 18. However, I emphasize the role of causality
specifically in relation to social phenomena, because it is necessary to establish
the legitimacy of functional relations as something other than an observer-
relative phenomenon.
Mind in society: causal patterns and the individual 165

pens; they can imagine the result, thus letting their hypotheses die instead
of themselves. For the purposes of CL, the representational aspect has not
been of focal interest, even if only the most dedicated internalists have
denied its existence (cf. Johnson and Lakoff 2002: 24950).
The crucial dimensions in a social cognitive linguistics is the third
aspect of mental properties, their constitutive role in sociocultural facts.
The word constitutive reflects the classic distinction in speech acts theory
(cf. Searle 1969) between regulative rules (which apply to activities that
already exist) and constitutive rules, which apply to forms of activity which
could not exist without them (without rules for contract bridge, you could
not play three no trumps).
Joint attention is the basic linchpin also in this context. It is the signifi-
cance that joint attention gives to other peoples mental states that creates
the fundamental link between the individual mind and the community.
Without the capacity for joint attention, meaning would exist in the indi-
vidual mind only. By bringing the human subject into a triadic relation
with the attended object on the one hand and a fellow subject on the
other, joint attention gives rise to a mental state that at the same time has
a constitutive role for social relationships with fellow subjects. There is a
relationship of increasing complexity from level one to level three: before
there is a mind, there cannot be mental representation; and before you
can mentally represent a national election, you cannot make it part of the
joint world you live in (as pointed out by Searle 1995).
At this third stage, the buildup from the mental end thus makes contact
with the account of social facts as described in terms of joint action in
Clarkes (1996) sense, cf. p. 155 above. Joint action requires that mental
awareness and social activity go together and link up the individual and
the collective level. The link can be described in terms of Searles concept
of status function. Just as an utterance can have the status of being a
promise or a christening, which is not derivable from the purely linguistic
properties, so a man can have the property of being President, which is not
derivable from his personal (embodied) qualities; the same goes for the
function of playing second violin in a symphony orchestra. To give some-
thing a status (function) is to give it a role in the construction of social
reality (the title of Searle 1995).24 This collective assignment gives the
object to which status is assigned some causal properties in the social com-

24 This is related to Clarks account of participant intentions, and Clark (1996:

61) cites Searle as a source for his account; Tomasello 2008, similarly, cites both
Clark and Searle.
166 Chapter 4. The foundations of a socio-cognitive synthesis

munity which have a magical flavour about them, such as the fact that
pieces of paper can buy you a house.
This collectively maintained extra value is crucial for understanding
linguistic meaning in a socio-cognitive picture. There is an important dis-
tinction between saying that meaning would not exist if it did not exist in
the individual mind (which is true) and saying that we can account for all
of the properties of meaning by looking at its properties in an individual
mind (which is false). The classic CL view is that meaning belongs in the
individual mind, and this view is also maintained by some of the pioneers
of the social turn. As discussed above, cf. also ch. 6, pp. 29394 Croft main-
tains that meanings can only be in an individual mind, and words under-
stood as social lineages therefore strictly speaking cannot have mean-
ings. But this reasoning depends on assuming that when we go outside the
individual mind, we have to disregard what is in the individual mind and
that is not the case. The speech community is built out of people with
minds, yet collectives do not have the same properties as the individuals
that make them up. If we see the existence of meaning at collective level
in those terms, the fact that meaning cannot exist without individual minds
is no argument against collective meaning.
I have previously spoken of traffic jams of meaning in cases when a
collection of individual minds formed larger configurations. The consti-
tutive role of mental constructs as part of social structure reflects a
higher level of organization, while the basic idea is the same. A car race
is more structured than a traffic jam, but is similar in that if you take
away all individual cars, nothing remains, while at the same time the
properties of the race do not boil down to properties of individual cars
(or drivers). Similarly, the formation of democratic governments is based
on the ability of citizens to conceive of a government as having execu-
tive power and members of parliament as responsible for legislation, but
uses these mental abilities as input to the formation of a larger structure
that goes beyond what is present in each individual considered on her
own. A democratic government therefore involves a large body of mean-
ing-in-society that is grounded outside as well as inside the individual.
The individual agent is always situated in a field of forces that includes
meaning assigned through social processes and meaning as a constituent
of her own mind.
This applies to language, too. The power of hello to create a greeting is
an observer-independent (and in that sense perfectly objective) fact about
life in the community, just as the fact that red light means stop! or the fact
that the current president of the US is Barack Obama. These just belong
to that subcategory of observer-independent facts that could not exist
Mind in society: causal patterns and the individual 167

without the existence of individuals with powers of mental representa-

tions. If the observer does not agree, he is simply wrong.
One objection to this objective, collective status is the role of variation.
It may be true that there is meaning out there among other people, and
not just in my own mind but all these other people are individuals, too.
And we might bring social constructionist approaches and their floating
signifiers from ch. 3 back in the discussion. Individual minds may assign
different statuses to hello, Obama and red light. Where is this hors-texte,
this collective domain outside individuals engaged in discursive processes,
that I claim contains objective meanings shared by all?
I go into detail with meaning in ch 5 and variation in ch 6. In this chap-
ter the point is to set up the general format for the answer to the question.
In this format, functional relations play a crucial role as links between
individual minds and the social process. The apparent difficulty in getting
out of the individual mind is a question of perspective. If you decide to
look at individuals one at a time, there is nothing left when you have been
through all individuals. If you adopt a panchronic, evolutionary perspec-
tive, however, it is uncontroversial that there are events that occur outside
the individual mind that affect meaning. It is most obvious in a historical
perspective. Changes of meaning due to functional pressures (to take the
englisch case again) do not occur inside the mind of an individual; and it
would make historical linguistics in general very cumbersome if we could
not talk about historical change in the meaning of an expression. But as
discussed above, synchronic states are part of the same panchronic whole:
Obama is President because that is the way he functions in society, and
hello is a greeting for the same reason. The nature of functional relations
as described in the previous section will allow us to do so without ideal-
izing in a Chomskyan way
Examples of status functions such as the US president and the mean-
ing of hello are cases of what I called equilibrium positions (those where
things match up) because there is a widespread consensus that is built
into the way the world works. Mental representations of the form we
take paper money to be a means of payment are an essential part of the
cause for money being what it is, just as taking someone to be president
is essential in order for him to be president. On the other hand, it is also
simply a true description of the way the world actually works: paper
money actually is a means of payment, just as a corkscrew is a means of
opening wine bottles. And presidents are what they are because when
they issue orders, lower executives have to carry them out. This follows
from the causal loop that is characteristic of a functional relationship.
Handing money over the counter has the effect of getting goods in return.
168 Chapter 4. The foundations of a socio-cognitive synthesis

And the effect, in turn, is what causes the persistence of money-making

behaviour as part of our lives (and thus of money). If money loses its
causal powers we will not continue to mentally represent it as money
and that in turn would cause the downfall of the constitutive role as car-
rier of the status assignment. The same goes for meanings of words for
pre-industrial agricultural practices as opposed to words for continuing
practices such as saying hello. This is the rationale for assuming that there
can be equilibrium states, but these have to be maintained by the prac-
tices that sustain them by functional feedback of the kind that can also
keep prices stable (or not).
However, equilibrium cases are not the only kind. Functional relations
have already been described as partial and many-stranded; they operate
over time in a complex causal chain. They are always only a small part of
what is going on. They arise out of the flow of causes and effects only
when stabilizing patterns emerge from the background of random and
criss-crossing events, and that background (unlike the functional rela-
tions) is always there. If the meteor theory is true, dinosaur extinction just
happened and had nothing to do with functional relations and if the
climate becomes warmer, heat conservation equipment such as heavy fur
that used to be functional may turn into an encumbrance and upset the
equilibrium between organism and environment.
Part of the aim of this book, therefore, is to emphasize the need for a
differentiated account of causal factors, focusing on the limited but inter-
esting role of functional relations. Knee-jerk causality, entropy, mutual
see-saw impact, chaos-theoretic trigger effects, etc., are also part of the
way the world works, but the historically variable form of systematicity
that characterizes functional relations is crucial in understanding why nei-
ther Platonic foundationalism nor indeterminacy has the final word.
Since mental status assignment depends on functional relations, there
will be divergences between social causality and mental representations
whenever there is a less than perfect equilibrium. Functional relations
could not be a factor in evolutionary development if they were unchange-
able. I think Searle underestimates this causal element in the foundational
underpinning of status functions; and this is where a jagged edge may
open between mental representations and social reality25. I am first going
to take his example, involving social structure, and then return to word

25 This is a development of the discussion in Harder (2003).

Mind in society: causal patterns and the individual 169

Searle (1995) discusses Maos statement that power grows out of the
barrel of a gun, and argues instead that real power comes from status
assignments; those with the guns are usually pretty low down in the social
hierarchy. Searle mentions the collapse of communism in Eastern Europe
as a case in point: people withdrew the status-assignments that supported
communist rule, and so it collapsed. While that account may plausible as a
description of a simple cause-effect chain in 198990, I believe the event
can only be fully understood if we think in terms of the more complex
type of causality that characterizes functional relationships. The role of
physical force, including the use of guns, must be understood in terms of
its function in the social reality of political power.
A textbook feature of being in power is to have monopoly on the use
of force. That is why the presence of warlords and militias in a nation is a
sign of governance problems. In well-functioning states, only the police
and the army use armed force, and only when licensed by the proper
authorities. One of the cases where use of arms is licensed is when indi-
viduals challenge the laws of the land, including the regulations main-
tained by officials of the state which in communist countries included
border patrols, secret police and lots more besides. Once citizens stop
abiding by the regulations imposed by such authorities, power such as that
wielded by communist government behind the iron curtain loses causal
efficacy. In other words, they cease to be power holders: people who have
no influence over what other people do are not in power.
This is where the men with guns may have a crucial functional role and
at the same time be very low in the pecking order. The withdrawal of the
status function of the communist party as the power in the land occurred
partly because the authorities, more specifically Gorbachev, refrained
from activating the men with the guns as opposed to the Chinese author-
ities who at roughly the same time stuck to Maos doctrines in response to
the demonstrations at Tienanmen Square (and who are still in power). A
well-functioning government may in theory persist forever without using
armed force on its citizens, if the collective assignment of status function
is maintained by other causal factors, such as popular support. Maos dic-
tum, therefore, shows the cloven hoof of the dictator. But there has to be
something that keeps the world working causally so as to maintain the
status. Collective function assignments understood as mental content,
which is what in Searles version keeps the system in place, do not operate
on their own, any more than other mental representations. Like other
functional properties, they depend on the way the world works. For the
same reason, green light can only mean go if cars by and large stop on
red. Status assignments depend on actual events; in other words, they are
170 Chapter 4. The foundations of a socio-cognitive synthesis

simultaneously beliefs as well as status assignments. If the belief is under-

lined, so is the status assignment.
Let us now look at situations where the equilibrium is not in place. An
example is the situation that obtained after the 2008 elections in Zimba-
bwe. If we assume with most observers that President Mugabe lost but
refused to abide by the results, what was the social reality immediately
after the election? We may suppose that some citizens represented Mor-
gan Tsvangirai as the new winner and others represented Mugabe (the
hero of the victory over colonial rule) as the only true leader. Which of
them were right? Well, both and neither. At that point, there was no oper-
ational social equilibrium that could be used as a criterion of what the
situation was really like.
When causal structure is uncertain, representations can float around
freely. This also means that there is more scope for simple causal chains
from representation to reality, i. e. direct social determination of the kind
that social constructionists are after. If all the people of Zimbabwe had
represented the situation as a transfer of power, there would have been a
transfer of power. As it turned out, Mugabe continued to wield the actual
power (via his armed men). Some citizens of Zimbabwe might withdraw
the collective status assignment whereby Mugabes government was the
legitimate power in the land, but it would be unwise to forget about the
causal structure of Zimbabwean social reality. Even now that Tsvangirai
has been sworn in as Prime Minister, it remains to be seen how that will
affect the causal structure of government in the country.
That kind of uncertainty has causal consequences both for actual social
processes and for individual participants. The years after the collapse of
communist rule in Russia are an example; we know from Durkheim that
anomie is a stressful and depressing condition, and we may speculate that
the support for Putins exercise of power, also when it involves blatant
violations of civil liberties, which is hard to understand for citizens in well-
entrenched democratic societies, has something to do with this. Lack of
stable status-assignments and lack of a reliable sense of how the world
works go hand in hand in undermining the sense of having a secure foot-
hold in the world.
The situation where representations lack causal anchoring has been
seen as an anomaly. This is fortunately true for national governments. But
if we look at most other domains of social reality, there is less clearcut
causal structure and more freedom for competing representations. Rela-
tions with people you meet in the street have little inherent causal machin-
ery to support them, and accidental mutual (mis)representations may have
free play. Processes of marginalization, which constitute core areas for the
Mind in society: causal patterns and the individual 171

hermeneutics of suspicion, work basically in the same way (although there

is a formal tip of the iceberg when it comes to legal issues such as gay
rights). In general, the kind of well-structured equilibrium cases where
mental representations and causal structure match up neatly, as in the case
of Barack Obama being President of the US, are surrounded on all sides
by vast amounts of much messier relations between mental representa-
tions and social processes. We return to the issue in ch. 8, p. 417 below.
This also goes for word meanings. There are processes going on in dif-
ferent groups all the time that influence understanding of meanings which
generate multifarious variants that never get into a dictionary. My wife
has a few rarely used expressions in her repertoire that are conventionally
understood as carrying a positive load but I know that when she uses
them I have to tread carefully. I doubt that anybody else is party to that
nuance. The point is that variation does not refute the existence of a col-
lective level in fact it occurs on the collective level. The issue of collec-
tive vs. individual meaning is orthogonal to the distinction between con-
sensual and variational dimensions of meaning. Functional relations select
from the variational spectrum and maintain lineages across all relevant
communities and subcommunities; knowing about communal lexicons is
part of participant abilities (cf. Clark 1998).
But if things vary all over the place is it not a more correct description
to focus on the variation between what people mean by words than to set
up a collective level that has to be qualified so heavily that almost nothing
is left of the objective, collective meaning anyway?
The reason why that conclusion would be badly wrong is that linguistic
meanings exist ONLY in collectives (cf. also Wittgensteins argument
against private language). Meanings may form all sorts of complex con-
figurations of variants, but the variants exist as part of a collective-level
pattern. Even in the smallest possible collective group, a word only has a
meaning if it can be used wrongly. If a word meaning does not exist in a
sociocultural niche (however fleeting and emergent), the word does not
exist at all. The individual is simply the wrong unit for understanding lin-
guistic meaning.
This also illustrates another important point. I have emphasized the
importance of causality in order to get behind the mirror cabinet of repre-
sentations and interpretations in radical social constructionism; but this is
not to say that brute causal power overrules all mental factors. The whole
normative, rational, interpretive dimension of mental experience cannot
be reduced either to neuron activation or to social pressures, and remains
a constitutive dimension also after it is realized that mental phenomena
are bound up with causality in various ways.
172 Chapter 4. The foundations of a socio-cognitive synthesis

This account of the constitutive, but non-exhaustive role of mental

representations in social reality has a methodological consequence:
because mental representations have a constitutive role in the wider
socio-functional picture of reality, we can get at part of social reality by
virtue of (and only by virtue of) the fact that we have participant access to
the function-assignments. This means that we can talk about language,
although not all of language, merely based on our experience as partici-
pants. That is convenient, because it means that we can hope to get at
some of the crucial causal powers of status-imbued objects (including lin-
guistic expressions) by means of intuition-based analysis. Intuitions should
be supplemented with other sources of knowledge, but without it the
other sources would be meaningless to us. Even sociology can go part of
the way by intuitive methods based on participant awareness of what is
going on; for instance, Goffmans work is based on participant experience
in a wide variety of communities.
The visible hand in joint action allows participant awareness to extend
upwards beyond the sphere of immediate participation (as in democratic
elections). Emergent effects may also enter into participant awareness and
trigger collective action by the visible hand, so that the joint activity is
adapted to take previously unrealized effects into account (when a win
was upgraded from two to three points in soccer football, it counteracted
the trend towards increasingly effective and boring defensive play that had
been an unwanted emergent effect of the previous system). The borderline
is permeable, but we cannot hope to catch up mentally with all invisible-
hand-type effects: the flow of social activity is a broad and muddy river.
When we talk intuitively, based on participant status in shared practices,
about large-scale processes like education, family relations, restaurant vis-
its and the monetary system, the basis for our observations is only the
mental printout of current experience of participation, and is necessarily
incomplete. Empirical, quantitative, mind-external facts play an essential
role. A full description thus depends on both sides: participant understand-
ing, without which social experience would be meaningless, and aggregate
social dynamics, which is beyond the reach of individual intuition.

7. Summary: the socio-cognitive world

I have now introduced the basic furniture of the socio-cognitive universe

as I see it. The new home ground contrasts in some ways with assumptions
in classic CL, without necessarily contradicting them. In this final section
I will attempt to sum up the essential features, as a background for filling
Summary: the socio-cognitive world 173

out the picture in the rest of the book. The discussion above has implica-
tions for language as an object of social cognitive description which can be
summed up in three main features. In a nutshell, the new universe
(1) works by two interdependent forms of causality, one emerging from the
individual, another operating at the aggregate level of the social community;

(2) involves a complex co-existence between mental, intentional understand-

ing and factors inaccessible to consciousness; and

(3) contains three interdependent objects of description instead of the classic,

unitary object, language as an integrated part of the mind

The co-existing features manifest themselves in three dimensions of the

same complex social object of description: language as a flow of activity, as
a feature of the sociocultural niche, and in individual minds.
The flow is the most basic dimension, without which neither of the
other two could exist. Life is flow (to coin a phrase); when the flow stops,
life stops. And in social animals, such as human beings, social interaction is
a major part of the flow.
The niche arises as a result of evolutionary dynamics. In the case of
human beings, the dynamics operates in two time scales: biological and
sociocultural. The niche consists basically in a specific set of causal regu-
larities that impinge upon the life of those beings whose ecological envi-
ronment it constitutes. In Denmark, niche properties include facts such as
the high prestige of ability to play football and the fact that ja means yes.
Niche properties are not stable, but continue to evolve (in both the evolu-
tionary and the sociocultural time scale). Niche-constructional processes
are at work whenever there is a two-way causal pattern of adaptation and
modification between individual and environment going on. In order to
understand the niche, neither nature nor nurture will serve as sole explan-
atory factors on their own; thus objectivism and social determinism are
equally inadequate.
The properties of participants emerge as part of the same process that
brings about the niche. Participant abilities are shaped by adaptation:
selection works against individuals who are bad at coping with the
demands of the niche and favours those who are well adapted. The dynam-
ics of niche construction means that there is constant interplay between
participant ability and the affordances and constraints of the niche.
Two sets of causal patterns are at work in the interplay between par-
ticipants and niche: the individual can only perform in ways enabled by
her own causal wiring, but at the same time that wiring is shaped by the
174 Chapter 4. The foundations of a socio-cognitive synthesis

way the world (= niche) works. Both the world and the individual have a
role in shaping what is possible, or necessary. In order to understand what
happens, we need both sets of factors.
Finally, the relationship between mind and mechanics has become
rather more complex than in Descartes world. Both internally and in the
community, shades of grey have taken over from black and white. Central
to the position of this book, however, is that the two sides always need to
be considered together: mental processes enter into the dynamics of the
social process only by means of their role as constituents of overt, causally
efficacious behaviour, not by inner mentalization alone; and on the other
hand, social causation is mediated by the mental and normative dimen-
sion of the human niche.
Specifically with respect to language, the three-way ontology comes
out as follows:
First of all, flow corresponds to actual usage, Saussurean parole. The
basic phenomenon is the dynamic activity, as stressed by Clark (1996). The
products you can tape and transcribe and put into corpora are important
as evidence, but they are not in themselves the fundamental object of
Secondly, I adopt the term competency for the properties that enable
human subjects to handle the demands of the niche with linguistic com-
petency as the central example. The term is chosen to indicate that I think
this should be seen as the (sociocognitive and meaning-imbued) successor
concept to a Chomskyan competence.26 My hope is that competency will
sound less exalted and ideal than competence. As an educational achieve-
ment (cp. Smith 1996), competencies are defined in relation to a certain
required standard, which is reminiscent of the evolutionary satisficing
perspective that my account is based on. Competency, therefore, simply
refers to the ability to cope in relation to environmentally defined selec-
tion pressure, such as the ability to take part in the linguistic communica-
tion that goes on in the speech community.
Basic to individual competency, in something like the seat of honour
that Chomsky wanted to reserve for innate formal grammar, we find the
cognitive capacity and emotional motivation for joint attention (Toma-
sello), intersubjectivity (Sinha, Zlatev) and joint action (Clark). The expe-

26 I am inspired by Langacker, who moved the other way and adopted depend-
ence instead of the more usual term dependency, cf. Langacker (1987: 306) for
the special twist he gave to the term.
Summary: the socio-cognitive world 175

rience of jointness is the primal scene for a developmental trajectory that

includes altruism, culture, society and underlies the role of communica-
tion as the basis of normative Habermasian features of communicative
life as opposed to the features of social life in Hobbes Leviathan.
It follows from the basic status of the flow of activity that if usage did
not exist, there would be no niche (speech community) and no compe-
tency either. On the other hand, when the whole going concern of linguis-
tic communication is up and running, the kind of flow that occurs is
dependent both on the competencies of the participants and the
affordances of the niche.
Thirdly, language as a property of the niche may be regarded as the
successor concept to Saussurean langue and from now on, by langue I
mean language in the niche. There is a fairly direct continuity with the
Durkheim-inspired views of Saussure, in that langue is a quasi-objective
feature of the way the world works in the speech community (as discussed
in relation to Croft (pp. 294295). [Tak] means thank you in Denmark,
yes in Poland, and branch in Holland and therefore the sound
sequence has different causal powers when spoken in Gdansk, Copenha-
gen and Amsterdam; the world simply works differently in the three com-
munities. The relations between linguistic signs have thus been relocated
from the abstract immanence of structuralism to the causal structure of
the social world. The reason for setting up the three dimensions as sepa-
rate but interdependent objects of description is that we cannot assume
that properties observed in one of them carry over to the other two. There
is variation across the speech community that is not matched by variation
inside the mind of every individual; there are actualized instances of use
that do not reflect what the individuals involved actually want to say; and
there are individual competencies with features that are at odds with
socially recognized status functions for the relevant linguistic items.
Therefore we need to be able to distinguish between the three different
objects of description.
The three objects of description represent different perspectives on
the same complex object, and the fact that you may get different descrip-
tions from the three different perspectives does not prove that one
description is as good as the other, only that all perspectives are needed to
capture the joint object of description adequately.
This proposal may appear very general and vague. I am therefore going
to offer some illustrations of why this expanded picture makes a differ-
Within the expanded picture, we can place classic CL as the approach
that takes its point of departure in the individual participant and describes
176 Chapter 4. The foundations of a socio-cognitive synthesis

properties of language from that vantage point. Like Newtonian mechan-

ics in physics, this type of description may be both fully adequate and
exhaustive under the appropriate set of boundary conditions. This is the
case when there is full co-ordination and matching competencies between
participants, and if we limit our universe to that type of joint activities
which operate in basic settings, cf. Clark (1996), we can act as if language
is complete in the individual (with a phrase borrowed from Paul Hopper
in discussion 2008).
There is obviously nothing wrong with taking one thing at a time. That
is the way science works. The question is what happens to the relation
between the different parts of the whole story. There is no automatic
mechanism in science for ensuring optimal collaboration between them,
and there is always a risk that different approaches see their own subfield
as the whole story. The issue in practice does not generally take the form
of explicit reductionism, but manifests itself in something that is almost
inevitable in the academic world, i. e. a prioritization of those aspects of
the whole issue that you are working on. As an illustration, I am going to
return to the neural theory of language (cf. above p. 32). We may take the
following as the credo:
Ultimately, the meaning of any utterance is its effect on the (physical and emo-
tional) well-being of the person saying or hearing it. Of course, human societies
have developed a vast array of intermediate structures (i. e. culture) that affect
meaning, but everything that matters is represented in each individuals brain
mechanisms. (Feldman 2006: 283).
The role of culture is recognized, but that part of it which is not inside the
individual does not ultimately matter. What is inside the individual can in
turn be captured at one privileged level of description, by what is explic-
itly called a process of reduction:
The systematic reduction to connectionist, and ultimately neural, realiza-
tion .is the key to grasping ECG [= embodied construction grammar], and,
for me, is central to understanding language and thought. (Feldman 2006: 294)
As already emphasized (p. 34), a research program focusing on the indi-
vidual described in neural terms is an essential dimension of the full story;
but if everybody followed the path outlined in this passage without turn-
ing back, the pursuit of bodily grounding would end up reducing culture
to cognition and cognition to neuronal activity.
The ontology that this chapter has drawn up is designed to provide a
map of the whole project that constitutes a social cognitive linguistics,
enabling us to place such more specific approaches as parts of the whole,
essential but not exhaustive. I would like to emphasize that this is not a
Summary: the socio-cognitive world 177

fence-sitting or ecumenical League of Nations position that leaves all

conflicts unresolved. It asserts uncompromisingly that purely mental
accounts are incomplete, and that the causal dimension needs to be
included; and further that all one-way causal explanations, if they are
taken as the whole truth, are simplistic and misleading, whether they take
neuronal activation, cognitive mappings or social pressures as their point
of departure.
Colour categories may serve as an illustration of how a gradual widen-
ing of the perspective (to which this book tries to contribute) can be seen
as one continuous path of progress.
Classic universalism assumed that objects in the world were the same
for all and were directly reflected in the categories of language. For col-
ours this implies that there is a fixed inventory of colours out there in the
world, which correspond to the inventory of names for colours. Structural-
ism discovered (as discussed p. 18 above) that the colour terms differed
between languages, so that it is necessary to distinguish between the col-
our spectrum and the different distinctions imposed by the net of lan-
guage. But structuralists over-interpreted this finding by suggesting that
it was the structural imposition of distinctions that was solely responsible
for creating the categories. Berlin and Kay (1969), followed by Rosch ((=
Heider 1972), then showed that the variation was not random, and there
were features of the different patterns that could be ascribed to the exist-
ence of an underlying universal cognitive map of the territory, which was
shared across languages. And finally Davidoff and associates (cf. Davidoff,
Davies and Roberson 1999) showed that you cannot account for all prop-
erties of colour categorization by referring to the bodily (perceptual)
grounding that underlies all languages. Rather, the underlying cognitive
abilities are subjected to shaping forces that derive from cultural prac-
tices, and linguistically encoded colour categories play an essential role in
that shaping process, because it is the linguistic categories, not the basic
neurocognitive wiring, that mediate the sociocultural practice.
Seeing this path as a continuing unresolved disagreement about the
same issue would miss the point, cf. also Kay and Regier (2009). To put it
bluntly: Aristotle was right that the difference between green (unripe)
cherries and red (ripe) cherries is out there and the same for all languages
and species; Hjelmslev was right that you cannot use such differences to
predict distinctions made in languages, and thus colour terms arise as a
result of the imposition of linguistic structure; Rosch was right that colour
space is not randomly divided by languages, but reflects innate perceptual
biases in the human species; and Davidoff is right that colour space is ulti-
mately defined by language and culture.
178 Chapter 4. The foundations of a socio-cognitive synthesis

In the context of this book, the general moral is that this necessary
complexity is built into the panchronic, tripartite ontology proposed
above. Its purpose is to enable the individual cognitive linguist, who nec-
essarily pursues her own specific project, to consider where that project
belongs in an expanded picture that includes the joint social world. Most
classic issues, like colour, can only be satisfactorily captured if you look in
more than one place: if you ask where the issue of colour ultimately
belongs, the answer is wrong no matter whether you choose mind, lan-
guage or the environment. There is no way to abbreviate the role of men-
tal, environmental, situational and linguistic factors into a single underly-
ing truth.
Functional causation plays a crucial role in linking up processes in the
individual and the community. It is not enough to say that there are causal
factors going in both directions the loose sense in which the word dia-
lectical is sometimes used. This idea of mutual influence reflects the
interdependence that is also captured by functional relations, but lacks
the precision and potential for prediction that is associated with the evo-
lutionary scenario in which functional relations belong, raising the risk of
circularity.27 Complex systems persist by causal mechanisms that tend
towards preserving a functional equilibrium; they create a development
over time which is kept within certain limits by the way the feedback
mechanisms work. There is nothing circular about selection processes in
evolutionary dynamics or in the market mechanism.
Based on the foundations that have been established above, the rest of
the book will now try to fill out parts of the new socio-cognitive picture. In
the next chapter, we look at the understanding of the dynamic dimension,
focusing on the basic dimension of flow. Chapter 6 discusses language as a
specific object of description, arguing that the recontextualization of cog-
nition from the mind to the community also provides the key to under-
standing the partial autonomy of language from individual cognition.
Chapter 7 presents a theory of the sociocognitive universe, focusing on
the relations between mind and society. And chapter 8 takes up a central
and problematic case, the multi-ethnic issue, as a core challenge for a
socio-cognitive theory.

27 Frank (2008: 244), taking the bull by the horns, speaks explicitly of a kind of
circular causality between the local and the global level. Functional systems,
however, could not survive if there was nothing further to be said than causal-
ity moving in two directions at the same time, like the force of gravity that goes
both ways between two bodies in space.
Chapter 5. Meaning and flow: the relation
between usage and competency

1. Introduction

In Part One I have given an outline of CL in the process of moving into

the social domain and tried to show that this requires a full-scale rethink-
ing of the foundations of the enterprise. In ch. 4 I introduced the main
features of the extended framework that I envisage setting the stage for
the social cognitive linguistics that I am going to present in Part Two.
One salient feature is a more prominent role for the functional dimen-
sion. As stressed by Langacker (1999: 13), CL is part of the functionalist
tradition, and I see the functional features that I propose as a spelling-out
of a dimension that has always been implicitly there. The functional
dimension the importance of the role of the item under investigation in
a wider context already permeates the whole CL framework in the form
of the element of recontextualization (cf. p. 3) that also involved a con-
text-oriented reevaluation of the nature of meaning. If the overall con-
text is presupposed in understanding language, meaning can only be
described by starting out with a larger whole that has to be taken for
granted. In CL the larger whole is understood as being cognitive in
nature but the reason why meanings need to be understood in their
larger cognitive context is a special case of the more general principle
whereby meaning needs to be understood in their larger functional con-
text the ultimate source of meaning being, with Wittgensteins term, the
form of life.
The first item on the agenda is the way in which the dynamics of usage
fits into the framework that has been proposed. The increasing focus on
usage is central in the ongoing social turn in CL that was described in Chap-
ter 2. From one point of view, reflecting the terminology where function
and actual usage is more or less the same thing, this may appear to fit
directly into a functional agenda. However, the concept of function pre-
sented above is somewhat more complex, and depends on more than a spe-
cific actual usage event: functions arise through feedback patterns that
operate across several reproductive cycles. Since the core object of descrip-
tion in CL is meaning, a key issue for an emerging social cognitive linguis-
tics is the relation between meaning and usage. This will be the first chal-
180 Chapter 5. Meaning and ow

lenge for the descriptive format presented in ch. 4, and the subject of this
The threefold object of description (flow, competency and niche) con-
trasts with the traditional CL approach, which is implicitly based on an
undifferentiated, unitary object of description, roughly describable as
language from a usage-based, cognitive point of view. This contrast gives
rise to the chief claim of this chapter: if the social turn gradually relocates
meaning entirely from the mind to the flow, it will not give an adequate
account. The ongoing development is rightly concerned to redress the tra-
ditional imbalance and give appropriate primacy to the flow of usage as
the lifeblood and raison detre of language but the issue of language in
the mind does not go away. Human language is too complex to be under-
stood solely and exhaustively as online flow.
This chapter addresses the relation between usage and competency
(the langue dimension is postponed till ch. 6). Two points are important: to
capture the pervasive primacy of the flow dimension, and to show what
(distinct and irreducible) role this flow-reorientation reserves for lan-
guage as an offline feature of the mind. Even if the mind is demoted to
secondary position, and the flow takes over the primacy, there is still an
interface that must be an explicit part of the account. Interfaces are two-
way streets: how does the competency reflect the primacy of the flow, and
how does the flow reflect features of competency (which is the condition
of membership/participant status in a speech community)?
It may appear surprising that the discussion of meaning in a usage-
based model raises the issue of competency, in view of the Chomskyan
connotations of the word. But competency as understood here is totally
severed from all Platonic ties and is strictly a property of the individual
participant in linguistic interaction. The general format of the argument
against Platonic essentialism has in fact often invoked a combination of
flow and the competency of individual participants. Conversation Analy-
sis is possibly the most uncompromising approach when it comes to insist-
ing on online actual usage as the object of description, but the essential
role of pre-existing competency is emphasized by Schegloff (1991: 152),
the key figure and Garfinkels term ethno-methodology was meant to
indicate the peoples own method for making sense.1 But what is this

1 Integrative Linguistics, another anti-system approach, cf. Harris (1998), simi-

larly stresses both the importance of the flow of practice rather than pre-exist-
ing structures, and also the essential role of the participants abilities in manag-
ing the flow of linguistic interaction. Edwards (2008, in discussion at the
Brighton conference on Language, Culture and Mind) has emphasized that
Introduction 181

individual competency, when it is detached from its traditional Platonic

First of all, the priority relations between static offline competency and
the flow of usage are the opposite of what was traditionally assumed.
Usage is basic and competency has to be adapted to the demands of
usage. This also reflects the individuals trajectory: the newcomer into the
community is confronted with a flow of activity but lacks participant
access and the success criterion is to achieve a competency that grants
that access.
Secondly, this creates a need to clarify exactly how dynamic and static
properties interface. We can no longer say that actual performance is sim-
ply a poor and error-ridden reflection of ideal competence. Without an
ideal Platonic substratum, what speakers possess must be something that
qualifies them to construct utterances, and utterance meanings, on the fly,
with all the variation that this involves.
There is one key consequence of this usage-based competency which
plays a pervasive role in the argument in this chapter: we cannot base the
account on complete utterance meanings alone. Otherwise all possible
utterance meanings would have to be available in advance, and there
would be no possibility of saying anything new (at least if you wanted to
be understood!). Language users must know how to handle individual
semantic contributions, not just whole finished messages.
This is an important point, because the usage orientation in CL is also
an orientation towards emphasizing the foundational role of actual usage
events. If the emphasis on actual usage events is not combined with a clear
recognition that the object of description also has other, less fundamental
aspects (in this case competency aspects), it involves a risk of what I call
usage fundamentalism: the assumption that only actual instances of lan-
guage in use are real and have meaning. This is a form of reductionism,
and as such different from the usage based approach. The usage based
approach (as adopted in this book) assumes that usage is the basis for
everything, but that human language use is so complex that it depends on
minds adapted to usage. Offline properties of minds thus also have to be
part of the story.
The role of units prefigured in the mind means that language use is not
a homogeneous and undifferentiated flow like water coming from a tap.
In addition to pre-existing categories that need to be imposed on the flow

also in discursive psychology, the ability to create social reality online must be
understood as dependent on participant competency.
182 Chapter 5. Meaning and ow

in order to make it decodable, there are also units that must be constructed
during the flow. This includes the basic unit of pragmatics, the utterance
corresponding (roughly) to what fills a turn-at-talk.
This inherent complexity means that language users have to operate
with a non-continuous dimension that can be captured in terms of three
sub-elements of the overall flow: input, processing and output. From the
production point of view, the speaker starts off with input in the form of
an intended message (of a kind), and generates an output in the form of
an articulated utterance. From the reception point of view (which will be
the preferred perspective below), the hearer gets a linguistic input in the
form of a flow that has to be assigned to recognizable linguistic units, and
then converted to an output in the form of a conveyed meaning or mes-
sage. (This process is what we discussed as meaning construction, in ch. 2).
This chapter explores the implications of this reversed role of compe-
tency and flow. The general direction is that in the beginning the focus is
on showing the importance of the flow features (input, directionality, and
the process aspects of meaning) and later the focus is on describing the
key properties of competency (as a necessary part of the whole picture).
Section 2 discusses the distinction between meaning as input and
meaning as output.
Section 3 explores the consequences of this process-oriented view of
competency for linguistic meaning. The main idea is that conventional
meanings have a directional dimension: just like utterance meanings,
they have a point of interest and a point of departure.
Section 4 discusses the relation between the procedural and the repre-
sentational aspect of participant competencies and argues for a clear dis-
tinction between skill and mental content (while emphasizing that the
two dimensions are in constant interplay).
Section 5 discusses what this distinction implies for meaning construc-
tion on the instructional and procedural premises proposed
Section 6 sums up the implications of the process dimension for under-
standing meaning, including the relation between conceptualization (as
flow) and concepts (as aspects of competency).
Section 7 sums up the argument.

2. Meaning as process input

The traditional understanding of meaning identifies it directly with the

message, the final product of the process of understanding that is at the
same time assumed to be identical with the speakers intention, cf. Reddy
Meaning as process input 183

(1979) on the conduit model. One of the major trends in the past genera-
tion has been a gradual movement away from this position. Instead, the
addressees role has gradually increased. Meaning works as part of a
dynamic event sequence: instead of a complete picture that is carefully
packed and then carefully unpacked, the addressee enters into a process
of (co-)constructing the meaning, based on the linguistic input.
In understanding the relation between dynamic and stable features in
cognition, a major influence in cognitive science generally is the impact of
the computer metaphor, which was at the heart of the first cognitive revo-
lution in the 1960s and 1970s. Computers and computation provided a
direct model for thinking about language in dynamic terms. A computer
program does not represent a situation in the world it consists in a series
of linked operations that specify a path from an initial state to an end-
state. The meaning of a computer program, in other words, is dynamic in
the sense that it changes the state of the system instead of merely repre-
senting it.2
This procedural dimension of computation, however, is inherently
bound up with a declarative dimension, since the procedural changes can
be exhaustively characterized in terms of the input and output states with
which they are associated. In the computational context, the procedural
approach thus does not make much difference.3 But there is no guarantee
that the operations which human addressees carry out in order to get at
the intended output are effective procedures that ensure equivalence. The
procedural dimension may therefore have more significant implications
outside a computational context.
Among the fields that were inspired by the procedural thinking that
came with the computer metaphor is psycholinguistics (cf. Johnson-Laird
1977, 1983). The program analogy (cf. above) gave rise to the phrase pro-
cedural semantics, which corresponds to a development in the psycholin-

2 The possible perspectives of this procedural dimension were taken up in a

number of contexts,also linguistic. Utterances as programs (Davies & Isard
1972) describes how an utterance, understood as a structured complex of lin-
guistic signs, can be viewed procedurally as something that enables the receiver
to reach the canonical informational end-state by carrying out the instructions
associated with the linguistic signs and their manner of combination.
3 Moreover, there is mutual translatability between an account in terms of rep-
resentations and an account in terms of procedures: an effective procedure is
one that leads to an output state that is uniquely specified by the program (cf.
Johnson-Laird 1983: 247). The excitement that the procedural dimension gave
rise to therefore gradually died down.
184 Chapter 5. Meaning and ow

guistics of memory retrieval and language comprehension, cf. the follow-

ing quote from Zwaan and Radvansky (1998: 162):
Rather than treating language as information to analyze syntactically and
semantically and then store in memory, language is now seen as a set of process-
ing instructions on how to construct a mental representation of the described
situation (see also Gernsbacher, 1990).
This approach captures the representational content of utterances in
terms of an instructional approach, leaving it to the participants to do the
The role of language as encoding processing instructions means that
speaking to people is basically a way of influencing them (impinging upon
them) often but not always with the specific purpose of providing them
with information. The overall role of linguistic meaning as performing an
operation on the world can thus be further analyzed in interactive terms:
the operation is not just one performed by the speaker, but one performed
in virtue of a constitutive co-operative relationship between speaker and
hearer, acting jointly as described by Clark (1996): the speaker ideally
does not programme the hearer, but co-constructs the situation with him.
Viewing meaning as input or instruction has been brought up from
time to time (cf. Acta Linguistica Hafniensia vol. 39; Harder 2009), but
has yet to seriously capture the imagination of linguists. I believe it has a
key role to play in the more dynamic and user-oriented conception of
linguistics that is now emerging in Cognitive Linguistics. Among cogni-
tive linguists who include this angle, Fauconnier (1985: 2) uses the word

4 The development towards focusing on what the linguistic input results in is

also reflected for instance in relevance theory (Sperber & Wilson 1986/1995).
Relevance theory makes a point of the extent to which a code view of under-
standing is misguided, and asserts the role of relevance-driven inferences in
both getting to the relevant proposition and in signalling what the relevance of
the proposition is (cf. also Blakemore 1987). Relevance theory is a develop-
ment of Gricean thinking about meaning, and occupies the same interface
position between a logical-propositional tradition and a pragmatic approach.
Like Grice, it offers a proposal for how to integrate the two sides. Unlike
Grice, however, relevance theorists are focally committed to dissociating lan-
guage in itself from interaction, associating it with information instead. This
orientation is bound up with the approach to meaning and understanding in
terms of informational profit (cf. Harder 1996:142). The product that is gen-
erated as a result of the linguistic input is understood in terms of cognitive,
informational content although of a different kind than the mental models
or structure-building type.
Meaning as process input 185

instruction about the underspecified meanings of linguistic items in

calling upon the fully specified cognitive representations. At the time,
however, Fauconnier rightly argued that it was necessary to move towards
the full cognitive perspective rather than limit ones attention to the
purely linguistic input.
Now that the encyclopaedic scope of meaning is well-established, and
meaning construction is coming into focus, I think it is time to pay more
attention to the input dimension of meaning. The role of linguistic mean-
ings as potential input to the flow of interaction becomes much more cru-
cial in a linguistics where actual usage is recognized as basic. This is the
context where we really see meaning at work. In this chapter I am going
to argue that from the linguists point of view, this is where we can most
meaningfully address the question of the interplay between the conven-
tional code and the situational context.
I begin by considering what a linguistic concept is, understood in this
dynamic perspective. There is an affinity between the input perspective
and the functional dimension, in that a very basic functional role for con-
cepts is to serve as a tool for assigning the multifarious stuff of actual
experience to a more limited range of kinds. A point of departure is a
functional interpretation of Lakoffs analogy (1987: 283) between the
container scheme and the way we think about categories. From an input
and a functional perspective the container as such is merely the prerequi-
site, not the function itself. In a dynamic perspective, the crucial thing is
the operation that is triggered when you evoke a concept as part of lin-
guistic communication. In terms of a pervasive metaphor (cf. p. 38),
putting something in a particular container can be described as an act of
grasping incoming experience, cf. the Latin etymon (con)capio: concepts
are handles on reality.
To possess a concept is thus to have a principle for aligning individual
instantiations under the same overall concept, i. e. to put a Porsche, a Ford
T and a Toyota in the same conceptual container labelled car.
One point on which this dynamic perspective makes a difference is
when it comes to the relation between concepts and instances of the con-
cept. If you understand a concept (= a conceptual category) purely as a
conceptual unit, it highlights the issue of how to treat divergences between
instances and concepts (after the neat Aristotelian distinction between
essential and accidental properties has been given up). While the polysemy
cline can handle part of the issue (more on this in section 6), it does not
entirely solve the question of boundaries between concepts, which are
often fuzzy. If conceptual territories overlap, it is not clear what the strictly
conceptual reality may be in such segmentations. How can we maintain
186 Chapter 5. Meaning and ow

the existence of a mental lexicon with separate bundles of conceptual

properties associated with them, if they merge into one another?
A functional approach, however, may provide the notion of concept
with a clear-cut role that makes fuzziness less problematic. The function of
dividing up a collection of phenomena is external to both the phenomena
themselves and the way we conceptualize them. Faced with the task of
dividing them into groups, a human agent has to decide what her criteria
are going to be this is not given in advance by the inherent properties of
the object. For the purposes of getting through the incoming business of
everyday life, the utility of having ways of grasping is like the utility of
cupboards with different shelves when somebody dumps a load of stuff on
you: it is useful to be able to put things somewhere. How precise the sort-
ing criteria are will depend on the circumstances, not just on inherent
properties. Pointing to experience is not always enough: mothers rightly
worry about providing children with the conceptual equipment that is
necessary to deal adequately with stuff that is currently outside the basic
and familiar sphere, such as vipers and strangers offering candy.
In the input perspective, this means that the evocation of a concept
must be understood as the first step in a process that concludes with an act
of categorization, where the instantiation has been placed in a container
and the concept has done its job. If communication is successful, at the
output stage there is a fusion of instance and concept (cf. also Langacker
2009: 232233 on the target), but at the input stage they are distinct. Cop-
ing by means of concepts depend on handling the input stage, while the
output stage is just a fragment of an actual usage event. Below we will
take up some of the complications of post-Aristotelian approaches to con-
ceptualization, and then in section 6 we return to what this implies for a
functional and input-oriented view of concepts.
Another central CL area where the functional and input-oriented per-
spective makes a difference is that of frames and framing. Framing, cf.
p. 27, involves an act of placing what you are trying to understand in the
appropriate context and a functional and dynamic construal shifts the
focus from the conceptual frame itself to the framing operation. Concep-
tually speaking, frame is a metaphor, based on the relation between a
picture and its frame and the structure that is transferred to the target
domain (word understanding) is the relation between the focused ele-
ment and the surroundings in which it belongs. You understand any given
instance of meaning by viewing it against its background an operation
that is built into the process of interpretation, no matter what the circum-
stances are. (In cases when there is no frame around a picture, the wall
itself must do duty as the perceptual fame; Traugott, personal communica-
Meaning as process input 187

tion). In this, it also reflects the basic reorientation of semantics towards

the process of human understanding (Fillmore 1982, cp. Croft and Cruse
2004: 8). The problem of understanding culture-specific terms is often best
accounted for not by trying to explicate the concept in itself, but seeing it
in the context of the whole cultural frame of understanding (cp. Croft &
Cruse 2004: 21, quoting Geertzs account of the Javanese concept of Rasa).
But what difference does it make if we view it in terms of input?
The point can be illustrated in relation to the role of frames and fram-
ing in meaning construction. Frames play a crucial role in the construction
of online interpretation, as demonstrated in great detail by Dancygier and
Sweetser (2005). They show how in understanding conditionals, a very
large part of the conveyed meaning is created by reconstructing frames
based on very sparse coded input. For instance, if I were you invokes a
situation that contains many more assumptions than a neat switch of iden-
tity. We therefore need to consider what the relation is between semantic
instructions and semantic frames. Are frames the same as meaning, for
instance, as the term frame semantics might well suggest?
In chapter 1 I suggested that the division of labour between frames,
domains and idealized cognitive models in a theory of meaning would be
clearer if we used the term frame as part of the dynamics of interpreta-
tion rather than as a static construct (which would be better designated as
an idealized cognitive model). In those terms, the act of selecting the
relevant frame for understanding is part of the situational activity of
meaning construction, but it is prompted by the conventional meaning
rather than being identical with the conventional meaning.5 In other
words, the conventional meaning serves as an instruction that triggers
framing activity. In terms of the basic frame metaphor, the linguistic
meaning of a linguistic item is analogous to the picture itself rather than
the frame around it
This view can be profiled by comparing with Hansen (2008), who uses
both the instructional and the framing approach, but parsimoniously sug-
gests that the instruction simply consists in evoking the frame. Following
Fillmore (1985), Hansen (2008: 23) gives the following example:

5 In the accepted terminology, frames are evoked if they are associated with
the coded meaning, but invoked if they are brought into the picture by infer-
ential rather than conventional mechanisms. Since in ch. 1, p. 27 I suggested
that conventionally obligatory frames were identical to domains, I do not need
to use the term frame about aspects of coded meaning, but can reserve it for
the situated event of inserting the situational background against which a
meaning-in-use is understood.
188 Chapter 5. Meaning and ow

Severus is economical: hell make an excellent husband

Severus is stingy: hell make a horrible husband

She argues that

the use of economical semantically instructs the addressee to interpret the
utterance as evoking a frame in which the urge to limit spending is conceived of
as a positive quality, and to identify some set of consequences potentially accru-
ing from this quality which may be beneficial to the spouse. The parallel use of
stingy on the other hand, evokes a very different frame, in which generosity is
an absolute virtue, and instructs the hearer to identify a set of undesirable con-
sequences of its absence.
Although I think the analysis in terms of output interpretation is correct,
the paraphrase of meaning in terms of frame evocation is potentially mis-
leading if it is understood as an instruction to evoke the frame per se. The
linguistic key elements encode the predication of a property to Severus:
the word economical means something like disposed to limit spending,
rather than fritter ones money away in an irresponsible manner. Framing
may come in as a background cultural scenario, for example the virtuous
script of making ends meet and saving up which may be invoked as part
of the background against which an economical husband is what you want.
Stingy is in the same structural position, as a property predicated of the
subject NP, Severus. The property is actually not the same as economical,
however. It means, roughly, to be reluctant to fork out the money for no
other reason than wanting to hang on to it. Again, potential cultural
frames come to mind in the form of the icon of the miser, an example
being Scrooge McDuck. Again, framing is part of the process of meaning-
assignment the particular frame is just not part of what the word instructs
you to evoke.6

6 The difference between meaning-as-instruction and meaning-as-framing also

comes out if we look at the complete utterance meanings. I do not think the
formulation literally instructs the addressee to identify some set of conse-
quences potentially accruing from this quality which may be beneficial to the
spouse (and I should add that this is not what Hansen claims in the passage,
which is meant for illustrative rather than precise analytic purposes). Rather,
the addressee is explicitly instructed only to understand Severus as an excel-
lent spouse in virtue of being economical, full stop. As a general feature on
meaning-making, this calls for locating a frame where this makes sense but
how to specify the frame is optional. A very nave addressee might in fact take
these words on trust, and understand the utterance as a piece of information
about married life (the blank wall acting instead of a proper frame, as it were);
Meaning as process input 189

In relation to Hansens (2008) own specific topic, however, I think she

argues convincingly for the identification of frames and instructions. Her
subject is phasal adverbs such as still and they are special precisely
because their privileged function is to specify the temporal background
rather than the temporal location. This is why he is still alive has the same
truth conditions as he is alive. What is added by still is the explicitly evoked
temporal background for his being alive. But that is a special case: usually
background aspects are not profiled as the specific contribution of an
The advantages of the instructional view may be most striking as an
account of purely procedural meanings with little conceptual content (cf.
Blakemore 1987), as exemplified with conjunctions like although and
focus particles like only. However, it has a number of advantages also for
lexical meanings, cf. Harder & Togeby (1993). In the context of Cognitive
Linguistics, I align myself with Tyler and Evans (2003: 40), who use the
word prompts about the contribution of words to utterance meaning,
and with Evans (2006, 2009) in distinguishing lexical concepts from con-
ceptual models.
A major consequence of this view is that linguistic meaning viewed as
an element of competency is never identical to an actual mental state or
process at best, there is just a satisficing fit. On the face of it, this is a
radical departure from traditional assumptions in CL. But I believe that
rather than a heresy it is a necessary clarification in making the social
turn. You cannot ask what it is to know the meaning and maintain that a
single event is the full answer. The distinction between online and offline
properties pinpoints a key aspect of participant competency: the ability to
store sediments from the flow in a manner that will work next time round,
too. Meaning is constructed in the flow that continuously produces con-
ceptual states including construals, mental models etc but participant
access to all this requires a competency that cannot be reduced to online
flow alone.

and we would not want to deny that she had understood what was actually
said. Generic frames (= domains), of course, are by definition evoked with the
words: the word husband thus obligatorily evokes the marriage frame.
7 Unlike Hansen (2008:32), however, I do not see the presuppositions as parts of
the frame. Instead, presuppositions (in the sense adopted below p. 195) are
rather the result of combining clause meaning with the frame that (in this spe-
cial case) provides the content of the instruction. Thus he is still a fool, by
combining the continuation frame of still with the content John being a fool
gives rise to the presupposition that John was a fool at some earlier time.
190 Chapter 5. Meaning and ow

This is in accordance with a view of encoded meaning as semantic

potential (cf. Evans 2006) or meaning potential (Allwood 2003), which
is translated to actual meaning in the process of utterance understanding.
From a competency point of view this means that knowing a word means
that you know what range of output interpretations it can be used to trig-
ger, not what it means on any given specific occasion.
The input-based approach is in harmony with a major development in
literary interpretation. According to the classic belief, there was a deter-
minate textual meaning hidden under the surface, and it was the job
of the literary scholar to uncover that meaning. This belief was widely
abandoned and replaced by a focus on the processes whereby meaning
is produced by readers. Many modern literary analysts have flirted with
the anything goes social constructionist position which may sound
rather more plausible in relation to literary interpretation than in the
theory of science (cf. above p. 108). Here, too, however, the issue is more
Although Stanley Fishs book title (1980) Is there a text in this class?
seems to throw doubt on the existence of any shared source of meaning,
his point is actually different. While reader response is variable, it is not
random, and Fish focuses on that variability of understanding which is due
to different sets of assumptions and practices in interpretive communi-
ties in our terms, a type of niche. The rejection of ultimate underlying
truth in literary understanding, and the focus on processes of meaning-
creation, thus do not rule out certain forms of stability. But it may be hard
to avoid overstating your point when you are up against the full might of
two thousand years of Platonic tradition. Fish himself comments on the
way he has generally been misunderstood, and one case in which he mis-
understood himself, thus getting a share in the responsibility for the mis-
This response to a response to Interpreting the Variorum contains the most
unfortunate sentence I ever wrote. Referring to affective criticism as a superior
fiction, I declare that it relieves me of the obligation to be right (a standard
that simply drops out) and demands only that I be interesting. I have long
since repudiated this declaration along with the relativism i[t] implies. The only
thing that drops out in my argument is a standard of right that exists independ-
ently of community goals and assumptions. Within a community, however, a
standard of right (and wrong) can always be invoked against a prior under-
standing as to what counts as a fact, what is hearable as an argument, what will
be recognized as a purpose, and so on. The point, as I shall later write, is that
standards of right and wrong do not exist apart from assumptions, but follow
from them, and, moreover, since we ourselves do not exist apart from assump-
Meaning as process input 191

tions, a standard of right and wrong is something we can never be without. (Fish
1980: 174).
The central point is that texts, like linguistic expressions, must be under-
stood as input to an interpretation process and there are norms in oper-
ation for how to carry out such processes. Reading does not yield totally
predictable outcomes, but nor is it the case that anything goes. When read-
ers are competent according to the standards for reading that obtain in a
given community (=niche), the coded text can instruct them to produce
outcomes with a systematic, hence describable, relationship with the lin-
guistic input.8
The habit of thinking that accords output states automatic primacy
also systematically overlooks the possibility that the process itself may be
viewed as a dimension of the successful result of the instigated action as
also pointed out by Stanley Fish. An illustration is given in relation to a
passage from Miltons Paradise Lost (IV, lines 912):
Satan, now first inflamed with rage came down,
The Tempter ere th accuser of man-kind,
To wreck on innocent frail man his loss
Of that first battle, and his flight to Hell

My contention was that in formalist readings meaning is identified with what a

reader understands at the end of a unit of sense (a line, a sentence, a paragraph,
a poem) and that therefore any understandings preliminary to that one are to
be disregarded as an unfortunate consequence of the fact that reading pro-
ceeds in time. The only making of sense that counts in a formalist reading is the
last one, and I wanted to say that everything a reader does, even if he later
undoes it, is a part of the meaning experience and should not be discarded.
One of the things a reader does in the course of negotiating these lines is to
assume that the referent of his in line 11 is innocent frail man. (Fish 1980: 3)
The same principle applies to metaphor understanding. If you look at
what precisely the distinctive outcome of metaphor in actual communica-
tion is, it is typically less than clear. The emphasis has been on the appara-
tus consisting in the mappings from source to target, corresponding to the

8 This applies not only in areas of poetic license; the same discussion has taken
place in relation to interpretations of the law, where the continuing role of the
interpretive community constituted by changing memberships of the US
Supreme Court has been recognized since the Marbury v. Madison case in
1803. Todd Oakley (ongoing research, personal communication 2009) has used
that as an illustration of the role of historical events in redefining the shape of
cultural reality.
192 Chapter 5. Meaning and ow

formula of understanding one thing in terms of another. But what pre-

cisely, if anything, follows from saying that MORE IS UP?
Because of the target domain override principle (Lakoff 1993: 216),
not everything from the source domain gets transferred to the under-
standing of the target domain (MORE may not always be UP, e. g. if a hori-
zontal puddle is spreading out). But if a target overrides whatever does
not fit in the source conceptualization, it remains open exactly what fol-
lows from applying the metaphor to that target. Davidson (1978) draws a
radical conclusion from this problem. He argues that metaphorical mean-
ing is too indeterminate to qualify as part of the description of language:
one can be precise about non-metaphorical meaning, but one cannot be
precise about a metaphorical meaning until the metaphorical mapping
is complete and the source domain meaning has become part of the target
domain and hence flattened into a determinate, literal-like meaning. To
take a familiar example: Going from London to New York is a journey in
a determinate sense, but we do not know exactly what LOVE IS A JOURNEY
means until we have completed the work of mapping the metaphor onto
the target domain. A primitive result of such a mapping would be some-
thing like love is an experience where, after the relationship has come
into being, you move through a succession of new situations, rather than
remaining in the same initial position which is again determinate and
Against this, Collin and Engstrm (2001) point out that the under-
standing of metaphorical meaning as something both non-literal and
determinate can be defended provided one takes an explicitly process-
oriented view of meaning. The whole process of recruiting metaphorical
meaning has a determinable content, even though not all of the potential
of using the metaphor is determinable at a given stage. With Turners and
Fauconniers (1995) analysis of the argument is war metaphor as an
example, Collin and Engstrm argue that beginning with war and impos-
ing it on argument takes you through a blending process with determi-
nate properties. The ending point is indeterminate, but the history of the
application is part of the meaning of the metaphorical expression: the war
path, as it were, is ineradicable from the total signification.
This underlines the foundational status of flow meaning for good and
for bad. Push polls, where members of the public are asked about their
reactions if a politician were to do or say something they did not like,
would not work if some of the dirt did not stick. But this also makes it
obvious why it matters what one takes on board from the flow. Plato was
right to be worried about letting the social process alone have the last
Presupposition and the directional nature of linguistic meaning 193

In all three domains discussed above, the instructional approach makes

it possible to unpack a competency dimension that becomes invisible if
you assume ideal convergence between offline and online meaning. Mem-
bership in a speech community depends on the ability to follow the path
from explicitly coded input, via participant meaning-construction activity,
to full-fledged contextual meaning. Whether in hard-core psycholinguistic
experiments, in interpretation of literature or in metaphor understanding,
the instructional understanding upgrades the significance of competency
as manifested in meaning construction to the constitutive participant
property in the speech community.

3. Presupposition and the directional nature of linguistic meaning

The instructional approach, however, does not exhaust the consequences

of the dynamic view of meaning. To illustrate what is missing, we can
return to the mechanical, computational version of procedural semantics.
The view quoted from Zwaan and Radvansky above would be compatible
with assuming that the meaning instructions are analogous to the instruc-
tions for assembling a kitchen cupboard. If you buy a cupboard under the
condition some assembly required, what you buy is not the instructions
but the cupboard (analogous to the finished mental representation), and
the dynamic stage would be just a tiresome complication.
What you get out of language, however, is not totally analogous to a
cupboard whose stable properties are the main thing. The output of a
process of utterance interpretation is itself a dynamic entity a change in
the interactive situation. The point is made by Verhagen (2005) with argu-
mentation as a key example: the aim is to change the addressees position,
not to fill him with information.
This is generally accepted in the case of utterances, the pragmatic units
of linguistic analysis. A speech act, following Austin and Searle, is designed
to change the situation rather than represent it. This property carries over
to individual elements: functions, also for words, are privileged effects, and
if meaning is functional, it is a kind of effect.9

9 In some cases, this is obvious. Thus when you say hello to someone, you achieve
the function of greeting them, thus bringing about a certain interactive rela-
tionship with them. This makes sense only as an interactive move, and cannot
plausibly be understood as an expression of a mental state in an individual
mind. This is why such greeting expressions can be slotted directly into a net-
194 Chapter 5. Meaning and ow

This action-based approach has a consequence for conventional mean-

ings which is the main point in this section: linguistic meanings are direc-
tional, i. e. they have a point of departure and a point of interest.10 The
point of interest is the business end, oriented towards the purpose that is
to be achieved. Thus when you say No!, the point of interest is the rejec-
tion. For obvious reasons, this is what gets the attention, just like the act of
cutting when you use a knife.
But there also has to be a point of departure for the utterance, i. e.
something that offers the necessary basis for an act of rejection, just as
there has to be something for you to cut in the knife example. Thus in actu-
ally using the word No! you are necessarily depending on the existence of
something that you can reject, just as you cannot perform an act of cutting
without something to cut, such as a sausage. This has consequences for
meaning construction: if you hear someone emitting a no!, you cannot
understand the utterance unless you know what is being rejected. Just like
(other) acts, utterances have a point of departure and a point of interest.
The significance of the point of departure for an encoded meaning can
be illustrated by a comparison between two similar meanings. The two
words repeatedly and again do not differ a great deal if viewed from the
external, instructional point of view they are both contributions to the
process of meaning construction which indicate iteration. However, they
differ interestingly from the internal, directional point of view: again takes
its point of departure in a previous instance and predicates a new one,
while repeatedly predicates the whole iteration sequence.
To the extent there are features of word meaning that require a spe-
cific point of departure, the word may potentially give rise to presupposi-
tional effects. The point of departure has to be available in order for the
word to be used and these demands are encoded as part of the conven-
tional meaning, spelling out what is taken for granted in using the word,
and may get a role in meaning construction. Whether it actually does so

work of interactive options, cp. Halliday (1973: 83), without intermediate con-
ceptual stages.
10 Viewing meanings as instructions already places them as parts of the dynamics
of interaction, so it might be asked, what is the difference between the instruc-
tional and the directional property of meanings? The difference is that the
instructional view looks only at the meanings as a whole, from an external
point of view. An encoded meaning works in toto at the input stage of under-
standing rather than as features of the output. The directional analysis looks
inside each individual meaning and asks: what does it do, and what needs to be
in place before it can do it?
Presupposition and the directional nature of linguistic meaning 195

depends on whether the use of the words fits into the context in such a
way that the demands are unproblematically satisfied. Not everybody
would accept that this was the case in classics like dont make a fool of
yourself again!
Presuppositional effects take the form of presuppositions of the classi-
cal kind that has been discussed in philosophy and linguistics, when words
in combination impose demands on the situation that can be spelled out
in the form of propositions. The classical example the present King of
France requires the addressee to identify a referent answering to the spec-
ifications, and thus presupposes that there is one and only one relevantly
identifiable King of France available. The dynamic implications of this
directional view are clear: the phrase is designed to change the situation of
the addressee, in that he has to move to the position of having identified
the referent of the expression. Inevitably, you thereby presuppose that it
is available for being identified. Presuppositions are the most overt and
salient manifestation of the directionality of meaning, because they signal
something propositionally explicit about the speakers point of departure.
The reason presuppositional effects are most often invisible is simply that
in the default case, the demands are satisfied: why invoke linguistic tools
in situations where they cannot work? Hence, the information that might
potentially be conveyed, such as the existence of one and only one King of
France, is already available in a default context of use, and so it is not con-
veyed but merely invoked. The phrase the weather has the same presup-
positional potential, but since the weather is always there, the presupposi-
tional potential remains dormant.
However, situations are always only partially specified. There is a con-
siderable grey area which is capable of being evoked as necessary back-
ground for utterances, but which might not be saliently available in the
absence of such invocations. This means that you can insert information
that is not taken for granted (such as making a fool of yourself last time)
with the linguistic status of being actually taken for granted.11 This can

11 This belongs in the area of presupposition failure, the subject of a book I wrote
in 1975 with Christian Kock. (Herb Clark has terrorized me out of referring to
this book as Harder & Kock!). Presupposition failure takes many forms (which
are the subject of that book), including deception (as when you pretend to
have beliefs that you do not (your munificence is overwhelming) and bullying
(stop making a fool of yourself). Generally, adeptness in distributing the status
of being taken for granted vis--vis the status of being explicitly addressed is
central in all strategic language use (not just tact and politeness). This is so
because by taking things for granted you assign them a position as being
196 Chapter 5. Meaning and ow

also be used non-aggressively: if an old friend turns up unexpectedly, how

you feel about it may not be given in advance, but if you explicitly thank
him for coming you have signalled that you treat his visit as a favour. In a
Truffaut film, the strategy is exemplified as advice for a gentleman who
finds himself in the situation of opening a bathroom door that had inad-
vertently been left unlocked, finding a naked woman behind it. The thing
to say (while hurriedly closing the door) is not pardon, madame, but rather
pardon, monsieur because the necessary point of departure for the term
monsieur is the assumption that you are addressing a man (hence it can be
inferred that no instance of consummated peeping has taken place).12

already in place in social reality, rather than taking it upon yourself to intro-
duce them into unfolding social reality, with the risks and liabilities that such
an action involves.
12 These strategic implications of the directionality-of-meaning are also relevant
in relation to framing. This makes it useful to consider what the difference is
between strategic use of presuppositional effects and strategic use of framing.
George Lakoff, as discussed in more detail on pp. 343 and 398, exhorts his
readers to actively frame the debates they take part in. Is that the same thing
as deviously smuggling your own preferred presuppositions into taken-for-
granted reality?According to the definitions I have advocated this is not (nec-
essarily!) the case. Framing is an inherent and constitutive part of the process
of understanding; for each new introduced meaning, you have to figure out
what background to understand it against. The phrase Joes girl could be
understood as invoking a father or a partner frame. You can frame a debate
quite openly by announcing that you view the issue of energy in the context of
national security or, alternatively, in the context of global warming. Depend-
ence on oil acquires quite different properties depending on whether you
frame the issue one way or the other.The use of presuppositional effects con-
stitutes a possible form of framing, but it is a much narrower category. Its
special flavour is precisely that it comes with the taken-for-granted status of
a point of departure for the utterance in question. Thus the King of France is
bald presupposes, in the accepted reading, that there is a uniquely identifiable
King of France to refer to, and thus introduces the frame of France being a
Kingdom as a presupposition while the frames that are invoked may include
many things, possibly absolute monarchy, with trappings including courtiers
and an irresponsible nobility, with shades of Louis XIV. In the case of again, a
directional analysis would say that its point of interest (=front end or cutting
edge) has the predication of an additional event. This carries with it a point of
departure (= presupposed background) in a previous event of the same kind.
Saying that Lee called the office again thus carries a presupposition (not just a
frame) that he has called the office before. The evocation of an iteration frame
in itself does not distinguish between he did it repeatedly and he did it again
The procedural nature of competencies 197

This ends that part of the chapter which focuses on the ways in which
meanings reflect their basic role as part of the flow. The moral is that lin-
guistic meaning is best understood if it is caught in the act of making a
contribution to the flow it is not an eternal idea, but neither is it part of
the flow itself.

4. The procedural nature of competencies

If meanings are input, what is the nature of the competency to handle

First of all, the competency is responsible for the process that con-
verts input to finished interpretation. Usage meanings, the finished prod-
ucts, are accessible to intuition; input meaning (I have argued) is what
the speaker knows between utterances. This leaves the process that is
needed to bridge the two as the domain of competency. I am now going
to argue that this makes a difference in relation to traditional ways of
thinking about the speakers knowledge of language as the object of
Conscious knowledge has to be applicable at the output end. Unless
we can become consciously aware of the meaning of an utterance, speak-
ing would do us no good. The how, on the other hand, we do not have to
be aware of. This applies not only to language competencies, but to all
other skills. Knowing-how, as familiar from Ryle (1949) onwards, is a dif-
ferent type of phenomenon from knowing-that. The implications of this
familiar fact, however, need to be made explicit in relation to the dynamic
reorientation. I am going to tackle the issue first with reference to the abil-
ity to evoke lexical meaning, instead of discussing it in relation to gram-
mar, since grammar is a much more complex and controversial issue. So
when I speak of skill, ability or competency in this section, I am referring
to the competency to use lexical items, which is clearly part of the speak-
ers know-how. Grammatical competency will be discussed in ch. 6.
This issue of knowledge of language is difficult enough in itself, and the
darkness became almost impenetrable after the introduction of the Chom-
skyan concept of tacit knowledge. This idea was part of the cognitive
revolution against behaviourism, and thus associated with an insistence

and therefore the directionality of meaning, as well as its manifestation in the

form of presuppositional effects, cannot be captured simply by the general
point associated with frame semantics.
198 Chapter 5. Meaning and ow

on human reason as opposed to mindless causal triggering. Lakoff &

Johnson (1999: 472) generously say that Chomsky deserved enormous
credit for this idea, which in classical CL survives in the form of back-
stage cognition inaccessible to consciousness. There is thus broad agree-
ment that we have a realm of tacit knowledge, although CL and genera-
tive grammar disagree on its nature: generative grammar thinks of tacit
knowledge as computational in nature, while CL thinks of it in terms of
cognitive mappings imbued with content.
I believe this way of thinking about it is fundamentally flawed. In the
context of CL, Zlatev (2007) has argued that Searles (1992: 156159)
argument against tacit knowledge should be taken seriously by CL, even
if it can be ignored by the generative world view (because of overall gen-
erative assumptions about competence-level entities). Basically the argu-
ment is that if tacit knowledge is so tacit that we can never become aware
of it, it cannot in any coherent sense be understood as a mental entity. We
only know mental phenomena from our subjective awareness if we can-
not access them in that arena, we have no way of recognizing them as
mental. Thus a pin prick may be the cause of our conscious feeling of pain,
but the prick in itself is not a mental event it is not the conceptual dimen-
sion that causes the pain or the bleeding. Similarly, a dislike of someone
may be due to mentally stored memories of him, or to physical factors like
his unpleasant smell but we can only know that it is due to a memory if
we can bring that memory to conscious awareness. This has implications
for the practice of inferring backstage cognition on the basis of conscious
mental phenomena how can we know it is really there? Postulating a
language competency that exists at the level of backstage cognition is
therefore problematic for precisely the same reasons that make Chom-
skyan tacit knowledge problematic.
Based on the approach that takes the flow of interactive activity as
basic rather than cognition, a different approach to the description of
competency suggests itself. The success criterion for behaviours is whether
they work in the social and natural environment or not the mental
dimension may be present in some cases and absent in others. The piano-
playing competency is judged by the version of the moonlight sonata
that members of the audience hear how much of the actual execution
is accompanied by explicit intentions is beside the point. This means that
the first question we must ask about a given skill, or competency, is
whether the organism is wired up so as to be able to perform satisfic-
ingly. If we look at a skill such as playing football, the question is whether
a person can produce satisficing behaviour in the kind of situations that
a football match faces her with. Doing the right thing without thinking is
The procedural nature of competencies 199

considerably better than having the right explicit intentions without

doing anything.
This means, in the terms discussed above, that basically an up-and-
running competency is a causal setup in the organism just as a function-
ing market is a causal setup in society. Both in the organism and in the
society, mental dimensions may play a variable role in relation to such
causal structures, but the success criterion is causal efficacy, and not the
degree of mental activity. This creates both a descriptive and a methodo-
logical dilemma: how do we tell the mechanical and the intentional
dimensions apart?
Searle (1992: 81) quotes a description of the digestion system, which
systematically uses a mental and intentional metaphor (italics by Searle):
The gastrointestinal tract is a highly intelligent organ that senses not only the
presence of food but also its chemical composition it is often called the gut
No one would presumably claim that these operations are mental. They
are clearly part of the way the organism is wired up at the causal, physio-
logical level. In Searles critique of first-generation cognitive science, this
was a key point. It was directed at syntactically oriented simulation, and in
the Chinese room experiment Searle (1980) showed that syntactic organi-
zation was not enough for meaning. His point was that the whole simula-
tion paradigm failed to address the real issue of mental representation,
because simulations ignored this distinction. But it affects also the con-
cept of backstage cognition.
The problem applies not only to the neural approach of Feldman, who
explicitly wants to ground explanation at a non-mental level, but to all
continuist accounts that are less than explicit about where the borderline
is between mental and non-mental phenomena. It is clear from the argu-
ment in the beginning of the section that in order for something to qualify
as a competency, its output must be consciously accessible. This is true
not just of the competency to use words, but also for football and sky-
diving; otherwise, it could not serve as a medium for intentional action. We
would not speak of the intestine as providing the individual with digestive
competency, because it is not an intended act, while football, piano playing
and speaking produce results that we can check against intentions and

13 Automatization is standardly understood to involve a process whereby what

used to be laborious and conscious gradually becomes fluent and unconscious.
Searle (1983) talks about the process of learning to ski whereby operations
200 Chapter 5. Meaning and ow

The two aspects are separable in terms of neural wiring. The purely
procedural setup is associated with implicit memory, while the consciously
accessible knowledge that is associated with language goes with explicit
memory, which works by separate pathways as demonstrated by both
brain imaging data and clinical data (cf. Feldman 2005: 78; Ellis 2007:
2223). Ellis refers to an anecdote about a patient who had lost the ability
to form new explicit memories. She had once been pricked by a pin while
shaking hands with the consultant, and at later meetings, she refused to
shake hands with him while denying that she had ever seen him before.
She had lost the explicit representational skill, but remembered well
enough implicitly, or procedurally i. e. in terms of what forms of action to
Patients with this form of amnesia have not lost the ability to form new
procedural skills (including clearly cognitively imbued skills such a mirror
reading), but cannot acquire new consciously accessible mental represen-
tations. Avoiding behaviours that give rise to unpleasant experience
clearly is one such skill type and this has implications for the role of
adaptation. Blindsight (cf. Weiskrantz 1986) may be a glimpse of the same
type of mechanism, whereby the environment can impinge on your orien-
tation without generating any conscious awareness.
If we follow the principle of caution, there is thus no justification for
assuming that the neurocognitive underpinning of skills, including lin-
guistic skills, have any similarity at all with the kind of mental representa-
tions that are consciously accessible to us. This might appear to support a
generative, purely computational approach to grammatical processing.
But this is due to a variant of the same fallacy as the gut brain example.
Not only are computational operations insufficient for meaning there is
no such thing out there in the world that is inherently computational
(except a computer). It follows from Searles (1980, 1983) argument that
it is empirically empty to describe these processes as computational
what exists is a causal pattern with certain properties that can be simu-
lated. The workings of the gastrointestinal tract, the workings of the brain,

that used to be conscious gradually recede into the background. But on closer
inspection, this turns out to be a conflation of two processes: one consisting in
the fading away of explicit representations, the other in the buildup of reliable
neurocognitive routines. There is never a point at which the actual skill is con-
scious: at all the different stages, the sensorimotor skill as such (like all other
sensorimotor skills) is hermetically sealed. The conscious representations
stand outside and prompt the action, but they are not part of the actual execu-
tion, not even at the fumbling stages.
The procedural nature of competencies 201

or the workings of the railway system, can with equal justification be

modelled by computational procedures but this does not mean that
either railways or digestive tracts exist at a computational level in addi-
tion to a digestive or a transport level, respectively. Hence there is no
argument for assuming that linguistic processing intrinsically possesses a
computational level.14
This does not entirely eliminate backstage cognition in the CL sense
but in order to avoid mystification it has to be assigned clearly to one of the
two following kinds: the first category is neurocognitive ability, which we
may choose to call cognitive but is not mental in the full representational
sense of being available to conscious inspection (and that means never, not
a low percentage of the time). This is knowledge only in the knowing-how
or procedural sense. One way of making that point is to say that tacit
knowledge is procedural knowledge; but as also pointed out by Langacker
(2008a: 459) it is better to avoid describing it in terms (such as knowledge)
with associations of explicit and declarative mental content and use the
term ability instead (thus explicitly going against Chomskys position as
quoted in note 14). A procedural interpretation appears to be compatible
also with some strands in the generative tradition, cf. Ullman et al. (1997).
The second category is the kind that is potentially conscious. Although
it is temporarily inaccessible, it might at some point pop up in conscious
form. Although percentage counts are no more than suggestive, it makes
sense to say that at any given time we are only conscious of a very low
proportion of the total potentially available knowledge we possess. This
also applies to the kind of representations that occur at the output and
input ends of linguistic processing. The procedural work of producing

14 Chomskys definition of competence therefore ends up in no mans land:

The term competence entered the technical literature in an effort to avoid
the slew of problems relating to knowledge, but it is misleading in that it
suggests ability an association I would like to sever (Chomsky 1980: 59,
quoted from Brown, Malmkjr and Williams 1996: 3)
Chomsky would like to posit something that is just like knowledge but without
being consciously accessible. At the same time, it must not be identical with
the infrastructure of ability. For the reasons outlined, there can be no such
thing. To the extent there is any reality in tacit knowledge of the Chomskyan
kind, it is the causal wiring of a complex behaviour. Chomskys own metaphor
of the language organ points to the same thing: the formally representable
processes that he is talking about are analogous to the causal mechanics in the
gut or the liver (the liver is Newmeyers 1998 illustration analogy in arguing
for linguistic autonomy).
202 Chapter 5. Meaning and ow

utterances prompts a great deal of concomitant representations, which we

may be only fleetingly aware of.15
In neuroscientific terms, part of the story is that the two systems inter-
act intensely in normal brains. In terms of language learning, this is
reflected in the fact that explicit learning has turned out to play an impor-
tant role for L2 acquisition. In continuation of the generative theory of
language acquisition, Krashen (1981), put forward a hypothesis that there
was a total separation between real, implicit language acquisition and
consciously accessible language knowledge but this theory has not been
borne out by the evidence (cf. Ellis 2007). The fact that you can wake up
in the morning and realize that you now have the solution to the problem
you thought about last night shows that there is an intimate relation
between dynamic unconscious processes and conscious outputs. This
interaction remains a tantalizing mystery. Another intermediate-type
phenomenon is the form of consciousness that may attend the processing
(which is in itself unconscious) qualia-type sensations like the sense that
something is wrong or that something is eluding you, or the tip-of-the-
tongue phenomenon: a message from the machine room that they are
working on the retrieval problem that you have set and any moment now
the word may be available.
It follows from the above that the ultimate goal of linguistics cannot be
a complete description of language competency in the form of declarative
representations with psychological reality at least if we mean this is
what really and truly is there in the native speakers brain as part of his
language competency. This ambition only makes sense (regardless of
whether it is practically possible) if we aim at describing flow, i. e. a con-
crete actual sequence of language use a stream of consciousness type of
object. This type of phenomenon is clearly distinct from the object of
description that most cognitively interested linguists, including generativ-
ists, have traditionally taken for granted (with a recent formulation, cf.
Langacker 2008a: 20): a fluent native speakers conventional knowledge
of language.
As already pointed out, the central claim is not new, and since it is very
hard to disentangle procedural and declarative elements anyway, it may
be asked what precisely follows from this discussion. I think it points to a
necessary reorientation that would be difficult to get ones head around in

15 I have elsewhere argued (Harder 2007d: 12551256) that much of what Lakoff
and Johnson (1999: 1011) see as completely inaccessible backstage cognition
is actually of the second, potentially conscious kind, an example being antici-
pating where the conversation is going.
The procedural nature of competencies 203

the absence of a clear recognition of the difference between skills and

representations. A social cognitive linguistics needs to integrate a tradi-
tion focused on representations into a larger context where adaptation to
social structures plays an important role. Adaptation means coping, and
coping means skill. Familiar concepts like entrenchment and automatiza-
tion must be seen in the dual light of the representational dimension and
the ability dimension.
An example that may illustrate the two perspectives is a proposed
reinterpretation of the figure-ground issue in relation to language. Krifka
(2007) suggests that there is a source which is not based on perception (as
generally assumed in classic CL, cf. p. 18), but rather in bimanual action:
the topic is held constant (with the non-dominant hand), while the com-
ment is the elaboration (by the dominant hand). Both ideas are in har-
mony with embodiment, but they are different in terms of the role of men-
tal representation and the skills element is more strongly profiled than
in the traditional perception-based account. The proper descriptive strat-
egy is to assume that there are two different dimensions and then seek to
integrate them, rather than assume that there is one object that can be
subsumed under the concept of knowledge or psychological reality.16 It
is interesting that at a seminal point, in the introduction to Holland and
Quinn (1987: 6), the distinction between operational and representational
knowledge was explicitly confronted and rejected. Instead, it was
assumed (in our view) that underlying models of the same order

16 This also means that the terminology of hypothesis-testing that persists in the
language acquisition literature should be replaced by skill-testing: only lin-
guists test hypotheses about language, but all learners try out their skills and
learn from the outcomes (some children, of course, may act like linguists!). The
much-discussed issue of negative evidence also falls into two subcategories:
although few children get corrected explicitly, all children experience varying
degrees of success in accomplishing the goals of their utterances, and we must
assume that they adapt according to the pin-prick type of mechanism along-
side their adaptation to conscious feedback. If the goal includes an integrative
component, as we may assume it does in all cases of L1 acquisition, all per-
ceived deviances between own output and other-output will feed into this
process, whenever the speaker senses that she did not get it quite right. Imita-
tion in L1-learning as described by Tomasello might plausibly be understood
as driven chiefly by unconscious adaptive mechanisms. Conscious representa-
tion typically sets in in breakdown situations, where the procedure does not
come off: consciously driven readjustment may then get the process on the
rails again.
204 Chapter 5. Meaning and ow

are used in operational and representational contexts. For the reasons I

have given, I think it is time to reverse that decision.
In the light of the history whereby tacit knowledge/backstage cogni-
tion provided a step forward from behaviourism, this might appear to be
a step backwards. Causal wiring of language activity, indeed! But that
would be a misunderstanding based on accepting one of the behaviourist
assumptions, namely that there is a chasm between overt behaviour and
complex cognitive performance. Seeing skills as more primitive than
conceptual representations would reflect this obsolete way of thinking.
Rather, for some purposes it makes more sense to see the skill as the fin-
ished product, while for other purposes the explicit declarative represen-
tations are the finished product. Each side would be severely hampered
without the other. For instance, thinking is itself a skill, and we have no
more conscious access to the procedural operations that go into thinking
than to the operations that go into walking. The point is related to the
point made by Verhagen on the importance of the regulatory as opposed
to the purely representational role of language (cf. p. 87), and he reports
similar reactions (personal communication). There is a very ancient pres-
tige structure at work here, in which the attempt to interfere with actual
social processes is vulgar compared to the glory of purely mental activity.
But fairly obviously it would be no less reductive to see mental processes
as working solely in a world of their own than it would be to go back to the
behaviourist idea that they are under direct stimulus control.
This leaves the question of the exact role of truly mental phenomena,
when it comes to understanding the process dimension. I refer to those
representations that form the mental content of the human mind the
inner sanctum of the black box that was declared open at the demise of
behaviourism. One aspect of the answer is that there is a functional rela-
tionship: we use conscious mental representations in orienting our actions
and adapting our action skills.17 A central functional feedback process is
the one that operates when we invoke the unconscious linguistic proce-
dures and evaluate the results for instance in accessing a lexical item.
When recruiting, for instance, the words left or right in giving directions,
many people (including myself) have an astonishingly high error rate.

17 Adherents of first-generation cognitive science occasionally claimed that con-

sciousness was an epiphenomenon with no causal significance a computer
screen that makes no difference to the execution of the programs. But as
argued by Searle (1992:125), there is no compelling argument for it. Also, it is
obviously implausible as applied to skills acquisition: think of learning how to
handle a computer with the screen turned off.
The procedural nature of competencies 205

Regardless of the neuroanatomical divide between skills and conscious

awareness, the integration between the two creates a functional system
which can only be described by putting mental representations and causal
wiring together in the same model (as done in e. g. Levelts processing
theories, cf. Levelt 1989, 1999). This is plausible in terms of the evolution-
ary satisficing criterion. If adaptive correction of skills based on con-
scious understanding was not an option, my wife and I would still be driv-
ing around looking for the hotel we had booked in Morocco last March.18
From a functional approach, the output end is criterial: only the actual
response enters into the feedback process that determines what persists,
not the preparatory stages and their variable degrees of conscious acces-
sibility. The output end, however, is also central from the mental point of
view: intentional human action is only possible if the act we end up per-
forming is consciously accessible (while again, preparatory stages do not
have to be accessible). Yet the two properties are still dissociable: even the
output stages are mentally conscious to a variable degree. When I perform
routine actions, such as riding a bicycle, my conscious mind may be occu-
pied for much of the time with matters entirely unconnected to the ride.

18 This is also the basis of the professional linguists use of intuitions: you draw
inferences about the properties of the linguistic input based on conscious eval-
uation of the output. These properties may themselves be consciously acces-
sible, but even if that is not the case, there is frequently no alternative avenue
of exploration. We can only get at these things via participant access. When
used responsibly, it is an experimental procedure, not simply introspection in
the sense of looking inward and describing what you find. In the so-called
commutation test, you exchange one sound segment for another (e. g. sit vs.
set), and then see whether the result makes a semantic difference in terms of
conscious understanding and that does not entail an assumption that the
procedure for articulating vowels is consciously accessible. When you describe
syntactic and semantic variation, you look at a number of different examples
and describe your conscious understanding of them and then relate your
conscious understanding to properties of the linguistic input. For semantic
properties, more may be accessible to participant consciousness than in the
case of vowels but we have to live with the fact that underneath the con-
sciously accessible level there is always an ability level where the object drops
out of sight. The difference between empirical and pure armchair linguistics is
not the use or non-use of this procedure without it, linguistics would simply
be impossible. The difference is in how much empirical input you bring to the
intuition-imbued experiment from large scale empirical data collection to
describing the object literally off the top of your head. Intuition, i. e. the use of
consciously accessible participant representations of the meaning of linguistic
expressions, is indispensable at all points on this cline.
206 Chapter 5. Meaning and ow

On very preoccupied days, I can get home before conscious attention sets
in, and I may have no recollection of what happened during the trip
because all the decisions were delegated to the procedural system. If lots
of interesting things happen, on the other hand, the whole ride will be
filed away in conscious memory. Only the actual decision to start off the
whole complex procedure needs to be consciously available (in order for
human rather than zombie activity to have taken place).19
Competency in handling conventional meaning therefore has to be
defined in relation to the whole span from the almost totally inaccessible
procedural choices to the fully accessible output meanings. When I say
almost totally inaccessible, it is because the link to consciousness consti-
tutes the borderline between meaningful and zombie activity. Lakoff and
Johnson (1999: 17)s description of the amoebas ability to distinguish
between food and toxic substances as an instance of categorization can
only be correct if the amoeba has conscious access to the result, and this
conscious awareness may drive feedback to later triggering. Most people
would side with Johnson-Laird (1988: 24) in putting this down to purely
hard-wired responses. A meaningful choice potentially triggers feedback
of the kind that defines functions, and the stored feedback from the out-
put to the shaping of future response is therefore linked up with meaning
in a way that makes it part of semantic competency. The motor procedures
in themselves are not semantic but the loop from conscious output back
to the conditions for triggering action subroutines in the future, and eve-
rything that takes part in that dynamic process, are constituents of seman-
tic competency from a functional point of view.
In order for semantic categorization to be involved, there has to be a
degree of interplay between consciousness and purely neural wiring. This

19 In previous sections I have argued that conventional meanings are more

directly related to the input end of the interpretive process than to the output
end that consists in the completed interpretation. But if only the output end is
guaranteed to be fully accessible to consciousness, how can semanticists get at
conventional meaning?The answer is that they are left in the same position as
ordinary speakers: they have to infer (and abduct, cf. Andersen 1973) the
potential-for-use via concrete instances-of-use. Semanticists have a task that is
in one sense harder, since they have to make their description explicit (in spite
of the fact that the meanings they describe are not inherently fully explicit
mental entities). But that is no different from any other discipline: molecules
are not fully explicit mental entities either. Linguistic categories are meta-
level containers, designed to capture features of the described object. They
are not identical to categories inherent in the described object. Not even lin-
guists should confuse the map with the landscape.
The procedural nature of competencies 207

is not necessarily always the case even in human beings. If you respond to
a tap on your knee by jerking you leg, it is not an act of categorization; and
this carries over to ways in which you respond to objects around you in
everyday life. From when I was six to when I was about eighteen I disliked
a subset (but not quite a conceptual category) of fruits. The reason was
that at six I had stuffed myself with a kind of not-quite-plums from a tree
on the way to school to such an extent that I was violently sick, and like
rats in Skinners experiments I steered clear of anything remotely remi-
niscent of them. The category (which gradually shrank over the years)
was defined not in terms of conventions in the niche, but due to gut feeling
(quite literally). But of course I had conscious awareness that there was a
type of fruits that I avoided and thus there was a thin conceptual layer
on top of the gut response.
This can be used as a stepping-stone to responses that form part of
Bourdieus area of habitus (cf. p. 110 above), linked by Searle himself to
what Searle (1992: 177) calls background. Searle points out the link back
to Wittgenstein, whose concept of forms of life also gets at the non-rep-
resentational background for meaning. The background contains all
embodied states such as feelings and routines that work by purely proce-
dural skills, and especially all states of being attuned to a world with par-
ticular features such as having a walking skill that depends on the ground
remaining solid under your feet. Bourdieus habitus is designed to capture
especially the kind of attunement that is the result of social pressures.
Such attunement is caused by a mixture of direct impact (acts of physical
oppression and punishment, for instance) and adaptation: if they prick us,
not only do we bleed, we also adapt our reactions so as to avoid future
prickings, thus developing embodied responses caused by social oppres-
sion. The posture of a serf in the presence of his master might be inter-
preted as indicating respect, but on a habitus interpretation it is merely
an entrenched bodily response to a potential danger. It is tempting to call
this secondary embodiment to distinguish it from the kind of embodied
responses that develop spontaneously as a result of normal human matu-
ration but since all human responses are the result of interaction between
the individual and environment, it is difficult to tell the two apart in prin-
Is habitus-driven activity zombie behaviour? Well, yes and no. It repre-
sents a response to life in the niche that is attended with conscious aware-
ness at the margins an awareness that may expand in ways that will be
discussed later (p. 316). Well-entrenched routines that are usually dele-
gated to the neurocognitive autopilot system and thus recede into the
background may sometimes erupt into consciousness, for instance in
208 Chapter 5. Meaning and ow

problematic situations. The social processes that in Bourdieus theory gen-

erate habitus as an unconscious set of bodily dispositions work via proc-
esses that also have consciously accessible aspects (such as education).20
A recent line of investigation which is beginning to throw light on this
issue is the investigation of embodied meaning via body specificity by
Daniel Casasanto (Casasanto 2009). The association whereby the domi-
nant side is conceptualized as the good side is reflected in widespread
associations such as right contrasting simultaneously with left and with
wrong, with the negative association of words for left such as sinister
and gauche, etc. There is no way to tease apart the cultural and the embod-
ied dimensions in explaining why up is good (because the predictions
would be the same, whether you saw it as biologically or culturally
grounded) but the existence of left-handedness offers a chance to see
whether there was a difference in embodied and cultural patterns. And
sure enough, in a series of experiments, Casasanto and associates have
shown that there is a significant bias correlated with the handedness of
subjects when it comes to what is the good side in a series of arbitrary
and balanced choice tasks, items came out as good in ways that reflected
a reverse bias in left-handed persons, such that items were evaluated more
favourably if they were on the good side. (The pattern extended to ges-
tures by US presidential candidates, who uniformly put problems on their
bad side!).
One would assume that such reverse response types in left-handed vs.
right-handed subjects are at the level of what you would call gut feeling:
they defy cultural patterns and show the robustness of responses grounded
in your own individual body. In interactive negotiation of meaning, the
bias in terms of categorisation into good and bad has to interface with
other peoples personal biases and hopefully be overruled by public and
joint deliberation in cases involving choice: Casasanto posed the question

20 This raises a methodological problem for experimental approaches to concep-

tualization: How do you tell conceptual categories from embodied, but non-
mental response patterns? There is a long history of formulations that blur this
distinction, as when Cheney and Seyfart (1992) suggest that vervet monkeys
would be excellent primatologists (with their precise awareness of the features
being studied), Harris (1998) talks about each speaker being his own linguist,
Chomsky uses the term grammar about both what is in the mind/brain and the
grammar book, etc. However tempting it is to sit comfortably down on the
fence, ultimately we need to devise ways of telling whether what we find is an
operational skill or a conceptual representation.
Usage, competency and meaning construction 209

of how much depended on which side of the bosss desk your rsum
ended up on.
Work in progress (Casasanto, personal communication) suggests that
in some experimental conditions you are more likely to respond based on
cultural patterns and in others the embodied responses win out. This is a
neat example of how the investigation of divergence and convergence
goes hand in hand: body-specific experience gives rise to convergence
with personal categorization on the one hand (as it were) and divergence
in relation to cultural categories on the other.

5. Usage, competency and meaning construction

The emphasis on sprawling semantic potentials, on dynamic meaning con-

struction in context, and on non-semantic skills all entail a development
away from neat mental representations and towards a larger and more
varied universe of structures and processes. However, I now turn to the
question of whether and to what extent we can still see the meanings
of linguistic items as separate objects of description. The emphasis I put
on competency meanings as a separate object of description also entails
emphasis on this possibility. I will first discuss the position of Croft and
Cruse (2004), then the position of Taylor (2006).
The following quotation from Croft and Cruse (2004) can serve as a
point of departure for the role I see for the instructional perspective.21
In many approaches to meaning, there is a determinate starting point for the
process of constructing an interpretation, but an indeterminate end point (.)
The present model of comprehension has an indeterminate starting point (a
purport) and a determinate end point. ().
Each lexical item (word form) is associated with a body of conceptual content
that is here given the name purport () purport is continually developing:
every experience of the use of a word modifies the words purport to some
degree (p. 100)
It is by a series of processes of construal that an essentially non-semantic pur-
port is transformed into fully contextualized meanings (p. 103)
The extracts make a valid point, but I am going to argue that a balanced
view of meaning also requires a sense of the term meaning that applies to
the input stage.

21 A more detailed version of the argument is found in Harder (2007b)

210 Chapter 5. Meaning and ow

Word meaning is indeterminate outside of an actual utterance in the

sense that the potential in its entirety is rarely relevant for any actual
utterance, and we cannot know in advance what area is central in relation
to a particular case. More senses are evoked than are actually used (as in
the case of bug explored by Swinney 1979). But it is not surprising that
you cannot know a precise utterance meaning except in relation to an
actual utterance. In a sense, the indeterminacy of the potential that is
extracted from utterances for future use is the whole point of having a
human language as opposed to a pre-determined set of calls, cf. Deacon
(1997): meanings in human languages are symbolic rather than situation-
Input meaning emerges from the whole range of use, rather than
being definable as invariant essence. Once it has emerged, however, it
constitutes something else than merely the sum total of actual usage
events. Langackers term centrality (1987: 159), closely related to what
Geeraerts discusses as salience (cf. p. 67), captures what happens on the
path of emergence from raw usage to (input) meaning: certain semantic
properties are highlighted at the expense of others, and knowing a word
entails knowing what features you centrally invoke when you use it. Cen-
trality is thus a good way to allow for gradual and subtle differences. But
there are also absolute differences. The meaning of the word computer is
distinct from that of grow and that of dirty in ways that centrality cap-
tures only awkwardly. Speakers need to know those distinctions, and
describing them as indeterminate and essentially non-semantic does not
give the full story.
A similar point can be made in relation to the position adopted by
Taylor (2006). Taylor lines up the hard problems associated with the mon-
osemy-polysemy spectrum: if you go in for polysemy, how do you indi-
viduate meanings? If you go in for networks, how do you choose between
competing versions? His own proposal is based on a rejection of what he
calls the dictionary + grammar model and the assumptions about compo-
sitionality on which it rests. Echoing familiar CL assumptions about par-
tial compositionality, he suggests that the whole idea of starting with a
word and searching for its properties is mistaken. Instead, meanings essen-
tially belong in the very large inventory of multi-word expressions
which speakers have learned as such, i. e. in concrete usage situations.
Beginning at that end, he suggests, means that the semantic contribution
of component words will become less and less of a concern (2006: 63).
This approach has a whiff of usage fundamentalism about it: only
chunks of actual flow are real. But I would like to begin by emphasizing
an important and under-appreciated point that Taylor makes. A crucial
Usage, competency and meaning construction 211

process in the rise of word meanings is top-down differentiation: you

abstract the contribution of the word out of your understanding of a whole
utterance. In the beginning was not the word, but the utterance (cf. Harder
1996: 265). Atomic elements, including word meanings, are not basic (cf.
also the discussion of Crofts Radical Construction Grammar p. 245).
Some words are more independent than others; but all words are some
extent syncategorematic.
For the more syncategorematic words, whatever you might wish to see
as the contribution of the individual words would need to be studied in
close connection with a study of the constructions they form part of, as
suggested by Stefanowitch and Gries (2003). Cases like all over and
cases like over the table would then come out with different indexes,
and thus with different status as descriptors of the word over as a unit in
itself. It is not an accident that the key discussion of polysemy has been
about a preposition. In terms of the theory proposed above, prepositions
have the job of locating entities, not of assigning them to a conceptual
category. When you say the lamp is over the table, you are not primarily
engaged in subsuming an object under a general category by assigning it
to the concept over. Rather, in combination with its complement, a prep-
osition is used to locate an object in the concrete situation. In the hypo-
thetical extreme case where a word is totally submerged in a web of col-
locations, with no perceivable contribution of its own, a network diagram
of its senses would of course be a pure artefact.
In all other cases, the issue of contribution would remain. Even if
meanings of smaller units are abstracted out of larger units, speakers still
need to know how to deal with those smaller units. The question of
whether it makes sense to look for the semantic contribution of a lin-
guistic expression is orthogonal to the question of what size unit you
choose. I think Taylors examples are persuasive, as far as they go: one
would not want to base a description of the expression all over on its com-
ponent parts, for instance. But the issue of semantic contribution also
applies to the whole multi-word expression all over. And if with Taylor we
take the next step to larger expressions such as all over the paddock (Aus-
tralian/New Zealand usage), the question remains. The only way you can
avoid the semantic contribution issue is by going all the way to whole
utterances with fully contextually specified meanings a list of actual
utterances, in other words. If we follow the argument against the concept
of semantic contribution all the way to that conclusion, it rules out mean-
ing construction (on which point it differs from the position of Croft and
Cruse as discussed above). Expressions learned as such are retrieved, not
constructed. Meaning construction requires semantic contributions.
212 Chapter 5. Meaning and ow

The instructional and directional approach offers a format which pro-

vides a clear distinction between usage based and usage fundamentalist
assumptions. Meaning as something attached to linguistic expressions
would not be possible without a process of abstraction not even for rote-
learnt holophrases like hurrah or fuck. This is a direct consequence of the
domestication process that deprived hominids of most innate calls but
made it possible to invent expressions with meanings that are not directly
triggered. Without abstraction abilities, it would not work (cf. also Stjern-
felt 2007 for a pursuit of this point).

6. Conceptual categories and the flow. Messy

and precise semantic territories

Above, I have argued that concepts are better understood as ways of trig-
gering an act of categorization than as a static construct but that in order
to understand what a concept can do for the speaker, it is necessary to
understand the input to the act of categorization as a determinate entity,
even if it is not as neat as Aristotle assumed. We have seen that there is an
issue in usage based linguistics with respect to how or whether one can
postulate meanings that exist in abstraction from usage; and in this section
this debate is taken up with respect to what used to be the centrepiece of
discussions about mental content: conceptual categories as such. As
pointed out by Langacker (2001: 11), the equation between meaning and
concept in the countable sense is not an assumption that is generally
taken for granted in CL: the uncountable and process sense of conceptu-
alization is more basic.
The role I see for a concept does not apply to all linguistic meanings. But
for certain meanings, I think it makes sense to understand a word meaning
as being directly linked to a mental concept. For example, I think it makes
sense to say that a central function of the expression dog is to evoke the
concept of dog. In this respect, I see the word class of nouns as having a
privileged position (for reasons which will be discussed in greater detail in
ch. 6, p. 257). The function is especially central when the noun stands as
head noun in a complex NP, which is the grammatical slot for indicating the
category to which a designated entity belongs: a Japanese American is an
American, while an American Japanese is a Japanese. Logically enough, in
discussions about the appropriateness of different concepts, it is tradition-
ally nouns that are considered. Categorization as a sub-act thus has a privi-
leged association with nouns (while verbs predicate and prepositions
locate, etc). Outside the head position, it becomes more complex: a com-
Conceptualcategoriesandthe ow.Messyandprecisesemanticterritories 213

pound such as computer science, for instance, does not directly classify an
object as coming under the category computer, and therefore it evokes a
wider part of the potential, including the activity of computing, than the
point-blank categorizing statement this is a computer.22 The following dis-
cussion presupposes that we are talking about head nouns.
Part of the functional utility of concepts as part of an individuals cop-
ing skill depends on concepts being adaptable to the categorization tasks
that they are called upon to assist with. It follows that concepts are not
immutable over time, even though they preserve their identity as line-
ages in Crofts terms (cf. p. 89). In terms of the framework of this book,
this is understood in terms of the evolution-inspired concept of function:
each act of categorization impinges on the concept which adapts to the
process of use. It is stable in a relative sense, however, in that it persists
between events, and is never reducible to any actual instantiation. The
first time I saw a red-skinned potato in my fathers garden, it expanded my
concept of potato without destroying the concept that was there before.
Concepts are useful precisely because they reduce variation by subsuming
individual instantiations under general categories.
The dynamic coping context can throw some light on the discussion of
classical Aristotelian concepts in relation to prototypes and family resem-
blances. From a functional perspective, Aristotelian concepts constitute a
maximally simple and reliable solution to the abstract question of catego-
ries a neat separation between properties to abstract away from and
properties that make a difference when we move from instantiation to
concept. Simplicity and reliability are important when you are interested
in the properties of the categories for their own sake, as philosophers and
scientists are: for them it is central to get the categories under control,
whereas dealing adequately with a single specimen can be entrusted to
the research assistant.
However, as discussed above on p. 17, the everyday functional mecha-
nism of dividing up the world based on what is relevant to the individual
does not have to work in such a tidy manner. For practical purposes we

22 A full discussion of this issue would take us beyond the scope of this book.
Part of the problem involves the issue of what exactly a concept is including
the question of the relation between concepts as parts of mental competency
(what is discussed in this chapter) and concepts as part of the niche (cf. ch 7,
p. 318 f). The claim made here is only that (in languages like English), nouns
inherently have a semantic element that links them to a category, which makes
the link closer than for other word classes, and that this property is partly
shared with so-called classifying adjectives, as in musical (instrument).
214 Chapter 5. Meaning and ow

choose categories that we find useful how precise and reliable they need
to be depends on the circumstances. In the crime story that I am reading,
the stereotypical stupid policeman complains about all the different cat-
egories of modern forensic psychology and says that the category nut
was good enough for him. Necessary and sufficient conditions are of no
interest to him. We may disagree for reasons of principle (political and/or
scientific), but this is beside the point if we want to understand the role of
categories in his everyday life.
For the same (functional) reason, the properties of a concept do not
have to match the properties of the instantiations precisely. In fact,
demanding a one-to-one match would ruin the whole point: there have to
be properties that we disregard when we align two different instantiations
under the same category otherwise the concept as such would duplicate
information that was already in the instantiation and be functionally
superfluous (cf. also Langackers use of the categorizing relationship dis-
cussed above on p. 50). When you grasp experience by means of concepts,
your grasp will be no more precise than the actual practices that it arises
from. Neatness and consistency are potential encumbrances, and fuzziness
is just a practical problem.
There is an orientation in the CL discussion towards local forms of
organization those that are as close as possible to the individual event.
On a path that starts out with classical concepts, you can then move via
prototypes to family resemblances and all the way down to a swarm of
individual exemplars, cp. Croft (2007), and this path can be seen as mov-
ing gradually closer to reality, away from abstract idealization. This is a
natural conclusion if you look for cognitive organization in the mind of
the individual, down to neurocognitive traces of individual events. How-
ever, if you view conceptual constructs as part of a social practice, you
have to look at the functional dimension. If we assume that linguistic
practices are responsive to selection pressures, what forms of conceptual
organization can this give rise to? Before we answer the question in any
detail, we have to begin by looking at the different kinds of practices that
are involved.
Roschs original distinction (1975) between basic-level concepts and
superordinate concepts already implied an important adaptive point: only
those categories that enter into everyday life on a regular basis develop
prototype effects. It takes time and processing power to build up a con-
ceptual container organized with reference to a selected cluster of proper-
ties, with a centre and a periphery. Unless you have a sufficient number of
instantiations and these are felt to be worth the effort, you simply do not
bother. For more abstract or schematic categories we therefore develop
Conceptualcategoriesandthe ow.Messyandprecisesemanticterritories 215

less rich representations, defined only in terms of a single or a few features

(cf. also the discussion of Geeraerts on diachronic prototype semantics,
p. 65).
As pointed out by Johnson-Laird (1983), these properties can be linked
to a functional division of labour that accommodates prototypes and clas-
sical concepts in different roles. In everyday life, there are many cases
where it only makes sense to deal with instantiations in a rough-and-ready
manner. Food, traffic, types of people for most purposes there is an obvi-
ous relation between the approximative nature of satificing and the
approximative nature of categories with more or less central members,
just as with the concept of rule of thumb. Basic-level concepts get their
rich but imprecise nature as containers for the same reason that our
kitchen shelves end up with somewhat heterogeneous stuff on them: they
may not fit too well, but you dont know where else to put them and you
cannot stand there forever before dealing with the next thing that draws
your attention.
On the other hand, there are cases which call for an either-or type of
decision. Johnson-Laird mentions legal distinctions as a case in point. If a
particular crime renders the culprit liable to a term of imprisonment of
two years, it is problematic to say that his act falls roughly into that cat-
egory: in order to put him away, we would like to get a clear answer is
this grand larceny or what? Medical diagnoses are another case where
precision may be important. A doctor may respond to symptoms by
assigning the category meningitis, which calls for a somewhat drastic
response, or she might not. The dilemma is there: tertium non datur. If it
turns out later that she did not assign the category and should have, a cri-
tique of the Aristotelian theory of concepts is no help.
This is similar to the special type of practice that consists in scientific
description. Aristotle was oriented towards clear-cut containers because
he wanted to get deeper than everyday rough-and-ready categories. Sci-
ence, like the law, is also a normative practice: you want to put the speci-
mens in exactly their right place. Thus it makes sense to see if you can set
up new categories that are clearer than the everyday ones for just that
Both classical categories and prototypes have a property that distin-
guishes them from family resemblances and exemplar clusters: they
involve a significant reduction of complexity. The category bird, which is a
prototype for many people, covers close to 10,000 different species in the
world, with untold numbers of specimens all together. If the generaliza-
tion associated with bird is significant for practical purposes, then all the
other differences may be irrelevant, once you have put something in the
216 Chapter 5. Meaning and ow

bird category (i. e., it is not a bomb, or a plane, or a childrens kite). Cas-
sowaries in the Karam Papua New Guinea community, on the other hand,
do not belong under the category, because their significance (as a large
ostrich-type species) is totally different from the significance of all other
kinds of bird (cf. Bulmer 1967 as quoted in Aitchison 1994: 65).
Family resemblances are thus less useful as the basis for categorization:
you cannot infer very much significant information from them. They
match the cases where the cohesive link of the term is fairly feeble, and
the centripetal forces are correspondingly strong. The extreme end of the
scale is the view of categories as constituted by a cluster of exemplars. This
approach is interesting, because from the perspective of direct relation to
psychological reality it is so attractive. As discussed by Goldberg (2006), in
non-linguistic literature on categorization evidence has accumulated that
people in fact not only store exemplars, they are also able to extract statis-
tical information from the collection of stored exemplars. This is why the
view that categories are equal to a collection of stored exemplars until
very recently dominated work on categorization in cognitive psychology
(Goldberg 2006: 46). Yet a pure collection view leaves the whole issue of
generalizations unaccounted for. As Goldberg points out (2006: 47), quot-
ing Ross and Makin (1999: 8), it takes away the categoriness of catego-
ries: only instantiations remain.
The functional properties of this cline of variability can be demon-
strated in mathematical terms. Warglien and Grdenfors (2007) have
shown that as long as their domain in conceptual space is convex, concepts
can start out as rather different in the minds of language users, and still
enable a process of homing in on the same meanings. This is because suc-
cessive rounds of attempted matching, in the course of working out an
interpretation, will make them converge on the same fixpoints. In real life,
the process can be understood in terms of Piagetian development with its
twin mechanisms of assimilation and accommodation (cp. p. 71): when-
ever a term is applied to an instantiation, it brings about an adjustment
both of the term and of the perception of the instantiation.
Of course the individuation of conceptual territories depends on vari-
ational and pragmatic circumstances also for nouns. I am not saying is
that there are really eternal invariant concepts after all: the precise nature
of the container depends on the circumstances, as all other forms of mean-
ing do. What I am saying is that noun (input) meanings are for putting
tokens in containers, and the meaning potential of a noun therefore
includes containers. This means that concepts are part of speaker compe-
tencies, even if there is no one-to-one relationship. The (niche) word dog
may designate the species or the male only (dog vs bitch) and that means
Conceptualcategoriesandthe ow.Messyandprecisesemanticterritories 217

that there are two well-defined potential containers in the semantic terri-
tory of the word. The more of those a speakers has, the more competent
she will be also for purposes of language use. If you do not happen to
know the male sense, a conversation about dogs and bitches would be
hard to follow.
Verbs are less categorial than nouns, and less sprawling than preposi-
tions (reflecting their position in the syntactic hierarchy, cf. p. 257). In gen-
eralizing about verb meaning, one should not look for containers, but for
elements that can be predicated about tokens in them.This bears on Tay-
lors argument against Searles (1983: 145f) well-known defence of mon-
osemous literal meaning for a verb like open. Searle claims that we know
perfectly well what a phrase like open the sun would mean we just dont
know what specific action it would refer to. Taylor, in contrast, suggests
that the only meanings we really know are the meanings of entrenched
combinations such as open the window, and the word open it itself has no
independent meaning apart from such usage-entrenched cases. It appears
that Taylor is to some extent presupposing a more rigorous view of seman-
tics than a cognitive linguist would adopt in other contexts. As an objec-
tion to Searles claim that we can know the meaning without knowing the
truth conditions or entailments, he argues (2006: 54):
This raises questions about the usefulness of unitary representations. If they
are not involved in determining truth conditions or entailments, what are they
for, and why should we be bothered by them?

While this argument might be the expected one from Lewis or Davidson,
it is not in the general spirit of CL in its increasingly social and dynamic
phase. Verb senses vary with different types of landmark elements and
this is part of the reason why it is difficult to define a truth-conditionally
unitary conceptual category for the verb open. The relevant form of
abstraction comes easier if you look for what the verb could predicate
about its landmark. One might tentatively propose (borrowing Talmys
concept of barrier) a core aspect of its meaning potential that involves
the removal of a schematic barrier at the boundary of the object nominal,
making its interior accessible. The nature of the barrier would then be free
to vary with the object in question (window, meeting, bank account, etc).
Sometimes it would not be easy to think of how that barrier might look or
what access to the interior would imply but that would be a task for
meaning construction, not an objection to looking for an input meaning
that is as well-defined as the data allow.
Precision, of course, also has a pragmatic dimension. As often pointed
out, you can reduce a prototype to a classical concept by defining the con-
218 Chapter 5. Meaning and ow

cept in terms of the prototype case alone and then treat instances beyond
the pale as errors in terms of the norm that you describe. A statement is a
lie, cf. Coleman & Kay (1981), if (1) it is false, (2) the speaker knows it is
false, and (3) makes the statement with the intent to deceive and benefit
from the deception. This leaves us with non-core cases like white lies
which do not confer illegitimate benefits on the speaker. For everyday
purposes it does not really matter whether they are viewed as marginal
cases of lies or as non-lies but there may be external social reasons why
the distinction has to be drawn sharply in actual cases. A vow of truthful-
ness, with sanctions in case of violation, is a narrative classic instance. For
the purpose we are discussing, i. e. adding an extra layer on top of every-
day criteria for conceptualization, it functions just like the scientific ambi-
tion to get the description 100 % right by imposing, through social pres-
sure, a higher degree of order on the world than we usually have time for.
This illustrates a dimension of offline competency that is at the same
time fine-tuned in relation to the actual situational flow. The speakers
conceptual apparatus for handling potential instances of lies can be
adjusted to different degrees of strictness: the container function offers
more than a one-size-fits-all construct. This is a very non-Platonic prop-
erty of semantic competency; but it does not inevitably translate into
sophistic irresponsibility. A lawyer who is unable to adjust his legal cate-
gories to the actual circumstances of the case is not going to make it to the
Supreme Court. The situational adjustment capability, whether deviously
pragmatic or ineptly inflexible, is in all cases understandable only as an
offline property of conceptualization. If we had only the flow of actual use
of concepts, there would be no possibility of even raising the issue of flex-
ibility, ineptness or honesty.
The same basic division into online dynamics and offline resources for
coping can be applied to other basic notions of CL. It would be too cum-
bersome to go through the argument in detail, but let me briefly outline
how the account can be generalized.
In addition to the basic, non-count stream of conceptualization we
have two kinds of units of representational content, concepts and ideal-
ized conceptual models, and two kinds of presupposed background-for-
concepts, domains and frames. We also have two kinds of locations for
conceptual content: one is the default location of the current discourse
space, the other arises when there is an alternative to the basic reality
space, and we thus have two mental spaces to allocate meaning to.
In dealing with an instance of conceptualization, all these can contrib-
ute something distinct to the understanding of what goes on. If there is a
family problem, for example, the concept mother singles out someone as
Summary 219

an instance of a salient category, while an evocation of Lakoffs nurturant

parent model brings a particular idealized conceptual model into play.
Understanding may also be abstractly considered in terms of the generic
domain of family relations. Finally, the person may be framed against a
presupposed model of classic (or postmodern) family structure, evoking
possibly an alternative mental space in which things would have been dif-
In order for all these operations to be available to the speaker, there
has to be an offline store of evocable mental constructs. But their basic
mode of being is when they are actually used as input to ongoing linguistic
interaction, and this is when they serve their functions which I claim is
their true identity. Functionally speaking, concepts serve to reduce incom-
ing experience to recognizable kinds, while ideal cognitive models provide
partial default models of how the world works, enabling agents to shortcut
the process of figuring out what the world is like solely based on the actual
usage situation a slow and error-prone process. Framing gives scope for
selecting different parts of reality as the background for understanding
what is on the agenda, with implications for what acquires presupposed
status. Mental spaces, finally, enable speakers to escape from the tyranny
of thinking only of what is currently accessible.
In this dynamic perspective, we thus retain the rich inventory of con-
ceptual resources that was discovered by classic CL but we view it as
constituting the offline sediments of dynamic online activity, and with a
complex relationship between purely procedural and habitus dimensions
and consciously available conceptual dimensions.

7. Summary

This chapter discussed the relation between two of the three aspects of
language introduced in chapter 4: flow and competency. It did so with ref-
erence to meaning rather than structure, because it would be unwise to
take on two confrontations at the same time: structure vs. cognitive con-
tent, and mind vs. usage. In the case of meaning, the question of offline
properties of language is less contaminated, since no one seriously doubts
that members of the speech community have something that they carry
around in their minds between utterances, which includes an ability to use
a number of different words.
In a usage-based perspective, flow and competency meanings have the
reverse relationship of the inherited pattern of Platonic tradition. Instead
of a collection of eternal ideas that are soiled by actual use, the flow came
220 Chapter 5. Meaning and ow

first and then members of the speech community developed the ability
(competency) to take part in it. The meaning of the whole enterprise is in
the flow, too: competency meaning is just the prerequisite for getting at
the meaning in the flow. And competency meanings are shaped so as to fit
into the flexible process of meaning construction that mediates between
the coded input and the output that is always necessarily constructed
online as reflected also in the literary theory of interpretation..
This perspective situates speaker competency rather differently from
the tradition, and therefore it involves different claims about what the
offline properties in speakers minds are. For instance it suggests that
meanings associated with words are in between the indeterminacy of
floating signifiers and the rigidity of Aristotelian categories. They are
untidy in the way things are untidy in a working environment but not so
chaotic that no work can get done.
Another implication is that language competencies have a considera-
ble proportion of non-mental elements. Abilities are never accessible to
conscious awareness, and without the ability dimension there would be no
competency, only unusable representations. The time-honoured concept
of the speakers knowledge of language, and with it psychological real-
ity, have to be abandoned as constituting the unitary object of description
for CL. A large chunk of the reality is not psychological at all, and knowl-
edge has to be unpacked into two entirely different things, unless cogni-
tive linguists want to make the same error as Chomsky: to postulate a
mental object that can never be accessed by the speakers mind. The two
different things are (1) knowing-how routines (such as lexical retrieval
and phonetic articulation) which can be modelled without raising the
spectre of the Chinese room because in themselves they do not need to be
interpreted, only to get done; (2) intentional representations associated
with words. The relationship between them are functional: it is the job of
the procedural ability to produce utterances to evoke representations. But
a description that flattened the whole object out into knowledge of lan-
guage would be like having an exhibition of banks that included both
river banks and commercial banks.
The new concept of competency was also used to argue against the
tendency to focus so strongly on the actual flow that the competency
dimension was too much reduced as in case of proposals suggesting that
words in the mind had no meaning, or that there was no semantic contri-
bution from individual words. The key idea was to see the competency
side of meaning as input to meaning construction: speakers have to know
what words can do for them. Meaning construction then produces output
meanings based on the competency input plus the situational circum-
Summary 221

stances (including acts of framing), which entails that output meanings are
different from one utterance to the next, because words do the jobs that
situations call for. So while it is true that words do not have output mean-
ings, i. e. meanings in the sense of message elements, and these are the
meanings that everybody else but linguists are interested in, precisely for
a linguist it is practical to assume that words have meaning.
The role of offline meanings as potential contributions to dynamic
online meaning construction also entails that meanings are directional
and may give rise to presuppositions. But they also retain their link with
the classic role of evoking concepts for the purpose of putting aspects of
reality into situationally relevant containers.
Because the division of labour between implicit and explicit dimen-
sions of action is variable and hard to track, categorization in the fully
mental sense thus has a variable and hard-to-capture relationship with
autopilot responses to situational input some of which are shaped by
social processes of the kind that generates Bourdieu-style habitus pat-
In the next chapter we look at structure and variation, and turn our
attention more in the direction of the niche dimension of language.
Chapter 6. Structure, function and variation

1. Introduction: the social foundations of structure

Meaning was the first topic to be addressed in terms of the new social
cognitive framework. Now the turn has to come to structure. Because the
account of structure builds on the account of meaning, this chapter contin-
ues some themes from the preceding chapter. The distinction between
offline and online features of language, and the status of offline features
as ancillary to the flow, are also essential for understanding structure. For
the same reasons that ruled out Platonic meaning as the basic object of
description, we also have to rule out underlying Platonic structure.
However, the issue of structure involves additional complications of its
own. The discussion of meaning addressed mainly the flow and the com-
petency dimension, but in order to understand structure a closer consid-
eration of the niche dimension is necessary. It is not individual mental
content that is responsible for the way language is structured: language
structure exists in the community before it is acquired by an individual
newcomer (and we were all newcomers once).
The special attention devoted to structure may appear to be a surpris-
ing development. The social turn was introduced as a continuation of the
process of recontextualization that began when CL moved beyond purely
structural description. Why, then, do we suddenly run into structure in
making a social turn?
The reason is twofold. First of all, there is a link between the social
dimension and structure that has been consistently underemphasized in
the tradition. Secondly, in order to understand those functional and vari-
ational aspects of language that are more usually associated with the
social dimension, a revised understanding of structure is not a step in the
wrong direction, but a precondition for moving forward. The traditional
association between social facts and chaos distorts both the understand-
ing of structure and the understanding of meaning and usage. This means
that if it is used as a frame for understanding the place for structure in
a social CL, the recontextualization narrative involves a risk of distor-
tion. Putting pure structure into the dustbin of history means that struc-
ture as a property of language has to be reconsidered. Otherwise you
would misunderstand not only structure, but also usage. An adequate
usage-based understanding of structure is necessary in order to get an
Introduction: the social foundations of structure 223

adequate understanding of functional and variational properties of lan-

Two issues are crucial to this updated understanding: the functional
and the variational dimension of linguistic structure. I address them in
that order, which follows the logic of the overall argument in the book.
The understanding of the functional dimension of structure (cf. section
3) is closely linked to the account in the previous chapter of meaning as
input to interactive flow. The main point is that the structure of complex
expressions must be understood as the structure of a complex act, i. e. a
structured intervention in ongoing interaction. Just as the meaning of an
individual expression is constituted by the job that it does for the speaker
in the usage situation, so does the structure of a complex expression con-
stitute a structured set op operations to be performed on the usage situa-
tion. The structure of language inherits general features of the structure of

1 The Platonic idealized cognitive model of the relationship is manifested in

Carnaps (1942) influential account in which pure formal structure is in the
centre, semantics can be added as the next layer (if you ingiclude what words
stand for), and then finally you can include pragmatics by adding language
users to the picture (thus letting in variation and flux).
This conceptual model has been widely used as the frame for discussion
about relations between these areas, also by non-formalists, who mistakenly
try to change the emphasis instead of rejecting the model. Semanticists have
emphasized descriptive meaning and downplayed the purely formal core.
Pragmaticists have emphasized communication and downplayed the role of
both formal structure and descriptive meanings. But as Lakoff has often
pointed out (e. g., 2004), by invoking a model as the frame of the argument, you
tend to reinforce it, even by arguing against it. From a usage-based point of
view, this model is not just misleading, but dead wrong.
First of all, moving from structure towards use gets it backwards: beginning
with formal structure only makes sense if first formal structure and then
semantic structure can be understood on their own, isolated from use. But on
a usage based account the opposite is true. When it comes to structure, I am
thus echoing the proverbial Irishmans reply to a question asking how to get to
Dublin: If I was going to Dublin, I wouldnt be starting from here.
Secondly, the model places descriptive meaning as the middle man between
structure and usage. This gives the impression that social forces are at the most
remote position from linguistic structure, while descriptive meaning is its clos-
est potential ally. But there is no privileged relation between descriptive
meaning and structure. It is a basic tenet in CL that the conceptualizations that
are evoked by language are not specific to language. Rather, language draws
on everything that is available in the cognitive system, and therefore specifi-
cally linguistic structure cannot emerge from conceptualization alone.
224 Chapter 6. Structure, function and variation

complex tool use, cf. Greenfield (1991), and this procedurally imbued,
functional dimension is not reducible to (but integrated with) the concep-
tual aspect of linguistic structure (cf. also the discussion of Krifka (2007)
on the link between topic-comment organization and manual co-ordina-
tion, p. 203 above). Since this places language structure as part of the abil-
ity to act, of the competency of speakers to cope in a particular niche, an
understanding of this dimension is central to the social turn as I under-
stand it.
Accounting for the variational dimension of structure (cf. section 4)
shifts the main focus from speakers-in-the-niche to language as a constitu-
ent of the overall social niche itself. In this, it is closely linked to the
account in the following chapter of the role of language in constructing
social reality. Here, too, it is necessary to highlight the role of structure in
a new way, in order to illustrate the way in which an account of the social
significance of linguistic choices presupposes an awareness of structure
itself as embedded in social reality. For that reason also, a reassessment of
structure is a necessary part of the social turn.
It is a problem for carrying out this task that although structure is
clearly an integral part of CL, as a theoretical issue it is a contaminated
area: for cognitive linguists, it goes against the grain to focus on the nature
of linguistic structure. Since the account below includes suspicious con-
cepts like partial autonomy and arbitrariness, there is a risk of past mis-
takes creeping into the perception of what is being claimed. Care must be
taken in demonstrating how an expansion into the social domain can at
the same time introduce a more profiled role for structure and also for the
dynamic and variational aspect of language. I am therefore going to give a
brief outline of the argument in order to link up these three aims and pro-
file them in relation to classic CL, hoping to avoid such misunderstand-
In classic CL, complex structure is understood basically in terms of
complex conceptualization. A syntactic combination of two units basically
inherits the conceptualizations of the individual units, while the combina-
tion itself also has unit status (often with properties not found in the con-
stituent parts); the whole of language can therefore be described as a col-
lection of units. In the approach presented below, this is understood as one
of the two dimensions of linguistic structure, which I will refer to as the
bottom-up or unit-based dimension; it is central to construction grammar,
the dominant approach to syntax in CL.
As discussed in ch.1 (p. 49), the distinction between bottom-up and
top-down can be understood in more than one way. In relation to genera-
tive grammar, CL differs by taking the (bottom-up) path from actual
Introduction: the social foundations of structure 225

instances to generalizations, where generative grammar works (top-down)

from general rules to instantiations. In another sense, the distinction can
be understood in relation to units as opposed to larger combinations,
where the bottom-up path goes from units to combinations, while the top-
down path begins with the larger whole and then subdivides it into smaller
constituent types and here, classic CL is mainly but not exclusively bot-
tom up, cf. the discussion on p. 49. The issue is further complicated by the
fact that in addition to the descriptive procedure that I have just discussed,
there is also an issue of bottom-up vs. top-down directionality associated
with acquisition (as discussed in relation to Tomasello p. 265), and with
processing (cf. the discussion in relation to Functional Discourse Gram-
mar, cf. p. 259). There is thus considerable scope for confusion and misun-
derstanding. In order to keep both to a minimum, let me try to summarize
what I see as the point, and why this does not entail that this point is
absent but merely under-appreciated in classic CL.
The sense in which I use the distinction is very elementary, and based
on the case of addressing a complex task: building a cathedral, organizing
Olympic games or, as I am now, writing a book. There is no way of doing
this successfully without approaching it from two opposite and comple-
mentary perspectives: top-down means starting with the overall goal and
considering what it takes to achieve it, dividing it into major components
etc. Bottom-up means starting with the individual elements that are going
to be necessary. The difference is not just in terms of how big units you are
considering: at all levels there is a top-down approach based on what you
want to do: what is the goal or function that needs to be served. At the
very top, the first thing is to decide whether the goal is a cathedral, Olym-
pic games, or a book. At all levels, there is also a bottom-up approach
based on what you have available. From that perspective you start with
raw materials: bricks and marble for the cathedral, sports arenas for the
Olympic games, and a list of points and ideas to be considered for inclu-
sion in the book. The two ends have to meet, not only at the end, but most
of the time, since all (bottom-up) component units have to be evaluated
based on whether they are adequate for their (top-down) purposes.
In relation to linguistic structure, the top-down dimension that I think
needs to be highlighted more is the role of functional relations between
linguistic units. This works top-down because functions are defined in
relation to an overall purpose, and a unit considered in isolation cannot
have a function: for example, the unit Joe cannot be a subject, merely con-
sidered as a unit (cf. below). The (sub)functions served by linguistic units
thus have to be described with reference to the overall whole. As with the
cathedral and the Olympic games, this does not conflict with a considera-
226 Chapter 6. Structure, function and variation

tion that begins with the units you are going to need for your purposes
which (as argued below) can be described as a list of available linguistic
The claim that I want to highlight, therefore, is that language structure
is both a hierarchy of functional relations and a network of constructions.
There is no contradiction, only a difference of perspective. In classic CL,
the functional roles (such as trajector and landmark) arise as properties of
units (in this case, as properties of relational predications). Thus in follow-
ing a bottom-up compositional path, structural configurations arise, in
which units enter into relations with other units. Construction grammar
makes a point of this unit-based format. As illustrated in the diagram in
Croft and Cruse (2004: 256), the constructional reinterpretation of struc-
ture backgrounds the structural levels (phonology, syntax, and semantics
fade out of the picture) and instead foregrounds the inventory of units. As
you move upward in the hierarchy, more abstract units occur, which pro-
vide slots for other units to fit into. Whether you start with the units or the
roles/slots, in the end you have to include both sides but the perspective
you choose makes a difference for what you focus on (see section 2 below).
It may not be obvious that functional relations between linguistic units
have anything to do with functional relations between language and social
context. The connection is this: a whole utterance has a function in rela-
tion to the interactive context, as when Joe died functions as an informa-
tive statement, for instance as an answer to the question what happened?
In order for the whole utterance to function as an answer, its constituent
parts have to serve their functions inside the message. A central part of
those roles are that he refers to the subject (of predication), and died pred-
icates the relevant information about the subject. These two roles can only
be understood with reference to the whole utterance meaning, since a sin-
gle unit of meaning can neither be a subject nor a predicate. From this
perspective, slots/roles in relation to the linguistic context are special cases
of the slots/roles in relation to the social-interactive context (this is
explained in greater detail in section 3 below).
The interface between the larger social context and conceptualization
is also involved in the issue of variation. In terms of classic CL, linguistic
structure, like linguistic meaning, reflects conceptualization; and because
the locus for that is inside the individuals head (cf. Grdenfors 1998),
there is no obvious place for variation, which involves relations between
Based on these basic assumptions, the chapter will defend three claims.
The first is about function-based structure and is mainly competency-ori-
ented, the third is about variation and structure as a property of the niche,
Introduction: the social foundations of structure 227

and the second links up function-based structure with language in the

(1) On units and functional roles:
The bottom-up, unit-based dimension that is central to construction gram-
mar, must be understood against a more strongly profiled role for the top-
down, functional dimension. This manifests itself in the jobs that must be
done (= slots that must be filled) by linguistic units. From a purely concep-
tual, bottom-up point of view, these are simply abstract, schematic units
but from a top-down point of view they are functional roles.
(2) On structure as interactive options in the sociocultural niche:
Structure is a set of offline recipes for structured communicative action.
Online usage draws on these recipes in producing (co-constructed) situated
utterance meanings. This essential, but dependent role in relation to usage
means that the anti-structural disposition in CL (inherited from the revolt
against generative grammar) is no longer necessary. Structural description
in a usage-based grammar can be clearly distinct from usage description
(just as the recipe is distinct from the cooking) without being in conflict
with it.
(3) On integrating variation and structure:
Since structure in the sense of langue resides in the niche, not in the individ-
uals head, langue (= the whole set of available options including the lexi-
con) can include choices that are differentiated along variational dimen-
sions such as class, ethnic affiliation, gender, age, etc.
In all respects, I am building on developments that are already underway.
With respect to points (1) and (2), I am highlighting a dimension that has
recently been given a significant new boost by Langacker (cf. 2008b). The
link between function and structure, rather than specifically between con-
ceptualization and structure, is laid out in Langackers introduction of the
notion of functional systems. First of all, functional systems go beyond
the purely semiological function of symbolizing conceptualizations; sec-
ondly, they are characterized by members that fulfil specific functions,
including interactive functions two centrepieces of functionalism.
Thirdly, they are organized into mutually exclusive systems, thus giving
rise to closed paradigms a centrepiece of structuralism, especially in its
European version2. This new development illustrates how the achieve-

2 Langacker takes his point of departure in Chomskys (1957) analysis and

shows how the attractive structural features of Chomskys account can be cap-
tured better in an account that includes meaning which can simultaneously
explain some of features that appear ad hoc from a purely formal point of
228 Chapter 6. Structure, function and variation

ments of classic CL are being recontextualized in a way that also involves

structural features of language.
Specifically with respect to point (2), it may appear that in arguing for
a more profiled role for structure, I am arguing against a straw man: after
all, all linguists agree that structure has a role to play. But the point is not
to reiterate this platitude it is to show how the significance of structure
can only be adequately understood when structure is viewed as a socially
shared blueprint for language use. Moreover, the structure-promoting
point goes with an equally important structure-limiting point. The social
blueprint view of structure entails a debunking of the inherited. mono-
lithic view of structure: the language system is an open, partial order main-
tained by social pressures.
Point (3), finally, builds especially on work by the Leuven group (cf. the
discussion in ch. 2) and shows how the ongoing variation-oriented expan-
sion of CL must be understood as building upon a structural description
rather than rejecting it. The ultimate descriptive target is a structured sys-
tem, but one that includes the spectrum of variation that is available in the
The path of exposition starts with the relation between structure and
usage in a social cognitive linguistics, cf. section 2. Section 3 presents the
theory of the functional dimension of structure, as introduced above. Sec-
tion 4 discusses the relation between variationist linguistics and structure.
A major point is the expansion of the descriptive commitments of CL
that is entailed by a variationist understanding of langue. The section con-
cludes with a discussion with Croft, whose evolutionary synthesis (cf.
p. 88) has formed the background for the account in this book of the
relation between structure and variation. Section 5, finally, gathers the

2. Usage, structure and component units

The main point of the previous chapter was that meaning as a property of
human languages is too complex to be understood solely by reference to
online flow. If you ignore the role of offline competency, the full picture is
reduced to actual online events, which is tantamount to usage fundamen-
talism. Essentially the same argument applies to linguistic structure. A
usage based description is one in which actual usage is basic, but not the
only thing that exists.
In the literature on cognitive and usage based linguistics, it is not always
easy to see what the exact role for structure is. The question of the precise
Usage, structure and component units 229

nature of linguistic structure is not one that has kept cognitive linguists
awake at night: structure was yesterdays buzzword. Owing to the strong
emphasis on providing an alternative to the pervasive role of structure in
generative grammar, agreement has been much more well-defined when it
came to making clear what structure was not (e. g., formal and autono-
When you look for descriptions of the role of linguistic structure, there-
fore, you find a type of statements that have given rise to repeated misun-
derstandings in discussions with representatives of other linguistic schools,
because they come across as denials of the role of structure in language.
Let me quote a recent example (Glynn & Robinson 2008, p. 222 in the
book of abstracts)
Within cognitive linguistics, no distinction is held between linguistic and prag-
matic semantics or between lexis and syntax3.
In their contexts, statements of this kind generally make valid points such
as linguistic meaning does not constitute an autonomous domain, but is
integrated with encyclopaedic meaning; and syntax does not constitute
an autonomous domain inside language, but is integrated with the lexi-
con. However, because there is no canonical account of what the new,
integrated-but-real role for structure is, such statements are apt to be read
by outsiders as if structure has no other role than being an area that it

3 This is a very difficult discussion indeed, possibly also because cognitive lin-
guists take for granted both this way of talking and the descriptive practices
that (as a matter of course) make reference to structural categories in lan-
guage. I have previously (Harder 1996: 260) pointed to the discrepancy in con-
nection with a somewhat similar statement from Langacker (1987: 3):
There is no meaningful distinction between grammar and lexicon. Lexicon,
morphology and syntax form a continuum of symbolic structures, which dif-
fer along various parameters but can be divided into separate components
only arbitrarily.
I described this statement as implying that Cognitive Grammar denied syntax
the status of a definable area within the larger area of language, and went on
to point out that this description is not an accurate reflection of Langackers
own descriptive practice, Harder (1996:261). Langacker (2008a:6) flatters me
by taking the trouble to correct my interpretation, pointing out that definabil-
ity does not presuppose clear boundaries, and that syntax can be recognized
even if it seen as symbolic. Since that was more or less what I was trying to say,
there is substantial agreement apart from the fact that I still believe the for-
mulation no meaningful distinction is open to unnecessary misunderstand-
230 Chapter 6. Structure, function and variation

would be wrong to single out for special treatment4. The index in the
Handbook of Cognitive Linguistics, Geeraerts and Cuyckens 2007, has no
entry for structure and most cognitive linguists would be unlikely to
look it up anyway.
The quotation above rejects two distinctions and thereby raises two
points, both of which are central in the following: linguistics vs. pragmatics,
and lexis vs. syntax. I first discuss the distinction between usage and usage-
based structure; then I discuss the characteristics of bottom-up unit-based
structure as an integrated account of lexical units and syntax. This will
then provide the background for a discussion in the next section of the
functional dimension of structure that is backgrounded in the cognitive
The tendency to emphasize the flow of usage at the expense of struc-
ture is most marked in West coast functionalism (cf. Butler 2003), the
type of functional linguistics that is most directly based on actual usage, as
represented by Joan Bybee, Sandra Thompson and Paul Hopper among
others. Their work has affinities with conversation analysis in its rigorous
attention to the details of actual empirical data, and they have been pio-
neers in showing how grammatical properties can be understood by refer-
ence to patterns in everyday language use.5 Among the properties they
have documented are that syntactic phenomena evince the kind of grad-
ual, prototype or family resemblance patterns that also play a major role
in CL (cf. ch.1), thus making it a natural partner for CL in the broader
usage-based alliance that was described in chapter 2.
The point I am concerned to make is that just as one can endorse a
neural theory of language without believing it can provide all the answers,
so one can endorse the drive to give actual usage data the status they
deserve in linguistic description without believing that a description of
usage can take the place of the description of structure. The issue has been
discussed in relation to the understanding of complement-taking predi-

4 As pointed out in Nuyts (2007), there is a clear contrast between CL and other
functionalist approaches, especially in Europe, on this point. Function and
structure have co-existed peacefully and productively in European structural-
ism, as opposed to the polarization that took place in the US.
5 Further, they have shown how apparently basic and clearcut grammatical pat-
terns have unexpected frequency properties in actual usage. For instance,
expressions of subjective reactions, with low structural complexity, including
low transitivity, cf. Thompson and Hopper (2001), pervade ordinary conversa-
tion, making transitive clause constructions much less central in usage than
one might have thought.
Usage, structure and component units 231

cates such as think, believe, realize, etc. (cf. Thompson 2002, Boye & Harder
2007). Using conversational data, Thompson (2002) shows the prevalence
of the discourse or usage role of embedding predicates like think as for-
mulaic stance markers, arguing that I think it is a good idea is an epis-
temic modification of It is a good idea rather than a report on a mental
state. Thompson cites Langackers description in terms of which the main
clause imposes its profile on the complement clause (Langacker 1991:
436) as an example of descriptions that come out wrong in such cases. In
these examples, the speaker does not override the assertion profile of it
is a good idea by embedding it in I think. What is expressed by it is a good
idea is conveyed as asserted content, not as embedded in an elaboration
site associated with a main verb think: in spite of the apparent structural
position as matrix clause, it is really I think which is conversationally sub-
ordinate to the message conveyed by the that-clause. Thompson concludes
that this is not just a description of the statistics of actual usage, but also
the only really relevant linguistic description, rejecting the standard
account in terms of the structural relation of subordination.
Verhagen expresses a similar position, both with respect to comple-
ment-taking predicates, cf. Verhagen (2005), and generally with respect to
structure, cf. Verhagen (2002). The latter provides a compelling account of
historical developments whereby processes changed the meanings of indi-
vidual constructions in ways that would make it misleading to look for
overriding generalizations applying to for instance all resultative con-
structions in the language. However, at the end this is seen as supporting
a general conclusion, cast as a reply to linguists who are worried about the
unity of their object of description:
I suggest that the real problem here is the underlying assumption that the unity
of the field should somehow exist in the unity of the system itself. Liberating
ourselves from this structuralist prejudice, we may see that the source of the
unity of linguistic structure may well be external to it, that is, in the processes
giving rise to all these bits of knowledge. (Verhagen 2002: 434)
It is a point of this chapter to show why this reply is incomplete. Just as a
linguistic description must include an account of the potential contribu-
tion of a lexical item (pace Taylor, cf. ch 5, section 5), it must specify the
potential contribution of structural patterns (pace Thompson) but this
requires a rethinking of what the system is.6

6 As argued in Boye and Harder (2007), Thompsons account flattens out the
relationship between usage and structure, thus submerging the role of struc-
ture in the flow of usage. Among the arguments are that the diachronic rise of
232 Chapter 6. Structure, function and variation

The difference between Thompson and Langacker can be elucidated

by reference to a passage in Langacker (2008a), where the basic relation-
ship between structure and usage is explicated. The link is the act of
abstracting schemas out of usage: usage events form the input (Langacker
2008a: 458) to this process, and the output is a linguistic unit. Once linguis-
tic units have been formed, they have to be understood as operating not
just as usage flow, but also as units in the conventionalized linguistic sys-
tem (Langacker 2008a: 222).
The duality between usage token and conventional unit is the centre-
piece in the understanding of structure in classic CL, where language is
regarded as a structured inventory of conventional linguistic units (cf.
Langacker 2008a: 222). This unit format goes with a bottom-up orienta-
tion that is reflected in the notion of a compositional path (Langacker
2008a: 60). In combination, this approach reflects the component-based
dimension of structure (cf. Harder 1996: 154): the structure that arises
when units go together to form complex expressions.
Such bottom-up structures are found equally on all ontological levels,
including physical, biological and social structures. Component-based
structures in language and elsewhere essentially inherit properties from
components, but may include emergent features: a diamond consists
entirely of carbon atoms, but has properties that are due to the specific
way they are put together, i. e. the crystal formation. In order for two
component structures to be integrated, they have to offer a point of
overlap, providing the basis for correspondence in a way that can be
captured by the chemical metaphor of valence (cf. Langacker 1987: 94;
As discussed on p. 245, the system can accommodate holistic proper-
ties by adopting the construction format of description, which combines
the unit format of description with internal syntactic complexity. Con-
struction grammar created a new focus for the understanding of structure
in CL, in which Chomskyan abstract generalizations were removed from
centre stage and replaced by a description focusing on complex units

complementation can only be understood as the result of a process whereby

verbs like think come to be able to take clausal objects (as also pointed out by
Diessel and Tomasello 2001). It is true that the epistemic modification is espe-
cially frequent with first person subjects (I think, I guess, I bet etc), but it is also
true that there is a general pattern which allows speakers to have second and
third person subjects as well as past tense verbs (he believed, they thought,
Usage, structure and component units 233

between the word and the sentence levels. Such units play a massive and
previously underestimated role in understanding human languages. As
illustrated by the let alone construction (cf. Fillmore, Kay & OConnor
1988) the abstract generalizations of sentence grammar cannot account
for constructional properties such as the peculiar relation type between
the two clauses indicated by the phrase let alone itself.7
As described by Croft & Cruse (2004; 227; 247; 256) the development
from a purely generative view of grammar via a generative view that
includes constructions to a purely constructional model highlights the list
dimension of language even more strongly than classic CL.8 The general
format for language description, also for higher-level units, is an inventory
of units. As a successor concept to the narrow traditional lexicon, it was
called the constructicon (cf. Goldberg 2006, citing Jurafsky 1996): a list of
all units of form and meaning with special properties, such that a learner
needs to acquire them as individual units. Although the constructional
format focuses on larger units than words, it still reflects a bottom-up ori-
entation in the hierarchical sense: it takes the units (of whatever size) as
the basic elements, and looks on combinations as the work of the speaker
rather than the grammar. It is bottom-up also in the usage-based sense:
constructional patterns arise out of usage in a very transparent fashion (cf.

7 The concept of construction has always been used for phenomena like clefting
that cannot be accounted for either by the specific properties of the words or
by the general properties of the sentence. The Chomskyan approach, however,
was strongly oriented towards maximally general properties of sentences
(motivated by the search for innate features) and therefore did not pay much
attention to the less-than fully-regular cases. They were all put on the lexical
side of the strict dichotomy between syntax and the lexicon, which was
regarded as a lunatic asylum (cf. Goldberg 1995) of irreducible irregularities.
Construction grammar turned this priority around and used constructions as
lexical units that could be used as a format for eliminating the entire Chom-
skyan domain of totally general and abstract syntax.
8 The three stages are as follows: the first diagram illustrates the generative
architecture with its basic division into three strata: phonology, syntax and
semantics. The lexicon is the odd man out that cuts across the three strata.
Diagram two is an intermediate model, which adds another bridging sector
alongside the lexicon, consisting of constructions. At the third stage, the three
strata disappear entirely from view and all we have is the bridging type of ele-
ment: a lexicon plus a list of constructions (each with its set of phonological,
syntactic and semantic features).
234 Chapter 6. Structure, function and variation

Bybee & Hopper 2001; Bybee 2002), with frequency as a key factor. In
this, it aligns itself with another major trend, namely the collocational and
idiomatic turn represented e. g. by Sinclairs (1991: 110) idiom principle
as opposed to the open choice principle associated with generative gram-
mar: idioms are constructions too.
If constructions take over the entire territory which used to be popu-
lated by general categories, one may ask whether generalizations have
been lost from sight. This question has been taken up by Goldberg (2006:
45f; 2009), who emphasizes that the focus on constructions should not be
taken to imply a rejection of general laws in favour of idiosyncratic items.
Rather, the advantage of constructions is that they can accommodate any-
thing from the most general to the most local phenomena. As we have
seen, an emphasis on local cases as the basis of all generalizations is dis-
tinct from claiming that local cases is all we have: exemplars are stored,
but so are high-level generalizations (Goldberg 2006: 50).9
Construction grammar, in short, is a major advance for two reasons:
first, it provides a descriptive model that can accommodate intermediate-
level patterns that were homeless in both the traditional word-based and
the modern sentence-based approach to grammar. Secondly, it highlights
the direct pathway from usage to syntax by focusing on pattern extrac-
tion the fundamental mechanism for making structure possible. How-
ever, it does not follow that constructions are the whole story. Before I
turn to the function-based dimension of structure, I will briefly discuss
two points where the contribution of construction grammar is less defini-
First of all, considered as a general hypothesis about the nature of syn-
tax, construction grammar as such is a rather weak theory; it is hard to
think of a case that would falsify it. There is nothing to prevent generativ-
ists from using a constructional format indeed, Goldberg (2006: 205f)
presents her own approach (Cognitive Construction Grammar) along-

9 As an example of a not particularly general pattern type, Goldberg mentions

serial verb constructions in English. There are three subtypes, including the
GoVPbare, as in go tell it on the mountain. As one of the local restrictions, she
points out that this one cannot take tensed forms of the verb go, cf. *he goes
bring/brings the paper. However, in support of her general point, that gener-
alization is always an option, let me offer her the following example (from
Tony Hillermans Listening women), spoken about a forceful woman in the
story: Some of em go after married men. But a real tiger like that Adams she
goes gets herself a priest.
Usage, structure and component units 235

side four other approaches, including a type of generative construction

grammar. Even a Chomsky-style bipartition between lexicon and syntax
would not be downright incompatible with a construction-grammar for-
malism: a sentence is of course a construction, just as a word is, and no one
denies the existence of cases in between there would just be some patchy
stretches on the cline.10
Secondly, the analysis is basically static. If you analyse a larger whole
simply in terms of collection of components, the process can be compared
with a jig-saw puzzle done in reverse: you remove one piece at a time until
everything is gone, and you can do it in any order your like. When all ele-
ments have been assigned to the relevant item in the list (= the construc-
ticon), thus filtering out all the form-meaning relations that have been
evoked one at a time, the description is over.
To sum up: first of all, I have tried to demonstrate the nature and the
importance of the distinction between usage and usage-based structure.
Secondly, I have argued that although syntactic description in terms of
component symbolic units (constructions) is essential, it is not all there is
to say. I now proceed to the points on which a function-based approach to
syntax can complement the unit-based description.

10 Although that is not the main point here, I would like to mention one example
of where the convenient property of accommodating all types of phenomena
with equal ease has a downside. If we want to carve language at the joints, the
theory should be explicit about where those joints are. Construction grammar
makes an important point against the 100 % separation in Pinkers words-and-
rules dichotomy (cf. Pinker 1999) but the distinction between word-level and
clause-level phenomena is not an artefact. Fuzzy borders, as we have seen, do
not contradict the existence of fundamental distinctions. A case that illustrates
the point is the fact that in Germanic languages, cf. Klinge (2005), there is in
general a very strict division between word-level and clause-level mechanisms.
Compounds, as word-formation, and phrases are kept rigidly distinct in a vari-
ety of different ways, cf. rote Wein and Rotwein, Wortbildung and *Wort Bil-
dung in German. The sole exception to this is English, where the two types (as
in red wine and apple pie) cannot be kept clearly separate; thus word forma-
tion (as pointed out in a series of publications, cf. Klinge 2005:299300) cannot
be unambiguously placed as either compound or phrase. The fact that English
lacks a borderline here is difficult to describe as a significant feature within
construction grammar. Similarly, those cases (cf. Nedergaard Thomsen 1992)
where the syntax of a language like Danish behaves in a polysynthetic way
would be just another construction.
236 Chapter 6. Structure, function and variation

3. Function-based structure

3.1 Structured division of labour and arbitrariness as a functionally

motivated property

So what does a function-based analysis have to add to a unit-based anal-

The answer comes in two instalments. This section describes what func-
tion-based structure is, including the relation between function and struc-
ture in complex systems. Section 3.2 will discuss the dynamic side of syn-
tactic operators, following up the discussion in ch. 5 of competency and
the dynamic dimension of meaning.
As we have seen, functional analysis of an object or unit whether a
linguistic expression, a biological organ or a kitchen utensil is about the
role it fills in a larger system. As with the thought experiment of pinion-
ing the wings of the bird, functional theories predict what would happen
if the object was removed or replaced with something different. This
applies at all levels of analysis, non-linguistic as well as linguistic. A pure
description of your qualifications acquires a functional role as a job
application if you send it in an envelope marked job application and
if you do not get the job, you may submit it to functional analysis by
wondering what would have happened if you had changed aspects of
that description before you sent it. The commutation test for phonemes
for instance, replacing [i] with [] in the context [s_t] to see if it makes a
difference is also an example of functional analysis. With the generally
adopted tagmemic term, functional analysis takes its point of departure
in the slot rather than the filler. Slots are more central for understanding
structural properties than fillers, because new structure arises when new
slots come into being but not necessarily when a new unit arises (for
instance, the cyber-term twitter does not in any obvious way change the
structure of English).
It may not be immediately obvious what this has to do with the social
dimension, or what the relation is between the structural flavour of the
slot-based analysis and its functional underpinnings. The similarity with
formal categories is not an accident: the generalizations which have been
put forward as pure formal categories are typically abstracted out of
functional relations. It is not an accident that Langackers (2008b) func-
tional analysis can take over core parts of Chomskys auxiliary theory: the
devices generative grammarians call functional elements often are just
that (to the extent they are real units of language structure and not just
abstract figments of imagination, as in descriptions which suggest up to 17
Function-based structure 237

functional heads in the clause). Complementizers and tense forms, for

instance, are indeed functional elements which have (mistakenly) been
abstracted out of the functional-semantic relationships in which they
However, to get a more robust feel for the relation between function
and structure in the social world, I offer an analogy with a clearly sociocul-
tural construct: the structure of a business company (for a fuller argument,
cf. Harder 2006). The analogy is meant to illustrate the development from
unstructured holophrases to syntactically structured utterances. The stage
of a fully developed complex business company can simultaneously illus-
trate some of the synchronic properties of complex linguistic structure.
We begin with the narrative of a classic entrepreneur, going from one-
man company to large corporation. The central points include a two-way
developmental process: the aspect I focus on is top-down differentiation,
but the success of this operation depends on the bottom-up process of
integrating elements into larger, structured wholes.
The starting point for the process is the event of one man (lets call him
Jones) setting up a business for himself. As long as Jones works his com-
pany alone, all relations with the world meet in one and the same unit,
namely himself. The functional relations with the world may be multifari-
ous and complex, as they are in the case of communication, but they are
not matched by a corresponding structural complexity in the company.
Jones is the company; the company is Jones. A description of what the
company does externally, in relation to its environment, will be an exhaus-
tive description of Jones Inc.
Now Jones gets paying customers to such an extent that it is no longer
possible for him to do all the work on his own. He therefore decides to get
someone to join him in the company. What used to be one thing now splits
into two: previously the questions what does the company do? and what
does Jones do? boiled down to one and the same question. Now the two
questions are no longer the same thing, because it is necessary to decide
not only what Jones should do but also what the new person should do.
When the decision has been made to split the workload, the company has
been differentiated into two positions. But there are not yet two compo-
nent units; there is only one unit plus a vacant slot (an opening).
The optimal division of labour cannot be derived unambiguously from
the external function of the business company. It is a decision Jones has to
make before he advertises the position. In the initial phase it is a top-down
decision, dividing the whole workload. Once the actual component unit is
there, however, the bottom-up dimension will make itself felt. What divi-
sion of labour is best for the company will then depend both on what
238 Chapter 6. Structure, function and variation

Jones would like the new person (Smith) to do and also what Smith turns
out to be especially good at. In any case, it raises two sub-issues that are
linked but irreducibly different (i. e. one is partially autonomous of the
other): how do Smiths talents fit with what the company does, and how do
they fit with what Jones does? Answering one question is no longer
Up to a certain point of complexity, the internal organization of a busi-
ness company may be functionally transparent. That is the case if the
internal structure reflects external functions. If a two-man company is
organized so that Jones does production and Smith does sale, internal
structure reflects external function. An analogical situation in the case of
sentence structure would be if a sentence has two structurally signalled
constituents, subject/topic and predicate/comment, and these are always
the topic and comment of what is currently being discussed. In terms of
iconicity, the internal structure is then merely a shadow of the functional
organization of the communicated message. If we know the function of
the component, we also know its structural position; there is no need to
talk about structure as a separate issue.
As the organization becomes more complex, however, the link between
internal division of labour and external function will not continue to be
equally transparent and obvious. Jones and Smith might at one point want
to employ a secretary. Customers will not feel the difference, since secre-
tarial services are only designed for internal use. As business grows, inter-
nal (including managerial) functions will proliferate, because there is
more to do in order to keep the company together. The people who are
doing production in large corporations typically grumble about the prolif-
eration of management, but if (e. g.) co-ordination between sales and pro-
duction fails, it will be bad for the company. There has to be agreement
between various elements in the flow of activity in a business company
between production and sale, between contracts and pay, between what
the managing director decides and the job specifications of the people
who have to carry out the decision. If these things do not agree, the com-
pany will not be able to reproduce its own capital base and it will go out
of business. How much management is functionally motivated is an open
question; there is a very general consensus both that it should be kept to a
minimum, and also that this tends to be a very difficult thing to do. The
differentiation into sub-functions or internal functions (structural con-
stituents) will generate positions that may or may not be strictly neces-
sary from the point of view of external function.
In relation to language, this part of the analogy reflects the second
stage of grammaticalization in the theory proposed by Boye and Harder
Function-based structure 239

(submitted). Grammar does not inevitably have to contain grammatical

items, i. e. purely ancillary elements in principle there might be com-
pletely isolating languages (a frequently cited example is classical Chi-
nese). However, grammaticalization processes tend to bring about con-
structions in which one element is ancillary (auxiliary) and the other is
lexical, and this is part of an overall complexification process (cf. Dahl
2004). Agreement markers are standard examples of superfluous gram-
mar but analogously, certain organizations have uniforms that indicate
rank and affiliation. In language, once the meaning has been put together
in the right way, the elements that serve only a linking function, with no
external contribution, can be discarded. But it may be only in hindsight
that they are superfluous, just as it is easier to see whos in charge if she
wears a uniform to show it. If there were neither agreement signals nor
word order restrictions, it might be hard to put the words together in the
right way, and if police constables did not wear uniforms, it would take an
extra effort to establish their authority in a crisis.
The company analogy extends to the tripartite ontology: the flow of
business transactions is the basic mode of existence, without which noth-
ing else would be possible (with the cash flow as an important aspect);
individual competencies are required to handle those transactions and
niche properties manifest themselves both internally in the company (if
they want to survive, employees have to adapt to the company just as cacti
to the desert) and externally (is the company in a hostile or a friendly
business environment?).
What can we learn from this analogy? There are two key points: first,
there is an inherent link between structure and function in a complex sys-
tem. The division of labour between the managing-director and the pur-
chasing, sales, production managers is at the same time a functional and a
structural fact about the company. Secondly, when there is such a func-
tion-based structure, it entails an interface between the functional roles/
slots on the one hand and the component units that fill those functions.
The existence of an interface means that there is no seamless harmony, no
guarantee that things will match up as there would be if we could reduce
the slot and the filler to one category only. (Predicate nominal is an
example of such a hybrid category).
The lack of a seamless fit between slot and filler is a familiar point in
organization theory. (Human resource management would be ever so
much simpler if it wasnt!). An employee in a company has both his posi-
tion (managing-director, nightwatch) and his personal properties (40-year-
old ambitious university graduate, 60-year-old, tired highschool dropout).
The point is that however intelligently we may fill out a slot in the com-
240 Chapter 6. Structure, function and variation

pany, we can never be sure that there is an ideal match between the slot
and the filler this is why there is always a tension between formal (top-
down) and informal (unit-based, bottom-up) structure in an organization.
And this is a feature of all functional systems. The duality is obvious in
biology, as emphasized by Givn (1995, 2002), and is implicit in the tradi-
tion from Aristotle onwards. The pandas thumb, cf. Gould (1980) is an
adapted maxillary bone, not an anatomical finger, but it functions (more
or less well) as a finger; an apple crate may serve more or less well as a
bookcase, etc, etc.
In linguistic terms, we know this duality as the distinction between
functional role and structural realization. In the continental European
tradition, this is a standard feature of university grammars (examples are
Van Ek & Robat (1984), Bache and Davidsen-Nielsen (1997), and within
CL also Verspoor and Sauter (2000). In the American tradition, apart
from its manifestation as the slot-filler dichotomy in tagmemics, it is gen-
erally backgrounded: neither tree structures nor Langackers pictorial
diagrams explicate it, for instance (Chomsky 1965 explicitly eradicated it
from the heritage that he took over from Jespersen, by defining subject
based on the structural realization category NP).
The top-down orientation inherent in functional systems manifests
itself in two ways: first, because it begins with the role rather than the
filler as such and secondly because it looks first at the larger whole and
only then at the parts. Sub-functions within the company exist only by
virtue of the existence of the larger whole in contrast to unit-based
structures which exist only because of the smaller component units. A
molecule is a unit-based structure because it arises out of combinations of
atoms; in contrast, limbs are part of a function-based structure since they
arise out of the differentiation of a whole organism just as key account
manager positions arise out of a differentiation within the company. If
language was purely a unit-based structure, nouns, verbs and adverbs
could have existed for thousands of years before a bright person got the
idea of combining them into sentences. But just as the company is basic in
business, a whole utterance is basic to linguistic interaction. The utterance,
i. e. the meaningful whole contribution to interaction, has to be there
first before the sub-utterance component fragments.
A holophrase like hello or fuck, just like a one-man company, is in itself
a whole functional unit, because it may serve as a complete utterance.
From a purely external point of view, this is the minimum level: half a
greeting or half an assertion is no more useful than other half measures.
The rise of sub-utterance elements combinable into whole utterances is
evolutionarily significant, in that it is a key part of the developments
Function-based structure 241

whereby human language as a unique phenomenon came into being. Pre-

human languages, as we have seen, are characterized by a correspondence
between signals and acts. Alarm calls have no components: they slot
directly into whatever course of events the signals form part of.
Although the necessary background for the rise of human language
includes the ability to understand symbolic meaning (cf. above p. 150), the
most central fact from a structural point of view is the rise of this new level
of analysis: in addition to seeing each sign in the context of the situation to
which it contributes, we now need to see each sign in the context of the
other signs that are part of the utterance (just as Jones, when getting a
partner, needed to define his own role also in relation to Smith). Words
like the or of cannot be understood directly in relation to the situation in
which they are used they need to be understood as serving a function in
relation to those signs with which they collaborate in coding a whole lin-
guistic act.
The crucial step into syntax must have involved a pincer movement
that simultaneously combined two previously independent units and split
the holophrase. Which way you see it depends on whether you approach
the development top-down (the utterance becomes complex) or bottom-
up (items start to combine). The pathway is the same as the company anal-
ogy: the firm goes from one-man company to two-man company (from
above), while the two people combine to form a team (bottom-up). As
always, the bottom-up perspective is the most concrete way to look at it,
while the top-down presupposes a higher and more abstract level. (An
example of the development viewed in an ontogenetic perspective is given
below p. 267).
Functional innocence is lost when the holophrastic pattern gets frag-
mented: it is simpler when there is one indivisible utterance meaning. But
the functional potential of language obviously becomes vastly greater with
the loss of innocence and simplicity. There is an upside and a downside:
human languages come with the label some assembly required.
The need to include the top-down side of structure applies to all func-
tional entities, including artefacts. A knife has two chief components, the
blade and the handle, and if we want to understand the relationship, we
cannot hope to reach the right result by looking first at the properties of
the blade in isolation and then at the properties of the handle in isolation.
The functional role is the presupposed holistic property, in terms of which
it makes sense with a differentiation into two functional sub-elements.
Similarly, the two oldest structural concepts in linguistic theory, subject
and predicate, cannot be understood as component-based, only as based
on a differentiation of a whole statement into two functional sub-elements.
242 Chapter 6. Structure, function and variation

Functions define slots that need fillers: hence, moving top-down you
move from slot to filler. The process starts with the overall slot that is
filled by the whole utterance. There is something that may appear to go
against the grain in CL here, because it seems as if this implies that there
is something abstract and contentless which comes before the component
unit. But the contradiction is only apparent: at the level of the whole utter-
ance, the basic slot is the turn-at-talk and if we go back to what was said
about the rise of intersubjectivity in chapter 2, p. 77, it will be clear that the
turn as a slot for an utterance actually arises long before the linguistic
utterance, also ontogenetically. At four months, children know that there
are forms of interaction with other people which depend on both parties
taking turns to make their contributions, and the whole felt quality of
being part of what is going on depends on making such contributions
(within the slots that arise). There is thus nothing mysterious in operating
with a slot that may or may not be filled with a linguistically encoded
The internal differentiation of functional systems can throw light on
one of the perennial discussions between functional and formal linguists:
can syntactic categories be understood in functional terms? This discus-
sion is typically conducted with external functions on one side and formal
categories on the other without any reference to the bridging notion of
sub-functions within a complex utterance. This leaves a gulf that gener-
ally makes the discussions unproductive. Also for functionalists, the rise of
internal slots in an utterance necessitates a two-level approach. In terms
of a sub-utterance slot, the immediately relevant function is function in
relation to the whole utterance, not function in discourse. This means
that sub-functions cannot be explained directly with reference to utter-
ance function, except when structure is functionally transparent; the con-
tribution of a sub-unit to utterance function is mediated via combination
with other constituents. Syntactic sub-functions such as subject are there-
fore, in my terminology, partially autonomous. By the same token, they
are partially arbitrary in relation to external, discourse functions. Since I
suspect that arbitrariness, in spite of Croft (1995), may still give off bad
vibrations, I will try to show that arbitrariness is in fact a functionally
motivated property.

11 Sometimes the slot may be very open but sometimes the independent impor-
tance of slot properties can be seen in cases where interactive constraints are
very specific. If you stand poised to answer the $64,000 question, the slot pro-
vides basically only two options (1) right answer (2) wrong answer. Either you
give the right answer and take home the $64,000, or you do not.
Function-based structure 243

The account I have proposed of function-based structure represents an

intermediate position (not a compromise!) that avoids the central mis-
takes of both rigid structuralism and nave functionalism. The argument is
as follows: the Saussurean revolution, whereby structure became the cen-
tral property of language, established a rigid separation between pre-lin-
guistic and linguistic properties. Motivation almost by definition became
associated with the forbidden attempt to derive linguistic properties from
non-linguistic (=a priori) properties. Arbitrariness from this point of
view is an absolute, defining property of human language. The motivation-
oriented approach of cognitive linguistics is a revolt against this position,
and puts the emphasis back on motivation deriving from factors outside
language as a system, on human cognition as a whole. Based on this con-
frontation, the obvious way of understanding the dichotomy is to con-
clude that either language is arbitrary, or it is not or conversely, either
language is motivated or it is not.
In terms of the analogy above, however, both conclusions would be
wrong. In the company, being assigned to serve a function (such as truck
driver) depends on properties of the component, i. e. the person (such as
ability to drive a truck). But the presence of motivation is not the same
thing as being actually selected to serve that function as anyone who
feels he has been passed over for promotion can testify. The important
thing is that the function is served not exactly how it is served. Because
of that, some of the properties represent a choice between various pos-
sible options a kind of decision or judgement (Latin arbitrium). Moti-
vation is necessary, but never sufficient. Every time there is more than one
option, arbitrariness and motivation need to join forces in order to bring
about the relevant functional relationships. Without motivating proper-
ties, the filler would not fit into the slot but only the actual arbitrium,
making the choice between the available possibilities, will actually insert
it into the slot.
As in biology, motivation is a question of satisficing, not optimizing:
as long as organisms have enough of a given property to enable them to
survive and reproduce, selective pressure does not automatically work
towards perfection of the relevant property. Dysfunctional elements are a
fact of life in functional systems, such as teeth that cannot bite, words like
englisch that are increasingly misunderstood, and business leaders who
cannot lead. If the adaptive pressure is low, dysfunctional elements may
persist forever. This is one of the reasons why nave functionalism is doomed
to fail. Exactly what kind and amount of motivation is necessary cannot be
answered in general terms the only thing we can be quite sure of is that
biological properties cannot be fully predicted from motivation. Arbitrary
244 Chapter 6. Structure, function and variation

properties may come across as a remainder, the charred fragments that

are left when scientists have done all they could. But this is a misleading
picture in the sense that it suggests that we ought to provide a total, one
hundred per cent explanation of all the interesting properties. The harder
the environmental selection pressure, the more we can expect functional
pressure to drive adaptation processes. But for the reasons discussed above,
motivating properties, however plentiful, never suffice to set up an actual
functional relationship which inherently depends on an arbitrium.
This shows that arbitrariness cannot be avoided. What is more, it shows
that arbitrariness is in fact a functionally motivated property. As one of
my teachers used to say, thank God the order of the alphabet was fixed
before linguists got their hands on the issue. Linguists, looking for moti-
vations, would probably have quarrelled endlessly about the optimal
order. Such a quarrel about motivation would have stood in the way of the
function itself, i. e. an order of letters in which listable words should appear
when lists are needed. And arbitrariness, moreover, only exists in func-
tional systems. The freedom from knee-jerk deterministic forces is a pre-
condition for the complex functional form of causality that works by
selective pressure: some elements are reproduced in the next generation,
others are not. This is what allows arbitrariness to arise as a more flexible
mechanism than determinism. This occurs in biological systems, and in
social institutions, but not in purely physical or chemical systems. That is
why it does not make sense to try to distinguish between arbitrary and
motivated features of the Atlantic Ocean.
The issue raises itself both at the individual level and invisible hand
level, cf. ch 4. If a speech community has a relevant functional purpose
such as being able to refer to horses, it fundamentally does not matter if
the expression that ends up doing the job is horse, cheval, or Pferd as
long as speakers manage to agree on one of them. What is motivated,
above all things else, is that there should be a functional option any
option that will do the trick. The functional utility of arbitrariness can
also be placed in the evolutionary scenario that was invoked above. The
link between animals calls and the triggering has a deterministic element:
animals cannot choose which calls to use or what the relevant triggers are
going to be. The vast expressive potential of human language arises
because this deterministic link is severed and the change consists in the
emergence of arbitrariness instead of determinism.12

12 Arbitrariness can also be understood as a different dimension from the issue

of determinism or not as the lack of an inherent link between the causal fac-
Function-based structure 245

To sum up: top-down structure, based on division of labour in achiev-

ing an overall function, unites function and structure in one theory, with
function as the basic notion. Partial autonomy and (partial) arbitrariness
represent points of similarity with structuralism, in that structure cannot
be predicted from external function (cf. Newmeyer 1998, 2002); but since
structure depends on functional organization rather than being fully
autonomous, the theory is fundamentally functional, rather than struc-
tural in the formalist sense.

3.2 Slots and constructions: coercion as construction-internal

functional pressure

The issue of how to bring about a meeting between slots and fillers also
arises within the perspective of construction grammar, as a feature of the
unification process. Further, construction grammar also involves a top-
down dimension. In his Radical Construction Grammar, Croft (2001: 32,
46) shows that the primitives of linguistic theory cannot be atomic catego-
ries such as nouns and verbs we have to take our point of departure in
whole constructions. Smaller units arise as (different varieties of) sub-
components of larger constructions. As larger wholes, constructions also
impose top-down properties on elements that enter into constructions. A
famous example is Goldbergs (1995) she sneezed the napkin off the table.
The verb sneeze does not in itself carry a caused-motion meaning but
when it is recruited to serve as verb in the larger caused-motion construc-
tion (S V NP PP), it assumes a reading that fits with the specifications.
Part of the story I presented in the previous section is thus captured by
construction grammar. But the underlying design feature that was dis-
cussed above, i. e. the top-down functional dimension in syntax, is not
overtly recognized because the unit-oriented nature of construction
grammar means that syntactic relations appear to be a matter of fitting
two constructions together. Construction grammars involve a taxonomic
network of constructions (Croft 2001: 58), rather than a hierarchy of
functional relations. This can be illustrated by a recent discussion in CL of
the role of slot properties.
To cover the phenomenon of a slot-determined meaning, the concept
of coercion is increasingly used about this process, also in cognitive lin-
guistics (cf. Ziegeler 2007: 99). Ziegeler is critical of the idea and discusses

tor itself and the outcome. For my purposes, the element of a decision is the
important thing.
246 Chapter 6. Structure, function and variation

examples, e. g. from Michaelis (2003, 2004) with a view to providing an

alternative account. One example is the coercion associated with the
indefinite determination construction, i. e. the use of an indefinite article,
in cases of potentially uncountable nouns, as in she had a beer. One point
which Ziegeler raises, in conformity with the general assumptions in CL,
is that the syntax (i. e. the Indefinite Determination Construction) is
perceived as basic, and prior . From a CL point of view, why should the
syntax override the meaning?
As an alternative, Ziegeler argues that metonymy is fully adequate as
a mechanism of meaning construction. Metonymy as a mechanism that
treats an input word as a vehicle for a contiguous meaning in the same
domain describes essentially what happens in those cases that can also be
described as coercion. From beer as a non-count substance to beer as a
unit there is only a short metonymic step so why should we encumber
the apparatus of Cognitive Linguistics with this superfluous explanatory
tool(2007: 120)?
I think Ziegeler asks precisely the question that should be asked from a
classic CL position. The limitation in the idea of coercion as a feature of
fitting things together, making square pegs go into round holes, can be
described with reference to the origins of the term in formal (Montague-
type) grammar, cf. Partee (1987). An early version of the idea applied to
the understanding of NPs in Montague grammar, where the understanding
in terms of generalized quantifiers implied that there was a problem with
coordinated cases like the temperature is 90 and rising. The reason is that
the referential meaning cannot match both predicated properties at the
same time: if the referent of the temperature in the first compound is 90,
it follows that 90 is rising (from the second compound), which is not the
idea. Type coercion is then a mechanism that adjusts the meaning from the
concrete value to the abstract category (very roughly speaking). In compu-
tational linguistics, the idea has been generalized to procedures that iron
out all discrepancies of this kind. The idea in coercion is thus basically to
restore compatibility in cases of mismatch. As Ziegeler rightly points out,
this general ironing out idea, driven by syntax alone, is not attractive from
a CL point of view that already has a rich apparatus of cognitive mecha-
nisms to account for meaning construction in context and she illustrates
the idea by offering examples of metonymic adjustment that is not driven
by syntactic coercion, such as Table One is movin upstairs (p. 117).
However, in terms of the framework I am arguing for, there is no such
thing as purely syntactic mechanisms. There is instead a constant interplay
between top-down and bottom-up factors, which is ultimately bound up
with the duality of function-based and component-based mechanisms. I
Function-based structure 247

think this account does justice both to the intuition associated with coer-
cion and the preference for a cognitive account in terms of metonymy.
The central idea is that slot properties ultimately derive from context-
driven functional pressures: linguistic expressions have to do a job in the
context in which they are used.
The point of origin for this process is that whatever component content
you put into the overall mother slot, the whole turn-at-talk, has to be
understood in a way that makes sense. This is what is captured in Gricean
pragmatics. In the smaller, utterance-internal slots, precisely the same
mechanism means that the relevant filler will be interpreted in such a way
that it can do the job that fits into the utterance-internal, i. e. syntactic, slot
in which it is used. I have suggested the term syntagmatic implicature as
a cover term for all accommodation- and coercion-type adjustments, in
order to stress the continuity between the utterance-external pragmatic
mechanism and the utterance-internal content-syntactic mechanism. So
what appears to be purely syntactic coercion, is really an utterance-inter-
nal manifestation of interactive, functional pressure to adapt to the con-
text in which the coded meaning belongs.
For instance, the slogan easy does it requires the reader to understand
easy as something that can fill the subject slot. It is complicated to trans-
late the process into a fully explicit syntactic conversion, but there is no
difficulty in understanding what it means. The meaning construction proc-
ess can fruitfully be understood as metonymic, as rightly argued by Ziege-
ler but this component-based side of the matter does not render the
top-down side of the matter superfluous: it is the insertion of the material
in the functional slot constituted by the subject position that drives the
metonymic reinterpretation of the conceptual content of the component
unit. Thus it is not a matter of the priority of syntax over semantic content,
but of the ubiquitous collaboration between bottom-up conceptual build-
up and top-down assignment of function.
The issue of slots and constructions also involves the interface between
language as individual competency and language as a property of the
community niche. Slots are not stored units of language in an individual
mind they are openings for something, like the turn-at-talk. This
comes out clearly in relation to the issue of formulaic language, which is
an inherently unit-oriented phenomenon. It can therefore be used to
illustrate what is missing if you focus strictly on (competency) units in an
individual mind: it goes for both niche properties and flow properties ana-
lysable in terms of slots that they cannot be captured from that point of
view. A discussion with reference to Wray (2002) will illustrate what the
problem is (cf. Harder 2007a for a fuller discussion).
248 Chapter 6. Structure, function and variation

Formulaic sequences are good examples of constructions, both in that

they are units, and also in that they are internally complex. Echoing essen-
tially Sinclairs position as cited above, Wray presupposes a Chomskyan
analytical approach as one half of her theory: it is not only a question of
knowing the words that go together into strings, but also of knowing the
strings of words that go together (Wray 2002: 281). The compositional
and the formulaic approach are different processing modes which the lan-
guage user can alternate between.
The limitations of this purely individual, competency-level account
become apparent at the end of the book where she draws the whole dis-
cussion together. In her overall diagram (2002: 263), she lists three differ-
ent unit sizes: formulaic word strings (strings of words stored and proc-
essed holistically), formulaic words (polymorphemic words stored and
processed holistically), and morphemes (bound or free, including mono-
morphemic words). But as her theory implies that there are individual
differences that cannot be inferred from the general form of these
sequences, she adds (2002: 264):
Unit size is just an explanatory device for the linguist who is trying to make
sense of lexical storage in terms of standard language descriptions () but this
is for representational convenience only: The units are not intrinsically dis-
crete. Any internal structure, including word breaks, is externally and second-
arily imposed ().

If you follow this non-discreteness to its logical conclusion, as Wray does

(2002: 264), the crux of the issue is that all units in the lexicon are effec-
tively monomorphemic. Formulaic sequences thus conform to the stand-
ard constructicon format: a list containing basic unanalysable elements,
not compositionally complex forms.
What is missing in this account touches on both the distinction between
competency and niche and between usage and structure. Let us take an
instance that is a unified whole in the cultural niche as well as in the indi-
vidual competency, such as a quotation one has learnt by heart. Wray
mentions Hamlets soliloquy as an item that may enter as one unit into an
individuals lexicon. This makes sense as a description of an individual
person if it is learned as a chunk with no understanding of sub-units. It
does not, however, cover an understanding of the speech as consisting of
meaning-bearing elements in collaboration, and therefore it also does not
cover the relation between the soliloquy and the English language as part
of the cultural niche. Its status as a unit in the mind of an individual is thus
parasitic on its status as a stretch of meaningfully structured English in the
sociocultural niche hence it is a partial description only. The rise of a
Function-based structure 249

formulaic unit of this kind presupposes the existence of a syntactic sys-

tem, which is not just a representational convenience. Whatever else it
may be, Hamlets soliloquy is not a morpheme.
The quotation from P. G. Wodehouse below can demonstrate another
facet of the issue, i. e. that even where there is no dispute about formulaic
status in the lexicon, we cannot assume a straightforward relationship
with the slots that play a role in the flow of usage:
I noticed that the boys manner was sullen when I introduced him to Mr Bax-
ter, and said he was going to be his tutor. He disappeared into the shrubbery,
and just now, as Mr Baxter was standing on the drive, George shot him from
behind a bush.
Good! cried Lord Emsworth, then prudently added the word gracious.
P. G. Wodehouse, Lord Emsworth and others (Penguin edition), p. 23.

As illustrated, gracious can be an increment, serving a purpose in a very

ad hoc slot that opens up, in the hour of need, in the cognitive production
processes of the speaker while in all other respects it is part of an unan-
alysable whole unit. Because in the public code good is a recognizable unit
even when it occurs inside a formula, the speaker can reconfigure his
utterance in mid-stride by using the language (= langue) exactly as it is.
The moral is that the question of what constitutes an entrenched pattern
does not exhaust the issue of the structural options offered by the niche.
Also in another respect, the concept of construction is at risk of being
made to carry too great a load. In Crofts (2001) list of frequently asked
questions in relation to Radical Construction Grammar, the most fre-
quent question of all is how to individuate constructions if you cannot
characterize them structurally in terms of the atomic primitive units they
consist of. Crofts answer requires the concept of construction to be at
the same time both primitive and complex (Croft 2001: 52). In the
perspective of the individual learner you can argue that constructions are
primitive, by seeing whole utterances as instances of constructions as
Croft claims they are (twice in ten lines, Croft 2001: 52): an utterance is the
basic unit that occurs in natural discourse, that children learn, and that
linguists study (Croft 2001: 52). If an utterance is also a construction, then
construction in that sense does indeed constitute the primitive given that
is the starting point of linguistic description.
But that argument glosses over the difference between utterances as
given unanalysed wholes, such as the child meets them in discourse (anal-
ogous to Wrays formulas) and utterances as instantiations of pre-existing
patterns. There is actually a double dissociation. Not all utterances are
instances of constructions in the sense of entrenched patterns. Picking a
250 Chapter 6. Structure, function and variation

passage from the debate on p. 458, lets for the sake of the argument that
this might occur as an utterance (an uninterrupted turn surrounded by
other turns):
. this idea that jeez I dont know, Jon, definitions in society
As far as I can see, this is only a construction if anything that has actually
been uttered counts as a construction by virtue of having been uttered, in
which case the statement becomes trivially true. On the other hand, more
run-of-the-mill utterances are not identical to constructions, they are
merely tokens of (several) constructional types. An utterance like be a
good girl and go to sleep may be an unanalysable primitive from the point
of view of the child, but from the point of view of the speech community
it is a complex instantiation of several different constructions rather than
simply a construction.13 (Crofts statement is of course true but unsur-
prising in the sense that utterances are instances of constructions just as
they are instances of vowels, intonation patterns and speech acts).
What is primitive about the utterance is its role as minimal communi-
cative act in other words the utterance as a functional unit. The utterance
understood as an instance of complex constructional patterns cannot be
understood as primitive. The criterion for being a construction in the
structural sense is that there is a shared structural pattern that character-
izes all instances and no pattern properties are shared between all utter-
The point Croft makes, however, is similar to the point I have made
above about structure beginning with the whole utterance, which is then
functionally subdifferentiated downwards, rather than with atomic prim-
itives such as nouns and verbs. The idea of top-down functional differen-
tiation is thus a natural ally of the constructional approach, at least on that
point. Constructions just do not capture the whole story. Because a con-
struction is understood as a form-meaning pairing, it is in reality co-
extensive with the concept linguistic sign. Saying that a language consists
of constructions is like saying that a language consists of signs. This is a
much better description than saying that language consists of syntactic
devices, but there is still more to be said than that.
Some of the limitations of an account in which the distinction between
grammar and lexicon is dissolved into a continuum of units have been

13 The issue involves the same principle as the adoption of a distinction between
constructions (understood as abstract types) and constructs understood as
linguistic expressions occurring as actual usage events (including utterances)
cf. Traugott (2008).
Function-based structure 251

addressed by the so-called Lexical Constructional Model, whose principal

proponents are Ruiz de Mendoza and Ricardo Mairal, cf. Ruiz de Men-
doza & Mairal (2007a, b). Situating themselves (like this book) in the
interface position between the functional and cognitive traditions, they
develop an approach that addresses a number of the issues that have been
discussed above in relation to the role of the top-down approach as a nec-
essary supplement to the bottom-up approach. The name of the model
signals the basic dichotomy between the inventory (lexicon) and the com-
binatory operation that is viewed as having an independent and essential
role in understanding the constructional side of language. This combina-
tory operation thus goes beyond the unification process (understood as
essentially trivial) that makes it possible to conceive of the whole gram-
mar in terms of a constructicon. The model also stresses the similarity
between the extra-clausal Gricean mechanisms and the intra-clausal coer-
cion mechanisms (cf. above p. 247).
The processes of coercion that occur when an item is inserted in a slot
are described in a format that is inspired by Role and Reference Gram-
mar, and at the same time links up with the classic conceptual mapping
operations of CL. Thus the mechanism by which Peter loved Mary back
into life is not just a unification of constructional with lexical meaning, it
also involves a process of metaphorical meaning construction whereby
the metaphor an emotional state is an effectual action is triggered (cf.
Ruiz de Mendoza & Mairal, no date). The model also involves a detailed
account of the constraints that this process is under, building on but going
beyond CL principles such as target domain override (cf. above p. 38).
Metaphorical interpretation in this model is thus also understood in
dynamical terms, not as a fixed invariant conceptual mapping but as a
potential for constructing meaning with the linguistic input serving as an
instruction, cf. above p. 192. The model, which is under rapid development,
is thus well positioned to demonstrate the potential of the underempha-
sized top-down and combinatorial part of syntax for CL.

3.3 Functional upgrading: the dynamic dimension of syntax

Grammar (or syntax) is symbolic in nature, consisting in the conventional

symbolization of semantic structure
(Langacker 1987: 2)

A central point about syntax in a functional and cognitive picture is that it

is symbolic, in the sense that it deals with meaning-bearing elements. In
252 Chapter 6. Structure, function and variation

chapter 5, it was argued that content elements are inherently dynamic,

shaped by their canonical role as contributions to the online flow and
that also the language competency of individuals had a non-representa-
tional, purely procedural core. Both conclusions clearly have implications
for the understanding of syntax. The dynamic dimension associated with
the overall syntactic hierarchy is dependent on a feature which I call func-
tional upgrading the type of mechanism whereby one meaning-bearing
item operates on another to bring about a complex whole with enhanced
properties. This mechanism, too, can be understood both top-down and
bottom-up but it is easier to understand if we take it bottom up, so it will
be illustrated from the point of view of composition rather than differen-
The dynamic aspect represents a further step beyond the basic idea of
unification in which two items go together like pieces of a jigsaw puzzle
if their specifications are compatible. The key dynamic operation I am
thinking of is familiar from a number of contexts, including logic and
mathematics, but in a functional context the most well-known instance is
speech act theory. It is generally recognized that utterance meaning has
both a doing (functional) dimension and a representing (symbolic-
conceptual) dimension. The classic manifestation of this duality is Sear-
les (1969: 31) formula for illocutionary acts, F(p), where F assigns the
speech act function (= illocutionary force) to p, which in turn provides
the propositional (representational) content. However, the function-
assigning relationship between higher-order operators and lower-level
operands or arguments is not limited to the speech act (= whole utter-
ance) level; it is also part of the way linguistic expressions are related in
the code itself, inside the clause.
The clause as a structural unit is at the same time a format for a com-
plete utterance meaning. There are other formats for complete utterances,
including holophrases like hurrah or yes. Common to all of them is only
the contrast with fragments such as the or may often which are not suited
to convey a complete utterance meaning. The special status of the clause
in linguistics is due to the fact that it constitutes the most fully differenti-
ated format. This means that most other constructions can be understood
with the clause as the presupposed overall pattern of functional differen-
tiation, cf. the discussion above. The clause format, however, can also be
used for sub-utterance units (subclauses) so the functional role cannot
be derived from the constructional properties alone.
A considerable part of this differentiated structure can be understood
with reference to operator-operand relationships. Operators, as described
above, take arguments as input and create more complex semantic enti-
Function-based structure 253

ties. An example that illustrates the basic principle both in the local and
linguistic domain is the negation operator (NEG), which can be under-
stood as a semantic functor which applies to a proposition (P), creating a
more complex, negated proposition NEG (P). Faced with the proposition
(1) the war is over

you may choose to apply negation to it, yielding

(2) the war is not over

(00) is a more complex syntactic entity both on the semantic content side
and on the expression side. On the expression side, the word not has been
added at a particular point in the linear order (after the verb and before
the adverb). On the side of semantic content, in contrast, the change can-
not be captured as a matter of linear addition. Rather, the word does
something to the rest of the semantic content, changing the whole mean-
ing of the clause.
In a number of ways, negation is an illustrative example of this func-
tional type of syntactic interaction between meanings: a semantic functor
affects the whole, instead of merely adding a small new piece to the puz-
zle; it has an obvious dynamic element: whatever P builds up, NEG (P)
undermines and finally, NEG does not represent anything out there
rather, it performs a semantic task.
The functional, task performance character can also illustrate why it
makes a difference to insist on the terminology of content syntax (cf.
Harder 1996) rather than just semantics or symbolic relations. There is
an important difference between encoded, syntactic relations between
meanings and those semantic relations which may hold between two items
in the clause regardless of their place in the clausal structure. Consider,
e. g. example (3):
(3) The biological world can be divided into the animal and the vegetable

In this sentence there are semantic relations between the words biologi-
cal, animal and vegetable. These relations include the superordinate/hypo-
nym relations between biological on the one hand and animal and vegeta-
ble on the other. These have nothing to do with the syntactic structure, and
would exist regardless of how the words had been placed in the clauses
that contain them.
In contrast, the relation between the words biological and world is of a
different kind: it constitutes a structural, encoded relation between the
two words. On the expression side (at the phonological pole in Langack-
254 Chapter 6. Structure, function and variation

ers terminology) this relation includes linear precedence and contiguity.

But the expression-side relations do not exhaust the relation between the
words: there is also a relation that is semantic in the sense that it involves
the meanings of the two words, such that biological specifies what sub-
category of the meaning of world that is encoded by the whole expression
biological world. This relationship, in contrast to the relation between bio-
logical and animal is thus both a syntactic and a semantic relation, more
specifically it is a relationship at the syntactic level of analysis that obtains
between two encoded content units rather than between two phonological
units: hence the term content-syntactic relation.
The obvious but usually ignored sign relation at the syntactic level
instructs the addressee to make a sub-category biological world, once
she has followed the lexical instructions to invoke world and biological.
This is also why the question posed by generative grammar of the inter-
face between syntax and semantics is a misbegotten concept: since seman-
tics includes one half of syntax, there can be no interface between them.
Instead, there is an intersection: the area of content syntax is simultane-
ously a subdomain of semantics and a subdomain of syntax.
Content syntax contains other relations than functor-argument rela-
tions.14 However, they are essential in the hierarchical structure of clause
meaning. We begin by following the bottom-up path towards full sentence
Within the structure of the clause, the hierarchy of function-imposing
operations constitutes a layered structure whereby higher, functional
operators have scope over lower content elements. This has been a cen-
trepiece of the functional view of syntax for the past generation, cf. Butler
and Tavernier (2008). It plays an important role in Foley and van Valin
(1984); Dik & Hengeveld (1991); Van Valin and la Polla (1997); Dik (1997),
Hengeveld (2004); Hengeveld and Mackenzie (2008). In cognitive gram-
mar, the same principle is found: grounding predications that embed
clauses in their situational content have the widest scope (cf. Langacker
1991: 240). It is also reflected in the formal clause structure of generative

14 Langackers concept of trajector is also a content-syntactic role: it designates

the chief participant in a relation, corresponding at the expression side with
specific word order relations between the phonological poles of predicate and
trajector expressions. There are also different kinds of functor-argument rela-
tions. The relation holding between biological and world has a functor-argu-
ment aspect but also a more special feature that can be called subcategory
attribition, or subcategorization, which is a content-syntactic relation at
phrase level.
Function-based structure 255

grammar (cf. Siewierska 1992). In formal grammars, the operator-argu-

ment relation is not universally understood in semantic-functional terms,
but the overall hierarchical ordering between function and content ele-
ments is generally recognized (for a functional-semantic account, cf.
Harder 1996: 274285).
The emphasis on the doing dimension is warranted by the fact that
when we build complex expressions, we do not just add the content of
each new element to a growing heap of meaning. When we add elements
at the top end of the hierarchy since they are functional operators and
not just concepts they are designed to change what is already there. An
obvious example of that is the complementizing type of operator, which
takes a content element as operand and imposes its profile on it (cf. Lan-
gacker 1987: 309), thus converting it into something that is both conceptu-
ally more complex and has a new functional potential. There is a signifi-
cant part of clause structure that can be viewed in terms of a conveyor
belt metaphor (cf. Dik 1994: 361), with the output of previous operations
becoming the input to following operations. A very skeletal paraphrase of
the way it works if we compile a full clausal utterance bottom-up, such as
Jill killed Leo, can be given with reference to (4)(7):
(4) Jill, Leo: nominal expressions, denoting entities
(5) kill (Jill, Leo): the predicate kill operates upon Jill and Leo, yielding a
predication (or state-of-affairs)
(6) past (kill (Jill, Leo): the past tense operator applied to the state-of-affairs
yields a proposition about a past situation
(7) Declarative (past (kill (Jill, Leo))): the illocutionary operator declarative
applied to the propositional content yields an assertion (cf. Searles F (p)
The nominal constituents (4) constitute the input end of the scope hier-
archy. This is reflected in the property discussed in relation to the billiard-
ball model (cf. p. 51), the fact that they are conceptually autonomous (cf.
Langacker 1991: 14): they can be conceived of without depending on the
presence of other elements. In contrast, a verb meaning cannot be con-
ceived of without something to fill the argument slots/elaboration sites
there is no way of understanding, e. g., threw without someone throwing
The next major step upwards in building a clause meaning is a predica-
tion (5), which is the result of the verb (kill) operating on (predicating its
content of) the arguments (Jill and Leo). From Searle (1969) onwards,
predication is generally recognized as a propositional act (cf. Croft 2001,
Hengeveld and Mackenzie 2008). It is also, from Aristotle onwards, gener-
ally recognized that verbs have a conceptual content that is characteristi-
256 Chapter 6. Structure, function and variation

cally distinct from nouns in having a temporal, dynamic dimension (denot-

ing events or dynamic relations, designating with sequential scanning,
in Langackers terminology (1987: 248). As discussed in ch. 5, the verb
open does something to its landmark entity, namely remove a barrier,
which involves progression in time.
In terms of the instructional format, each operator serves as input to an
interpretive operation that is to be performed by the addressee. In terms
of the cooking analogy, the structure of a clause is like the structure of a
recipe, not the structure of food produced (~ the take-home meaning).
When the addressee follows up in the instructions in the recipe, he con-
verts potential to actual meaning as part of the same process that com-
bines meanings in the syntactically appropriate way. In the process of
executing the operation encoded by an operator, the semantic potential
is imposed on the operand, resulting in a take-home meaning that is pre-
liminary until all instructions have been followed.
In the linguistic discussion, the job (predication in the case of the
verb) and the conceptual content (dynamic relation) generally lead sepa-
rate lives, and the conceptual content tends to be in focus in linguistic
theories (not only in CL). However, it is only by the act of predication that
the things denoted by nominal constituents get converted into partici-
pants of the event denoted by the verb. The verbal meaning operates on
the nominal constituents, rather than just being concatenated with them:
as a result, the content of kill is superimposed on Jill and Leo, so that one
becomes the killer, the other the killee. The fact that this functional
dimension is not just emergent from component units (cf. the discussion of
the Lexical Constructional Model above) can be illustrated by cases when
unusual items are conscripted to serve in the functional slot.
(8) Girl (to herself): Oh no! Ive been Napoleoned! (http://www.urbandiction-

If we put a nominal stem such as Napoleon into the verb slot, then even if
you dont know what it means, you understand that Napoleoning has
been predicated of the girl. The name has thus been coerced into serving
an operator function.15
The predication that is the output of applying a predicate to the argu-
ments can subsequently be operated upon by a past tense form, which
once again converts the input to something more complex, namely a prop-

15 Apparently it means having a short guy come on to you in a way that suggests
he is trying to compensate for his size.
Function-based structure 257

osition (6). Past tense is a grounding operator (cf. Langacker 1991: 195),
the state-of-affairs is now viewed as applying at a particular time before
now, and the question of truth can be raised: does the descriptive content
match the situation to which it is applied?
As the last step illustrated here (7), the temporally anchored proposi-
tional content may be operated upon by an illocutionary operator (the
intra-linguistic variant of the F (p) formula). In the above example it
assigns the status content of a declarative statement to the proposition.
This is the skeleton of the obligatory clausal hierarchy in English.16 The
topmost function-assignment operation, as described above p. 242, is a
situational process of a Gricean kind, triggered by the utterance as a whole
or rather by its insertion into the situational slot, the interactive turn: it is
the interlocutors who determine (in a collaborative manner, cf. Clark
1996) what an utterance is ultimately to count as. So the bottom-up logic
of the process of constructing a (clause-formed) utterance goes from lin-
guistic input at one end via a series of coding operations in which each
step re-functionalizes the output of the previous step, to the final, situa-
tionally interpreted output.
While the dynamic (job) aspect of meaning may be easy to see in the
case of operators from the verb upwards, at the bottom end it is not obvi-
ous what the dynamic role of nouns (noun phrases) might be. That role, I
argue, is to carve out entities (= arguments) for predication to apply to.
Once they are understood, these billiard balls are conceived as static and
autonomous, as pointed out by Langacker, but the linguistic operation, cf.
also Langacker (2008a: 460), interactively instructs the addressee to access
them. Nominal meaning thus operates directly on the world as an object
of reference, cutting out those segments of it that we want to talk about.
This is also why the sub-act of categorization has a privileged relation with
nouns as opposed to all other items: nouns, unlike verbs, prepositions and
adjectives, denote things, and common nouns denote them as instances
of a category. The categorical container is part of the act of identifying a
thing as something that can be talked about. The only way not to see this
as a dynamic operation would be to assume that we talk about the world
as objectively given, with no interference from the human subject.

16 There are also optional operators: it is possible to add adverbial modifiers at

various points to specify additional functional properties. At the top of the
semantic hierarchy, one can add illocutionary or textual adverbials such as
admittedly, furthermore or to cut a long story short, which imposes the roles of
concession or conclusion on the operand message.
258 Chapter 6. Structure, function and variation

Compared to the approach to syntax in terms of a network of construc-

tions, the functional upgrading approach represents a difference of focus.
Two aspects are central:
(1) it highlights the general hierarchical features of clause semantics
(2) it expresses the different stages as dynamic options

The centrepiece is the compositional logic of successive semantic upgrad-

ing operations performed on input elements, until we reach the level of
the full conventional (clausal) utterance meaning The constructional
approach, in contrast, describes each item separately, including both gen-
eral and idiosyncratic properties.
The clausal hierarchy is a highly general feature of clause structure; but
its generality is not the main point. General syntactic principles of clause
formation, whose special status I have defended in one particular version
above, are often associated with formal Chomskian rules, also by those
who are interested in other types of linguistic phenomena (cf. e. g. Wray
2002). But as pointed out by Goldberg (2006, 2009), construction gram-
mar has no problem with generalizations. What is special about it here is
the dynamic, functional nature of the hierarchy as opposed to the static
jig-saw combination mechanism of unification.
The central rule-like operation that constitutes the dynamic aspect of
meaning presupposes an individual competency that enables human
subjects to carry out those operations in ways that are appropriate to the
situation, i. e. to construct the relevant utterance meanings. This reflects a
quite general functional principle that also applies (e. g.) to acts like cook-
ing and cleaning: a functional competency consists in being able to cope
with an input and operate upon it until it suits your purposes. Cooks are
people who can take uncooked ingredients and operate upon them to
produce food (rather than consume them as they are). Cleaners are peo-
ple who can take dirty clothes and operate upon them to produce clean
clothes (or people would have to simply re-use them in their existing con-
dition). Human speakers are people who can (be instructed to) take avail-
able expressions and upgrade them until they meet the standard of the
new communicative step they want to take instead of merely re-using a
fixed inventory of meanings.17

17 Langacker (2009) gives an account of the relation between lexical items and
the constructions they enter into which shares many of the points I have tried
to make above (including what he calls the skewing mechanism that is
involved in the verb to Napoleon, cf. example 8, p. 256). While my strategy
has been to profile the distinctive contribution of the top-down perspective,
Function-based structure 259

3.4 The interdependence of the top-down and bottom-up perspectives

The previous sections have argued for the distinctive contribution of top-
down functional operators to the understanding of syntactic relations
not as an alternative to unit-based constructional networks, but as a neces-
sary addition to it. The aim of this section is to show how the two sides
The operator-operand distinction and the slot-filler distinction express
two sides of the same fundamental relationship, which is built into all pur-
poseful action: the relation between the means (~ the filler/operand) and
the end (~ the functional slot). In one fell swoop acts such as eating a unit
of food to satisfy your hunger, the relation is direct and simultaneous. In
complex actions, however, it becomes indirect and opaque: although your
intention is to wash your clothes, you start very indirectly by putting them
in the bin. This is how the split between top-down and bottom-up comes
into being, and with it the problem of making ends meet: the bottom-up
means have to be supplied before the topmost goal can be achieved. The
bigger the task, the greater the distance and the co-ordination problem. If
the goal is to hold Olympic games, a very large number of things must
happen in a co-ordinated fashion before you can declare the games open
which requires simultaneous consideration of means and ends at all stages.
This applies to the case of utterances as a species of complex action.
Even in a strictly top-down procedure, such as the one embodied in Func-
tional Discourse Grammar (cf. Hengeveld and Mackenzie 2008), the
simultaneous presence of several stages is recognized (cf. Mackenzie 2004)
in that one stage can trigger the next before being completed. Speech
errors demonstrate that ends may in fact not quite meet when processes at
different levels are running simultaneously, as in she come backs tomor-
row; the quake caused extensive valley in the damage (cf. Aitchison 1998:
257), where the words have been recruited and also the syntactic structure,
but the words go into the wrong slots, short-circuiting strict top-down plan-

however, Langackers strategy is to provide an overall format of description

that covers both the bottom-up and the top-down dimension in the same uni-
fied mechanism. That mechanism is a generalization of the categorization
relationship and while Langacker recognizes that it may seem peculiar to
say that the preposition on categorizes the table in the composite expression
on the table, the sense in which he uses the term makes it perfectly accurate.
However, I think his account captures the rather schematic similarity that all
forms of functional upgrading share with categorization, rather than the ele-
ment of function-assignment that I want to focus on.
260 Chapter 6. Structure, function and variation

ning.18 To manage this process, it is necessary to be able to understand

structures in both perspectives at the same time the idea of saying that
one is more important than the other is incoherent. In planning complex
action, you have to move from ends to means in carrying it out, you have
to provide the means before you can achieve your end.
To illustrate the built-in bidirectionality, the structure of the statement
Germany lost the match yesterday can be analysed from both directions.
The rough top-down path is: the overall goal is to make an assertion (illo-
cution), about a past situation (past tense), more specifically yesterday
(temporal adverbial), describable by assigning the predicate lose to the
arguments Germany and the match.
Bottom-up it goes like this: the entities I am going to talk about are
Germany and the match, as participants in an event of losing; this event is
temporally located yesterday. In linking the state-of-affairs with this past
situation I construct a proposition (about the past), which I assert to be
true (declarative illocution).
A striking illustration of the co-presence of the two perspectives is the
issue of what precisely the head of a construction is. What is the most
essential and defining element in a phrase depends on whether you look
at it top-down or bottom-up. True to their general orientation, generative
grammarians and usage-based grammarians tend to differ on this point. In
the case of what Langacker calls nominals, generative grammar has
evolved to view the determiner as the head, renaming the phrase DP (for
determiner phrase). Viewed top-down, determiners take scope over the
entire phrase, including the noun that is the basic operand. With the ter-
minological convergence that was mentioned on p. 236, this gives them
the status of functional head in generative grammar. Representing the
usage-based, bottom-up perspective, Croft (2001) adopts the strategy of
naming phrases after their Primary Information-Bearing Unit (= PIBU),
which for NPs is the noun that denotes the core conceptual content.
A functional view of bidirectionality can accommodate both rationales
at the same time. Occupying the hierarchically higher slot gives one form
of primacy (the power to impose a functional operation on the target
operand). Supplying the content gives another form of primacy (the raw
materials define the possibilities for what kind of operation you can per-
form). Coercion is an example of the top-down upper hand, as exempli-

18 As an example of the same thing, Bakker and Siewierska (2004) also mention
that phonological elements may be recruited at a fairly early point in the over-
all procedure.
Function-based structure 261

fied in the case of determiners by the indefinite determination construc-

tion (cf. p. 246). The coercive power of the indefinite article reflects the
fact that the determiner has the highest scope, imposing a task on the
noun: in the combination a beer, the noun beer is employed to denote a
countable unit, which obligingly it does (cf. Harder 2008 on determiners).
Bidirectionality is also reflected in the existence of dual dependence
relations between higher and lower elements (cf. Harder 1996: 276)19: there
is a functional dependence of lower elements on higher elements, in addi-
tion to the conceptual dependence defined by Langacker (1987: 301). 20

19 Conceptual dependence goes downward in the hierarchy and reflects the fact
that operators cannot work without having some content to operate on. With
reference to the example above, if you want to predicate the concept lose, you
need a loser to predicate it about, one who can fill out the elaboration site in
Langackers terms. Similarly, in order to apply negation, there must be some-
thing to negate. But because of the functional embeddedness of utterances,
there is also a dependence that goes the other way. It applies already at the level
of the billiard-ball model, and can be illustrated with reference to the two par-
ticipants Germany and the match. Just as the verb lose cannot be understood
without reference to participants, the two NPs cannot stand alone and consti-
tute an utterance consisting in Germany. The match. You can conceptualize both
of them without needing to add a predicate to them but only when we link
them up by means of a predicate such as lose do they suddenly have a job to do
in a coherent message content. Higher in the clause, the proposition is a concep-
tually perfectly self-contained unit, which philosophers can discuss happily
without needing to attach them to a speech act operator. Nevertheless, a propo-
sition on its own does not constitute a roadworthy utterance content because
for message purposes we do not know what to do with a proposition, unless the
speaker either asserts it or raises the question of whether it can be asserted.
20 Dependency relations play an important role in understanding the continuity
as well as the discontinuity between meaning in language and meaning in the
situational environment. From a narrowly linguistic point of view dependence
can provide a rationale for two opposite situations: it can mean either than
there has to be an encoded linguistic item (otherwise the dependent item dan-
gles in the air), or that there does not have to be one (because if the depend-
ence relation is obvious, the addressee has to supply the missing item to satisfy
the dependence relation). Both cases are found in real life.
First, there are cases where dependence works so as to require an explicit
coded element. An upwards functional dependence that works in this way in
English is the proposition, as discussed above you cannot encode a proposi-
tion on its own, without an illocutionary value. Downwards, it can be exempli-
fied with a textual adverbial such as further, which needs to be followed by an
explicit textual operand that can constitute the addition that it announces.
262 Chapter 6. Structure, function and variation

Dependence is linked up with the slot-filler dichotomy in the following

way: top-down operators provide slots: a declarative operator opens a slot
for a proposition; a negator creates a slot for a negated item; a verb creates

Secondly, there are cases where dependencies allow items to work without an
encoded item to satisfy them: The potential utterance none as exemplified
below works both ways. It is conceptually dependent on a category specifica-
tion; the addressee therefore has to retrieve it from the previous discourse
context. (Otherwise he would not understand what there is none of). It is also
functionally dependent on the top end of the sentence hierarchy, including the
illocution. In the right discourse context, the addressee will recruit content to
complement both missing ends, cf:
A: How many passengers refused to pay?
B: None
In interpreting Bs elliptic reply none, we have no trouble supplying passen-
gers at the bottom end and assertion from the top end.
Languages differ with respect to the strategies they follow. Mandarin is
famous for allowing many things to be uncoded that have to be explicitly
encoded in clauses in Standard Average European languages like English. This
includes the arguments that are conceptually dependent on the verb (making
it almost impossible to classify Mandarin verbs according to syntactic valency).
This has consequences for the grammatical description, in the sense that the
strategies that you have to follow when understanding verbs in languages that
do not require explicitly encoded arguments are different and more compli-
cated: you have to do more interpretive work on your own when you are listen-
ing to Mandarin than when you are listening to English. Thus the verb gi (give)
can be used alone, leaving the addressee to figure out who gave what, and if
relevant, to whom. Mastery of these strategies is therefore part of knowing how
to use verbs in Mandarin; you simply have to make do with fewer instructions.
A somewhat similar mechanism applies in the case of agreement marking.
As pointed out in Croft (1995), it is often the case that gender classes can agree
either with classes of linguistic items and also with classes of referents. Nouns
have inherent gender in many languages, including French and German and
in the same languages referent entities have either feminine or masculine
(sociocultural) status. But the place where gender is marked is on indirectly
associated items such as adjectives. The word for crazy in French (masculine
fou, feminine folle) therefore links up the adjective with an item either with
linguistic or sociocultural gender, which do not necessarily match: thus majest
(= majesty) is feminine, also when used as the title in front of the King: sa
majest le roi. In terms of experiential content, the two cases of feminine mark-
ing are thus clearly different but functionally and instructionally there is an
equally clear parallel, cf. p. 184, because folle in both cases instructs the
addressee to pair off the property crazy with a referent that has the feature
feminine, thus reducing the scope for attaching meanings in the wrong way.
Function-based structure 263

a slot for participants. Bottom-up content units provide fillers: NPs pro-
vide fillers for participant slots, and a proposition fills out the content slot
of an assertion. This function-based structure is again generalized from the
properties of non-linguistic functional action: the labour market is a meet-
ing place between job openings and people ready to fill them, and as we
know, jobs are created top-down, while people apply for them bottom-up.
Here, too, there are bidirectional dependence relations: applicants depend
on jobs, and jobs depend on applicants; here the discussion about which is
more basic is a political issue, but again, neither side can be eliminated.
In relation to Cognitive Grammar, the dual directionality manifests
itself in the account introduced above of the English auxiliary in terms of
Functional Systems, cf. Langacker (2008b), which is a development of
the classic analysis: the grounding elements (tense and modality) stand at
the top of the scopal hierarchy (cf. above p. 255) and take scope over the
predication, with the lexical (main) verb as the head. 21 The analysis goes
beyond the classic CL account, in introducing a special apparatus for
approaching clause meaning from the functional top end rather than the
conceptual, lexical end.
The analysis also adds an essential dimension to the analysis of subject-
hood in Langacker (1999). In the linguistic literature, the top-down dimen-
sion of subjecthood (cf. Harder 2006) has been under-emphasized because
the subject has been seen chiefly as defined in relation to the verb (at

21 The grounding elements attach themselves to the topmost verb stem, which
Langacker calls the existential verb, because it specifies the reality status of
the clausal content. Subject and existential verb form a unit called the basic
existential core together with the subject, which can be separated by the rest
of the clause by an adverbial, as in (1)(2), (8a-b in the original):
(1) Will she, perhaps, be waiting for us impatiently?
(2) Has she, perhaps, been waiting for us impatiently?
The term core reflects an analysis in terms of layering, where will she and
has she is interpreted as being the centre of the clausal structure. Although it
may be confusing to have two sets of layering relations, one for functional
centrality and one for conceptual centrality, the status of diagrams as heuristic
devices means that this is a practical rather than a theoretical issue. At the
presentation, functional systems were said to have an indirect relationship
with grammatical organization, analogous to the decisive but intangible role
of dark matter (Langacker, personal communication) in the universe. As
with dark matter, this raises the challenge of providing an integrated account.
(The special status assigned to subject + topmost auxiliary corresponds to the
mood super-element in the clause in Systemic-Functional Linguistics, cf. But-
ler 2003: 173, with the rest of the clause called residue).
264 Chapter 6. Structure, function and variation

billiard-ball level, as it were). However, the whole discussion of the role of

subjects in English depends on the relationship with the top-level func-
tional choice of indicative (i. e. either declarative or interrogative illocu-
tion), as opposed to e. g. imperative or exclamative. Only if the speaker
chooses to make an indicative utterance does obligatory subject choice
arise, cp. imperatives like please keep your luggage with you at all times,
exclamatives like what a waste! and exhortations like down with creation-
ism! In indicatives you have to choose a subject, even in the case of zero-
argument verbs like rain functionally speaking because subject-verb
order in English encodes the distinction between interrogative and declar-
ative illocution. Encoding subjects that are referentially null with zero
expression would thus eliminate explicit status assignment at the top end
of the scopal hierarchy. In support of this interpretation, it may be pointed
out that in Spanish, for instance, absence of obligatory subjects goes with
absence of word order coding of declarative/interrogative illocutionary
choice; and I challenge the reader to find examples of obligatory subject
expression in languages where overt subjects do not have a role in coding
functional choices.22

22 Langackers (1999) description of subjects in terms of primary focal partici-

pant as the prototype, analogous to MacWhinney-type starting points, is fully
compatible with this account. In indicative utterances you have to define a
point of departure for the proposition, and in the functional systems analysis
(Langacker 2008b), this point recurs in the form of the role of the subject as
default anchor, whose integration with the illocutionary choice makes essen-
tially the same point as the one I outlined above. A fully explicit integration
between functional systems and the grammatical hierarchy, however, would
have to incorporate a description of the interface between (1) the top-down
choice of a subject that is bound up with the choice of indicative illocution
type and (2) the bottom-up introduction of potential fillers by the choice of
verbs and constructional patterns.
This would also make explicit the element of redundancy in assigning the
status of primary focal participant to the subject. There are many situations
where subject choice does not add any independent semantic content, and
where subjecthood is in that sense a purely formal or syntactic property.
When there is only one argument, as in intransitive clauses, that argument
wins the subject assignment by default primary participant status goes (vac-
uously) to the only argument there is. This contrasts with non-vacuous cases
like Jack resembles Joe versus Joe resembles Jack, where subject choice makes
a clear functional contribution to clause content.
An explicit account of this interface between the two directions would
thereby also throw light on the nature of the arbitrariness that is wrongly
understood in generative grammar as indicating that subject choice is funda-
Function-based structure 265

The co-existence of top-down and bottom-up forces is also essential in

relation to language acquisition. Tomasellos usage-based theory of lan-
guage acquisition (2003) shows in concrete detail how children work their
way upwards from actual chunks, without needing innate higher-level
rules. The emphasis therefore is on the bottom-up path from concrete
instances of unanalysed utterances and step-by-little-step towards more
and more elaborate and abstract patterns. The same basic orientation is
found in other functional approaches to language acquisition, cf. MacWhin-
ney (1975) on item-based patterns. This path matches the component-
based structure discussed in section 6.2 above; progress in acquisition can
be described in terms of constructions that become more and more
abstract, adding more and more complex components to the childs con-
There is no contradiction, however, between this pathway into lan-
guage and the simultaneous presence of a top-down element that is under-
stood not in formal, but rather in functional terms. Children come
equipped to move in either direction part to whole or whole to parts
(Tomasello 2003: 39); but on the same page, Tomasello wonders why chil-
dren start with one-word unanalysable expressions. I suggest that is to be
expected if you view the whole-to-part operation in terms of the top-down
functional differentiation. If the functionally relevant unit is the whole

mentally a purely formal mechanism (Pinker 1994: 42). Arbitrariness, as dis-

cussed above, is a functionally motivated property of languages, arising from
the need to make a choice between (motivated) options. In this case the choice
is how to organize indicative utterances, and English represents the choice of
organizing them around a primary focal participant. Once that choice has
been made, it must be imposed on all clause contents, whenever they appear
as the content of an indicative clause. In certain cases this is more motivated
than others but such is the nature of functional systems. As a final suggestion
for the project of integrating the functional systems with the grammatical hier-
archy, I see no reason why the illocutionary force operator (in this case, inter-
rogative or declarative) cannot simultaneously occupy a place at the top of the
scope hierarchy, as in the layered clause structure of functional grammar, and
also enter into the functional choice systems described by Langacker (involv-
ing also polarity and anchoring). The idea of functional choices operating top-
down presupposes that the material on which the choices must be imposed is
provided bottom-up. As pointed out in the introduction above, this also illus-
trates the link between functional and structural organization, adding a set of
paradigmatic choices to the previously chiefly syntagmatic clauses analysis in
cognitive grammar. There is every reason to see them as parts of the same
integrated grammatical system.
266 Chapter 6. Structure, function and variation

utterance, the functional pressure is to go for the top level and fill in the
utterance slot as soon as you can. Since the holophrase saves you from the
some assembly required condition on human language, that is the natu-
ral starting point (as it is also the evolutionary pathway, cf. above p. 240).
This again pinpoints the distinction between the utterance as the function-
ally primitive unit and the utterance as a complex construction. If only the
bottom-up path is considered, the function and the component unit end
up being conflated, as discussed above: we get the whole utterance under-
stood as a construction, cf. in addition to Croft (2001: 52) also Tomasello
(2003: 325).
From the top-down side, structure arises when you move from the sim-
plex utterance to the complex utterance. In acquisition, this works by an
emerging functional division of labour, which may take the form of a
pivot grammar stage, cf. Braine (1963). Pivot grammar does not hold
up if understood as manifesting a generative, formally inviolable subject-
predicate grammar, cf. Tomasello (2003: 9596, 124), and therefore Toma-
sello does not understand pivot schemas as syntactic (Pivot schemas
do not have syntax, Tomasello 2003: 115). According to Tomasello, the
transition into syntax proper has the following crucial elements:
Essentially what they need to learn is that whereas some linguistic symbols are
used for referring and predicating things about the world, others (including
word order) are used for more grammatical functions. (Tomasello 2003:125)

Tomasello thus distinguishes between pivot schemas and item-based con-

structions by the criterion of what he calls syntactic marking, and item-
based syntactic constructions are therefore only present when there is
consistent grammatical marking.
This reflects an understanding of syntax that is essentially taken over
from generative grammar, where syntax is reserved for grammatical
rather than contentful functions. In contrast, if syntax is understood as
reflected in the term content syntax, where the driving force is the
intention to recruit two meanings in order to produce a complex mean-
ing, the emerging formation of item-based pivot schemas qualifies as syn-
tax. In content-syntactic terms the basic relationship between reference
and predication is clearly part of grammar in the sense of content syn-
tax arguably the most basic syntactic relation there is. Moving from a
stage of pure cut-and-paste operations towards this form of combination
is then one of the partial, tentative steps that children take towards
imposing regularities on their utterances. For English children, it is a nat-
ural bid for adapting to the organisation of indicative clauses discussed
above to begin (inductively, item by item) to use utterances starting out
Function-based structure 267

with the primary focal participant and assigning a description to it, as in

Mummy gone.
The development can be understood as stage one in the process of
grammaticalization (cf. Boye and Harder submitted), where the rise of
grammatical items is the next step. If we start out with a repertoire of
holophrases with no relations between them, the rise of a complex unit
with a division of labour between them is the first step towards gram-
matical structure, while the rise of items with ancillary functions is the
next step. In the following diagram, the wugs example (cf. Berko 1958) has
been chosen because it shows that the word is not a holistic unit, but a
productive combination. The path reflects the same scenario as the busi-
ness company example above p. 237, where the first step is the combina-
tion of Smith and Jones, and the second is to employ a secretary):

1. Holo- Mummy! there!

(no rela-

2. Gramma- Mummy there

tical relation
lexical items

3. Relation Wug- -s
lexical and
cal item

In an acquisitional context as well, the top-down perspective that is bound

up with the utterance as the functional unit is always present. It has been
found that children show many signs of following not a rigidly bottom-up
path, but a pincer movement whereby they jump ahead to top-down
mechanisms, even as they painstakingly and conservatively build their
inventory of constructions bottom-up. As formulated by Israel (2002),
learners go for both local and global consistency (quoted with approval by
Goldberg 2006: 64). In the case of negation, similarly, children appear to
follow the dual pathway of on the one hand gradually unpicking the jig-
268 Chapter 6. Structure, function and variation

saw puzzle of negation types, and on the other using creative and general
strategies, cf. Cameron-Faulkner, Lieven, and Theakston (2007).
The triple ontology of language, where competencies are understood
as adapted to the niche, may throw light on this issue. Tomasellos trajec-
tory of language learning constitutes a rejection of the continuity assump-
tion that was made in generative grammar: if grammar is innate, it is the
same for all, and so from an innatist perspective the child basically has the
same grammar as the adult. As Tomasello points out (Tomasello 2003:
32324), there is no evidence for this assumption: children build up their
own grammar painstakingly as they go along. But if acquisition is a form
of adaptation to the niche that contains the causal affordances (i. e. pat-
terns) associated with adult grammar, then adult grammar has a role in
the whole process nevertheless as selection pressure exerted by the
environment to which children are adapting. If relocated from an innate
module to the sociocultural environment, there may be a place for conti-
nuity as one element in the sociocognitive, functional-causal process of
language acquisition.
I regard these acquisitional implications as an important argument for
the two first hypotheses defended in this chapter: the need for a more
profiled role for the functional dimension of structure and the contribu-
tion of this reprofiling of linguistic structure to a better understanding
also of the usage based nature of language. Below we shall see how the
functional approach can throw light on variation, the third hypothesis:
adaptation is a hit-or-miss type of process, and therefore it is not surpris-
ing that individuals should respond to selective pressures in different
ways, and actual utterances do not always reflect fully consistent and fully
internalized grammatical patterns.
The general moral is that the syntactic hierarchy must be understood
simultaneously in both perspectives. There is a bottom-up path that goes
from component units to more and more abstract and complex entities
the favoured path of approach in classic CL. There is also a top-down path
from the overall function (which is the raison detre of the whole thing) to
the sub-functions into which it can be differentiated. And this bidirec-
tional approach is necessary in relation to all relevant linguistic objects of
description: whether you are interested in structural, functional, concep-
tual or acquisitional properties.
Norms and variation 269

4. Norms and variation

4.1 Introduction: the interdependence of structure and variation

Some readers may feel that this chapter has taken such pains to demon-
strate the importance of structure that for all my claims to adopting a
social and usage-based perspective, I have actually opted for structure
rather than usage. But as announced in the introduction to the chapter,
the aim is to show that structure is essential for understanding usage prop-
erties. This applies also to the understanding of variation: not only do we
need both structure and variation neither side can be understood except
in relation to the other.
To put it schematically, variationist linguistics belongs at stage two on
a path of increasing complexity, not at stage zero. Stage zero represents
chaos, as found in Genesis on the first day of creation, where the world is
without form. An ontogenetic stage zero is the infants babbling period,
where phonetic variation is (in principle) unconstrained and sounds have
no meaning. Variationist linguistics is not about collective babbling. What
occurs at the stage zero may therefore be called fluctuation rather than
Linguistic communication requires a state of co-ordination, as reflected
in Bloomfields (1933: 78, 147) fundamental assumption of linguistics,
that in every speech-community some utterances are alike in form and
meaning.23 There must be usage events which are instances of the same
linguistic category before linguistics can get to first base. Stage one of the
scale is therefore one in which there is structure, minimally in the form of
a set of categories of which utterance exemplars are instances.
For that reason, it is logical that structural linguistics historically arose
before variational linguistics. Variational linguistics is stage two because
it builds upon but goes beyond structural description. Labov (1972) could
not have described the variation in realization of postvocalic r in New
York City if he could not take his point of departure in the category
postvocalic r. The necessity of the structural category comes out most

23 The term alike is imprecise, because what is really at stake is that two utter-
ances or signals have to be understandable as instances of the same category
in order to have meaning at all. Speak and speech are alike in form and mean-
ing, but that is not enough to provide them with the relevant form of same-
ness. The question is really whether there is a state of co-ordination such that
two occurrences of one lexeme (two different tokens of speak, for instance)
count as instances of the same signal.
270 Chapter 6. Structure, function and variation

clearly in the fact that one of the manifestations is zero, i. e. there was no r
sound in fourth floor in these instantiations. Unless Labov could presup-
pose a structural description in terms of which word forms are describable
as realizations of an abstract (meta-level) unit called postvocalic r, there
would be nothing to link the two manifestations, i. e. nothing that they
would be variants of. There are very good reasons for being sceptical of
units postulated at the meta-level, prefering to stick to what can be heard
on the tape, but this does not allow you to tell significant absences from
things which informants simply happen not to say.
The path from stage zero to stage two may suggest that structure
could occur without variation, but not vice versa, as assumed in structur-
alism. But in the real world, structure can only exist as a property of a
substratum, i. e. something that is structured and when structure arises,
what used to be fluctuation turns into variation within the structural cat-
egories that arise. In this way, variation comes into the world hand in
hand with structure. As the babbling child gradually adapts to the sound
categories of the community language, a certain range of sounds (which
used to be merely different) begin to acquire the status of variants of the
same sound category. The task of providing a structural description must
reflect this process, by studying how structure is imposed upon a cline of
differences. The structural linguist has to establish structural categories
by showing how categorial (emic) distinctions relate to non-categorial
(etic) differences. Hence, the structuralist conception of a virtual level
at which there is only structure and no variation is simply a daydream
an illusion created by projecting meta-level ideology onto the object of
The interdependence of variational and structural description can also
be illustrated with reference to the results of the sophisticated statistical
methods (cf. Gries 2003, Croft & Poole 2008) which have greatly increased
the descriptive power of corpus linguistics. Although these results are
achieved as a result of studying patterns of variation, what they do in
terms of the interdependence position I am arguing for is really to get bet-
ter at extracting even the most elusive structural categories (which exist
only to the extent they rise above random fluctuation).
Variationist description is thus not an optional extra, but a dimension
inherent in the overall task of describing a linguistic reality that is struc-
tured, but not just structure. The alternative to variationism is not struc-
tural description, but the misguided polarization in terms of which struc-
ture and variation are at loggerheads. Rather than being mutually
contradictory, variation and structure can only be adequately understood
as part of the same overall account.
Norms and variation 271

4.2 Variation and the linguistic system

In chapter 4 (cf. p. 175), a successor concept to Saussurean langue was

introduced: the linguistic system behind the individual utterances (=
parole) was understood as a property of the sociocultural niche. But what
happens to the notion of a system if we give up the idea of structure as
existing in a virtual world of its own?
Structuralism liberated language from having to mirror the ontological
structure of the world.24 Instead it envisioned a parallel but equally all-
encompassing immanent structure as the object of description: the whole
interconnected set of expressions, describable in terms of their valeur in
relation to each other rather than their link with language-external real-
world categories. Set apart from actual usage (parole), the language sys-
tem (langue) was thought to be first an ideal abstract object and secondly
in some way omnipresent in the speech community.25
I suggest that a social cognitive linguistics can take over the idea of
langue as a property of the community, merely by giving up the first
assumption. For structuralists it was a point in itself to aim for the most
abstract and general description, reducing language to a crystalline
essence that was as far as possible from actual usage. But letting abstrac-
tion run on until the vanishing point of human experience only makes
sense if you are allowed to posit a virtual haven, where the system can
reside in splendid isolation. If the language system is a feature of the soci-
ocultural environment, it can be no more abstract than the actual mecha-
nisms that are relevant in that environment.
The system, therefore, is the collection of expressive options that are
available for speakers to tap in producing actual utterances. Structural dif-
ferentiation means that these options are linked and subsumable in cate-
gories, which entails a degree of abstraction; but because linguistic struc-
tures survive by reproduction, linguistic systems tend to have roughly that
degree of abstraction which is functional for speakers. This is why a major
part of linguistic knowledge has list form, as discussed above in relation to
the constructional and lexical dimension of language. Most facts about
language structure are low-level (Verhagen 2002), bound up with items
and their more or less idiosyncratic properties, semantic as well as distri-

24 Cf. the term speculative grammar, from Latin speculum = mirror.

25 This was what gave linguistics the status of model science for the humanities:
other disciplines (literature, anthropology, etc) also envisioned autonomous
structures for their own domains of description
272 Chapter 6. Structure, function and variation

butional. The most abstract and general end of the spectrum is the struc-
tural principles for constructing complex utterances, i. e. the function-
based structure of the clause (cf. p. 251). The chief reason why these
syntactic properties are more tightly structured than others is the process-
ing factor, cf. Hawkins (1994): if you want to produce and understand
utterances at the rate of several words per second, you have to be able to
compile complex meanings according to a very simple and general set of
operations there is little room for superfluous idiosyncratic complexity
on that point.
As pointed out by Croft (2000: 21, citing Hull 1988: 417), it has been
assumed that the structural integrity of an animal was the paradigm exam-
ple but language structure is really more like that of a plant, where dif-
ferent parts can more easily be linked or severed without endangering the
whole organism. English could shed the remainder of the subjunctive
without problems, just as it has shed nominal gender. This looseness also
has implications for the understanding of variation. For the same reason
that the system as a feature of the sociocultural niche does not demand a
maximum of abstraction, it does not demand a minimum of variation. The
functionally regulated boundary condition is that categories have to be
recognizable: variation that goes beyond recognizability would be penal-
ized by not enabling the evocation of intended categories.26 Within that
condition, several different forms of variation can be integrated into a
usage-based reconstruction of language as system.
The most basic kind of variation is the one that is inherited from fluc-
tuation. It includes within-category differences between actualizations in
different cases. Such variation follows from the fact that superimposed
structure takes over the properties of the substratum. But fluctuations
also transgress structurally imposed boundaries, because speakers make
mistakes, leading to accidents de la parole, i. e. cases where there is a mis-
match between instantiation and relevant features of the category (note
that unless both levels exist, there can be no mismatch).
Beyond fluctuation we find the lectal kind of variation that goes with
social differentiation within the speech community. In a structural view of
language, this kind of variation is typically handled simply by transposing
the idea of a whole system to lower lectal levels dialects, sociolects and
idiolects. A lect thus inherits the same kind of structural integrity that is

26 The speech of schizophrenics may occasionally demonstrate how reduced

attunement to feedback mechanisms may cause speakers to end up outside
the boundaries of comprehensibility.
Norms and variation 273

traditionally ascribed to a language, as reflected, for instance, in the criti-

cal maxim that a language is a dialect with an army and a navy.
In a variation-based approach, however, the issue cannot be handled
by multiplying the number of monolithic systems. The social side consti-
tutes an extra dimension of the same overall picture: for every single
linguistic choice, the speaker situates herself in a social universe, in addi-
tion to making a choice of what conceptualization to invoke. To illustrate
this basic variationist point, let us take one of the most pervasive obser-
vations in sociolinguistics: the co-existence of high-status and low-sta-
tus alternative forms. Labovian sociolinguistics as well as matched guise
experiments (cf. Lambert et al. 1960) show the almost ubiquitous co-
existence of competing sets of criteria for what is the right way to speak.
One type is oriented towards overt prestige, and endows the speaker with
the status as a powerful, well-educated and high-status person. The other
is the community variant, which makes you come across as more loyal,
friendly and nice to be around than the prestige form. This duality is
independent of what precisely the linguistic differences are between the
variants associated with overt as opposed to covert prestige. It can be
found across a range from truly diglossic situations to cases where it is a
matter of subtle nuances in lexical and phonological forms within the
same language.
Before we expand this picture, it is instructive to consider its basic
implications. It involves a split that rules out any attempt to reduce lan-
guage in the niche to a single non-social or society-neutral structure, or
to filter off the social dimension as a separate issue. The existence of dif-
ferent social locations is reflected in the choice between linguistic forms
that situate speakers with respect to those social locations, in this simple
case with respect to a vertical dimension of power and status. In its classic
form, this axis goes with a horizontal dimension of geographical differ-
ences, which are greater the lower you go on the vertical dimension. Geor-
die, Scouse and Glaswegian vernaculars differ more between them than
regional varieties of standard English. Bernard Shaws famous dictum
that it is impossible for an Englishman to open his mouth without making
someone else despise him is a polemic way of saying that it is impossible
to speak English without placing yourself somewhere within this sociocul-
tural triangle.
But this is of course only a very basic sample of the expanded universe
that opens up for a cognitive linguistics that aims to include the social
dimension fully. As demonstrated by the work of the QLVL group (cf. the
discussion in ch. 2), including the social dimension involves a radical
expansion of the task of cognitive and usage-based linguistics. In addition
274 Chapter 6. Structure, function and variation

to the triangle presented above, there is the ethnic dimension, the age
dimension, the subcultural dimension, the gender dimension, etc., etc. It
follows from what has been said above that a division into separate lects
would be simplistic here, too: there is no such thing as a separate closed
system for old people, for immigrants, for hipsters and for women (etc.).
The opposite extreme would be a continuous cline on all dimensions, so
that you could place yourself anywhere you wanted on a gradient from
young to old, from autochtonous to immigrant, from male to female, etc.
That would be just as unrealistic as a complete lectal discontinuity.
What exactly the reality is like can only answered by empirical investi-
gation. For all dimensions, in all communities, a large number of questions
raise themselves about what options are available for speakers when it
comes to situating themselves by their choice of linguistic forms. Actual
occurrence, however, is not enough; in order to capture the social causality
of choosing particular varieties, it is necessary to include the matched
guise type of investigation otherwise we cannot know what the force
of these choices are. This is particularly relevant in case of the type of
variation that is experimental and explorative, including the kind that is
driven by processes of identity construction (cf. Eckert 2000 and the dis-
cussion in ch. 3). The study of this type of variation, which has to some
extent replaced the attempt to explain variation by reference to fixed
static parameters (e. g., class, sex and age), is a particularly clear example
of why variational dimensions need to be seen as co-existing in one socio-
cultural niche. If each variety was assigned to a separate lect, the use of
linguistic choices for purposes of identity construction would be beyond
the horizon. In order to study such processes it is necessary to map out
both the distribution and the force of all relevant choices in the relevant
Much of the variation takes the form of differences in salience and bias
within categories, cf. the description in Geeraerts (2000: 75) of salience as
the place where structure and use meet. Salience manifests itself in fre-
quency, for example of a given meaning within the overall potential. An
example is that containment is found to be more central than inclusion
in the semantic potential of in (cf. Vandeloise 1994, as quoted in Geer-
aerts 2000: 80). This kind of salience can be seen as a structural reflection
of pragmatic phenomena because instead of placing options simply as
alternatives, as is traditionally done, salience adds a weight factor, giving
rise to probabilities rather than possibilities (Geeraerts 2000: 76). This
simultaneously introduces an essential interface between semantic intui-
tion and empirical quantification, cf. Grondelaers, Speelman and Geer-
aerts (2007: 998999): quantitative features of the object of investigation
Norms and variation 275

require a quantitative methodology, but the measurement presupposes

the qualitative distinctions within which variation occurs.
Labovian sociophonetics, like the investigation of presentative er, cf.
Grondelaers et al (2002), illustrates how the interest in variational descrip-
tion is in the combination between the quantitative dimension and those
qualitative features that are relevant for variation. The qualitative dimen-
sions may include features of the coded meaning itself (e. g. the accessibil-
ity factor in the case of er). Further, however, it may involve the connota-
tive dimension of meaning i. e. the variational values viewed as meaning
rather than as contextual parameters, cf. Grondelaers, Speelman and
Geeraerts (2007: 99495). Under the traditional landscape of structural
criteria, therefore, a rich landscape of factors emerges that have different
weights in different circumstances but are part of the same overall system
of options.
The connotative dimension of linguistic communication and coding is
part of what I called the directional nature of meaning (cf. above p. 193).
In standard cases, connotations point to aspects of the situation that con-
stitute the point of departure, not the point of interest: you do not nor-
mally use, e. g., a French informal word (e. g. merde!) purely in order to
convey that this is an informal francophone situation. Connotative effects
can be almost totally dormant, especially in the case of speakers who stay
within geographical and social pockets that approximate homogeneous
sub-communities; they increase in importance the more socially complex
the speech community and the speaking practices are. As illustrated by
the case of honorifics, they may even enter into grammatical structure.
It may appear that including variation in langue is tantamount to giv-
ing up the idea of a system entirely. But the only thing that is really given
up is the idea of maximum structural abstraction and integration. Among
the remaining important aspects of the langue concept is the fact that in
the system, the choices are specified in advance and outside of the indi-
viduals control and that one choice has implications for others. If there
are no words for naming different kinds of butterfly (as in Mlabri, cf.
Rischel 1995) or no prospective future form available in the community
(as in Danish), you cannot choose these options, whatever your individual
communicative intentions might be. And if you have decided to use a
polite form of address, this has consequences for other linguistic choices
that you make, just as the choice of a plural subject may have implications
for the choice of verb form. Even in a social cognitive linguistics, the indi-
vidual cannot with impunity buck the system.
But, it may be objected, is not the only thing that remains the restric-
tions on individual elements? Since each individual linguistic item has its
276 Chapter 6. Structure, function and variation

own idiosyncratic distributional pattern, this appears to rule out any inde-
pendent role for higher levels of the system (because they can be no more
pervasive than the elements they apply to). Similarly, if individuals can
pick and choose among available forms, there can be no more coherence
than allowed by individuals choices. It may also be suggested that since
synchronic variation is closely associated with its diachronic counterpart,
instability, it entails a state of chronic fluctuation in the system. Does it not
follow from this expansion of the system that any apparent coherence in
the system viewed as a property of the sociocultural niche is really just
epiphenomal, i. e. an artefact of the linguists urge to impose an overarch-
ing order upon his world?
However, as pointed out by Dahl (2004: 3739) in his discussion of
structural complexity, more complex levels of systems have a life of their
own. Social patterns of behaviour typically persist regardless of the erratic
behaviour of individual members; and the speech community typically has
both more stability and more variation than the speech of an individual.
Analogously, regularities at more abstract levels may persist where prop-
erties of individual elements vary: across Germanic languages, the class of
strong verbs has been more stable than individual members that belong to
it. The concept of confirmation bias is familiar in psychology, denoting
the tendency to maintain existing interpretations even in situations which
would appear to undermine them; and Hutchins (1995: 239) describes the
circumstances in which a collective body will reinforce such a confirma-
tion bias. Using the properties of a constraint-satisfaction neural network
described in McClelland 1986, Hutchins show that if there are two (sub)
sets of units, each of which satisfies the constraints of the other units in the
subset, once a network has settled into a state that fits one of these subsets,
it will be very difficult to change the interpretation state of the network.
The simple reason is that if the network changes as a result of adverse
input, the immediate result is likely to be less overall consistency/harmony
inside the (individual) network. Each unit is kept in place by the others.
If you build a network of networks, integrating all units in a society of
networks, the interconnectivity may reinforce the tendency towards find-
ing and maintaining a shared interpretation: not only internal consistency
but also external consistency may drive such a process. A speech commu-
nity is one instance of such a network of networks. This is of course par-
ticularly clear in areas subject to functional feedback mechanism, which
work by allowing individual members to mutate, while biasing selection
against dysfunctional deviations.
Norms and variation 277

4.3. The role of norms

The kind of existence that language can have in the niche, i. e. outside indi-
vidual minds, was described in ch. 4. The key idea is that although mental
content can only exist in individual minds, minds can interact in ways that
create larger, collective mental constructs which depend on individual
minds but create a causal pattern that works independently of the indi-
vidual mind. Like beaver dams, collective mental constructs become part
of the niche out there, even if their existence is due to properties inher-
ent in the individuals themselves. The basic channel for the rise of such
mental constructs in social space is the intersubjective permeability of
minds, which operates via the joint-attention capacity. Functional feed-
back mechanisms channel the trajectory of arising collective mental con-
structs, so that some mind-dependent collective constructs are confirmed
and replicated while others die out again because they enter into joint
attention events too rarely.
The offline existence of such collective constructs is generally under-
stood in terms of the concept of norm. This goes back to Saussure, who
took over Durkheims view of norms as the basic glue that keeps society
together (cf. the discussion p. 141); in the context of cognitive linguistics,
the foundational significance of norms has been emphasized by Itkonen
(1978; 2008). What exactly are norms, when they are viewed as part of the
First of all, the quasi-objective status that Durkheim stipulated for
norms as part of social reality is replaced by the status as emergent prod-
ucts of interactive events. Norms exist to the extent they are confirmed in
action. As such, they work not by direct causal single-step triggering, but
by the functional mechanism that is associated with differential reproduc-
tive success. The causal power of invisible-hand level norms is that it is the
source of selection pressures. Norms (for speech, clothing, career choice
etc) therefore bias actions towards the normative standard. The mecha-
nisms are analogous to the aggregate market pressures that determine the
success rates of business companies in ways that may defy the best efforts
of the individual entrepreneur. The norm-based effect of a linguistic
expression in the community is analogous to the market price of a com-
modity: they are partly the result of forces that are beyond the purview of
the individual speakers flow of activity. The deselected variants (with
Kellers example, the angelic sense of englisch) go out, and centenarians
who have not adapted are at risk of being misunderstood if they use words
according to their own unadapted competencies.
278 Chapter 6. Structure, function and variation

Secondly, it follows from the description of variation above that lin-

guistic norms are a plural and multiple entity (cf. Harder 2003a) rather
than the monolithic singular presupposed by Saussure. As stressed by
Bartsch (1985: 67), there are no plausible grounds for assuming that there
is a single integrated set of structural norms that is copied as a whole into
the mind of each individual participant. Capturing this multiplicity of
criss-crossing and interlocking norms is part of the vast descriptive task of
mapping out the dimensions of variation in the langue of a community, cf
above p. 273)
Thirdly, the relation between norms and subjective evaluation needs to
be clarified. In the days of unified science, there was assumed to be a
chasm between ought and is, such that science could only address ques-
tions of what is, while talk of what ought to be was unscientific. This con-
ceptual model is well-entrenched in linguistics, which understands itself as
descriptive in contrast to the bad old days of normative statements about
the split infinitive, preposition stranding, etc. Especially in the context of
sociolinguistic variation, the issue is very much alive in the form of a battle
against normative statements about correct as opposed in incorrect lan-
guage. However, the clearcut distinction between ought and is has turned
out not to be tenable (cf. Searle 1964) for reasons that have to do with the
collective validation of shared values: if you are a member of a community
that recognizes the category of promises, it follows that if you have made
a promise you ought to keep it.
Because of the unclear status of social facts, it has been easy to conflate
the purely subjective component of norms (the attitudinal aspect) with
the social component. Once it is realized that it is vital for human develop-
ment to become a member of a normative community, sharing experi-
ences that are felt to be good, it is no longer possible to relegate the issue
of the norms for interactive behaviour to subjective evaluation. Norms
underpin the whole apparatus of shared practices, including the conven-
tions that assign meaning to linguistic expressions. If this is so, it is not
possible to be generally against norms. What does that imply for the anti-
normative tradition in sociolinguistics?
It is convenient to address this issue in two instalments. The first is tied
up with a particular, pervasive configuration of linguistic norms, the type
in which there is a standard variety alongside other, non-standard varie-
ties. A standard variety, roughly speaking, is one that is socially recognized
as appropriate for certain types of communication, including official busi-
ness. As such, it is linked with the written language, including spelling, and
with the education system. Part of the norm set associated with a standard
language is that it operates regardless of the regional or class affiliations
Norms and variation 279

of individual participants. This is what made promotion of the standard

language part of the French enlightenment, as discussed by Geeraerts (cf.
above p. 68).
Sociolinguists, as already mentioned, have generally been reluctant to
accord Standard English any special place. The main cause of the unease
is that the standard language is far from neutral in relation to speakers,
since the standard form is always closely associated with the language of
the ruling classes, and teaching and promoting the standard variety inevi-
tably involves a corresponding devaluation of the vernacular alternatives.
What is more, it was almost inevitable in pre-sociolinguistic days to iden-
tify the language as such with the standard variety and demote dialects to
being simply wrong forms. The structural correlate of this view would be
to identify the system with the standard variety.
In terms of the picture offered here, such an identification is of course
untenable: the system consists in all the options that are available in the
community, not just the standard. The sociolinguistic battle against sim-
plistic views of correct vs. incorrect language remains fully justified also
from that point of view. It is convenient here to introduce a distinction
between prescription and norm: a prescription is a speech act laying down
the law, while a norm is a fact of life in the community. Old-fashioned pre-
scriptivism can be fought partly by pointing out that its prescriptions are
without foundation in actual linguistic norms.
But the problem is only partly solved by recognizing the multiplicity of
norms. Part of the anti-normative stance of most of sociolinguistics is
directed also against norms that actually exist as a fact of life in the com-
munity most centrally the whole set of norms that are associated with
standard languages as they actually function. Like Eliza Doolittle, speak-
ers of non-standard varieties have problems getting the jobs they seek;
and matched guise experiments tell us the speakers of the standard vari-
ety come across as more powerful, well-educated and even intelligent.
Such real-life attitudes go naturally with the faulty assumption that the
standard is qualitatively better than the vernaculars.
Here again, we need to divide into two issues. One is the question of
how rigid norms for the use of the standard are or have to be. Such norms
can be changed, like other norms (such as norms against gay lifestyle). In
the BBC, linguistic variety has increased in the last generations; in Nor-
way, support of regional variety is part of the official social norm set. The
anti-normative stance is partly a plea for greater linguistic tolerance. Most
linguists would agree with such a plea. However, it must be emphasized
that in spite of the natural association between them, the argument against
restrictive norms is a different argument from the argument against
280 Chapter 6. Structure, function and variation

unfounded prescriptions. Informing people (preferably tactfully) that

their prescriptive views of language are without factual foundation is the
natural prerogative of an expert. Tolerance, however, is matter of cultural
attitudes, and experts have no privileged position when it comes to telling
people what attitudes to have.
The most difficult issue, however, still remains and that is the one
where the question of the standard is bound up with the foundational role
of norms. It is tempting, in trying to eradicate the divisive impact of norms
associated with the standard language, to go for a wholesale elimination
of norms: actual usage is the only reality, so why should norms be allowed
to rear their ugly heads at all?
Something like this ideal seems to be at work in the following expres-
sion of the scepticism against standard language that is representative of
the general attitude in sociolinguistics:
it seems appropriate to speak more abstractly of standardization as an ideol-
ogy, and a standard language as an idea in the mind rather than a reality a set
of abstract norms to which actual usage may conform to a greater or lesser
extent (Milroy & Milroy 1991: 2223)

If there is really no standard language out there, and it is only a norm in

peoples minds, why not just forget about all norms so that each individual
can speak the way they do?
However, it follows from what has been said above that this is in
principle impossible (cf. also Bartsch 1985: 149). Cheval only means
horse because there is a collectively recognized norm according to
which this is the case. If we eliminate the element of approximation to a
norm and regard language solely as a property of the individual, under-
standing would be a random accident. The functional pressure for con-
verging on a collectively recognized linguistic norm is therefore ineradi-
cable. Deviation from such a norm will therefore inevitably be recognized
as such.
One might therefore propose as the minimum amount of normativity
what Bartsch calls the highest norm (Bartsch 1985: 171172): that one
should express oneself in a way that the listener understands, and under-
stand utterances as intended by the speaker. Comprehensibility is often
the slogan that is set up against correctness in language education and
elsewhere. But on closer inspection, that is not enough because it is not
a norm of the kind that makes linguistic communication possible in the
first place. The norm of comprehensibility does not enable cheval to mean
horse. Its status is more appropriately understood in relation to intersub-
jective behaviour in general, as a variant of the basic attunement towards
Norms and variation 281

joint action in the sense of Tomasello and Clark (and also Habermas, cf.
p. 364). Hence, the so-called highest norm, viewed as applying to linguistic
communication, in fact presupposes the existence of other norms, without
which comprehension would be a matter of natural or inferential com-
munication alone, unaided by shared linguistic expressions.
We are back, therefore, with linguistic norms as necessary behavioural
targets in social space, to which speakers have to adapt in order for lin-
guistic communication to be possible. That does not logically entail the
existence of standard norms, of course. But if there is actually a standard
language in operation, the norms that underpin it are sustained by the
interactive practices that replicate them it is not just a matter of atti-
tude. Conformity or deviation in relation to the targets is therefore some-
thing that will inevitably be recognized in the same way as conformity
with other behavioural targets even if tolerance is increased, as we may
hope, with the principle of comprehensibility as the outer limit of toler-
If the status of the standard language is to be changed in more pro-
found ways than by extending the margins of tolerance, it therefore
requires a change in the practices that sustain the standard norm (more on
the process of social construction in chapter 7). From a functional point of
view, one may be sceptical of a battle against the standard ideology
referred to by the Milroys, if it would aim to eradicate the standard
entirely. If we recognize the existence of dimensions of variability as the
normal situation in a speech community, we simultaneously endorse two
positions: one is that individuals place themselves differently on those
dimensions, and also that their practices bring them in contact with each
other (otherwise we would be back with separate lects). Abolishing the
standard would mean that shared business would always have to be con-
ducted across gaps in individual linguistic locations. As long as the meta-
norm of comprehensibility would permit it, this is entirely plausible, and
in fact I personally support this idea for the purposes of face-to-face infor-
mal communication (for instance among the Scandinavian languages).
However, it is a different matter in the case of those practices for which
a standard typically evolves: official levels of communication like the law,
political agreements, and education. They would become subject to poten-
tial misunderstanding because of undetected subcultural differences (sali-
ence, evaluative difference, etc.) of the kind that are inherent in variation.
This may be preferable from a certain point of view to the hegemony of a
standard that reflects the position of the power holders. But it would still
raise problems. In the case of English, it is hard to ignore the argumenta-
tion of Preisler (1995):
282 Chapter 6. Structure, function and variation

The evolution of a multiplicity of culturally autonomous Englishes, far from

proving the irrelevance of Standard English, means that Standard English will
have to be maintained as an instrument of cross-cultural communication (Pre-
isler 1995: 355).
A point that may be overlooked is that norms do not actually block out
discrepant behaviour since they only influence events indirectly, via
functional pressure and adaptation. Among the choices that would not be
available without norms is the option of flaunting your own opposition to
norms, perhaps to impose your own subcultural norms on a subniche in
the community.
Whatever ones attitudes, from a social cognitive point of view there is
a central task that consists in mapping out the norms as part of the way the
world works. Much of that work can be done by extending familiar CL
concepts to the social dimension (cf. Kristiansen 2001, 2008, and the dis-
cussion below, p. 288). Norms are the stuff linguistic niches are made of.
You can be against specific norms in the same sense that you can be
against other aspects of the way the world works, such as political oppres-
sion, climate change or infotainment. But a descriptive sociolinguist has to
be aware that without linguistic norms he would be out of business.

4.4 Norms and individual competency

Putting the system in the niche makes the relation between the system
and the user less straightforward than assumed in the structural tradition,
where the two were supposed to match up. Saussure assumed that the
monolithic system was downloaded intact in the individual user; Chomsky
stipulated that the linguist could work with an idealization projected from
individual competence to a whole society of linguistically identical speak-
ers. The point of the distinction in this book between langue and compe-
tency (cf. ch 4) is to highlight the fact that there is no guaranteed match
and the relationship is one that needs to be investigated empirically.
The miracle of language acquisition can still be assumed to get every
normal child to a point of mastery of her mother tongue but what pre-
cisely the mother tongue is has become a more problematic issue. Adap-
tive pressure can, as always, be assumed to maintain a pattern whereby
(surviving) individuals acquire a satisficing mastery of the language in
the primary group that centrally includes the mother/caregiver. This is the
linguistic dimension of the basic socialization process that turns the child
into a member of a normative community (cf. ch 2). How does this funda-
mental competency fit into a wider social context?
Norms and variation 283

As an illustrative starting point, we may postulate an idealized Ur-

scene in which the primary group is identical to the whole society: a small
band of hunter-gatherers where everyone speaks to everyone else, main-
taining joint linguistic norms on a daily basis. In such a situation, system
and competency would match up except for purely idiosyncratic differ-
ences (as in the case of the field linguist whose first informant turned out
to have a speech impediment).
In less basic speech communities, however, the dimensions of variabil-
ity presented above will entail that individual competencies vary at least
when it comes to production for instance, young speakers will differ
from older members of the speech community. With increasing complex-
ity, reception will also tend to become less fluent, the further removed the
variety is from that of the primary group, and individual competencies will
tend to cover a smaller and smaller part of the totality of the linguistic
system that is operative in the niche.
To some extent, this can be captured in lectal terms, cf. above. In a
variationist perspective, the basic concept is dimension of variation and
each choice can then be described with respect to where it situates itself
(according to the norms of the speech community) on the relevant scales.
Common core elements (words like house, and grammatical choices like
simple present) will then be equally at home across all variational scales,
while lectally marked forms index specific positions in social space (such
as using pray instead of please, or omitting the copula, as in he dead, etc).
There is a degree of correlation between different sets of choices which
can be captured by stylistic levels, and the greater the correlation, the
closer we get to a situation in which it would be legitimate to speak of a
particular lect as a system in itself but it should be understood as the
extreme point on a scale rather than a neutral point of departure.
The competency angle on this phenomenon is captured in the term
linguistic repertoire. But repertoire differences are not restricted to lec-
tal differences. Also when it comes to different aspects of what used to be
considered one linguistic system, individual repertoires differ. Dabrowska
(forthcoming) gives an overview of emerging research on this issue, where
especially education stands out as a factor that discriminates between
competencies in different population groups when it comes to mastery of
certain forms of complexity. Like other aspects of variational cognitive
linguistics, this is an area that is only beginning to be subject to systematic
empirical investigation.
A thoroughgoing investigation of this area would also add a much-
needed empirical dimension to the issue of how to deal with standard
language as opposed to lectal variation. In the competency perspective,
284 Chapter 6. Structure, function and variation

linguistic repertoire must be viewed in connection with communicative

competence (cf. Milroy & Milroy 1991: 118). As we saw above, the tradi-
tional sociolinguistic position, that all varieties are equally good, is only
unproblematically valid if understood as expressing the debunking of the
belief that there is something inherently wonderful about the particular
linguistic forms which happen to have overt prestige. It cannot off-hand
be assumed that vernaculars are equally good ways of coping with actual
social challenges.
First of all, language attitudes are real, and even a well-founded convic-
tion that they ought to be changed is no substitute for ability to cope with
the world as it is, including actual attitudes. Secondly, if the standard lan-
guage as a matter of fact is the operative medium for certain types of lin-
guistic practices, there is also a matter-of-fact analogy between incomplete
knowledge of the standard when you are in the territory of standard lan-
guage practices, and incomplete knowledge of French when you are in
France. You might feel that if Frenchmen were more reasonable, they
would accept English on a par with French but that takes the issue
beyond a question of attitudes. In the perspective of individual compe-
tency as opposed to social practices, the practical question for the indi-
vidual is how well she can cope whether the practices should work dif-
ferently is a theoretical question.
Greater empirical knowledge of real linguistic barriers is a prerequi-
site for addressing this issue in a way that is both professionally and ethi-
cally adequate. Speaking on behalf of most of the sociolinguistic commu-
nity, Trudgill (1983: 99100) expressed the hope that one can change
attitudes by making people understand that varieties are just as good as
Standard English; but the point of the argument above is that the mistake
of inherent superiority is only part of the issue. In order to settle the argu-
ment with the adherents of standard English (including, most vociferously,
John Honey, cf. Honey 1997), it is necessary to consider what the realistic
grounding would be of a situation in which all normative ranking is sus-
pended between linguistic varieties (compare a situation in which all
clothing was ranked as equally chic).
Honeys claim that Standard English is better suited than vernaculars
for certain practices could do with more empirical documentation, cf.
Trudgills review (1998), but if we go beyond discredited generic claims of
the standard being more logical etc, the question translates into a pleth-
ora of individual issues about how specific forms of language competency
would in fact serve their bearers for different social purposes under differ-
ent assumptions about the state of norms in the community. We need to
know more about actual barriers as well as actual possibilities for remov-
Norms and variation 285

ing them; the job of the experts must be to tell others what they know
about the lie of the land, and try to keep that apart from the attitudinal
dimension. As Trudgill has also pointed out on several occasions (cf.
Trudgill 1998), the issue of empowerment remains beside the attitudinal
The problem is also relevant in foreign language teaching, most directly
in the question of what form of the language is going to be the learning
target. The general unease about hypostatizing the role of the Standard
language has also manifested itself here. Many teachers prefer to operate
with a goal defined in terms of comprehensibility rather than correctness
in relation to a norm but here again, we run into the problem that the
norm of comprehensibility (as opposed to norms that underpin conven-
tions for how to express a given meaning) does not provide a learning
target. An uncompromising variationist might want to replace the stand-
ard form with the whole spectrum of variations, as the only attitudinally
acceptable learning target. But that would render the learning target
effectively unattainable (for the same reason that a predator needs to iso-
late an individual from the flock before it can attack effectively). If you do
not have a clear target, how can you define a clear path forward? A broad
and imprecise anti-normative stance is no substitute for empirically based
knowledge about what the language requirements are for relevant soci-
etal functions and against whom they constitute a barrier, coupled with
knowledge about manageable strategies for those who want to climb
those greasy linguistic poles.
A general question about the relation between individual and society
in relation to language is: exactly how much power should be ascribed to
linguistic norms in the niche as opposed to entrenched patterns in the
individual mind? After all, field work with a single speaker works: you can
describe a (whole?) language that way. Is it not an exaggeration or a dis-
tortion to say that the individuals system is merely one among other
possible ways of being adapted to a whole linguistic niche with myriad
criss-crossing norms? And even if it is, why not say that the individuals
language is his own individual possession, and societal norms are external
to it once he as acquired it?
There is both a relative and an absolute issue. With respect to the rela-
tive issue, the only generalization that can be made is that the strength of
normative pressures is variable. The absolute issue involves the uncom-
promising fact that normative pressure, strong or weak, is always part of
the picture (for reasons discussed above). There can obviously be no
objection to taking an individual speakers competency as ones descrip-
tive target, as long as it is understood that it means to carve out an ele-
286 Chapter 6. Structure, function and variation

ment from a larger whole: the whole variationist approach entails that
language is not complete in the individual as pointed out in the quote by
Hopper (p. 176). Variation entails that the individual is placed in a field of
forces, which are therefore part of the picture. The role of the speech com-
munity is also manifested in the breach, by the familiar experience that it
becomes difficult to describe a language when there are so few speakers
that the mechanisms that maintain linguistic norms begin to crumble.27
In relation to the scope for individual deviation, it must be emphasized
that norms are no more precise than warranted by actual practices in the
community. In the days in which a homogeneous total system was taken
for granted as the point of departure, a slogan for practically minded lin-
guists was that all grammars leak. A more accurate formulation would be
that grammars are only partially specified. Although practices are con-
strained by the partly invisible hand of community norms, a constraint is
different from a full specification. In biology it is generally accepted that
you can only explain a limited part of the properties of individuals as
adaptations to the niche; the same applies to the properties of linguistic
utterances. Also, some forms may float more than others: the double
perfect in Danish (as in I have had borrowed the book) is subject to
more fluctuation than other compound forms and has not yet (as far as
the linguistic community knows) acquired a systematic place even in a
lectally flexible conception of what the system is. But even such floating
can only be described in a universe where the individuals usage is viewed
against the background of a social field of normative forces.
It should be mentioned that there is one exception clause. You can opt
totally out of the grip of the invisible hand both as an economic agent
and as a language user. However, it means you have to withdraw from
those practices that are under the sway of invisible hand mechanisms. In
terms of economics, you can grow your own vegetables and catch your
own fish, sail by the wind and opt out of the monetary system and the
market economy. In terms of language, you can retreat from the wider
community and establish a niche of your own, modelled on communica-

27 Chafe (1992, in discussion) reported how speakers of a native American lan-

guage were not sure how to respond to elicitations because we dont have any
old people any more. These speakers were themselves in their seventies, but
they were referring to the absence of the community of elders, whose language
functioned as the model for the community. Although they functioned as a
visible hand, their role in underpinning the cohesive force of the norm illus-
trates the same point: I as an individual can only know the language if the
language is there in the community for me to know.
Norms and variation 287

tion practices known among identical twins. In those cases, your economy
and your communication will be determined by your own local practice or
usage alone, with no interference from factors operating over your head.
But as soon as you sneak down to the supermarket and say hello to the
shop assistant, you are again in the grip of the invisible hand, of overarch-
ing mechanisms that define your activities partially behind your back.

4.5. The social dynamics of linguistic variation

Above, we have gone from describing the system in the niche to describ-
ing its relation to individual competency. But there is also another level
that is relevant to understanding the implications of a variationist redefi-
nition of language as a system, and that is the anchoring of variation in
social processes at group level. Lectal variation is the linguistic aspect of
social differentiation, and differentiation entails an issue of how the dif-
ferentiated subsystems interact within the shared macro-niche.
The social processes that influence linguistic norms involve an element
of power, cf. (in addition to Humpty Dumpty and Foucault) also Grden-
fors (1998) on the linguistic power structure. This element by no means
disappears when we move from one integrated system or standard to a
network of interlocking subsystems. Geeraerts (2008) has taken up the
issue in relation to Putnams (1975) discussion of the difference in mean-
ing between elm and beech, where ordinary speakers defer to experts. This
is an instance of meaning being settled as part of a social process in which
individuals have unequal status; and Geeraerts shows how this pattern of
deference is only one among several possible patterns. Citing Bartsch
([1985] 1987), Geeraerts associates the standard CL concept of prototype
with a co-operative relationship driven by the highest norm of establish-
ing mutual understanding, and views this form of collaboration as an
alternative to deference. As the last of the three models, he mentions con-
flict, citing Janicki as the one author who has taken up this issue in the
context of cognitive linguistics.
Conflict is a radical example of the consequences of giving up the idea
of one homogeneous set of norms in the community, since it puts a ques-
tion mark against all other basic assumptions, including common ground.
As pointed out by Geerearts, once this issue is raised, the matter cannot
be left there you have to address the question of how to resolve the con-
flict, or what the consequences are if you do not. This issue will be
addressed in chs. 7 and 8, where the Foucauldian notion of discourse will
be reconstructed on social cognitive premises as a format for describing
288 Chapter 6. Structure, function and variation

meaning-imbued social conflict. The general format of the issue is that

general social relations such as authority, collaboration, and competition
will also be involved in relations between linguistic utterances and their
characteristic features because linguistic norms as well as the actual
processes that sustain them are part of the general social process.
One other type of process that we need to know more about is the
dynamics involving differential social distribution of particular types of
meaning. A familiar example is the interface between lay and professional
terminology, involving terms for different types of saws for carpenters and
different types of plants for botanists. More tricky is the kind of perturba-
tion that is due to different types of meaning that may occur when both
the vocabulary and the area of experience is shared between the groups,
but where different groups have different meanings. Such differences
come on top of language barriers due to variational lectal competencies,
as discussed above.
An example from an educational context may illustrate the problem: a
young, progressive teacher from a modern city background invokes the
phrase doing your duty, cf. Hjort (1984). The text is about subservience to
authority, and the teacher asks how you can make things easy for yourself
by just doing your duty. The intended answer is that if you simply do your
duty, you dont have to think for yourself a cop-out, in other words. To the
students, who mostly belong in a traditional agrarian community, this ques-
tion does not make sense because doing your duty is hard work on a daily
basis. It would be much easier if you could just shirk, but that would be
selfishly letting the family down. There are myriad issues of that kind
involved in inter-group communication that we know very little about.
Because variation involves differential social identity, variational
dynamics includes not only semantic territories associated with word
forms but also the resolution of identity issues. In the case of lectal choices
in Britain, Kristiansen (2001) describes how status-relevant linguistic
choices tie in with a complex and variable sociocultural universe, follow-
ing Tajfel and Turners social identity theory. The centrepieces of the
theory are Categorization, Identification and Comparison (Kristiansen
2001: 134). The comparison element is defined in terms of the following
three assumptions, as cited from Tajfel and Turner by Kristiansen:
1. Individuals strive to achieve or to maintain positive social identity.

2. Positive social identity is based to a large extent on favourable comparisons

that can be made between the in-group and some relevant out-groups: the in-
group must be perceived as positively differentiated or distinct from the rele-
vant out-groups.
Norms and variation 289

3. When social identity is unsatisfactory, individuals will strive either to leave

their existing group and join some more positively distinct group and/or to
make their existing group more positively distinct. The basic hypothesis, then,
is that pressures to evaluate ones own group positively through in-group/out-
group comparisons lead social groups to attempt to differentiate themselves
from each other. [] The aim of differentiation is to maintain or achieve supe-
riority over an out-group on some dimensions. Any such act, therefore, is essen-
tially competitive. This competition requires a situation of mutual comparison
and differentiation on a shared value dimension. (Tajfel and Turner 1979:

The cognitive mechanisms include elements analysable based on CL,

including prototype effects and metonymy. A central metonymy effect is
the one whereby a stereotype is conceived as standing for the whole group.
However, the effect also depends on a process called accentuation, which
is not part of the common-or-garden CL inventory (as stressed by Kris-
tiansen 2001: 142), because it depends on the comparison and competition
element in social identity formation described above. In order to buttress
ones own identity in relation to outgroups, there is a tendency to enhance
similarity on both sides of the divide, thus accentuating differences
between the groups, while providing salient reference points for the cate-
gorizations that are essential to identification and identity formation.
This addition to the cognitive process, as pointed out by Kristiansen, is
essential to the specific type of idealized conceptual model that consti-
tutes a social stereotype, and which plays a significant role in the social
domain. Social complexity thus calls forth simplifying mechanisms that
are not only motivated by the question of what makes categories useful in
general, but specifically by what makes categories useful for social iden-
tity formation. Yet it is essential to understand these as driven by the same
mechanisms that are active in categorization in general, not as something
that is inherently pathological. On the basis of empirical experiments,
Tajfel (whose entire family was wiped out by the Nazis, cf. Tajfel (1981: 1)
as quoted in Kristiansen 2001: 134), thus stresses that exaggeration and
stereotyping (rather than being signs of pure evil) is a way to impose
order and to cope (Kristiansen 2001: 137). But the weights and biases that
skew categorization are part of a web of social causality at the same time
as it draws on the general mechanisms of categorization familiar from
classic CL.
Status as associated with linguistic choices is a niche-level property for
the same reason that status as a feature of career choice is a niche-level
property: it is what other people think that determines what status you
get, not your own mental representations. Status oriented criteria pervade
290 Chapter 6. Structure, function and variation

social reality: being more or less attuned to the kind of action that appears
to be successful and prestigious is relevant for all interactive choices
beyond your own back yard. It is also likely that the non-representational
adaptation mechanisms have a significant impact here, with a smaller role
for conscious awareness than in relation to conceptual content (at least in
spoken communication). This is so because status is more directly associ-
ated with situational force than conceptual content. For the same reason
that the procedural memory adapts to unconscious pin-pricks, we can
assume that the procedural memory adapts to socially successful versus
socially unsuccessful action, so that we tend to adopt the more socially
successful alternatives in recurrent situations.28
In sum, the factors underlying variation, with their differential salience,
weighting and biases (in all senses of the word), are at the heart of the
integrated sociocognitive landscape that is emerging.

4.6 Usage, structure and variation in an evolutionary framework:

a discussion with Croft

The aim of this book is to suggest an overall framework for the ongoing
expansion of CL into the social domain. A not unreasonable question
would be: Whats wrong with the overall format that Croft (2000, 2001,
2009) has already provided? As will be evident, my proposal owes a great
deal to ideas taken over from Croft, including the basic evolutionary for-
mat. In spite of the piecemeal sniping that is also found in various places
above, it may be difficult to see what I understand as the interesting dif-
In terms of basic orientation, the perspective I adopt attaches greater
importance to offline aspects of language; as discussed pp. 94 and 209
above, Croft understands meaning and syntactic structures more wholly
as usage-level phenomena. Very roughly speaking, the difference has to
do with the role I claim for the niche as the locus of adaptive pressure on

28 This is not to say that status is inaccessible to conscious awareness, of course.

One explicit spillover effect is the semantic pathway from terms indicating
social status to terms for inherent human qualities. We would rather be chiv-
alrous and noble than we would be vulgar or churlish which etymologi-
cally speaking means that we would rather be like knights and aristocrats than
like common people and serfs. This is likely to be due less to superior human
qualities in the upper classes than to a mixture of stereotypes and adaptive
deference to members of higher classes.
Norms and variation 291

language use. This role has no direct counterpart in Crofts theory, which
means that more of the action is understood in strictly local, online terms.
Because Crofts theory is so carefully worked out, it is possible to trace
this difference back to an explicit theoretical assumption. Although the
argument may be hard going, I believe it is worth while being precise on
this central point.
In the following extract, Croft (2006) explains where he places struc-
ture in his theory of language as a population of utterances. A key point
is to argue against positing units that are abstract in relation to utterances.
Structure is inherent in actual use:
the paradigm replicator is a linguistic structure in an utterance. This entity
has not played a major role in grammatical theories, which have dealt with ide-
alizations the phoneme /p/, or the periphrastic future construction, rather
than the specific realizations of /p/ or the future construction occurring in par-
ticular utterances. A significant exception is variationist sociolinguistics, in
which the basic data are tokens of linguistic structures; these are called the var-
iants of a linguistic variable. These variants are sampled from the utterances
in a speech community, quantified and correlated with various social and lin-
guistic factors. The term lingueme is coined in Croft (2000: 28) to describe this
(Croft 2006: 104).
Linguemes are centrepieces in his overall panchronic theory, because they
are the structural units that persist across generations (a position that I
take over from Croft). In my view, this entails that they are reproduced by
functional-structural causality impinging on participants conscious activ-
ity. According to the argument produced above, I then have to show that
you cannot individuate such structural units at the level of usage alone, as
Croft claims we should do.
So where is the problem? The way I see it, it is the following: the pas-
sage quoted above ends up with the NP this entity, which on the face of it
could have either the concrete variant or the abstract variable as its pos-
sible antecedents. However, if Crofts argument is to hang together, it
must refer to the concrete variant otherwise we would be back at the
abstract units that he wants to get rid of. A footnote in Croft (2000: 38,
note 4) is relevant here: Croft points to the ambiguity involved in talking
about genes as types or tokens in the biological literature and concedes
that he unfortunately carries over this practice to the way he talks about
his structural units (= linguemes).
This type-token ambiguity is very hard to avoid, because the criteria
for talking about the type always license use also about a token (otherwise
it would not be an instance of the type). In the case of genes, it is not a
292 Chapter 6. Structure, function and variation

problem because genes are instances of physical matter with just those
causal properties that are characteristic of the type. But it is a problem
when the whole point is to avoid presupposing the causal relevance of
structural types (such as the phoneme /p/) and replace them with actual
usage events (where /p/ may be represented by the aberrant variant [f]).
The problem is that linguistic usage events do not subsume themselves
under types with identical causal properties, unlike genes which have (as
instances) the causal properties of the physical type they instantiate. The
problem emerges in the following passage which comes just after the
extract cited above:
By taking the replicator as a lingueme, this evolutionary model is fundamen-
tally usage-based: replication is language use. Replication, that is, language
use, produces variation, namely first-order variation in form, meaning, and
their pairing in grammar (section 4). The recognition of this variation is the first
step in constructing a theory of change by replication.
Change occurs at two levels in replication processes (). Altered replication is
change that occurs in a lineage of specific replications. A linguistic example
would be replication of /p/ as [f] instead of the original [p].

The interesting thing is that in the formulation above, it is /p/ that is repli-
cated (either as [f] or as [p]). But /p/ is the idealized phoneme, or at best
the variable rather than the variant. There is no way it can be the physical
and embodied variant because there are two embodied variants, first [p],
then [f]. A concrete variant [f] is not a replication of a concrete variant
[p] there has to be some abstract categorization in order to make them
the same.
This follows from the argument that was made in the discussion of
postvocalic r in relation to Labovian variationism (above p. 270). Croft
borrows the terminology from Labov, and if we are speaking of the same
variant (type), it follows that a variationist investigation of the two rele-
vant utterances would have to class them as representing the same vari-
ant (rather than two variants of the same variable) which hardly makes
sense. As far as I can see, Croft in fact continues to presuppose the exist-
ence of units based on abstractions from actual usage as indeed he must,
in order to have a criterion for what counts as structural variation rather
than random fluctuation between elements in the flow of usage.
The problem of how to understand the interaction between actual
usage events and structural abstractions from them comes out also in
relation to Crofts theory of conventions. I understand his view as basi-
cally equivalent to what I have spoken of as linguistic conventions or
Norms and variation 293

Linguistic convention is central to the theory of language change. Normal rep-

lication, altered replication, and selection are all defined in terms of conven-
tion (as conforming to convention, not conforming to convention, and estab-
lishing a convention, respectively). (Croft 2000: 99)
Croft bases his notion of convention on Lewis (1969) and Clark (1996), cf.
Croft (2000: 95). Convention is a means of achieving co-ordination
between individuals, including when co-ordination involves linguistic
utterances. But although conventions are essentially collective, as proper-
ties of the speech community, their manner of operation is still individual-
based and therefore in terms of my distinction between individual-level
and invisible-hand-level properties, the same problem surfaces once again:
Convention is a property of the mutual knowledge or COMMON GROUND
of the speech community. Of course, common ground is found in the minds of
speakers, albeit shared with other members of the speech community (Croft
2000: 7)
The problem to my mind becomes relevant in the continuation of the pas-
Thus there is an interplay between convention and individual speakers knowl-
edge, or COMPETENCE as it is usually called.
If the convention is in the individual mind, what exactly is the mechanism
whereby it interacts with individual speakers knowledge which is also in
the individual mind?29 As far as I can see, the problem is insoluble because
conventions cannot interact with individual minds if they are inside indi-
vidual minds but according to Croft they cannot have meanings if they
are outside individual minds. Croft explains this in terms of the deficien-
cies of the conduit metaphor:
The obvious error is that thoughts or feelings cannot go anywhere outside of
the minds of humans, whether it is into words or into an external space The
Lewis & Clark model of convention and communication also avoids this error.
The whole idea behind a communication problem is that getting speaker and
hearer to converge on the same meaning is a problem precisely because our
thoughts cannot leave our heads.
A less obvious problem that follows directly from the primary one is that
linguistic expressions do not contain meanings. Meaning is something that

29 One theoretical possibility would be to put the common knowledge in a place

distinct from the rest of the individuals knowledge, but since Croft explicitly
denounces the idea that there is a separate dictionary entry distinct from gen-
eral knowledge, that is not an option.
294 Chapter 6. Structure, function and variation

occurs in the interlocutors heads at the point of language use (speakers mean-
ing), or something that represents a memory of a history of uses available to a
speaker, albeit organized into senses and sense relations all embedded in a net-
work of encyclopedic knowledge. Reddy emphasizes this fact, that for instance
a body of texts does not have meaning; there must be a basis for readers to
evoke meaning in their heads through shared knowledge with the culture that
produced the texts.
It is not so obvious that the Lewis & Clark Model of convention avoids this
error. Instead, in the preceding section I argued that one must interpret the
notion of a recurrent situation in the definition of convention quite loosely in
order to accommodate the fluidity of meaning.
Croft (2000: 111)
Crofts claim that linguistic expressions do not have meanings is expli-
cated in such a way that it is possible to compare quite precisely with the
position of this book: for Croft, meaning exists at the level of usage (at
the point of language use) and as competency (a memory of history of
uses) but it does not exist as langue meaning. Recognizing that this cre-
ates a problem for the notion of convention, Croft points to the need for
a loose interpretation of the notion of those recurrent situations which
underlie the rise of conventions. But fluidity does not really solve the
problem: if expressions cannot have meanings, they cannot have fluid
meanings. In practice, Croft also talks about conventions as belonging at
the community level:
speech communities arbitrarily pick one solution, say the string of sounds
butterfly to mean the insect, and stick with it (Croft 2000: 97)
And he also speaks of language in a way that sounds suspiciously essen-
tialist (cf. Croft 2000: 17), and certainly is not equivalent to the definition
in terms of a population of utterances:
At this point we have a precise definition of how a language system is a conven-
tional system for communication: language is a conventional signalling system
(Croft 2000: 99).

Although Croft cashes out these notions in ways that point back to indi-
vidual speakers, I think these formulations are entirely justified in the
form cited above, i. e. as properties of the community rather than the indi-
vidual. As argued throughout this book, social entities depend on indi-
vidual minds without being reducible to them: they constitute supra-indi-
vidual configurations of mental content in the minds of a collective of
individuals. Langue meanings are more structured than mere traffic jams
of meaning they constitute continuing regularities of meaning-assign-
Norms and variation 295

ment in the niche to which speakers have adapted. Because they give rise
to selection pressure, they are in terms of the causal structure of evolu-
tionary processes necessarily in the individuals environment, not inside
the individual.
As such, a langue is a social construction, and social constructions are
the subject of the next chapter. Anticipating the discussion slightly, a cen-
tral property that goes beyond the individual mind is efficacy (cf. p. 310).
A viable social construction stands on two legs: it could not exist unless
participants were capable of understanding it mentally, but in addition it
also needs to be actually working and this is what drives adaptation. The
langue of a given community can be described as the set of affordances
that would enable strangers to participate in communication, if only they
knew the efficacious expressions and their meanings. And that can only be
the langue meanings, because at that point, there is neither a usage mean-
ing nor a competency meaning in existence for the poor outsider only a
social configuration of meanings from which he is excluded.
As far as I can see, Crofts objections against assigning meaning to lin-
guistic expressions as such do not apply to this view. In the niche view,
moreover, variation gets exactly the status described by Croft: when dif-
ferent people enact what they see as the same mental operation, it need
not be exactly the same thing that happens in each case thanks to the
mechanisms of co-ordination with other minds, [f] can be assigned the
status of a variant of /p/. Fluidity and flexibility are therefore essential to
get conventions to work, just as Croft points out. Because of that, we can
also describe how Crofts promissory note on the interplay between the
individual and the collective convention of the community works in prac-
tice: it works by causally efficacious feedback when you try out the words.
If a speaker of a Germanic language goes to Italy, he might be excused for
believing that calda means cold. If because of that mental representation
he then asks for agua calda insistently enough, the waiter might actually
bring him a cup of hot water. If he tries to blame the waiter, other people
will inform him that he was in the wrong: his mental representation did
not accurately mirror the actual causal powers of the word. He may then
adjust his own mental representation for the very good reason that it did
not match the conventions that were actually in force in the community.
The fact that the two elements are understood as the same is due to the
concrete efforts of individual minds; unless people are collaborative in
such cases, things go wrong. But the whole mechanism of conventions
works because individual efforts are aided by causal powers stemming
from co-ordination between individuals. As pointed out by Hutchins (cf.
p. 276), you can change the working of the whole system merely by chang-
296 Chapter 6. Structure, function and variation

ing the way the individual networks are linked up, without changing any-
thing inside an individual subsystem.
This type of causality is also at work in speech communities. Back in
the days when Rask-Grimms law was getting under way, the [p] sound in
certain words increasingly came to be pronounced as [f] so that in the
same words (as understood by members of the Indo-European speech
community) the segment could come out in two different ways. Also dia-
chronically, sameness arises from matching the different segments to
something which is collectively the same. Without the causality associ-
ated with the convention, we cannot understand why there would exist
anything shared in terms of which [f] and [p] are variants of the same
thing. If we ask where that operation of sameness-assignment comes
from, we are back at the nine months revolution and joint attention as a
basic prerequisite of speech. Children are genetically programmed to
regard fellow subjects as potential sharers of sameness experiences: they
can attend to the same things as you do, and crucially they can say things
and mean the same thing that you understand by them. Langue is the
system of samenesses that children have to home in on in order to obtain
participant access to linguistic interaction.
It is important to stress that langue, thus conceived, does not come
under Crofts injunction against essences: a language understood as a set
of conventions that are in force in a given community is still defined in
terms of a population of speakers, as he claims. Conventions are just caus-
ally empowered abstractions which have a different life history than the
population of utterances itself. Conventions therefore live and die with
the population of speakers who enter into causal relations with them
just as the population of utterances does. Also, language does not interfere
with the causal pattern that Croft needs to have to achieve his parallel
with evolutionary change. Crofts concept of language as a population of
utterances also plays a role in the picture advocated here, but as usage
(parole in Saussurean terms); it cannot be the sole carrier of structure.
This difference has implications for the way Croft (2009) uses Chafes
pear stories in illustrating the importance of the variationist dimension in
the future social cognitive linguistics. Croft describes the old CL style of
description in terms of variation in construal: informants describe the
pear story in different ways because they construe it slightly differently:
putting, dropping and filling are different verbs conveying different ways
of conceptualizing what happens to the pears. But this is wrong in terms of
the new, social cognitive linguistics that Croft is introducing. If we take the
social-interactional dimension seriously, he argues, the variations in ver-
balizations must be understood as related not just to the speaker, but also
Norms and variation 297

the hearer, and thus against two different past usage histories with the
inevitable difference that this entails. Precise conceptualization is there-
fore not available instead we have to operate with indeterminacy as the
background fact of life in communicative language use: One cannot put
too great a precision on the shared semantics of linguistic forms. This is
in accordance with the fact that language users, relying on shared knowl-
edge, interpret a range of alternative formulations as more or less alter-
native verbalizations of the same scene (cf. p. 94 above).
Although that is true enough, it does not disprove that the formula-
tions convey alternative construals, or that there is a valid and real struc-
tural difference between the potentials they evoke. Any competent lan-
guage user can tell the difference between putting, filling and dropping,
irrespective of the fact that they may use them as alternatives in describ-
ing the same event just as coast and shore may be used about the same
border between sea and land. Retelling a scene is an onomasiological task,
and it can come as no surprise to anyone that such a task can be accom-
plished in different ways and still be the same task. What description you
get out of a variational analysis depends on what type of unit you choose
(within which you then look for variation). If you choose a unit defined in
terms the task you perform, you cannot use it to say anything about vari-
ation within linguistic units
The other side of the argument is that variation adds an extra dimen-
sion of precision, which presupposes but goes beyond shared conventions.
If you just evoke the whole potential and leave it at that, your understand-
ing will be very imprecise, cf. ch. 5 you have to superimpose the varia-
tional level on the structural level to get it right (as also argued by Croft
and Cruse 2004). Cases where it does not matter what word you choose
are those where the situational target is so well-defined that it will over-
ride differences of encoding.
As a consequence, there are two interpretations of the conclusion of
the following passage from Croft (2009: 418), one of which is true, while
the other is wrong:
Language is a fundamentally heterogeneous, indeterminate, variable, dyna-
mically unfolding phenomenon, just like the human society it constitutes part

It depends on how you read the adverb fundamentally. If it means at the

fundamental level of flow, it is true. If it means that this type of variation
has the last word, it is wrong.
298 Chapter 6. Structure, function and variation

5. Summary: function-based structure and langue

in a social cognitive linguistics

In this chapter I have argued that the social turn must include a rethinking
of structure that reflects the tripartite understanding of language as flow,
competency and langue. I have also argued that a function-based approach
is necessary to understand the social embedding of structure.
Two issues were presented as central: the function-based dimension of
structure, and the role of variation. Building upon the account in ch. 5 of
meaning as input to communicative action, the account of function-based
structure argued that one dimension of structure reflected a very general
feature of complex, structured behaviour: the operator-operand relation. A
complex utterance is a recipe for how to act upon the existing usage situa-
tion, co-constructing a new step in ongoing interaction. This dimension of
structure is analogous to the structured operations of a washing machine in
bringing about an intended new situation with respect to the laundry:
speakers act on the situation with a view to carrying the interactive process
forward. The argument attempted to demonstrate the existence and impor-
tance of this dimension of structure, which is integrated with, but not reduc-
ible to the conceptual dimension of structure that takes it point of depar-
ture in the units of meaning that enter into complex structure.
This argument constituted a defence of the first two claims announced
in the introduction to the chapter: that the top-down approach to complex
linguistic structure is significant in a way that does not automatically
emerge from a unit-based description, and that structure constitutes an
affordance (a set of offline recipes) in the niche/speech community for
interactive, communicative action. Language structure is both a hierarchy
of functional options and a network of constructions, and both sides need
to be profiled in a social-cognitive theory.
The third claim was that a rethinking of the linguistic system as a prop-
erty of the community niche means that the system must be partial, include
variation, and be relatively concrete, rather than monolithic and abstract.
This proposed alliance between variation and structure is perhaps the
most unfamiliar part of the conception offered above. There are several
explanations why variational linguistics is often seen as an alternative to
structural description, rather than an enrichment of it. The fundamental
reason, however, is that structure has traditionally been understood as
being inherent, underlying, and Platonic more basic than actual usage.
Hence, if you (rightly) believe that actual usage with all its variation is the
most basic manifestation of language, you almost automatically tend to
reject a structure-based description. However, once it is realized that
Summary 299

structure serves to (partially) organize usage, rather than constituting an

abstract ideal order, this fallacious inference loses its foundation.
This view depends on a panchronic, evolutionary perspective, which
allows for a complex set of presuppositional relations between structure
and usage. The basic status of usage is easiest to see in the diachronic per-
spective. In the history of evolution, structure presupposes a flow of com-
municative usage, since without it there would be nothing to impose struc-
ture on. There must have been communicative interaction at the start of
the complex niche-constructional process that gradually created human
speech communities and linguistic competencies. Ontogenetically, a child
cannot learn what structures are available in the community if there is no
communication going on. But once a community language is a going con-
cern, presuppositions go both ways: without a structured set of conven-
tions in the speech community (to which individual competencies are
adapted), articulated utterances would be impossible.
A similar complexity characterizes the status of variation as a specific
aspect of usage. On the one hand, variation is presupposed by selection
pressures, because selection works by favouring certain variants at the
expense of others. Socially imposed structure could thus not work unless
there was already a range of variant expressions to select from (as when
selection pressures operate on genetic variants in biology). On the other
hand, in language the only criterion for treating certain forms as variants
of each other is that there is already a structural pattern in terms of which
they can be assigned to the same category. Therefore we can only do vari-
ational linguistics if we presuppose the structural categories in terms of
which they constitute variants.
On this point the analogy between evolutionary patterns in language
and biology thus fails to hold. In biology, the status of genetic forms as
variants (alleles) can be ascertained by physical properties inherent in
the individual organism. Their causal power does not depend on social
patterns. In language, the causal power of linguistic units depends on the
causal-functional setup in the niche. What happens when you say [tak]
depends on whether you happen to be in Denmark, Poland or Holland,
and thus selection pressures work on variants as embedded in socially
constituted larger systems, not directly on discrete variants viewed as
existing on their own. Everything in language is co-determined by the
causal power of aggregate-level (community-level) selection pressures.
Because the language system is part of the cultural environment, it
includes the spectrum of variation that characterizes all cultural environ-
ments. The structure of linguistic choices is an aspect of the structure of
cultural choices, including what clothes to wear, what food to eat, what
300 Chapter 6. Structure, function and variation

careers to pursue, etc. What has made it appear that language was under-
lyingly uniform is its basic status as shared: a language cannot exist with-
out bridging the gap between people who may be different in terms of
other features (including clothing, eating habits and careers). The more
complex the division of labour, in fact, the more necessary it is to have a
shared language through which overall joint-but-divided labour can be
mediated. The position of langue-in-society is in the middle of this field
of tensions: between the pull towards fully shared understanding in order
to conduct interconnected tasks satisfactorily, and the pull towards an
understanding in terms of individually and subculturally different experi-
ential backgrounds.
Using the countable form a community presupposes that speech com-
munities can be individuated. This is more controversial than it may seem,
in fact very tricky once we leave the traditional idealizations behind. The
traditional approach that takes the homogeneous speech community as
the presupposed ideal condition is one extreme position; the opposite
approach would be one in which each individual was basically understood
as having his own private idiolectal variety. In order to avoid this impasse,
it is useful to view community, like other fundamental concepts such as
conceptualization and matter, as basically non-count rather than counta-
ble. Human beings depend on community for being able to survive, while
the question of exactly how to delimit a countable community-unit is an
open question in fact an empirical question about real social structure.
An isolated island tribe would be easily individuated, while anyone would
have a hard time describing the English speech community.
The basic claim of the niche theory is that there are social structures
in the community to which individuals adapt. The theory does not depend
on assumptions about the precise distribution of such social features into
neat individual boxes. Communities are not defined in terms of their
members, but in terms of collective practices; football clubs, political par-
ties, religious denominations. Individuals are free to define the scope of
their own membership ambitions: you can be a member of as many com-
munities as your social competencies, including language competencies,
allow. The speech community is therefore definable in terms of a set of
collective communication practices, which may be more or less easy to
individuate depending on the general organization of social practices in
the community. In a dialect continuum, community may be virtually
uncountable (although communicative practices may have natural units
in the form of villages and towns).
For speech communities of the kind we tend to presuppose in Europe,
this also means that there is an inherent link between the cohesion of the
Summary 301

langue and the cohesion of the whole set of practices that constitute the
society. Disruptions in the social order are likely to lead to disruptions in
the language system that constitutes one aspect of the way the world
(niche) works. As described by Dahl (2004), when channels of communi-