Representation of Language
Georges Rey
Reviewed by Gabe Dupre
Representation of Language: Philosophical Issues in a Chomskyan Linguistics
Georges Rey
Oxford: Oxford University Press, 2020, £65
ISBN 9780198855637
Analytic philosophy’s relationship with generative linguistics has been highly variable. For a couple of decades beginning in the early 1960s, generative grammar was a central topic in philosophy of mind and language, receiving extended discussions from major figures such as Dummett, Fodor, Putnam, Quine, and many others, with the Journal of Philosophy even dedicating an entire issue to recent issues in linguistic theory. However, attention seemed to wane somewhat in the 1990s, with philosophers of mind focusing more on neuroscience and consciousness, and philosophers of language drawing more on work in formal semantics than syntax. However, recent years have seen something of a resurgence in this area.
Georges Rey’s long-anticipated Representation of Language exemplifies this new wave, culminating years of Rey’s own novel contributions to these debates. The book synthesizes classical and contemporary debates in philosophy with empirical results from linguistics and psychology, providing a highly entertaining window into the often surprising philosophical commitments of, and puzzles raised by, this branch of linguistics.
The book has three main goals, roughly corresponding to its three major subdivisions. Part I aims to introduce readers to generative linguistics both as a body of knowledge concerning the languages of the world and their similarities and differences, but also, and more importantly, as a scientific research project with its own philosophical presuppositions and methodological strategies. In Part II (and Sections 3.3–3.4), the book provides a defence of this programme from a variety of attackers, drawn from across philosophy and linguistics. Finally, Part III presents Rey’s own original take on the nature of the psychological capacities involved in the acquisition and use of language.
The research programme of generative linguistics consists of a set of inter-related methodological and theoretical commitments. Straddling this methodological–theoretical divide is mentalism, the proposal that linguistics is a branch of cognitive psychology. The goal of a linguistic theory is to describe and explain the psychological capacities that enable humans, and only humans, to acquire and use natural language. Famously, Chomsky further argues that such capacities are largely the product of language-specific biological development rather than cultural learning (nativism) and that they are largely invariant across the human population (universalism).
In investigating such systems, Chomsky advocates for what he calls the ‘Galilean style’, according to which the scientific project consists not in the accumulation of massive amounts of generally highly messy data, which our theory must then be fine-tuned to accommodate, but instead in the search for those observations, typically rare or artificial, indicative of an underlying simplicity. The distinction between performance (actual linguistic behaviour) and competence (the specifically linguistic rules underlying performance) is a famous application of this methodological style (see Sections 1.2–1.3 especially).
Rey proposes ‘WhyNots’, interpretable expressions that native speakers can recognize as ill formed but without being able to explain why, as a paradigmatic class of such competence-reflecting data. English speakers, for example, recognize that ‘Who did Caoimhe hear the rumour that is on drugs?’ is not a suitable way of asking which person is such that Caoimhe heard the rumour that they are on drugs, although they can have the sense that this expression suggests such an interpretation.
These WhyNots are important because they seem inexplicable except with reference to highly non-obvious features of our psychology. Given that such expressions will never be uttered, and that adult speakers are usually unaware of these constraints until they are pointed out, it is hard to see how they could be learned. And given the apparent dysfunctionality of such arbitrary constraints on meaning, communicative intentions and conventions seem similarly impotent. So innate features of our psychology are implicated. If such innate constraints can be established, linguistics and psychology must then be integrated with developmental and evolutionary biology in explaining how such an innate system could itself come about.
Rey spends much of the book defending this picture from numerous antagonists. Platonists and nominalists rejecting mentalism, empiricists of various stripes rejecting nativism, Wittgensteinians and behaviourists rejecting the very idea of internal rule-systems, and many others are dispatched along the way, using arguments novel and familiar.
Thus far, Rey’s discussion is fairly typical of those in the generative camp. Where it gets more controversial is when Rey turns in Part III to internal disputes about exactly the nature of the internal linguistic system.
Rey argues that linguistic theories are representational theories. This comes out most clearly in linguistic perception. This, for Rey, consists in representation of environmental stimuli as having linguistic properties, such as being a noun, being grammatical, being aspirated, and so on. As Rey argues that these properties are rarely, if ever, instantiated in the environment, such representations must be systematically inaccurate.
This combination of representationalism with the claim that most such representations are inaccurate Rey calls ‘folieism’. It is in the development of this folieist view that Rey is at his most interesting and provocative. Rey very adeptly finds middle ground between those, like Chomsky, who argue that linguistic theory need not commit to representationalism, and those, like Michael Devitt, who are willing to populate our environments with genuinely linguistic entities. Folieism, Rey claims, can retain the representationalist explanation of how minded beings manage to be ‘sensitive’ to stimuli for which they lack physical transducers (for example, linguistic properties), without embracing Devitt’s inflationary ontology. And he compellingly shows that this folieist position calls for a substantial re-interpretation of the rules of linguistic theory, as governing representations, rather than the linguistic entities they are typically stated in terms of.
While I feel the pull of Rey’s picture, I think he underestimates the difficulties that such a combination of views can lead to. In particular, any representational theory immediately raises the question of how mental states come to have such representational properties: how their contents are ‘grounded’. This is a difficult question in every case, but seems to raise special concerns when the represented objects don’t exist.
Rey’s approach (dubbed ‘BasicAsymmetries’) has it that the content of a symbol (for example, a mental state) is determined by the properties of that symbol on which the behaviour or properties of tokens with that content ultimately depends. For example, that the term ‘Napoleon’ can be used as a common noun to describe hyper-aggressive short people is explained by the fact that this word is used to refer to Napoleon Bonaparte, a (purportedly) hyper-aggressive short person. But the use of this term to refer to the individual is not explained by its use as a common noun. Thus, this referential use is ‘explanatorily basic’, and so determines the meaning of the expression.
The difficulty then becomes figuring out what these explanatorily basic properties are for linguistic representations. The non-existence of the targets of such representations here becomes a worry, as the most obvious candidates for explaining the role of a mental symbol are these states’ typical causes. Standard externalist theories of content (for example, Fodor [1990]; Burge [2010]; Shea [2018]) build on the idea that representation begins with capacities to track environmental objects and their properties. Obviously, this can’t be what is going on when we represent non-existent linguistic items.
As Rey notes, we are of course able to represent non-existent and non-local entities (fictional characters, numbers, and so on), and so there must be ways of representing not dependent on causation in the straightforward ‘tracking’ way. One such way is dispositional: we can be disposed to respond to the presence of an entity by tokening a symbol, even if we never encounter such an entity. This seems plausible enough for representational contents like CUBE, which never fully accurately describe entities in our environment, but which would be tokened if their targets were locally instantiated.(1This dispositional proposal resolves what, to me at least, seemed paradoxical in Rey’s discussion of our perceptual capacities as ‘sensitive to’ non-existents.) Applying BasicAsymmetries, if the actual behaviour of our cube representations is explained with reference to their dispositions to be caused, counter-factually, by the presence of genuine cubes, but these counter-factual tokenings are not explained by their dispositions to be caused by non-cubes, then they represent cubes, despite not being in fact caused by them.
However, even this permissive account of content-determination seems to run into trouble. While I can, I suppose, imagine a world in which the environment contains perfect cubes, fit for content-grounding perceptual responses, it is much harder to know what the world would have to be like for me to accurately perceive a sentence as ungrammatical due to its containing a non-locally governed reflexive. Linguistic perception seems to consist in imposing linguistic structure on essentially non-linguistic stimuli, and so appealing to counter-factual dispositions to respond to ‘good’ cases seems less useful here.
If causal and dispositional approaches to grounding content thus don’t seem promising, what is left? While allowing for Fodor-style externalist approaches, BasicAsymmetries is more general, allowing also for internalist, or inferentialist, representation.(2Rey cites Fodor and Horwich as the two major inspirations for his account.) On these accounts, having a representational content is not a matter of interacting with some extra-psychological entity, but simply of having the right kind of internal, relational structure. Paradigmatic cases here are the logical connectives: the meaning of AND is exhausted by the role this symbol plays in licensing inferences, however such inferential processes are situated in an environment. So, perhaps we could individuate linguistic symbols in a similar way: there is nothing more to say about grammaticality, C-command, sonority, and so on beyond identifying the way linguistic (grammatical and perceptual) processes distinguish between, and differentially respond to, given classes of symbol.
To an extent, I think this seems correct. But, to that extent, this does not seem like a representationalist theory at all. The inferentialist proposal says: when a psychological type is treated in these sorts of ways by psychological processes, it represents. But the representational story then just seems like a third wheel. Nothing is gained by the stipulation that such-and-such computational system is, purely in virtue of these computational properties, also a representational system. All the causal and explanatory work is done by the computational story. There thus seems to be nothing to favour an internalist, representational theory over an account, like that developed by Collins ([2014]), that views linguistic theory as a purely computational, non-representational, enterprise.
Of course, these are big and complex topics. Too big to be dealt with to any degree of completeness in a book review. I have merely gestured towards the reasons for concern, and there are live avenues of response for each. And the non-representationalist view has significant problems of its own, highlighted at numerous points in Rey’s book, especially in his discussion of hypothesis-testing models of linguistic perception, which do seem best-served by a representationalist account. My guess is that no views in this area are likely to be without significant drawbacks. But the only way to figure out what these are, and ultimately which are worth living with, is to spell out each possible approach in as much detail as possible, and Rey’s work is an excellent example of just this.
Despite not sharing Rey’s optimism about the prospects for a meta-semantics for linguistic theory, I think this book is a great achievement. It is huge fun to read, and collects together the highlights of decades of debates involving generative linguists and analytic philosophers without sacrificing detail. Rey’s take on linguistics and cognitive science is often compelling and always stimulating. This book is also that rare creature which will be equally engaging for both the expert and the novice. For those new to philosophy of linguistics, I think there is no better book on the market. For old hands, aspects of the picture will be familiar, but seeing it all in one place generates novel perspective, and there is enough originality in the details to warrant careful reading. I thus recommend Rey’s book to anyone looking to understand why generative linguistics made such waves in philosophy in the mid-to-late twentieth century, and what has been happening in the philosophy of linguistics underground since, and hope that it will be the catalyst for renewed collaboration between these two disciplines.
Acknowledgements
I attended a seminar at UCLA in 2019 run by Mark Greenberg, in which we discussed an early draft of this book, and I organized a reading group at Keele focused on the final draft. My understanding of these issues has benefitted greatly from these discussion groups, and so I thank attendees at both. Thanks also to Georges Rey for e-mail correspondence concerning the book and broader debates in philosophy of linguistics. While Georges won’t agree with everything I say, where we differ has, I hope, been made much clearer by his engagement.
Gabe Dupre
Keele University
G.G.Dupre@Keele.ac.uk
Notes
(1) This dispositional proposal resolves what, to me at least, seemed paradoxical in Rey’s discussion of our perceptual capacities as ‘sensitive to’ non-existents.
(2) Rey cites Fodor and Horwich as the two major inspirations for his account.
References
Burge, T. [1989]: ‘Individuation and Causation in Psychology’, Pacific Philosophical Quarterly, 70, pp. 303–22.
Collins, J. [2014]: ‘Representations without Representata: Content and Illusion in Linguistic Theory’, in P. Stalmaszczyk (ed.), Semantics and Beyond: Philosophical and Linguistic Inquiries, Berlin: De Gruyter, pp. 27–64.
Fodor, J. A. [1990]: A Theory of Content and Other Essays, Cambridge, MA: MIT Press.
Shea, N. [2018]: Representation in Cognitive Science, Oxford: Oxford University Press.