The Science of Language

by Noam Chomsky

  The concepts expressed in natural languages are not, however, those found in the sciences. In the case of scientific concepts, it makes sense to speak of them as at least in part human inventions, creations, artifacts. No doubt there are innate constraints on the construction of hypotheses and theories, constraints attributable to the natures of our science-formation capacities (about which we know virtually nothing). There are also constraints of a sort set by ‘the world’; science is, after all, an empirical form of study that aims to provide an objective description of the things of the world. But that does not change the fact that they are ‘made.’

  Second, it does not follow from the discussion of the atomicity of concepts that Fodor's DOORKNOB (see Fodor 1998) is an atomic concept, nor even that DOOR and KNOB (and HOUSE and HOME, and so on) are atomic. All the concepts (or more carefully, the “mode of presentation” aspect of a Fodorian concept) that Fodor takes to be atomic could be composed of more primitive ‘meaning’ elements, where the composition takes place in the automatic operations of some kind of internal system or systems. That way, both elements (atomic, but at a different level) and products would still count as innate, and if the system(s) were distinctively human, one might be able to account for the distinctiveness of human concepts. All that follows is that our minds when doing science are ‘set up’ to seek the most primitive elements, and these are what get seen as atomic – at least until a more primitive set is found. This tendency is likely attributable to the science-forming capacity, a capacity that is unique in the animal kingdom about which we know little, except through its effects and the methodology it appears to demand.

  Chapter 5

  Page 35, Agency; the language faculty and what it provides

  Chomsky is not claiming in this discussion, and never has claimed, that there is a homunculus. However, there is no doubt that that is the way we humans think and talk about how the mind works. We apparently need some way to understand what appears to be a fact, that when people act, they manage to bring an extraordinary amount of information from different systems to bear in a coordinated way to produce what we think of as actions that a unified person does.

  Chomsky's remark about the ‘global’ nature of fixing intonational patterns is self-explanatory. For those unfamiliar with binding theory and Condition C, an informal discussion appears in Language and Problems of Knowledge (1988) and a more formal but still readily understood one in Knowledge of Language (1986). Some informal and also technical definitions and discussions can be found on the internet. Readers should not confuse Condition c of binding theory with what is called “c-command.” On the latter (an important matter concerning the structure of the computational system) see Appendix VII.

  A caveat: since views of binding and of variables are often provided by linguists and philosophers and others who in writings and in semantics, philosophy of language, and logic classes cheerfully speak of pronouns and nouns referring and of binding as binding variables to things, keep in mind that Chomsky takes the working assumption of the theoretician of language to be that binding is syntactic, although (as indicated above in the discussion of global facts involving interpretation) perhaps syntactic in a broad sense. It is explained by what is inside the head, including, among other things, Condition c. Whether this is strictly linguistic, or not, is not clear – as Chomsky indicates. However, it is still syntactic (internal, and intrinsic properties of “signs” or mental representations generally) and requires no appeal to anything outside the head. Taking this seriously, one should speak not of pronouns and nouns referring, but of them being in “referring positions” in sentences. Placing this in a broader context, think of people referring, and of their using what language offers them – including nouns and pronouns in referring positions – to refer.

  Chapter 6

  Page 37, On parameters and canalization

  I discuss variation in languages and parameters’ role in it in Appendix VIII; parameters appear too several times in the main text. Principles are the ‘natural laws’ of the language faculty, universal across the human species. Canalization – the fact that development yields a robust and distinct phenotype despite differences in gene, environment, and ‘input’ – is discussed briefly in Appendix VIII too, and is taken up again in the text below.

  What might parameters have to do – if anything – with canalization? Using for simplicity's sake the headedness macroparameter, assume, as is reasonable, that the child's language faculty's initial state, UG (as specified in the genome), for whatever reason, allows for either option. Assume further – as indicated – that choice is automatic, a result of some kind of ‘triggering’ (which may, however, involve something like experimentation). Assume, finally, that the options described by parameter descriptions are ‘written into’ either UG (in the genome), or into developmental constraints of other sorts fixed by nature, including physical, chemical, computational, and informational constraints. If so, the genetically determined initial state allows still for several courses of development and final state. In this sense, it restricts development to a limited set of courses. And thus parametric options, so long as they are written into the genome, or into other nature-based constraints, could play a role in canalization. That was the reasoning behind my question.

  André Ariew (1999) has suggested that claims about innateness should be read as claims about canalization. See also Collins (2005); Collins makes useful points against anti-nativist efforts on the parts of Cowie (1999), Prinz (2002), and Samuels (2002). The basic claims in favor of canalization are useful tools in the effort to say how language develops, although it should be kept in mind that canalization is not just a biological phenomenon, if one means by “biological” under control of the genome. As Chomsky's third factor constraints on growth and development make clear, there are almost certainly non-biological constraints on development/growth.

  Page 37, Parametric difference and understanding

  A film, Windtalkers, exploited the fact that different parametric settings yield languages of different groups and if one's parameters are set one way during development, very different settings can make a language appear out of reach to adults. The film dealt with the role of Navajo speakers in combat against Japanese forces in World War II. Although the Japanese could monitor US forces’ field radio communications with ease, the Japanese radio operators and translators in the field could not understand, nor get translations for, communications in Navajo by native Navajo speakers that were then translated by the Navajo into English for the US forces. The parameter settings for Navajo are too distant from those exploited in Japanese (or English, for that matter), although not for the Navajo infant growing up in an environment with both English and Navajo.

  Page 39, On Universal Grammar

  Universal Grammar, or UG, if identified with what biology (the genome) specifies, might be very small and remarkably simple (unlike earlier ‘format’ models of UG) – perhaps Merge alone. What about parameters? Perhaps some or all are due to non-genetic, physical and informational constraints – constraints that control and channel the way(s) in which the mind develops as it grows. Because it is not clear what is genetically fixed and what is due to other, ‘third,’ factors, it is not clear what UG is. Nevertheless, these other non-genetic factors can be said to be included in the assumption that language is innate. As Christopher Cherniak (2005) has pointed out, one must allow for a non-genetic notion of innateness.

  To provide a label for all those aspects of language that can plausibly be taken to be innate, whether genetically fixed or by other means – that is, all those aspects except for those due to what Chomsky calls the “second factor” (the role in growth of experience or ‘input’) – one might invent a label, such as “UG+.” It just puts a label on a set of phenomena and mechanisms, some of them very poorly understood, but it does provide a general term that can cover innate linguistic sounds and concepts, plus non-Merge notions.

  Page 41, On the difference between
‘meaning’ computation and ‘sound’

  A characteristic feature of several recent views of grammar in the Chomskyan naturalistic project is that they construe the nature of the computations that proceed from a selection of lexical items to the production of a complex ‘concept’ at the semantic interface to be uniform – in current terminology, Merge through-and-through. The same is not true of the computations that lead to the phonetic/sign interface. There are alternative ways to construe this difference, or represent it in the shape of the grammar or computational system one devises. According to one construal – perhaps the easiest to understand, and one that I implicitly assumed in other explications – lexical items in some sense ‘contain’ the features relevant to determining the set of phonetic/sign features that provide instructions to other systems that produce a sound or a sign (or perceive one). They also ‘contain’ the semantic features that lead to the semantic interface. But the lexical features (phonological features) that are relevant to the production of sound/sign are at some point stripped off and fed to a separate computational stream that leads to the production of phonetic/sign features. The point at which they are stripped off Chomsky calls “Spell Out.” Beyond Spell Out (or multiple Spell-Outs, should they be involved), the kinds of algorithms (Merge) that operate in ‘semantic’ computation no longer apply. In recent minimalist versions of grammar (Chomsky 2001, 2008), there might be several stages (“phases”) of computation in the construction of a sentence/expression, and at the completion of each of which both semantic and phonological information is transferred to the relevant interfaces.

  After Spell Out, sound/sign computation is different from that leading to the semantic interface. It is assumed to remain uniform in the semantic stream, which continues to utilize the two (or three, or . . .) forms of Merge. Because of this, it seems reasonable to suppose that – as Chomsky points out in the discussion above – sound/sign production is not an essential feature of the language computational system. Merge and the compositionality of thought that it provides is.

  For a recent summary of Chomsky's reasons to maintain that there are important differences between computation to SEM and computation to PHON, along with reflections on the secondary roles of communication and articulation in the evolution of language, see Berwick and Chomsky (2011).

  Page 42, On canalization: likely the third factor

  Chomsky seems to treat canalization as a third factor matter. It is not entirely clear what exactly Waddington had in mind by way of an explanation of canalization, although one dominant theme is his appeal to “buffering” due to epigenetic ‘networks’ – intuitively, interactions between alleles and the environment. A prominent example is the transformation of stem cells (which can be ‘made into anything,’ as the popular press puts it) into cells of a specific sort: their DNA remains the same, and the environment ‘specializes’ them. Are epigenetic factors “third factor” contributions? Plausibly, yes: they involve more than DNA coding.

  The phenomena themselves are in general obvious enough. ‘Canalization’ captures the remarkable fact that despite genetic variation and mutation within a genome and considerable environmental variation, plus a lot of variation in specific ‘input,’ the result of development is a stable and clearly distinct phenotype. It is generally agreed that canalization depends on fixed pathways of development. Waddington invented the term “chreodes” for these. Developmental biologists have not adopted his term, nor his related term “homeorhesis,” for biologically expressed processes that constitute such pathways of development. A lot has happened since Waddington's early (1940, 1942) work on canalization with drosophila wings and a ‘heat stress’ gene; the field draws a lot of attention now, and research continues. For a fairly recent review of developments and issues, see Salazar-Ciudad (2007).

  The contemporary research program known as “evo-devo” indicates clearly that development and growth are due to more than the genetic instructions contained in what are called “master” genes, the genes that specify that a creature will have, say, vision, or that some pattern will appear on butterfly wings, etc. The non-‘master’ parts of the genome, plus various environmental factors, physical constraints, etc., play a crucial role in timing of development, and in placement, symmetry, modularity of components, and other ‘form’ (and phenotype) factors. For an engaging and illuminating informal introduction to evo-devo, see Sean B. Carroll (2005). For a very useful review of the current state of evo-devo and its role in the understanding of evolution, see Müller (2007). For a useful although rather compressed discussion of the relationship between evo-devo and Waddington's early 1940s to 1960s thoughts about development and canalization, see Gilbert (2000). For an engaging overview of what difference the study of evo-devo makes in the study of organisms and minds, see Fodor and Piattelli-Palmarini (2010). For some discussion of Chomsky's notion of perfection and its relation to some evo-devo theses, see his “Some Simple Evo-devo Theses: How True Might They Be for Language?” (2010). See also Chomsky (2007c).

  Chapter 7

  Page 43, On ‘master’ or ‘control’ genes

  Gehring was one of the first to emphasize how improbable it was that the eye (or some kind of visual detection system that depends on rhodopsin) could have evolved independently in so many clades. It does not evolve separately, he claimed; it evolved once, and takes different ‘shapes’ in different species because of other features of the organism involved and the way vision and these other factors develop. His work on the role of the ‘master’ or ‘control’ gene PAX-6 has helped emphasize the importance of looking for genetic coding that appears in many species (that is “conserved”) and contributes to the development of some organ/trait or another that serves much the same function in each species.

  Chapter 8

  Page 48, On hidden or unconscious thought and free will

  These comments may raise in some readers’ minds doubts about whether humans have free will, as Chomsky assumes we do. For if decisions are unconscious, how can it be that a person makes them? Unconscious decisions seem to be out of a person's control, even though they appear to the decision-maker to be choices that he or she undertakes freely, and sometimes with difficulty. Notice, however, that there is no reason to believe that even if decisions of an organism are unconscious, that they are determined. To have reason to believe this, one would have to have in hand a decent naturalistic theory of an organism's decision-making, and getting that is extremely unlikely. As pointed out elsewhere, the mind's various systems operate relatively independently and have multiple inputs and outputs. Constructing a deterministic theory of an organism's action/behavior would require coming up with a solution to an n-system problem. This is a far more complex task than coming up with a solution for the extremely simple cases found with the systems discussed in n-body problems. There are far more variable factors, assuming even that we had a completed list of the mental systems involved and their specific contributions.

  The disconnect between our usual assumptions concerning action, planning, deliberation, and decision-making and the descriptions and explanations offered by naturalistic sciences of the mind may continue to bother the reader in the discussion that follows. If we are biological creatures, as we seem to be, we have limited cognitive resources. We know we have common sense and science; if we have others, they have not made an appearance of which I am aware. If limited to these two resources, very likely we cannot overcome the disconnect. The concepts that we have for speaking of the actions of whole organisms are those provided by common sense, a conceptual scheme virtually ‘designed’ to serve the interests of speaking of agents (actors) and agency. Those available to the sciences are those created by scientists when they invent theories of systems that are construed – generally – as deterministic. Some philosophers such as Fodor with his “computational theory of mind” seem to believe that the gap can be overcome. It is not clear to me that he addresses the basic problem created by what appear to be very different ways of
understanding the world. Common sense places agency front and center and it deals with an endless number of anthropocentric concerns; natural science of the sort available to us appears to completely lack the tools to deal with it. Fodor (1998, chapter 7; 2008) broach the issue. However, he continues to follow Putnam and others in declaring that the property WATER is identical with the science-defined property H2O.

  Page 52, What's wrong with externalist dogma (again)

  No one should assume that Chomsky expects that the theory that can accomplish this is a learning theory based on some kind of generalized learning procedure. It will be a theory of the (largely) ‘devoted’ machinery, the operations of which explain how the child's mind recognizes relevant patterns in the speech of (mostly) other children and manages to bring these patterns to bear in the child's own speech production. Speak of imitation if you wish, but keep in mind that that is just a label for a job to be done, one that develops a theory of the relevant system(s) that takes into account the biological, physical, computational . . . considerations that constrain development. For recent discussion of some current efforts to undermine the nativist foundations on which biolinguistic study proceeds, see Berwick and Chomsky (2011).