Appendices

Appendices

Appendix B: Comments on ‘About I-Language’

i. The excerpt opens with a frank but immediately confusing statement: it seems odd to talk of the internal capacity of a person, since humans possess a multitude of internal capacities, only a few of which have anything to do with language. However, given that the audience has come to hear about linguistic capacities, let us allow that only the linguistic capacities of a person are relevant. To call these capacities “internal language” is natural but otiose: for over 50 years, Chomsky has rejected the idea that language in any theoretically interesting sense could be external.

ii. So what is “internal language” (“I-language for short”)? The next phrase makes it clear: it is ‘whatever you have in your head.’  Whatever is not a term of art: in English, it only means anything. Yet this cannot be what Chomsky intends by I-Language: no-one — be they linguists, philosophers, psychologists, neurologists, or poets — has any use for such a vague notion, and Chomsky certainly doesn’t mean all of the linguistic knowledge contained in the mind of an individual speaker. Far from it.

iii. More confusing universal statements occur in the next phrase. In the first place, it’s not clear what is 'presupposed by everything': Is it the "concept"? Or "the study of the concept"? No natural or biological phenomenon presupposes its study. Nor is any natural phenomenon “presupposed by everything”. Further confusion lies in the notion “somehow logically prior”. Logical priority is not a fuzzy notion: "somehow logically prior" is as incoherent as "somehow pregnant". If Chomsky means by this only that I-language should be accorded a priority in a larger theory, it needs to be said what that theory is such that everything else is secondary to I-Language and why I-language deserves this special attention. (Normally, in theories of biological systems, we give priority to the parts of the theory that do the heavy lifting; the lungs, for example, in a theory of the respiratory system. By this criterion, I-language would hardly be a frontrunner.) Moreover, it is simply not clear that the notion of logical priority plays any role in explaining either a biological system such as the cardiovascular system, or a biological organ such as the heart, since what is important or even relevant depends entirely on the level of analysis: genetics is not logically prior to biochemistry, anatomy is not logically prior to haemodynamics. Even at the systems level, the cardiovascular system is not logically prior to the digestive system or lymphatic system, or the metabolic system, nor do any of these other systems have priority. Use of the expression logically prior makes I-language seem intrinsically important, but it might be a tale…signifying nothing. (The theory of phlogiston was ‘somehow logically prior’ in earlier theories of combustion: see Epilogue.)

iv. Here is the first segue from open discussion into theoretical commitment. ‘If we investigate [the study of I-language], we enter into an inquiry, into language as essentially part of biology’. Notice that here and elsewhere, Chomsky uses the word essentially in two contradictory ways: first, in a literal sense, meaning ‘in its essence’; second, in its looser sense, to mean ‘roughly, approximately, as it were’. In scientific discussion, the latter usage is unacceptable—in student writing, at least. A bird is not essentially a plane, a carrier pigeon is not essentially a message transmission system; an eye is not essentially a whisker (though they are both involved in proprioception, in similar ways). Leaving this quibble aside, we may ask: is this necessarily so? Does the study of I-language commit us to a biological approach to language? While it would be ridiculous to deny that knowledge of language is represented in our minds, and that minds are biological at some level, it doesn’t follow from this that an explanatory theory has to be stated in biological terms. It is possible that this is the case, but it is not logically necessary: in the fields of computer science, especially AI, mathematical genetics, biophysics, and medical physics—including sub-fields as diverse as nuclear medicine and prosthetics—it is generally assumed that purely biological factors can be abstracted away from at little or no cost to the explanation; see also I is for Internalism. What makes this more ironic is the fact that Chomsky is no eliminativist: he avowedly does not propose handing over linguistics to biologists.

v. ‘Whatever the linguistic capacity of a person is, it’s something internal to them.’ [repetition of a previous tautology]

vi. ‘It’s essentially [sic] a kind of an organ of the body on a par with the visual system, or the immune system….or organs of the body.’ This assertion is incoherent. Systems are not organs (in any specialist, or other sense, of the words); organs are not systems. [See O is for Object of Study, for  elaboration of this point.]

vii. (viii)-(ix). So I-language is such a system. No empirical evidence is given in support of this claim, though since the construct is Chomsky’s, he is free to declare it to be whatever he wishes — just as long as he does not impute it to biological systems. I-language as a system with no wider function, but with a ‘property’— the property of discrete infinity. There are two distinct problems here. The main problem lies with the claim that a system can display discrete infinity. Chomsky claims that it does, apparently in virtue of the fact that I-Language generates sentences, and sentences — it is claimed — exhibit this property. (In this lies the second problem: as discussed in v is for von Humboldt, it is not obvious that sentences have this property, either). Yet even if one accepts the assertion that sentences display the property of discrete infinity, nothing follows from this fact about the system that generates these sentences. Take plant growth or cell division (mitosis): you may have a twig with five leaves, or one with six, but there are no twigs with ten half-leaves; a cell may divide into two identical cells, each nucleus further dividing into two, but not will not find a structure with 2n-1 cells; there is, at least in principle, no longest twig or vine, a tree can always grow another millimetre. Hence, the products of mitosis display discrete infinity. But this does not mean that the mechanisms responsible for mitosis themselves display discrete infinity: these are bounded, finite systems. (It is like claiming that a milk factory displays the property of discrete infinity, because it only produces milk in quantum units (bottles, packs, etc), and continues to produce bottled milk for as long as there are cows, and a sustainable market for its products.)

viii. (xii) ‘You can have a hundred word sentence, a 10, 000 word sentence, and so on…’ This is at least debatable; most probably, it is false. You can’t [have a 10, 000 word sentence], or any sentence even approaching that length: 8-10 words is a good ball-park average. (You can, conceivably, have a 10, 000 word utterance, but that is something else). See v is for von Humboldt. It is reasonable to claim that if Chomsky’s grammar generates something that is impossible in natural language, it is the theory, not the phenomenon, that is at fault.

ix. (xiii)-(xvi) ‘It’s like the numbers, the natural numbers.’ It may be like the natural numbers, but this is not obvious, especially since research has shown dissociations between language and calculation ability; see especially Varley et al. (2005). (xiv) 'You don’t find it in the biological world, above the level of maybe DNA', (xv) 'but it seems to be a unique property of human language…' It is not clear that it is a property of human language, let alone a unique one. …(xvi)‘You find it in the number system, but that’s probably an offshoot of language.’  All this is controversial at best: given Varley et al.’s results, Chomsky’s assertion can only at best be understood in evolutionary terms, not synchronically.

x. (xviii-xx) ‘the general theory of discrete infinity is pretty well understood...some kind of generative procedure.’ This is a nice example of Chomskyan argumentation: assume the truth of the premises, then conflate one accepted notion with another that is much more controversial. The theory of 'discrete infinity' is not pretty well understood; it is largely Chomsky’s own term of art. Support for this assertion is given by the fact that a Google search for the string “discrete infinity” yields only 17, 400 hits, all associated with Chomsky; the search results for the string only extend back to Chomsky (2000), before which the term is unattested. By contrast, a string search for “recursive function” yields 335,000 hits, very few of them related to linguistics. Moreover a—not the—theory of algorithms involves much more than consideration of recursive functions (making it singular only serves to reify the notion). Hence, to say ‘the theory of algorithms, the theory of recursive function, various other names,’ is akin to saying, “the theory of agriculture, the theory of cherry cultivation, various other names.” (xix), 'and the theory of language is going to fit in there somewhere'. It does not follow from the fact that there is a theory of recursive function that 'language is going to fit in there somewhere', or 'that the core I-language is going to be some kind of an algorithm, some kind of a generative procedure.' This is pure argument by fiat: the claims are entirely independent of one another.

xi. (xxi)-(xxii): '… that constructs an infinite array of hierarchically structured expressions…all internal to us.' As flagged above, it is questionable whether there is anything infinite—as opposed to extremely large—about the output of grammatical knowledge. But even if were the case, on Chomsky’s own theory, no single instance of sentence generation produces an infinite array. This is because generation of a sentence through Merge operates on an initial array of elements, selected from the lexicon. That array may be large or small, but it is finite: every grammatical sentence starts with finite set of selected items (words and features). Furthermore, the elements within the initial array that can be combined with one another must be constrained by selectional information—one cannot start off by merging is and the, for instance, or crawl and steal; selection further reduces the set of generable structured objects that can find legitimate expression. Even discounting syntactic selection, the set of structured expressions is finite: for an eight word set of elements in a numeration, if one allows all possible combinations, there are 8*7*6*5*4*3*2(*1) linear combinations (= 40, 320). This is a large number, but it is not infinite. And the number of non-trivial hierarchically-structured expressions, where ‘hierarchically-structured’ means anything in terms of propositional content, is significantly more restricted than this. Categories are adjacent for a reason, as Hawkins (2001) discusses. Nor is it clear, in Minimalist terms, that these expressions so generated are ‘all internal to us’. Syntax delivers chunks of words to the PF interface, where they get pronounced: they are not complete sentences until they are pronounced as utterances. And phases are not infinite either. But once pronounced, they are — by definition — external objects. On the other hand, a much larger set of computational possibilities are never pronounced (spelt out), so never complete; by definition, these cannot count to the array of expressions, since they are never 'expressed'. Once again, it is not clear that this assertion is coherent.

xii. (xxii)-(xxv) (xxii) '...which are then transmitted to what are called interface systems, to other systems of the cognitive system, the physical system, (xxiii) and there are at least two of these: the one is the Sensory-Motor system—because it’s externalized somehow—(xxiv) and the other is systems of thought, planning, understanding, perception, interpretation—loosely called the Semantic Interface.'

The distinction between Sensory-Motor and Semantic Interfaces seems confused here. Intuitively, perception must be part of the sensory-motor system (perception is, after all, what sensory means, in common usage). Also, in language production, it must be the case that semantics interfaces with syntax prior to the computation of grammatical expressions, in order to generate the initial candidate-set (the numeration, initial array); compare Levelt’s (1989) ‘Conceptualizer’. If we intend to order pizza, for example, the generation of the set of evaluable expressions presumably shouldn’t include ‘You inadvertently took the dog for a walk’; nor should it distinguish between this string and the 42, 319 other combinations of these particular lexical items. These latter word combinations may be part of an infinite set of utterances based on the English lexicon, but presumably they are not part of the set of possible combinations of the initial array at hand (which may or not include the words "with anchovies".). If syntax took the totality of the lexicon as input, nothing could ever be computed, given any amount of real time. But unless this information is relatable to the syntactic component prior to computation, there is no way to exclude all irrelevant sentences from the set. It seems at this point as if Chomsky is (implicitly) treating the Sensory-Motor component as the conduit of speech production, and is equating the Semantic Interface with comprehension. Yet it should be clear that both ‘interfaces’—assuming they have any psychological reality — are equally implicated in both comprehension and production (albeit in different ways).

xiii. (xxvi-xxix) ‘And the core properties of language are [sic] that it has such a generative system… If any part of this statement is true, there is next to no empirical evidence for it. As discussed below, see, e.g. v is for von Humboldt, syntax is neither necessary nor sufficient for the successful transmission or interpretation of propositional content: a significant proportion of spoken utterances are not complete sentences; furthermore, L2 speakers, and atypical language learners, regularly successfully interpret what they cannot parse by means of pragmatic strategies; they also pronounce wholly interpretable strings that are not grammatically well-formed. Given this, it seems more doctrinal than scientific to claim that a generative system is the core property of language: it is as if a Christian would claim ‘the core property of (any) religion is the Trinity.’

xiv. (xxvii)-(xxviii) 'We’re interested in this generative procedure…"in intension" not "extension"…meaning that we want to know what it actually is, not just what is [sic] the class of structures that it generates.' The expression "in intension” does have a meaning: in v is for von Humboldt section 3, I try to illustrate the difference between an intensional characterization of the n-times tables and its extension: from this discussion, it should become clear that intensional does not entail internal, as Chomsky implies here. Thus, even if one could arrive at an intensional characterization of the generative procedure, there is no reason to think that this is psychologically, or even neurologically represented.

xv. (xxx) ‘because what’s represented in the brain somehow is a particular algorithm, not a class of algorithms, which are equivalent in the sense that they have the same category of structures that they produce’ There is no reason whatsoever to suppose that I-language involves a particular algorithm, as opposed to a class of algorithms. Quite the opposite. If any part of the discussion in Sections I and II of this volume is correct, it must be the case that languages (and possibly different groups of language learners, quite possibly different individuals) vary in the kinds of algorithms they employ to map meaning to form (and back): for example, speakers of verb-framed languages must make use of a different set of mapping procedures to express manner-of-motion from speakers of satellite-framed languages [Section 2 Case #2]; speakers of ‘wh-in situ’ languages call on algorithms to interpret interrogative utterance which are different from those speakers of languages with ‘wh-movement’; speakers whose phonology is described by a different feature-geometry must employ different algorithms to realize the same phonetic segments. More significantly, if the distinction between Level 1 and Level 2 is valid, then I-language (being part of a Level 1 theory) should be wholly indifferent to the particular choice of algorithms adopted by languages/learner groups/individuals: as long as they satisfy Level 1 constraints, any algorithmic solution is equally valid. Level 2 algorithms are simply not the business of a Level 1 theory, on a levels of explanation approach.