The Richness of the Stimulus

B.F. Skinner in  1950, at the height of his prestige, a few years before Chomsky used the "poverty of the stimulus" to debunk Skinner's claims. Now Chomsky's argument is collapsing, although that will not be enough to restore Skinner.

This blog has obtained a landmark paper on language origins that will appear in a forthcoming issue of Behavioral and Brain Sciences. It provides details for the great reversal that is underway in our understanding of the origins of syntax, and confirms Babel’s Dawn’s report last March that the Chomskyan paradigm of generative grammar no longer provides the guiding ideas for the effort to understand language origins.  In my series of reports after Barcelona I worried, “The question the conference has left unanswered is whether a new paradigm will appear or whether the study of language evolution will simply fall back into the chaos and confusion that once made taboo all inquiry into speech origins.” See: Paradigm Lost.) Now we have a distinctly encouraging sign. A detailed, theoretical and empirical understanding of how to reverse old linguistic assumptions is becoming clearer. A strong candidate for this reversal's founding paper is the forthcoming, “Language as Shaped by the Brain,” by two psychologists, Morten H. Christiansen and Nick Chater. (Various drafts are circulating on the web. One of them is here and another here.) The next several posts on this blog will discuss this paper, but let's begin by its overthrow of the "poverty of the stimulus," the original argument that carried the day for the Chomskyan paradigm.

Readers familiar with Thomas Kuhn’s account of paradigm shifts already know that these changes are not just readjustments of existing ideas. They change the fundamentals, often by reversing what was previously believed. The new paradigm's central reversal denies that the human brain evolved the capacity to create structured sentences. Instead, sentence structure evolved the capacity to be used by the brain. It is a Copernican-style denial of an obvious fact, and can only come to dominate if many explanations based on the old perspective can be accounted for. The reversal itself has been discussed before on this blog (see: Co-Evolution Idea Won Big in Barcelona) and can be traced to Terrence Deacon and Rudi Keller. Christiansen and Chater's critical contribution is to undermine the supporting explanations of syntactic origins based on the old commonsense view that the rules are in our heads.

Fifty years ago B.F. Skinner was America's leading psychologist and his book Verbal Behavior sought to explain language usage without requiring knowledge of the rules. Chomsky destroyed his claim by appealing to the "poverty of the stimulus," the argument that there is not enough information in the linguistic input to account for linguistic output. I see that this argument resurfaces in criticism of Christiansen and Chater. Steven Harnad has written a commentary, “Why and How the Problem of Universal Grammar is Hard” (available here) that rejects their reversal restates the classic Chomskyan argument on the poverty of the stimulus:

It is not that children speak flawlessly from birth. But the little the child experiences during the relatively brief period of transition from being unable to speak to being able to speak does not involve any errors (or error-corrections) in the rules of UG, either from the child or from the speakers that the child hears. There are conventional grammatical errors and corrections aplenty, but no UG violations produced, heard, or corrected. UG rules are never broken, never corrected, hence never "learned": Therefore they must already have been inborn.

I discussed this argument in a report from Barcelona on experimental work by Simon Kirby. The work did not refute the logic of Chomsky’s complaint about “the poverty of the stimulus,” but did refute the fact of it.

Chomsky famously argued that passing along language in [the manner of Kirby’s experiment] cannot work because of the “poverty of the stimulus,” that is to say there is not enough information in the output to explain any underlying structure used to express meaning. And yet Kirby found that after ten “generations” of this passing along of output the subjects are using a language that has become structured to optimize both learnability and expressivity. (See: Language Structure is Cultural, Not Genetic)

At that time (only three months ago, but things are hopping in the study of speech origins almost as fast as in the American presidential contest) the facts refuted the old paradigm, but there was no new argument to account for the observation. The Christiansen and Chater paper attacks one of the old paradigm's premises: the claim that the innate, universal grammar is arbitrary. Its rules could just as easily have taken some other form; there is no functional explanation for doing it one way rather than another.

Christiansen and Chater cite the example of reflexive versus accusative pronouns. Reflexive pronouns link the object to the subject. Thus, in John sees him the subject (John) and direct object  (him) must be two different people, but in the reflexive John sees himself the subject and direct object are identical. It might seem that the rule for reflexivity is simple: replace the standard pronoun with a reflexive one. Often, this rule works: Joe shot him/himself in the foot but modern syntacticians have excelled at finding complications for simple-seeming rules. A sentence like John said he won can be reflexive or not. Unless there is some larger context, he refers to John. The reflexive meanings are apparent in a sentence like Bush said he won while Gore insisted he won. But why does it work that way? Meanwhile, a sentence like He, said John, won cannot be interpreted reflexively.

If you write a computer program to handle these peculiarities, their arbitrariness becomes quite obvious. Why are the rules this way rather than another way? As kids like to argue, Just because. So that is another assumption of the poverty of the stimulus argument: our brains are just like modern computers. Christiansen and Chater do not challenge directly the notion that the brain is some kind of a computer, but there is no reason to assume that it works like your desktop machine. Quite obviously, it does not. We know enough neurology to know the architecture is very different, and we know enough psychology to know that the brain’s proper functioning depends on perceptual activities like paying attention that have no direct counterpart in the input/output activities of desktop computers.

So there is another way to put the poverty of the stimulus argument: Let’s assume that the human brain functions in the same way a mainframe computer works, and let’s assume that the rules of universal grammar have no functional explanations. In that case, it is perfectly obvious that the examples of language encountered by children are far too few and skimpy for them to figure out the rules of grammar that they master so quickly. Thus, the rules must be built in to them.

A skeptic might ask why must we assume those things, especially since we have Simon Kirby's experimental evidence that shows something else? The skeptic in this case is not rejecting reason itself, only assumptions that have some plausibility but are not holding up to empirical testing.

But, demands the traditionalist, what non-arbitrary, non-mainframe functions might explain the syntactical structures? Christiansen and Chater reply, “Pragmatic processes [are] crucial in understanding many aspects of linguistic structure, as well as the processes of language change.” These processes are peculiar to humans and have nothing to do with mainframes or computer networks. Looking there for answers means breaking decisively with generative grammar and its assumptions.

So there is the first element in the new account of the origins of syntax: the stimulus is quite rich enough for children to organize their sentences. More next week.

Comments

[Michael]

I must say I habe not yet read the paper by Christiansen and Chater because I wanted to wait for it publication in the Behavioral and Brain Science along with the commentaries, but here are my two cents:
I must say that I have trouble grasping why the non-arbitrariness of certain aspects of language would refute arguments made for nativism and/or UG. It makes sense that syntactic patterns are influenced or even shaped by functional constraints and processes.
Also, I do think a lot of (I think in fact most)Generativists would agree that the brain is not just a simple desktop computer.
As Jackendoff (e.g. Linguistics in Cognitive Science: The State of the Art 2007, available at hsi webpage) points out, explanations for language can take place on three different levels of inquiry:
the formal level
the functional/psychological level
the neural level

generativist formalists do not claim that the brain exactly computes these formalized rules, but that the formalized rules have some psychological reality in that they "symbolically approximate" (a term coined, I think, by Smolensky) what is going on in terms of neural computations in the brain. Thus I do not think that "our brains are just like modern computers" is in any way an adequate description of generativist or nativist viewpoints.
The kind of argument the are advancing against UG to me seems a bit as if they are attacking a straw man (and if there are people proposing exactly such a theory, the critique would nevertheless leave generativism in general unfazed).

One further comment regarding Kirby's experiment. AFAIK it hasn't been published next so we'll have to wait for further details a bit, but one question I asked myself:
The Poverty of Stimulus argument doesn't say anything about people learning or creating a new language who already speak one. This is second language acqusition or creation somehow akin to what took place with Tok Pisin, if I'm not mistaken. Unlike in real first language aquisition, in Kirby's experiments the cognitive architecture of language (whatever its structure may be) is already up and running, fully developed, and operational, and thus this language creation process is vastly different from the one a human infant encounters. The experiment surely has some important and profound implications, but i do not see it refuting the fact of PoS (if there is such a fact.

Much stronger case four UG, which ought to be considered seriously, are, e.g., the examples of Nicaraguan Sign Language and home sign, which bears certain morphosyntactic features prior to any linguistic inputt (see e.g. Susan Goldin-Meadow. 2005. The challenge: Some properties of language can be learned without linguistic input, available at her home page)

I'm looking forward to reading your next post, maybe there are really some good arguments for this new approach, but the arguments against UG didn't convince me.

[watercat]

Poverty of the Stimulus is nothing other than the creationist's Argument from Incredulity: whatever patterns in the data remain unexplained are the work of a Universal Grammarian who lives in our heads. Instead of seeking the origin of the Unknown Gnome who created the patterns, the scientific method seeks functional explanations for their apparent arbitrariness.
Nicaragua/Sinai prove an innate “syntax”, but cognitive structures like agent-actor and lineo-temporal sequencing also underlie realms such as mathematics. The syntax of 2-1=1 is [[SVO]VO], and that this structure is attested in both pre-linguistic children and languageless adults disproves any dedicated language module.
--------------------------------
BLOGGER: It may be true that " the scientific method [begins by seeking] functional explanations for their apparent arbitrariness," but sometimes things are arbitary. Einstein never accepted quantum mechanics because it says that, at bottom, there is no reason for the phenomena. Most physicists made their peace with that. Evolution says many things are the result of drift, i.e., arbitrary chance. So these days you cannot just dismiss a claim a priori on the grounds that it reveals no underlying reason for the nature of things.

[watercat]

Sheer random chance could be an explanation, but PoS has never offered it. The claim is that there cannot be an explanation—a claim that requires the maker to be omnisciently aware of every possible factor that has been or ever will be discovered.

In the absence of any alternate theory, one might well settle for a "symbolically approximate" model, but C&C show there are numerous empirically based alternatives offering explanation right down to the neural level. These successfully generate aspects of language from pragmatic and non-linguistic input, so the traditional argument—that the Primary Language Data alone is insufficient—could be true, but is also irrelevant.

The authors deconstruct PofS pretty well I'd say. And it's refreshing to see 'evolution' not taken too literally, even though they waterboard this 'useful metaphor' until it seems to muddle their main point that there is no 'genetically specified linguistic endowment'.