35.1623, Review: The Constructicon: Diessel (2023)

[The LINGUIST List]

LINGUIST List: Vol-35-1623. Fri May 31 2024. ISSN: 1069 - 4875.

Subject: 35.1623, Review: The Constructicon: Diessel (2023)

Moderator: Francis Tyers (linguist at linguistlist.org)
Managing Editor: Justin Fuller
Team: Helen Aristar-Dry, Steven Franks, Daniel Swanson, Erin Steitz
Jobs: jobs at linguistlist.org | Conferences: callconf at linguistlist.org | Pubs: pubs at linguistlist.org

Homepage: http://linguistlist.org

Please support the LL editors and operation with a donation at:
           https://funddrive.linguistlist.org/donate/

Editor for this issue: Justin Fuller <justin at linguistlist.org>

LINGUIST List is hosted by Indiana University College of Arts and Sciences.
================================================================


Date: 31-May-2024
From: Vsevolod Kapatsinski [vkapatsi at uoregon.edu]
Subject: Syntax: Diessel (2023)


Book announced at https://linguistlist.org/issues/34.3277

AUTHOR: Holger Diessel
TITLE: The Constructicon
SUBTITLE: Taxonomies and Networks
SERIES TITLE: Elements in Construction Grammar
PUBLISHER: Cambridge University Press
YEAR: 2023

REVIEWER: Vsevolod Kapatsinski

SUMMARY

The Constructicon is a brief survey of recent developments in
usage-based construction grammar, a major current approach to grammar
in which the grammar and the lexicon form a unified system of
form-meaning mappings (constructions) called the constructicon, and
these form-meaning mappings emerge from language use. As summarized in
Chapter 2, the constructicon was initially conceived of as a simple
inheritance network in which constructions inherited the properties of
more general constructions, with some possibility of override (a
semantic network in the sense of Collins & Quillian, 1969). However,
recent research has reconsidered the structure of the constructicon as
a spreading activation network in which constructions can be linked by
a wide variety of relations, not necessarily taxonomic, and the
relations are reconceived as associations, which means that positing a
relation generates predictions for language processing. The shift from
a taxonomic view of memory to an associative network one generally
mirrors the shift undergone in the study of semantic memory between
Collins & Quillian (1969) and Collins & Loftus (1975) and brings CG
into better agreement with network models in psycholinguistics. This
book reviews these developments, and surveys the evidence for a wide
range of relations, balancing evidence from language structure and
evidence from language processing.

The bulk of the book’s content is in Chapters 2-5, with Chapters 1 and
6 serving as brief bookends. Chapter 2 discusses the basic assumptions
of construction grammar (the grammar and lexicon forming a continuum,
and being formed of the same ‘stuff’ – form-meaning pairings) and the
shift from a taxonomic view of the constructicon to the associative
view. It also spells out what the term association means by listing
salient characteristics of associations in the psychological
literature. Chapter 3 points out that constructions themselves are not
unitary entities but can themselves be seen as networks.
Construction-internal associations include not only associations
between form and meaning but also sequential associations between
forms specifying serial order within a construction and filler-slot
associations specifying what elements can fill the open slots of a
construction (e.g., the fact that donate can fill the verb slot of the
prepositional dative construction, I donated time to the cause, but
not of the double object construction, *I donated the cause time).
Chapter 4 then presents a novel view of syntactic categories such as
Noun, Verb, or Adjective. Syntactic categories are argued to emerge
from networks of associations constituting the constructicon: forms of
the same category are linked by filler-slot relations to the same or
similar set of constructions. This view builds on the conception of
syntactic categories as emergent from the network of constructions in
Croft’s (2001) Radical Construction Grammar but explicitly shows that
both constructions and syntactic categories are emergent objects of
the same type, forming subnetworks of the same global network. This
proposal thereby brings the view of syntactic categories in
construction grammar in line with the view of categories in
connectionist psycholinguistics (Rogers & McClelland, 2005). Chapter 5
then turns to the global network to introduce paradigmatic
associations between constructions, which link together whole
constructions by similarity and contrast. These include associations
between (near-)synonymous constructions as well as members of
morphological or syntactic paradigms (singular vs. plural;
prepositional dative vs. double object) that are exchanged for each
other based on specific features of context. Among the highlights of
Chapter 5 are an innovative extension of the notion of a paradigm to
syntax (see also Zeldes, 2012, for related ideas), and an interesting
discussion of the differences and similarities between constructional
families, neighborhoods, and paradigms.

EVALUATION

This book provides a remarkably concise and insightful exposition of
recent research in usage-based construction grammar. It gives much
food for thought about the nature of our mental representation of
linguistic structure, and should be of interest both to construction
grammarians and those not as familiar with the framework. There is
abundant exemplification (though mostly from English) and most
representational claims are illustrated by clear diagrams. A strength
of the book is that it integrates psycholinguistic evidence for its
representational claims throughout the exposition. For example,
Diessel draws on syntactic priming for evidence of constructional
similarity, and on the visual world paradigm for evidence of
predictive sequential relations between the parts of a construction.

Where the book can be faulted is in the lack of comparisons between
the proposed representations and alternative theories. This makes it
unclear how the representational proposals in The Constructicon can be
falsified. It is evident that the proposals are intended to be in
principle falsifiable by psycholinguistic experimentation. For
example, on p. 15, “the multidimensional network approach (of
usage-based linguistics) is solely motivated by psychological
considerations. In this approach, the links of the constructicon are
defined as specific types of associations shaped by specific
domain-general processes,” and on p. 59 “the main reason for positing
horizontal relations is that they are psychologically real. [That is,
they are] needed to describe a particular type of association that
must not be ignored in a psychologically motivated theory of grammar.”
I share the view that the grammar is a network of learned
associations, and that individual constructions are also networks of
associations rather than unitary entities (Kapatsinski, 2018, 2021,
2022). However, it is often possible to account for the same
behavioral data (such as priming between related constructions) with
many different associative network models. In particular, it is often
the case that it is possible to explain the same data without
reference to one or more of the association types proposed to underlie
the behavior in the Constructicon. This raises an important question
that is not addressed in the book: how, or whether, one should decide
between alternative analyses or even representational frameworks.

For example, Section 3.2 proposes that constructions contain within
them sequential relations that specify serial order between the
elements of a construction (pp. 20-22). These relations include
left-to-right associations between forms (e.g., I→don’t→know) and more
abstract categories (e.g., P→NP), echoing the proposal of Osgood
(1963).. Evidence for such sequential relations is provided by studies
of sentence processing in the visual world paradigm, where
participants watch a visual display while hearing a spoken sentence.
Participants will often look at upcoming referents that have not yet
been mentioned. This is good evidence for prediction, but prediction
can be implemented without storing direct left-to-right associations
between forms. For example, recurrent artificial neural networks
predict upcoming words from preceding context without storing direct
word-to-word associations. Instead, the next word is predicted from a
representation of the preceding context in which the representation of
the preceding word may be impossible to isolate (Elman, 1990, et seq).
In Interactive activation neural networks (which look much like the
proposed conception of the constructicon), sequential prediction is
accomplished instead by the first word activating larger units that
contain it, which in turn activate their other parts (e.g., I → I
don’t know → know; McClelland & Elman, 1986; McClelland & Rumelhart,
1981). In constructionist terms, hearing some part of a sentence would
activate the constructions that are likely to be present in the
signal, which would activate their upcoming parts. In production,
theories of serial order often avoid the positing of item-to-item
associations in favor of mechanisms that impose order via top-down
mechanisms or filler-slot associations. For example, Page and Norris’s
(1998) model of serial order would have I don’t know impose an
activation gradient on its parts, such that I is easiest to activate
from I don’t know, don’t is second easiest, and know is hardest,
ensuring that the three words come out in the right order. Gomez et
al. (2008) argue for representing serial positions of letters within a
word as distributed positional codes in which nearby positions (slots)
are represented by similar codes. One can easily imagine similar
representational schemes for representing serial order in
constructions that would dispense with word-to-word associations. This
is not to say that these alternative theories of prediction and
representation of serial order are superior to the proposal in the
book. For example, there is evidence that preceding context biases the
selection of upcoming words above and beyond the influence of top-down
selection (Bannard et al., 2019; Harmon & Kapatsinski, 2021; Lindsey &
Logan, 2019). However, the present book does not make a case for
word-to-word associations over alternative representational devices
posited in psychology and psycholinguistics.

Similarly, Chapter 5 suggests that there are paradigmatic associations
linking together allostructions (e.g., Agent Action Recipient Theme ~
Agent Action Theme to Recipient, as in I gave my cat a book vs. I gave
a book to my cat). It is said that “Like allomorphs, allostructions
are members of the same category or schema, but since their properties
are not directly predictable from the shared mother node, it seems
reasonable to assume that they are horizontally related” (p. 59). It
is a standard assumption in interactive activation models of
comprehension that competing alternatives develop mutually inhibitory
associations that make it easier to select one over the other (e.g.,
McClelland & Rumelhart, 1981). However, these associations are not
strictly necessary, and there are models that do away with them – for
example, alternatives might be competing for a limited supply of
spreading activation, or dividing up probability mass given the
present cues (Norris & McQueen, 2008). In production, even though one
construction might be rephrased as the other, suggesting an
association between them, it is also possible to accomplish this kind
of toggling between allostructions by top-down means alone
(Kapatsinski, 2022). Furthermore, relations of similarity are usually
no longer represented by associations between nodes (e.g., nodes
representing similar constructions) but by overlap between distributed
semantic representations (e.g., Rogers & McClelland, 2005). For
example, syntactic priming can be accounted for by horizontal
associations between similar constructions, or by semantic and
structural overlap between their representations. The representational
overlap account is generally preferred in modern psycholinguistics
because encoding similarity as association strength conflates
similarity with co-occurrence by encoding both in a single number
(strength of association) and thereby creates rather complex problems
for learning.
The lack of argumentation in support of specific types of associations
over models that omit them may be theoretically justified, given the
‘maximalist’ orientation of usage-based construction grammar. To many
constructionists, the parallel nature of the brain/mind may imply that
the burden of proof rests largely on those that would argue that a
particular connection does not exist, rather than on those that would
want to argue that it does. Indeed, Langacker (1987), in what might be
considered the founding document of usage-based linguistics, correctly
notes that the traditional dichotomy between words and rules is
ill-founded and the fact that something is stored does not imply it is
not also computed. In this view, following Householder (1966), “the
brain makes little use of parsimony”, and it is therefore tempting to
encode every real-world relationship that speakers seem to know about
as a corresponding mental association. Thus, if A and B are in some
salient relationship to each other, an A←→B association might be
posited to account for how the speaker gets from A to B, or from B to
A, even if it is possible to get from A to B through other, indirect
routes.

This direct encoding approach contrasts, however, with the rest of
cognitive science, where one has to argue for the pathways and
associations one posits, based on them being necessary to account for
a particular behavior or experimental finding. The alternative
parsimony approach is also well motivated by the use-it-or-lose-it
nature of synaptic pruning and the fact that neural redundancy is
metabolically extremely expensive – brains are very energy-hungry
organs (Sterling & Laughlin, 2015). More importantly, it is unclear
how a maximalist approach can guide methodology – when are we
justified in positing an association, or even an entirely new type of
association? As a matter of scientific practice, there seems to be no
real alternative to showing that the representational mechanisms we
posit are necessary to account for the data. There are places where
the Constructicon takes this approach, and convincingly argues against
alternative representational theories. For example, I am quite
convinced that filler-slot associations between individual words and
constructional slots are necessary, and cannot be explained away by
semantic overlap between words and constructions because there are
word-construction co-occurrences unaccounted for by semantics (pp.
29-34). A particularly insightful example is the fact that forgive
favors the ditransitive construction despite not having the semantics
of transfer (p. 33). This case shows clearly that a filler-slot
association can persist even as the meaning of the filler changes.
However, the book does not take this approach of disconfirming
alternatives consistently, which leaves the reader without
methodological guidance on how one can or should argue for a
particular association type as being psychologically real.
In sum, The Constructicon is an excellent guide to current thinking
about cognitive representations in usage-based construction grammar
that brings it into much stronger, though incomplete alignment with
connectionist / neural network models of cognition. At the same time,
it does not defend the proposed representations over plausible
alternatives afforded by the broader connectionist/associationist
framework and therefore falls somewhat short of the stated goal of
establishing their psychological reality. By bringing construction
grammar closer to neural networks, the Constructicon does, however,
succeed in begging new questions, which is in itself valuable: by
articulating a concrete proposal, it makes questions about the
structure of the cognitive network underlying productive language use
askable within the constructionist framework, and expands the research
horizons for cognitive / usage-based (psycho)linguists.

REFERENCES

Bannard, C., Leriche, M., Bandmann, O., Brown, C. H., Ferracane, E.,
Sánchez-Ferro, A., ... & Stafford, T. (2019). Reduced habit-driven
errors in Parkinson’s Disease. Scientific Reports, 9(1), 3423.
Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory
of semantic processing. Psychological Review, 82(6), 407-428.
Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic
memory. Journal of Verbal Learning and Verbal Behavior, 8(2), 240-247.
Croft, W. (2001). Radical construction grammar: Syntactic theory in
typological perspective. New York: Oxford University Press.
Elman, J. L. (1990). Finding structure in time. Cognitive Science,
14(2), 179-211.
Gomez, P., Ratcliff, R., & Perea, M. (2008). The overlap model: A
model of letter position coding. Psychological Review, 115(3),
577-600.
Harmon, Z., & Kapatsinski, V. (2021). A theory of repetition and
retrieval in language production. Psychological Review, 128(6),
1112-1144.
Householder, F. W. (1966). Phonological theory: A brief comment.
Journal of Linguistics, 2(1), 99-100.
Kapatsinski, V. (2018). Changing minds changing tools: From learning
theory to language acquisition to language change. Cambridge, MA: MIT
Press.
Kapatsinski, V. (2021). What are constructions, and what else is out
there? An associationist perspective. Frontiers in Communication, 5,
575242.
Kapatsinski, V. (2022). Morphology in a parallel, distributed,
interactive architecture of language production. Frontiers in
Artificial Intelligence, 5, 803259.
Langacker, R. W. (1987). Foundations of cognitive grammar: Volume I:
Theoretical pre-requisites. Stanford, CA: Stanford University Press.
Lindsey, D. R., & Logan, G. D. (2019). Item-to-item associations in
typing: Evidence from spin list sequence learning. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 45(3),
397-416.
McClelland, J. L., & Elman, J. L. (1986). The TRACE model of speech
perception. Cognitive Psychology, 18(1), 1-86.
McClelland, J. L., & Rumelhart, D. E. (1981). An interactive
activation model of context effects in letter perception: I. An
account of basic findings. Psychological Review, 88(5), 375-407.
Norris, D., & McQueen, J. M. (2008). Shortlist B: a Bayesian model of
continuous speech recognition. Psychological Review, 115(2), 357-395.
Osgood, C. E. (1963). On understanding and creating sentences.
American Psychologist, 18(12), 735-751.
Page, M., & Norris, D. (1998). The primacy model: a new model of
immediate serial recall. Psychological Review, 105(4), 761-781.
Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A
parallel distributed processing approach. Cambridge, MA: MIT Press.
Sterling, P., & Laughlin, S. (2015). Principles of neural design.
Cambridge, MA: MIT Press.
Zeldes, A. (2012). Productivity in argument selection: From morphology
to syntax. Berlin: Walter de Gruyter.

ABOUT THE REVIEWER

Vsevolod Kapatsinski is Professor of Linguistics at the University of
Oregon in Eugene, Oregon. He works at the intersection of usage-based
linguistics and learning theory (primarily between morphology and
phonetics). He is the recipient of the 2020 Bloomfield Award from the
Linguistics Society of America (for Changing minds changing tools),
and has written over 30 journal articles in linguistics and cognitive
science. VK currently serves as area editor for computational,
experimental and cognitive linguistics for Linguistics Vanguard, and
on the editorial board of Cognitive Linguistics, and is excited to be
directing the next Linguistics Society of America Summer Institute
(Language in Use, Eugene, 2025).



------------------------------------------------------------------------------

Please consider donating to the Linguist List https://give.myiu.org/iu-bloomington/I320011968.html


LINGUIST List is supported by the following publishers:

Cambridge University Press http://www.cambridge.org/linguistics

De Gruyter Mouton https://cloud.newsletter.degruyter.com/mouton

Equinox Publishing Ltd http://www.equinoxpub.com/

John Benjamins http://www.benjamins.com/

Lincom GmbH https://lincom-shop.eu/

Multilingual Matters http://www.multilingual-matters.com/

Narr Francke Attempto Verlag GmbH + Co. KG http://www.narr.de/

Wiley http://www.wiley.com


----------------------------------------------------------
LINGUIST List: Vol-35-1623
----------------------------------------------------------