Rough Theory

So it’s been a hot and smoky weekend in Melbourne. The cool change has just come through – not much help unfortunately, I think, for those on the massive firefront. But a signal for me to shake off my heat-induced sluggishness, and get a bit of thinking done.

I’m well and truly past my self-imposed deadline for writing something substantive on the reading group discussion of the debate between Pinker & Jackendoff and Chomksy, Hauser & Fitch, over the evolution of the language faculty – the trajectory of which is conveniently outlined at Language Log. I’ve hesitated to post in part because I was trying to work out a way to break through what seemed to be the main issue that arose in the reading group discussion: the perception that these articles were highly technical pieces, written by and for specialists, such that deciding between the various “they said-they said” arguments would be essentially impossible for a lay reader. I wanted to work out whether there were some way to approach these readings that could at least minimise this reaction – since the reaction, after all, tends to become a self-fulfilling prophecy.

What I do below is therefore not summarise the debate – as I suspect that even a summary of these already fairly condensed and economical texts would mire us in minutiae. Instead, I provide some suggestions that might help someone read the debate a bit more easily – mainly by locating the various empirical skirmishes in the context of what I take to be the overarching theoretical conflict that motivates the empirical battles. I’ll say at the outset that I very much doubt the reading framework I outline is the only – let alone the best – way of working your way into these texts. I offer it more as an example of how I personally went about trying to make sense of this discussion, without a specialised background in any of the scientific fields referenced in these texts: hopefully, your personal path through these texts will substantially improve on mine.

I want to start where Chomsky, Hauser & Fitch begin: by noting that the purpose of these articles – a point agreed by all sides in the debate – is to make a case for the value of interdisciplinary work when investigating the evolution of the language faculty. This point is more important than it may seem, particularly to non-specialist readers: it tells us that, in spite of first impressions, these articles will not assume that readers have any specific disciplinary background – they may assume a sound scientific knowledge of some sort, but they won’t be assuming a socialisation into any particular scientific discipline: the nature of interdisciplinary work is that you cannot assume such things. This therefore holds out hope that, in principle, a non-specialist reader ought to be able to make sense of these debates.

Where I’d like to go next is to make the suggestion that, in the beginning, readers bracket the empirical skirmishes. This may sound a bit perverse, as these empirical conflicts make up the overwhelming majority of the exchange – and, in fact, mark the least contested points of contact between the two sides: the existence or nature of any theoretical argument is disputed within these texts; there is more consensus over where the empirical fault lines lie. Nevertheless, I suspect we’ll find more light if we step back a bit from the empirical heat, and take a closer look at the strange, half-denied theoretical debate that runs through these articles.

I’ve characterised the theoretical debate as “half denied” for a specific reason: Chomsky, Hauser & Fitch don’t admit that their position is motivated by any specific theoretical perspective. Instead, their repeated claim is that they are posing the only possible scientific questions one could pose about the evolution of language at this moment in time – a point to which I’ll return in a moment. Pinker & Jackendoff then argue: no, these aren’t the only possible scientific questions that could be posed – and, in fact, you have only posed these questions because you are presupposing the validity of a particular linguistic theory: the Minimalist Program. Chomsky, Hauser & Fitch then say: we are doing no such thing, our questions have nothing to do with the Minimalist Program – they are instead the only possible questions a scientist could ask. Pinker & Jackendoff then come back and say: actually, a scientist could ask many other questions, if not already inclined to believe the Minimalist Program were true: here, look! – we’ll show you some…

Essentially, then, in venerable academic tradition, the debate boils down to a set of “did not!” – “did too!” exchanges over whether Chomsky, Hauser & Fitch are pushing the Minimalist Program on the sly. The empirical exchanges all fit within this context – which is why so many of the empirical debates are not primarily about what the facts are – not about who did what, when, where, why and how – but about which specific facts matter, and in what ways they matter, for understanding the evolution of human language. I am suggesting, in other words, that this is not an empirical quarrel, but a philosophical one.

Because what I’ve written is somewhat long, I’ll tuck the main body below the fold. I’ll have to apologise in advance for the length, and for what will almost certainly be inadequate copy editing – I’m writing this on borrowed time, so to speak, and I haven’t been able to give this piece the thorough proofing it undoubtedly desperately needs. Read more of this post

Back in October, when I originally posted the forward projections on the reading group’s upcoming choices, I had left the exact selections for the coming week a bit on the vague side, just referring to the Language Log archives on the general theme we would be discussing, which relates to an ongoing debate between Pinker, Jackendoff and Chomsky. L. Magee did, though, piece together a specific list of recommendations for the group, which I thought I’d post in case anyone is curious exactly what we’ll be discussing next Monday.

L. Magee’s suggestions are:

Perhaps start here:

Mark Liberman’s outline of the Pinker, Jackendoff, Chomsky discussion at Language Log

Then:

Chomsky et al: The Faculty of Language

Pinker et al: The Faculty of Language: What’s Special About It?

Chomsky et al: The Evolution of the Language Faculty

and Pinker & Jackendoff’s Reply

Also of interest:

http://itre.cis.upenn.edu/~myl/languagelog/archives/002423.html
http://pinker.wjh.harvard.edu/articles/index.html
http://www.wjh.harvard.edu/~mnkylab/

We workshopped a number of suggestions for where to go next – with the general idea of staying with the linguistics theme for a while longer. We’ll have an email round to solidify these suggestions, and then I’ll post another forecast list…

Too sick today to write anything substantive, so of course I’ve spent two-thirds of the day dozing, and the remainder digging around for articles on the poverty of the stimulus argument… For those who might have missed this particular obsession, the poverty of the stimulus argument is one of the claims discussed as part of our all-Chomsky-all-the-time reading group discussion. The basic claim is that children learn grammatical rules that are so massively underdetermined by their environments, that the pattern of early language acquisition provides strong evidence for the existence of an innate, specific faculty for language acquisition. Since I’m sure all my readers share my fascination with this argument, I thought I’d post the link to the best article I stumbled across in my admittedly somewhat fever-impaired reading today: Geoffrey Pullum and Barbara Scholz’s “Empirical Assessment of Stimulus Poverty Arguments” The Linguistic Review 19 (2002): 9-50.

The gist of Pullum and Scholz’s argument is that, in the cases most often cited in the nativist literature, the stimulus might, in fact, not be so impoverished after all. They challenge linguists to engage in more extensive empirical investigation before making strong claims about the rarity of children’s exposure to particular sentence structures – and they also point to the need to establish more explicit and well-reasoned statistical measures for how much environmental exposure would be “enough” to undermine the claim that children could not possibly deduce a grammatical principle from their environmental exposure.

The article, it should be noted, is not itself a critique of nativism – it remains agnostic on the issue. The authors’ intention is simply to point to the weak empirical base for existing claims, and to challenge the discipline to become more serious about empirical investigation – and also epistemological clarification – if it intends to persist in viewing the poverty of stimulus argument as pivotal for the cognitive science case for universal grammar.

Being an inaugural guest’s post, I hope that, while not up to the usual standards of this blog, the following content is not too far short either. By way of introduction: I am involved in the same reading group as N Pepperell, and have been invited to write some notes on several of the texts we cover. Our reading so far has aimed at doing a potted history of 20th century linguistics, and our path has arrived at Chomsky’s “Language and Mind”.

My apologies in advance for what might seem a naive and anachronistic take – none of our group are linguists by training, and therefore we have stumbling through what is no doubt a fragmentary and selected course. And I apologise for the prolixity of this, an inaugural guest’s post… In any case, what Chomsky presents in this 1968 version of his account of language is roughly as follows: there are 6 lectures, split into 2 groups of 3. The first group of 3 were based on lectures delivered at Berkeley in ’67; the second group were delivered to different audiences at different times. In this light, the didactic purpose of the book becomes clear in moving through it: Chomksy is presenting, to different audiences, in more or less technical detail, 3 distinct ideas.

Firstly, he presents in a number of guises the idea of universal grammar – where grammar is a set of rules which determine the matching of a given sound to a given meaning. Meanings belong, according to the universal grammar thesis, to a deep semantic structure; sounds belong to a surface phonological structure; and what connects the two are syntactic rules.

Secondly, he presents the related but not, in my view, essential idea that such a system is both required by all human languages, and acquired too quickly by human individuals to be accounted for by a range of cultural, social or environmental stimuli. Instead, using the Poverty of the Stimulus argument commented upon elsewhere in this blog, this system must be innate.

Thirdly, the “innateness” thesis connects his, that is to say Chomsky’s view, with a broad, important but largely forgotten tradition of rationalism (long since eclipsed by the rise of empiricism in general, and behavioralism in particular, in the human sciences).

At this stage I want to comment about one aspect of Chomsky’s argument: that semantic deep structures exist prior to their transformation into phonological surface structures. At least in his account here, it seems that every sentence – even assuming for now that in a normative account, sentences can be taken for meaning “informational statements” – must exist in some ready-made semantic form before it can go through some transformational process. However this account seems to me to miss a key aspect of language production – that most sentences develop in time. While a well-formed sentence may be reverse-engineered into into its constituent parts – into noun phrases, verb phrases and so on – it seems unlikely to me that this is how such sentences are always generated in the first place. Such analysis misses the essential fact of language utterances – that they exhibit a linearity, insofar as the start of a sentence is always before the end of the sentence. The following is a tentative stab at how I think this argument could unfold:

When I make a sentence, I can proceed in a number of ways. I can start with some logical proposition I wish to convey, then determine which is the best form for the conveyance (moulding some semantic content into phonetic or othographic form via syntactic transformations, in the Chomskyan way). But this is only one of a number of ways I can proceed. I can start with a particular phrase (“Notwithstanding Chomsky’s insightful analysis…”), then proceed with what may seem a logical extension (“… I disagree with his theory of the innateness of language”). I can also recant: “… well, actually, I agree with his analysis”). I can also forget where I ‘was’, or simply allow my semantic content to remain, perhaps wistfully, unexpressed: “…”. I can perform a number of acts, but none of those acts are irrevocably determined by some original semantic content which remains intact throughout the duration of the act. Even the original phase (“Notwithstanding Chomsky’s insightful analysis…”) does not establish a determinate meaning for the sentence in which it is found – my subsequent words can serve to undercut – with sarcasm or irony; to frame – by quoting or paraphrasing; to emphasise, castigate, or perform a number of acts which alter the meaning of an existing phrase or sentence. Chomsky’s examples always assume that a speaker who has considered what she or he wants to say before saying it, and remains committed to the saying of it at least until the saying is said – but of course there are many examples, mundane and famous, which run counter to this.

It might seem that this sort of criticism echoes those of pragmatists generally, and Searle in particular. However what is striking to me is not that sentences can do things other than make statements of fact, but rather that even when making a statement of fact, I can alter the meaning of the statement significantly at any time during its utterance. This is trivially true, in that I can always proceed as teenagers do when they append “…not!” at the end of sentence. So the point would be that it is empirically contingent for any given well-formed sentence as to whether its meaning had been determined entirely in advance of its utterance – and therefore could have undergone the sorts of sentence-wide syntactic transformations (active to passive, and so on) that Chomsky proposes. On the other hand, it seems equally plausible that for a number of sentences the syntactic form is prior to the semantic content – as in question-and-answers routines, where the answer’s syntax mirrors that of the question (“who saw Bill” – “John saw Bill” versus “who was Bill seen by?” – “Bill was seen by John“). Of particular importance in this regard is that it is possible for me to be committed to the syntax of a sentence I am about to make, before I know what I am going to say. How could I then, consciously or otherwise, apply rules to get to a known syntactic construction (for instance, a passive construction) from some as-yet unknown semantic meaning?

This is not to say, at the level of a phrase, and indeed for many sentences, there may not be processes at work transforming meaning into sound the way Chomsky describes – indeed this may even be the normative case. However it seems to me that there are many other ways sentences can be formed, and these show that a more complex relationship exists between the semantic, syntactic and phonological parts of a sentence than what Chomsky – circa 1968 – will allow.

Quite possibly, later Chomsky or other more recent accounts resolve this problem in some way – if indeed it is a problem. Meanwhile, we are now moving on to How to do Things with Words…

I had mentioned previously that the reading group intended to take a break for a month, while two of our members tour the US. At yesterday’s meeting, though, we decided that a month was simply too long to go without a substantive (or, depending on your point of view, insubstantial) discussion of linguistics. We’ve therefore decided that, for the next several weeks, we’ll move the reading group online. We’re still working out the exact details, but the general concept, I think, is that we’ll open periodic threads for discussion of specific works on this site, reading group members (and anyone else who wants to dive in) will contribute and discuss, and general enlightenment will ensue…

The works proposed for the virtualised reading group include some lighter and some heavier material – if only to allow our travelling members some opportunity for light entertainment while they’re away. The specific works proposed, in the order we’re currently intending to discuss them, are:

Austin How to Do Things with Words

Pinker <a href="http://www.amazon.com/Language-Instinct-Creates-Perennial-Classics/dp/0060958332/sr=1-1/qid=1161811475/ref=sr_1_1/002-5294189-5521642?ie=UTF8&s=books&quot; target="_blank"The Language Instinct

And then as much as you care follow of the debate between Pinker, Jackendoff and Chomsky, outlined in this Language Log summary of the state of play. (Note: for brevity’s sake, I’ve posted only the Language Log link here, rather than the more detailed set of recommendations that were circulated via email within the reading group – reading group members will know the more specific recommendations; anyone following from a distance can make their own selections, which I’m sure will be no worse than ours…)

I’ll try to post a few observations on our previous reading selection, Language and Mind, in a separate post, some time during the next couple of days.

Rather than doing something rational like reading the works the reading group will actually discuss this week, I’ve been drawn off onto a tangent, looking at works on what is called the “poverty of the stimulus” argument. Unfortunately, most of the key pieces I want to read are temporarily unavailable – needing to be recalled from other students, or in journals I’ll need to request via interlibrary loan – making the attempt to trace the intellectual history of this argument more frustrating than it otherwise ought to be. The discussion that follows is therefore unlikely to reflect the best possible presentation of the poverty of the stimulus argument – a better understanding of which might well obviate most or all of my questions.

As I understand it, the “cash value” of the poverty of the stimulus argument is to establish that nativist theories of language acquisition, which posit that children possess an innate predilection for grasping key grammatical principles, are more plausible than empiricist theories, which posit that children decode grammatical principles via exposure to speech in practical situations.

The argument relies on the counter-factual notion of how an “unbiased” learner might conceivably approach language acquisition – where an unbiased learner is understood as something like a computer, tackling the problem of language acquisition by brute computational force, without any organising principles at its disposal other than its own observational analysis of statistical patterns: capable of hypothesising all the various ways in which words might conceivably be combined, and having no particular a priori conditioning to organise words in any particular order. (There are prior issues about how an unbiased learner gets to the point that they understand the difference between a meaningful and unmeaningful sound, how an unbiased learner develops an understanding that streams of meaningful sounds can be chunked down into words, etc. – I’ll leave these issues aside for the moment, although they can also be brought into the nativist-empiricist debate.)

The poverty of stimulus argument maintains that the number of hypothetically-conceivable principles that could motivate the observed regularities in any given language is much greater than the number of principles that children appear to entertain, when their process of language acquisition is studied empirically: there are certain hypothetically plausible interpretive errors that children never seem to make, even though, in some cases, these errors might involve simpler “hypotheses” about how language works than the hypotheses that seem to inform children’s linguistic practice. The theory further argues that environmental stimulus is inadequate to account for children’s linguistic practice: that children do not receive sufficient reinforcement, particularly negative reinforcement, to account for their “correct” interpretations of the principles underlying spoken language. The theory can also point to a range of empirical evidence – for example, the observation that children tend to learn their native language (including sign languages) in consistent stages, with characteristic errors and achievements at each stage – that is highly suggestive that children’s language acquisition is not sensitive solely to environmental conditioning. The conclusion is therefore that nativist explanations for language acquisition are more plausible than empiricist ones.

I have a few questions about this argument, but I should perhaps indicate at the outset that I don’t have a dog in this fight – I’m not completely certain that I even understand, at this point, why this fight draws such strong emotions from the opposing positions (I’m sure I’ll learn…). I find empirically persuasive certain key claims about children’s language acquisition (that it progresses in clearly observable stages, with characteristic errors, etc.) because I’ve worked professionally in roles where these sorts of observations could be easily confirmed. To that extent, I suppose I’d be claimed by the nativist side – but only because, and this is where my questions begin, this argument seems to assume that the key opposition is between something like the following positions:

– a form of empiricism that views humans as unbiased learners, with no inbuilt predilection for making sense of their world in one direction, rather than another, and for whom environmental stimulus must therefore account for any regularities we observe in language acquisition;

vs.

– a form of nativism that argues that, because humans can empirically be observed not to be unbiased learners of language, we can therefore draw the conclusion that humans possess an innate, domain-specific facilty for language acquisition.

My curiosity – and I should stress that, at this point, my position is nothing more than a curiosity – is whether these two positions exhaust our options? The version of empiricism presented here seems somewhat straw-mannish to me – and I worry that this straw man element may have the result of obscuring an overbroad conclusion on the nativist side. It sounds superficially impressive to note, for example, that there might be a statistically enormous number of ways that an unbiased learner might choose to order English auxiliary verbs (“50,000 times the number of seconds since the beginning of the universe, and about a hundred billion times the number of neurons in the human brain”, according to one source), but that children somehow manage to hit on precisely the 99 combinations that are grammatically possible in English. But this assumes that the “real odds” of children’s performance can validly be calculated by comparing them to a hypothetical unbiased learner who can freely select any of a number of equally-weighted options from a universe of choices.

While this might sound plausible on first glance – surely you calculate odds against the universe of everything that it possible, etc. – I’m not convinced that this kind of reasoning applies in complex, interactive systems in which the actual choices available might be highly dependent on a wide variety of conditioning factors – one of which, certainly, might be some kind of nativist specialised language acquisition subsystem, but which might equally – since we’re speaking about hypothetical entities that, as I understand it, are not currently considered subject to direct empirical investigation – be hypothesised to include conditioning factors that, while not specialised to language acquisition, nevertheless have the effect, individually or in combination, of making certain kinds of linguistic interpretations more plausible than others: for example, feedback from a wide variety of environmental stimuli, not limited to the “environment” provided by spoken words, but with implications for language acquisition; neurological functions that are not specialised to language acquisition, but that shape language use; cross-fertilisations that result from the fact that children are not working only on a “problem” like auxiliary verbs in isolation, but in conjunction with other kinds of puzzles that must be also sorted during language acquisition – the combined effect of which may be to make certain grammatical extrapolations more likely than others; etc.

I don’t mind, of course, if someone wants to conclude that their preferred option among these choices is a nativist concept of a specialised language subsystem – as long as this is acknowledged as a working hypothesis for a period when we can’t yet have evidence to differentiate among a variety of possibilities. I become uncomfortable only when the argument seems to be framed as though the only conceivable alternative would be some form of hyper-Skinnerian notion of a tabula rasa, whose behaviour is determined in a very straightforward, mechanistic way by environmental stimulus. I admittedly don’t know the field of linguistics very well at all, but I have a difficult time believing that this Skinnerian fever-dream is anyone’s notion of a serious alternative. (But, yes, I’ll acknowledge that the poverty of stimulus argument offers a devastating critique of such a perspective, if anyone out there holds it…) It seems to me, though, that the actual problem is far more complex – and I’m somewhat concerned that the nativist approach might be slicing this particular gordian knot a bit too quickly: how much do we really know, at this point, about the complex array of factors that make us biased learners? About how these factors interact with one another? About how these interactions limit the plausible ranges of learning we are likely to acquire? Do we know how to interpret the empirical evidence we have about children’s language acquisition, until we have looked at these issues in more detail?

I am painfully aware that these questions may be simply naive – I am so early into this reading, and so new to this field, that I’m happy to be pointed to work of which I am all-too-likely to be unaware… These were, however, the questions that struck me, as I was reading things I wasn’t intended to read, for the discussion my reading group is about to have on something else entirely… ;-P So now back to that copy of Language and Mind…