A Critical Study

Ray Jackendoff, Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press, 2002. Pp.477. H/B.

In Foundations of Language, Jackendoff synthesises his recent thinking on language and its place in the mind/brain into a coherent and accessible thesis. The book is written for linguists, psychologists and philosophers, as well as non-specialists; it should prove to be of interest to all who seriously think about language.

Jackendoff’s broad aim is to get the non-linguist to appreciate that linguistics (generative grammar and its off-shoots, properly speaking) may genuinely speak to wider concerns in cognitive science and philosophy, and, concomitantly, to get the linguist to appreciate that if their discipline is really a branch of psychology (and thus, ultimately, biology), then a proper integration of linguistics into the other sciences of the mind/brain should be pursued rather than foreclosed in the name of autonomy. Jackendoff does not merely wish for such integration; he actively pursues it with a wealth of data and elegant theory. The focus of the sequel will be on the aspects of this work that directly impact on some of the current concerns of philosophers. The book divides into three parts and, each will be considered in turn. As will become clear, each of the parts is in a mutually supportive relation with the other two.

The first part deals with the battery of concepts, distinctions and hypotheses that constitute the theoretical hard core of generative grammar: nativism, universal grammar, poverty of stimulus arguments, competence vs. performance, mental representation, knowledge of rules, etc. For Jackendoff, while the seminal arguments for the existence of innately structured competencies in syntax, phonology and semantics remain intact, their traditional elaboration carries an illicit commitment to mental representationism, as signaled in the notion that generative linguistics is essentially concerned with the knowledge a speaker/hearer possesses. Jackendoff seeks to exorcism linguistics of this representationism.

The philosophical interest in generative grammar, at least from the time of Chomsky’s psychologising of linguistics in the mid 1960s, has focused on the issue of innate knowledge and, correlatively, mental representation - the putative object of knowledge. This approach has divided philosophers. On the pro side, generative grammar has been seen as a basis from which traditional philosophical concerns to do with representation, the mind and knowledge of meaning may be naturalistically pursued.[1] On the con side, the ‘Chomskyan revolution’, especially in its eschewal of performance and social aspects of language and its explicit Cartesianism, has been seen as philosophically naïve, beset with deep confusion over the a priori normative and communicative dimension of language.[2] Jackendoff, with Alexandrine panache, cuts through this tangle by expunging the notions of knowledge, information, representation and even mental from the theoretical lexicon of the linguist. Of course, a rose by any other name, it will be thought, would engender the same philosophical controversy. However, Jackendoff does not merely substitute terms, but removes the underlying representational connotation. Central to this notational overhaul is the use of ‘f-mind’ in place of ‘mind’ (and its cognates). The f-mind is understood to be the high-level structural/functional organisation of the brain. This approach effects a double dissociation from the philosophical traditions precisely by removing the object of contention, ‘representation’. There is no naturalisation program to follow, for there is nothing that is not straightforwardly ‘natural’ to begin with; linguistics does not commit us to entities that suspiciously straddle the natural and semantic worlds, as it were. Nor is there, a fortiori, anything to preserve in the face of naturalisation or eliminative programs. The real issue, for Jackendoff, is whether particular theories of high-level organisation prove to be explanatorily indispensable, and, if so, how they might be implemented and integrated into the wider sciences of the mind/brain. The pressing question, therefore, is whether there is a loss of explanation if we descend to a more direct study of the brain and eschew specifically linguistic concepts. Jackendoff argues that, where language is concerned, abstract descriptions of phonology, syntax and semantics are indeed necessary to capture the required generalisations of both linguistic development and mature competence. In this defense of the integrity of linguistics, the notions of knowledge and representation are not doing any explanatory work; the work is being done by the network of concepts and distinctions our theories posit, for this network belies low-level theorising independently of whether the subject is taken to know or represent the structure. As goes the mind, so goes representation, and so goes knowledge. Syntactic structure, for example, is not represented in or by the f-mind, it is simply part of it: the structure is internally individuated by the set of distinctions the syntactic component of the f-mind may realise, where these differences are reflected in the component’s interfaces with other cognitive components of, principally, phonology and semantics. Equally, the linguistic f-mind does not represent anything ‘external’, for there is nothing out there to represent. Nouns, inflectional heads, empty categories (phonologically unrealised pronominal-like elements), etc. are not to be found in the world. Indeed, phonology is not even in the world (acoustics is simply not phonology), nor are words themselves in that they consist of clusters of features drawn from the three components. To say, therefore, that a speaker S knows a language is, if intended theoretically, just to say that S’s f-mind realises such and such structures that support such and such interfaces between syntax, phonology and semantics. The details are to be filled in by developing theories of these three components. If intended non-theoretically, the idiom is fine, but we should not be led into thinking that there must be some ‘object’ of the knowledge.

Jackendoff does not explore the philosophical response this reconstrual is likely to attract, although in Part III he considers issues to do with conceptual analysis and theories of meaning that have been assumed to presuppose a representational understanding of the mind. Still, in broad terms, Jackendoff’s stance may be understood as an attempt to elaborate a genuinely naturalistic framework, one constrained to ‘best’ explain identifiable phenomena free of a priori or metaphysical stipulations as to what language or mind or reference must be.

Those philosophers who have always spurned the insights of the generative program, the con party, as it were, would perhaps welcome Jackendoff’s approach as showing that linguistics does not really encroach on epistemological and representational issues: ‘As we have always said, these issues are normative, not descriptive or theoretical’. Such a reaction would grievously miss Jackendoff’s point. Let me quote him: “If one wishes to join the conversation about the nature of language, one must recognize and acknowledge this complexity [the interfacing of detailed phonological, syntactic and semantic structure]… one may not willfully ignore it and still be expected to be allowed in the game” (p.18). Thus, Jackendoff is not inviting the philosopher to ignore linguistics because, pace Chomsky, it has nothing to say to him. On the contrary, with Chomsky, Jackendoff is admonishing the philosopher (and the psychologist, the psycholinguist, etc.) to constrain their theories to be flush with the structure of the linguistic f-mind. For example, the popular philosophical conception that linguistic understanding consists in some capacity for propositional communication or a set of complex social conventions, must be assessed by its ability to answer the most rudimentary of questions. Why do competent English speakers recognise that himself in Bob expects to wash himself is co-referential with Bob, while in Bill wonders who Bob expects to wash himself, the reflexive is co-referential with neither Bill nor Bob? What convention might govern this? How might one be inducted into this practice? How did you recognise that the stated relations hold, even though the sentences were new to you? Such questions arise with every construction and philosophy has not been forthcoming with answers. The moral here is not that the study of language should just amount to the investigation of the cognitive structure behind reflexives, wh-movement, etc., but that an account of language that does not factor in such structure will be demonstrably inadequate. Far from the f-conception letting philosophy off the hook, then, it challenges it to find a proper place for notions of knowledge and representation in theories of linguistic competence.

What of the pro party of philosophers, those who see generative grammar as promising to ground notions of representation and linguistic knowledge? Jackendoff’s clear message is that the promise is empty. As with any other real phenomenon, the scientific study of language is not constrained by our intuitive conception of what the phenomenon is. It bears notice that some well recognise the simple fact that words and grammatical properties are not to be found in our environment. How, then, might they be represented? It is said by some that our minds are still representational, for language is an abstract or Platonic object; psychological questions (f-factors, as it were) arise only when we turn to the question of how a given brain manages to represent a given language.[3] (This understanding is much more prevalent than is explicit talk of Platonism; the ready appeal to propositions betrays it.) Jackendoff briefly considers this sort of notional externalism in Part III, but the appropriate answer is to be found throughout the book. The ‘Platonist’ position only makes sense if a “hard line” is taken on the competence/performance distinction, i.e., if it is a matter of principle that the conception of what the speaker/hearer knows is ideal, independent of factors of implementation and interface with other f-components. Jackendoff resists this position. One of the key effects of the tripartite (phonology, syntax, semantics) interface architecture Jackendoff offers in Part II is that there is no clear separation of competencies - all three effect one another. If this is so, then the putative Platonic object is very queer indeed. Are we to imagine that the set of stress patterns and syllabic structures associable with English are waiting to be realaised? Are the structures of gesture that comprise the set of possible sign languages similarly awaiting realisation? Further, Jackendoff marshals much data that suggests that performance shares the shape of the architecture, i.e., the competence architecture is actually predictive and explanatory of performance factors - the difference between competence and performance is only “soft”, a methodological distinction that may potentially give way. This, suffice it say, is a bold empirical claim. My present point is only that Platonism is quite unstable if how we understand and use language is so flush with what we are supposed to represent. This counter-riposte, though, is somewhat academic, for it is difficult not to see Platonism as a fifth wheel employed to provide a notional represented asked for by the ‘representations’. ‘It is better’, Jackendoff would say, ‘to have done with both.’ It is certainly difficult to see what possible explanatory role a Platonic object might play.

Curiously, Jackendoff’s notational overhaul of generative grammar is somewhat presented as if the anti-representational substance behind the substitutions is not appreciated by Chomsky. This is quite wrong. Jackendoff only tangentially mentions Chomsky’s long-standing use of the ‘I-‘ prefix, which Jackendoff himself employed in previous work, to signal that the prefixed term (‘language’, ‘concept’, ‘belief’, etc.) is marking an internal aspect of cognitive structure rather than a representational notion. Indeed, Chomsky’s general internalism, vigorously defended in recent writings, is, as far as one can see, equivalent in both detail and import to Jackendoff’s ‘f-mind’ approach. Chomsky, for sure, continues to speak of ‘knowledge’ and ‘representation’, but there is no substance behind these notions that eludes the ‘I-‘ prefix.

There is much else in Part I. There is a limpid discussion of the poverty of stimulus considerations, which Jackendoff extends to words. That is, just as much of syntactic structure must be innately encoded because there is not the data/stimulus for the child to acquire it, so the child’s acquisition of a lexicon cannot be explained by the data available to the child. Once set out, this result is obvious, given Jackendoff’s position that words just are clusters of features drawn from syntax, phonology and semantics. This threatens to slam the door shut on the not uncommon philosophical position that concedes that syntax might be innate, internally structured, but keeps to the notion that words are learnt by a training regime of ostentation and negative feedback (e.g., Davidson’s (1997) model of triangulation.) Jackendoff also mounts a splendid defense of the f-conception against challenges from “cognitive neuroscience” (read, neural networks). This discussion is commendably fair-minded. Further, a lovely new analogy is the comparison of universal grammar (UG) with a universal “tool-kit”; that is, the resources of UG may be selectively employed to effect different mappings between syntax and the other f-components to produce distinct languages on the ‘surface’. This analogy should at last sink the mistaken idea that UG is a descriptive notion that picks out a set of features common to all languages. Although the notion of a tool-kit might appear not to sit well with the principles and parameters UG model (P&P) of mainstream generative grammar, still less the earlier metric based selection model of the standard theory (Chomsky, 1965), the core anti-descriptive message is shared. After all, P&P does not entail that there will be descriptive universals.[4] Nor is UG guided by the search for universals; Chomsky has long argued that results are forthcoming from the study of one language. Jackendoff’s discussion would have benefited from this kind of comparison. One point of genuine discord is that, as noted above, Jackendoff does view Chomsky as “unfortunately” adopting a “hard line” on the competence/performance distinction. This leads us to Part II.

Part II is given over to a presentation of Jackendoff’s tripartite parallel interface model of the architecture of the linguistic f-mind. The basic idea is quite simple. The three parts of the model are phonology, syntax and semantics. These components may be looked upon as independent generative systems, i.e., they possess their own primitive cognitive items and formation rules; the application of the latter to the former produces cognitive structures that serve to realise the different aspects of our linguistic competence. The model is parallel in that each of the components is linked to, or interfaced with, the other two. There is not, however, a chaotic free for all. The interfaces are constituted by ‘interface rules’ that may be viewed as conditions on what one component can see of another. For example, assume that the syntax component generates empty categories to serve, inter alia, as understood subjects of infinitives. (i) may serve as a rough example:

(i) The boat was sunk (by s) PRO to collect the insurance.

‘PRO’ serves, pronominally, as the understood subject who collects the insurance; after all, the boat cannot collect insurance. Further, whoever collects the insurance is who sank the boat, but no such subject is given; thus, the sentence also requires an agent of the sinking to bind PRO. This all quite complicated, and is demanded by the semantics of the sentence, but none of it makes it to the surface, as it were. In short, these items must be visible to the semantics component, but, they must not be visible to phonology. For example, (ii) is gibberish:

(ii) The boat was sunk by Bob he to collect the insurance.

Otherwise put, what semantics demands is distinct from what phonology demands from syntax. The difference, though, is not exclusive. The syntax and semantic components may be able to see, say, properties of stress that effect interpretation of modifiers (e.g., ‘French TEACHER’ - a teacher of French - and ‘FRENCH teacher’ - a teacher who is French), but they will not see features of, say, voiced or aspirant as realised by a particular vowel sound. The fundamental idea, then, is that the cognitive structure that underlies a given sentence may be viewed as a complex set of relations between products of the three components.[5]

Jackendoff defends his model on three fronts. He suggests how it may cater for the extant data; how it may explain data which has traditionally been problematic (especially constructional idioms: e.g., gave X what for as opposed to kick the bucket); and how it may productively bear on questions of performance and parsing, a point raised above. Much of this discussion recapitulates Jackendoff’s (1997) The Architecture of the Language Faculty, and may be read as an introduction to it. Jackendoff’s position is genuinely novel and raises a host of problems for recent theories of generative grammar, especially as regards the lexicon and idioms. In one respect, however, the focus of Jackendoff’s fire is unfairly directed at the linguist when it could with equal potency be aimed at the philosopher.

The chief motivation behind the tripartite model is to challenge what Jackendoff dubs syntacocentrism, the thesis that all generativity in language is due to syntax, with phonology and semantics merely being interpreting mechanisms that work on the infinite products of syntax. This conception has certainly dominated generative linguistics up to late government and binding. Curiously, though, the syntacocentric epithet better describes the dominant philosophical conception of a compositional level of logical form than it does generative grammar of the past decade. The recent incarnations of the generative tradition certainly retain the notion of a unique level of syntactic structure that interfaces with semantics. This level is dubbed LF. But to think that there is a finished LF product is not, eo ipso, to think that semantics is not independently generative. Under the minimalist construal of LF (e.g., Chomsky, 1995, and Hornstein, 1995), semantic principles cannot license syntactic items at all. That is, LF is not the syntactic level that determines formal features of meaning, it is just the level that interfaces with semantics, where the conditions on a syntactic structure meeting the interface are internal to the features of the lexical items that make up the structure, not what thought, if any, might be expressible by it. This is part of what Chomsky means when he says that LF might code for “gibberish”: semantics has to take care of itself, which is precisely what Jackendoff urges. What minimalism insists on is that there is a single interface between syntax and semantics, but it does not say that the two sides may not be independently generative. (This point is obscured by the common assumption that the interpreting semantics is truth conditional; yet there is nothing in the syntax that so much as suggests that this must be so.) On the other hand, the philosophical conception of logical form amounts to the idea that, in some sense or other, there is a specification available of the formal features of lexical and phrasal meaning that, with the items under interpretation, compositionally determines (generates) sentential meaning or truth conditions. This bald idea, of course, is independent of any particular view of syntax (e.g., for Quine and Davidson, logical form is relative to a choice of the logic of the metalanguage), yet on the assumption of the f-conception, the logical form thesis would simply be syntacocentrism, i.e., it would be a level of formal features that determines all generativity and which, under interpretation produces full meanings, as it were. It is this sort of idea that Jackendoff seeks to refute by his tripartite model. The suggestion, then, is that Jackendoff runs together logical form with LF; Jackendoff’s critique is not wasted, however. I should say that most philosophers and not a few linguists think of LF as a ‘psychologically real’ logical form, i.e., LF is the object of non-generative semantic interpretation. This thought is problematised by both minimalist methodology and Jackendoff’s model. The problems Jackendoff presents for syntacocentrism, therefore, are at least as pressing for philosophy as they are for generative linguistics. There is a further problem with the philosopher’s logical form: the notion that logical form, via a truth conditional semantics, presents the structure of reality. LF has no such ambitious role. Moreover, as should be evident from Part I, Jackendoff rejects any such idea. This stance is buttressed in Part III.

Part III is concerned with issues in semantics and conceptual/lexical analysis. Its two principal burdens are to remove all traces of intentionality or representationism from the theory of semantics/thought and to defend and articulate a decompositional notion of lexical meaning. As to the first desideratum, the discussion will perhaps appear naïve to most philosophers not otherwise sympathetic. Jackendoff glosses the referential/truth idioms with an as the subject conceives the world rider, an approach easily lampooned as more hy-phen-a-ted-anti-realism. Jackendoff, however, is not proposing any metaphysical thesis, still less an anti-realist one. As far as one can tell, his metaphysics is respectably scientifically realist. His claim, rather, is that thought and meaning do not constitutively answer to the way the world is anyway; a fortiori, one cannot read the way the world is from the conceptual structure underlying sentence meaning, only the way the world is for us with our particular f-minds. This is a radical break from the long tradition in philosophy - from Frege and Russell and (early) Wittgenstein to Quine and Davidson - that does its metaphysics via its semantics. The break, though, enables us to sanction the full range of ‘objects’ and properties our thought apparently tells us there are without concomitantly committing ourselves to semantics’ ability to answer questions about the structure of reality. As far as the scientific study of thought is concerned, this ontology is a projection of the f-mind, it is not an external parameter which constrains the investigation of the f-mind. Spun the other way: if we want to know what reality is like, we have to do science, not semantics. So, for example, semantics cannot tell us whether there are events or not (whatever precisely that means), it can only say that our thought employs the category of events, independent of whether or not events exist in some other respect. In this light, ‘concrete’ events and objects are on a par with fictional and abstract objects, directions, senses of humour, purposes, etc.[6] Again, it is not an apt philosophical response to say that Jackendoff is just doing psychology. Indeed he is, yet for those who would talk about aboutness there is an obligation to show how the notion is essential to the explanation of linguistic competence; it is not enough to propound or reject naturalisation projects. This is Jackendoff’s challenge. It is, however, perhaps in the nature of the matter that no consensus will be reachable here in the foreseeable future. Greater immediate advance may be expected from Jackendoff’s discussion of lexical meaning.

Jackendoff argues for a decompositional theory of lexical/conceptual meaning. The basic idea is that our range of concepts are generative in much the same way as our sentential competence is: the range is determined by the combinatorial options defined over basic innate features or functions; the job of lexical semantics is to discover our words’ decomposition into such features. Jackendoff’s discussion borrows from Pustejovsky’s (1995) work and may be read, in part, as an introduction to it. Such a position is controversial and has been strenuously attacked by Fodor and others.[7] Jackendoff’s defense should put the debate onto a different footing. In the first place, Jackendoff dismisses some standard queries that are taken to be devastating: What is the full set of basic features? How would one determine what is and is not a member of the set? How would one tell when a decomposition is complete? Jackendoff’s response is just to say that these are not a priori questions, they are empirical ones to be answered by the confirmed theories at which we arrive. Just as physicists do not, as chemists before them did not, throw out basic elements because fundamental problems attach to the notion, so the semanticist should not throw out basic features because of outstanding problems.

A more serious problem, which Fodor has long stressed, is that there appear to be no successful cases of decomposition. Jackendoff’s response is to say that Fodor is quite right given his conception of decomposition as definition, i.e., an analytical equivalence between the target concept/word and a phrase or sentence. Note that on this understanding, the above problems are pressing; for any word or phrase, surely we can ask if it enters into the definientia of other words or phrases. But, Jackendoff asks, why should we think that the salient question is whether or not a language non-trivially translates itself? The idea that it may does indeed appear to be false, yet it still might be that, at the level of the f-mind, lexical items and concepts decompose into basic features, features which have no clear phrasal individuation. This is just what Jackendoff, following Pustejovsky, proposes (the features may be looked upon as a system of categorical differentiation based on notions of constitution, origin, form, purpose, etc., with words realising different permutations of the characteristics falling under each category.) Indeed, Jackendoff suggests that some concepts may be linked to perceptual models or paradigms of objects or actions that have no phrasal correlate in principle. Jackendoff also explains in detail how such clusters of features compose into lager units to form the underlying structures of meaningful phrases and sentences. This position is a serious empirical hypothesis that offers an explanation of a range of semantic phenomenon; e.g., how prepositions pattern to indicate spatial, directional and possession relations. These explanations are not dented by refutations of definitions, and the proposed non-decompositional alternatives appear unable to explain the data the decompositional theory explains.

There is much else in the book not touched on here; the brief of the foregoing is only to highlight those aspects that should be of greatest interest to philosophers. In broadest terms, though, the chief virtue of Jackendoff’s splendid volume is its exemplifying the interdisciplinary effort that is necessary if we are to gain a deeper understanding of linguistic cognition.

Notes

---

[1] See, e.g., Katz (1966), Harman (1973), and Moravcsik (1975). The contemporary incarnation of this tradition is illustrated by Larson and Ludlow (1993), Larson and Segal (1995), Laurence (1996), and many others.

[2] See, e.g., Searle (1974), Bennett (1976), Dummett (1989), Wright (1989), and, by implication, many others.

[3] See, e.g., Katz (1981), Higginbotham (1989, 2001) and (perhaps) Soames (1984).

[4] The P&P model of UG views the set of possible languages as, in effect, the values of the set of possible permutations on a finite set of innate parameters. Transparently, no subset of parameter settings will be visible in every language.

[5] Jackendoff also tentatively suggests that the tripartite model may answer Chomsky’s concerns over the evolution of language: it is easier to understand the co-evolution of three components than the evolution of one ‘perfect’ component.

[6] See Ludlow (1999) for a strenuous effort to keep to a ‘metaphysics via semantics’ methodology while also commending an internalist ‘Chomskyan’ position on language. The result, however, is a Kantian empirical (constructionist) realism which is perfectly in sympathy with the view that thoughts do not represent the way the world is anyhow.

[7] See Fodor (1998), and Fodor and Lepore (1998).

References

Bennett, J.: 1976, Linguistic Behaviour, Cambridge University Press, Cambridge.

Chomsky, N.: 1965, Aspects of the Theory of Syntax, MIT Press, Cambridge, MA.

Chomsky, N.: 1995, The Minimalist Program, MIT Press, Cambridge, MA.

Davidson, D.: 1997, ‘Seeing through language’, in J. Preston (ed.), Thought and

Language, Cambridge University Press, Cambridge.

Dummett, M.: 1989, ‘Language and communication’, in A. George (ed.), Reflections on

Chomsky, Blackwell, Oxford.

Fodor, J.: 1998, Concepts: Where Cognitive Science Went Wrong, Oxford, University

Press, Oxford.

Fodor, J. and Lepore, E.: 1998, ‘The emptiness of the lexicon’, Linguistic Inquiry, 29:

269-88.

Harman, G.: 1973, Thought, Princeton University Press, Princeton.

Higginbotham, J.: 1989, ‘Knowledge of reference’, in A. George (ed.), Reflections on

Chomsky, Blackwell, Oxford.

Higginbotham, J.: 2001, ‘On referential semantics and cognitive science’, in J.

Branquinho (ed.), Foundations of Cognitive Science, Oxford University

Press, Oxford.

Hornstein, N.: 1995, Logical Form: From GB to Minimalism, Blackwell, Oxford.

Jackendoff, R.: 1997, The Architecture of the Language Faculty, MIT Press, Cambridge,

MA.

Katz, J.: 1966, The Philosophy of Language, New York, Harper and Row.

Katz, J.: 1981, Language and other Abstract Objects, Rowman and Littlefield, Totowa,

NJ

Larson, R. and Ludlow, P.: 1993, ‘Interpreted logical forms’, Synthese, 95: 305-55.

Larson R. and Segal, G.: 1995, Knowledge of Meaning: An Introduction to Semantic

Theory, MIT Press, Cambridge, MA.

Laurence, S.: 1996, ‘A Chomskian alternative to convention-based semantics’, Mind,

105: 269-301.

Moravcsik, J.: 1975, Understanding Language, Mouton, The Hauge.

Pustejovsky, J.: 1995, The Generative Lexicon, MIT Press, Cambridge, MA.

Searle, J.: 1974, ‘Chomsky’s revolution in linguistics’, in G. Harman (ed.), On Noam

Chomsky: Critical Essays, Anchor Press/Doubleday, New York.

Soames, S.: 1984, ‘Linguistics and psychology’, Linguistics and Philosophy, 7: 155-79.

Wright, C.: 1989, ‘Wittgenstein and the central project in linguistics’, in A. George

(ed.), Reflections on Chomsky, Blackwell, Oxford.