I recently picked up Geoffrey Sampson’s Empirical Linguistics. Sampson has a bone to pick with Chomsky– he wrote an earlier book called Educating Eve: The ‘Language Instinct’ Debate where he engages with Steven Pinker’s version of Chomskyanism, without apparently ever getting a response. But he’s actually a pretty good bone-picker. He thinks Chomsky carried some very unlikely propositions by sheer force of personality rather than argument, and he makes a good case.

Speaker intuitions

The target here is the idea that models of language should be based on speakers’ intuitions of grammaticality. Historically, probably Chomsky insisted on this as part of his reaction against Skinnerian behaviorism, which had little role for the mind; Chomsky asserted that speakers do have a model of language.  Indeed he reified this out of all proportion into a mini-organ precisely specified by genetics– though he never says how.

For decades, syntax was conducted by linguists consulting speaker intuitions– meaning their own. In effect they were making up their own data, which is an invitation to trouble. Sampson mentions William Labov scathingly documenting how Chomsky treated his own intuitions as scientific fact, those of others when they disagreed with his as fallible opinions.

More seriously, speaker intuitions can be demonstrably wrong. People can be quite sure that they never say something– one example is a speaker who convincingly insisted that he never used any more positively, as in John is smoking a lot anymore; but then he was caught spontaneously saying Do you know what’s a lousy show anymore? Johnny Carson.

Intuitions are all right for basic matters; the problem is that syntax today is so sophisticated that the sort of sample sentences people are asked to judge are so complicated and unlikely that it’s unlikely a pre-existing rule covers them. Only nativists like Chomsky can really maintain that the grammar covers all possible situations in advance. It’s simpler to maintain that as in other cultural domains like law or fashion, people creatively approach new situations when they’re confronted with them.

These delusions are especially dangerous when theoretical edifices have been built on top of them. Sampson recalls giving a seminar which covered center embedding; he referred to the conventional wisdom that multiple center embedding was impossible. Anne de Roeck asked, “But don’t you find that sentences that people you know produce are easier to understand?” Sampson was well into an extended answer to the question before he realized that de Roeck’s question was in fact a counter-example.

The experience spurred a new interest in empirical investigation of linguistic claims. He started to work with linguistic corpora, using computers for searching and analysis; much of the book is a set of reports on how such work is done and what sort of things come up. Not surprisingly, what people actually say and write is more varied and interesting than the somewhat artificial constructs linguists make up.

This is heresy from a Chomskyan point of view– isn’t a grammar supposed to generate all possible sentences and divide them into acceptable and unacceptable? Well, no, that’s just Chomsky’s pet idea. All a grammar has to do is tell us what people say and write– their positive performance. We don’t need to posit a mechanism to deal with the sentences people don’t actually utter or encounter. (It can be a convenient shorthand, of course, to say “We don’t say XXX”. It can eliminate pontes asinorum— or just contrast the dialect being described to others where XXX does occur.)

One chapter of the book departs from the overall topic to consider The Logical Structure of Linguistic Theory, a huge work Chomsky wrote in 1955 and didn’t publish till 1975, but frequently referred to.  It had semi-mythical status when it could only be consulted by the priestly elite of Chomskyanism; once he could finally read it, he found it extremely disappointing.


Curiously, when it comes to meaning, Sampson takes the opposite tack. He thinks meaning can’t be investigated scientifically (that is, empirically, subject to falsification) at all. That doesn’t mean it can’t be handled at all, but the approach will be that of the humanities: essentially narrative documentation of human creativity.

He blisteringly attacks Fodor and Katz’s semantic primitives– the analysis of “bachelor” as [+HUMAN] [-MARRIED] [+MALE], for instance. As a minor point, he shows that the idea only works for nouns and adjectives anyway– it’s useless for verbs. For verbs you’d might as well just use inferences: e.g if John buys trout from Sally, then Sally sells trout to John. This can be extended to nouns: If John is a bachelor, John is human, unmarried, and male.

But worse yet, the primitives don’t hold up under analysis. Meanings are too fluid. Does a cup have to have a handle or not? Does it have width or height requirements? Labov tried to approach this by showing people specially constructed objects which were designed to test aspects of the definition of ‘cup’; he found that people’s responses weren’t atomistic, but probabilistic. A certain width-height ratio might produce a given probability that people would judge the object a cup. And he didn’t even get into (say) the usage of the objects. Other researchers concluded that far from coming to a consensus, people could come up with “a myriad” of possible common features.

Adam Kilgarriff, working on computational linguistics, concluded that “I don’t believe in word senses”.  Actual usage is so fluid and vague that it makes no sense to ask in the abstract how many meanings a word has; you can only ask what meanings it could be convenient to distinguish for a given task.

Barbara Partee apparently once found young children asking her whether she had a father. This confused her till she realized that they were asking if she had a husband. They didn’t understand the adult notion that fathers were related to conception; they understood them as male heads of household. Presumably once we understand the facts of life, we adapt our definitions. In other words, faced with new information, we either follow the social consensus or create a new one.

As an example of the latter process, Sampson notes the new reality that people can change sex. If a man fathers a child, then becomes a woman, is he now the child’s mother? Until society faces the question, there is no answer, except in terms of individual creativity. This can’t just be handled by fiddling with the [+MALE] node.

Sampson chides linguists for ignoring the philosophical debates on meaning… linguists’ books on semantics often barely consider Wittgenstein, White, or Quine.  (The sources I’ve read do mention them, but I think I should probably check them out directly.)

On the whole I think Sampson is mostly right about empirical verification of syntactic claims, and probably right about semantics. He obviously dislikes Chomsky’s work very personally, but he’s not a crank– he gives good arguments for his skepticism.