The Return of ‘The Poverty’

This Nativity Scene by Bosch could inspire a great many sentences describing its contents, but the nature of attention and perception puts a limit on the form those sentences might take.

A bit of hubbub appeared on the blog this past week in response to my previous post. The author of the paper discussed in that post, Steven Harnad, did Babel's Dawn the honor of posting a couple of comments. And for his trouble he got kicked around a bit.

His first complaint was that I had not really addressed the subject of his presentation in Montreal. My reply is to plead guilty, with an explanation. First, full discussion of a 6,000 or so word presentation doesn't squeeze too easily into a blog post, even one of the wordy posts common on this blog.

Second, frankly I wasn't much interested in the paper's general thesis (the role of lexical categories and propositions in language) but I didn't feel like denouncing it either. People interested in language origins get to pick the aspect of language that interests them. I don't feel like insisting my interests are right. So I might have let the paper go, but there was something that intrigued me and I riffed on it.

In his comment Harnad summarized his thesis, and then, as an aside, said that the origins of universal grammar (UG) was a much harder problem than the origin of language "because of the problem of the 'poverty of the stimulus.'" That last bit provoked our resident Skinnerian to recommend the work of Ernst Moerk "for solid empirical evidence against the validity" of the poverty argument. Harnad replied to that and was rebuked again for scorning Moerk.

Part of the dispute seems to be a disagreement over just what the Poverty of the Stimulus argument is. I took a quick skim of Moerk's The Guided Acquisition of First Language Skills (2000) and was surprised to see that it does not address the syntactical issues that give the argument so much force. Thanks to a follow-up comment I see that behaviorists consider the poverty argument to refer to Roger Brown's book First Language, while Harnad refers to a more abstract argument.

The basic problem that crops up in the general argument is this:

1. We can imagine languages from now 'til doomsday, but many of them never appear in the real world.
2. Some of the imaginary languages turn out to be unlearnable by the ordinary methods of language acquisition—hearing, imitating, using.
3. Existing, learnable natural languages all share certain very abstract properties. And artificial languages without those properties are, at best, very hard to learn.
4. We need some explanation of why points 1, 2 and 3 are true.

What are some possible explanations?

1. The universals are present in the stimulus and thus turn up in all learnable languages.
2. The organism is built to learn languages with certain properties and if those properties are not present, the language cannot be learned.
3. The universal is imposed on experience by the perception of a stimulus and are, thus, imposed on language too.

The first solution was the one favored by B.F. Skinner and denied by Chomsky. That's why the problem is called the "poverty of the stimulus." Chomsky argued that the stimulus is too poor to provide the information needed to make a grammatical distinction. The form of the sentence does not include all the information needed to understand it. Thus, He ran home, He ran away and He ran fast all have the same form but differ syntactically. Can you stick in just any word after ran?

You cannot say He ran farm. The speaker has to say either He ran a farm (direct object in which case ran has a different meaning from the model sentences) or He ran to the farm (indirect object). How, by examining the stimuli (the sentences) does a person learn these distinctions? As far as I have determined neither Skinner, nor Moerk,nor anybody else has put the explanation in the stimulus. The 50 year old complaint remains unanswered.

The second explanation was proposed by Chomsky as an alternative to the failed stimulus explanation. It says that there is a Universal Grammar built into the brain which is critical to the brain's internal algorithms which produce outputs that cannot be predicted from knowledge of the stimulus alone.

This explanation has never been subjected to the kind of brutal analysis the first explanation received, but after half a century of trying, universal grammar remains merely a posited explanation, not a successful one. Universals have been identified, but they are at so high a level of abstraction that they cannot do any better than Skinner at explaining how we learn the difference between He ran a farm and He ran to the farm. (I stress explaining because UG students do a much better job of describing the problem.) If there is a UG, it too is too poor to produce the result found.

That brings us to the third hypothesis, the universals are built into perception. Language universals run in two forms: syntactical categories and phrase combinations. (1) He hit the big red ball can be said in many ways in different languages such as (2) ball red big hit he and (3) he big ball red hit. But there are some ways you cannot say this sentence in any language; e.g., (4) Big red he hit the ball* and (5) He's hit the big red ball*.

The problem with sentence 4 is that a phrase 'big red ball' has been broken up and you cannot separate parts of phrases, although sentences 2 and 3 show that you can move phrases into all sorts of different orders and you can organize phrases differently as well. Sentence 5 changes the nominative form (he) to its genitive (he's). You cannot move words into new categories.

Why do these rules exist? The UG people seem to think they are quite arbitrary, but they make good sense if you think of language as a tool for piloting attention, and sentences as a form of perception by other means. A phrase like 'big red ball' defines a unit of attentional focus. Sentences 1-3 have three such units: (a) he (b) hit (c) the big red ball. In English, this sentence shifts attention from the doer to the action done to the recipient of the action. Other languages can manipulate attention in other ways, but all of them must take the focus as a unit. We cannot scatter the focus so that parts of 3 are in one place and parts in another. It is not an arbitrary rule, but one that follows naturally from the nature of attention.

The consistency of syntactical categories is a little different because categories change from language to language. That is one of the issues that can make translation difficult. None the less, they only change within limits, drawing attention to different aspects of the same perception. Thus, if an English speaker perceives a person as a doer in an incident, requiring the nominative case, a Masai speaker is not going to perceive that person as a possessor requiring the genitive case.

The richness of the perception can explain the structural observations that baffle those who favor the first two hypotheses. That's why this blog's heroes are Deacon, Dessalles, Chater, Christiansen, Marchetti, Tomasello, and others whose focus is on neither Stimulus-Response explanations of language origins nor the UG snipe hunt.

Comments

[STEVAN HARNAD]

THE POWER (AND POVERTY) OF WORDS

I think this will be my last posting, because I am beginning to suspect that you are not really serious about trying to address the problem of UG and POS, Edmund.

The examples you give of uses of the word "run" have nothing at all to do with UG. They are examples of the constraints of conventional grammar, which are learnable, and learned.

As I said, across the decades, many, many would-be critics of Chomsky and UG have aired their opinions without ever confronting, let alone understanding the actual evidence (in the systematic starred and unstarred utterances that are actually at issue -- not arbitrary, isolated ones of our own choosing.)

The utterly pointless example of "run" is a case in point.

Not is the question about whether the rules are "in the stimulus," to be "perceived." The question is whether the examples of utterances that the child hears and says, and the corrections the child receives, are sufficient for the child to learn the rules UG from -- as they are, for example, in the case of learning the rules of chess, or of arithmetic.

To show that the Brown database is sufficient for this, you have to actually show how those data are enough so that the child can figure out from them all the rules that it has been taking teams of linguists decades to begin piecing together.

The problem of POS is that child's database (e.g., the data in the Brown corpus) is not sufficient. There is nowhere near enough there for the child -- or any learning system -- to induce the rules of UG on the basis of those data. (The rules of UG -- not a layman's pet proxy for that complex set of rules on which linguists are still working. Otherwise I could state with confidence that, say, both Goedel's theorem and Quantum Mechanics are false, based on my own pet examples.)

(I expressed no objection whatsoever that you did not review my paper. I just said that you had missed the point about UG. And you're still missing it. I have great faith in words for bridging any and every conceptual gap -- but for there to be a way, there has to be a will.)

[M. Nestor]

I'm disappointed, have had to skip all of Mr. Harnad's lengthy comments due to the insulting levels of condescension and snark that became immediately apparent. I can only assume they're stimulating, or would be with some, ah, peer review.

Meanwhile, I'm puzzling out the nebulous retroactive significance of the recent issues with Marc Hauser and retracted/flagged papers...

[Paul Strand]

In reviewing Harnad’s earlier posted comments on the Poverty of the Stimulus (POS), I was drawn to Fiona Cowie’s book, “What’s within? Nativism reconsidered”. In it she notes that there are really three different POS arguments made by nativists. Importantly, only one draws on data or seeks support from the real world (i.e., is a true scientific hypothesis). The other two are philosophical and draw on thought experiments and logical argumentation. See Cowie’s book for a complete exposition.

When I queried Harnad in my previous post about data relevant to the POS, I was inquiring about the empirical, a posteriori, POS. His response, however, invoked the non-empirical, a priori, POS. That is, he did not respond to the issue of the quality of the empirical evidence (Brown, 1973; Moerk, 1989, 1990, 2000). Instead, he noted that the difficulty within linguistics of identifying all the possible UG structures is evidence of the impossibility of their being learned by a child. This response, of course, has nothing to do with data. Instead, it moves the discussion toward philosophical argumentation.

It is fine to carry on with philosophical discussions of POS. But POS has for years been identified by nativists as an empirically supported construct. This has been happening despite the fact that the only empirical data relevant to the POS, of which I am aware, provides no support for that construct.

Obviously, POS is unnecessary for arguing for the existence of UG. But one cannot claim that UG is innate without evidence for POS. Do nativists define POS scientifically—meaning that it is a proposition that can be tested empirically? Or do they see it as empirically irrefutable?
-------------------------------------------
BLOGGER: I suppose if you favor a third way, you must expect to be caught in a crossfire.

First, the UG side: I was completely startled to have the sentences about ‘running’ challenged as I thought I was just repeating an argument made in Chomsky’s famous review of Skinner. I had not read that piece in a few years, but after Harnad’s dismissal of the line of argument I pulled the review off the web and read the whole thing again. The piece has several illustrations using ‘run,’ but more in agreement my point is this passage:

“’ Struggling artists can be a nuisance’ has the same [grammatical] frame as ‘marking papers can be a nuisance,’ but is quite different in sentence structure, as can be seen by replacing ‘can be’ by ‘is’ or ‘are’ in both cases.”

Chomsky’s point is not that this is an example of the UG, but that you cannot figure out this structure by studying the stimulus (i.e., the two sentences) alone. As Chomsky put it, “It is evident that more is involved in sentence structure than insertion of lexical items in grammatical frames; no approach to language that fails to take these deeper processes into account can possibly achieve much success in accounting for actual linguistic behavior.” That is exactly what I intended to illustrate as well.

I believe these various forms are learned, but they are not learned in any manner that can be explained by recourse to a learning theory based exclusively on stimulus, response and reinforcement.

I also resent Harnad’s part about my not being serious. I may be an idiot, but I am a serious idiot.

Next, the learning theory side. I disagree with the suggestion above that “thought experiment and logical argumentation” are not “true science.”

Yes, science advances through observation and experimentation, but logical arguments can steer us clear of false solutions. A fine example comes from physics in the early 1920s. At that time Einstein’s proposal of the existence of photons was generally dismissed by other physicists. Einstein proposed an experiment which he said would settle the matter. At first some of the world’s most important physicists were very excited, but then a couple of them realized the experiment would produce the same result no matter whether or not the hypothesis was true. Einstein’s proposed experiment was never conducted. Although, this moment in science history did not advance the cause of understanding the nature of photons, it did keep people from chasing a red herring.

I’ve long considered Chomsky’s review of Skinner to be that sort of anti-red-herring contribution. He never uses the term ‘poverty of the stimulus’ but he makes the case for what has come to be known by that term, concluding, “It is easy to show that the new events that we accept and understand as sentences are not related to those with which we are familiar by any simple notion of formal (or semantic or statistical) similarity or identity of grammatical frame. Talk of generalization in this case is entirely pointless and empty.”

Before I’m willing to plow through Moerk in detail, I want to see that logical point rebutted (‘refudiated,’ to use a charming recent coinage).

What are children learning when they learn to speak?Are they learning to recognize a generalized stimulus? Or are they learning how to apply a Universal Grammar to a particular linguistic instantiation? Or are they learning how to direct the attention of others and of themselves? Or is there a fourth answer?