The Role of Intuition in Linguistics

In doing science, there are certain principles that are widely believed to be inviolable. One is that the science (and the scientists that practice it) be objective. Another is that the results obtained in any investigation be replicatable. These two principles help to put sciences on as solid a foundation as possible.

I originally was trained as a theoretical linguist, specializing in syntax, the theory of sentence structure. The original goal of syntax was to construct a theory of sentence formation and within that theory, provide descriptions of the sentences that comprise each language we want to describe. In this way, our theory was to have universal applicability. Chomsky argued that the set of sentences in any language is unbounded, which is to say, infinite in size. He also argued that it would be wrong to gather our data from observation by, say, transcribing the sentences we hear. One must, instead, use our linguistic intuitions as to what is and is not a sentence in the languages we are native speakers of. In short, he argued that it would be wrong to use actually occurring linguistic data, data most naive observers would say is "real data."

Chomsky noted that the gathering of data from observation would be very unlikely to contain data critical to theory construction. There is some merit to that view. We all recognize that

Which girl did John tell the boy who kissed to get out of town.

is utterly uninterpretable. It is unlikely that we would encounter it in real speech but suppose we did. What would we do with it? Throw it on the scrap heap and continue on without inferring anything at all from it? That doesn't seem to be a very good idea. Indeed, speakers of languages that are sufficiently like English to allow for construction of such a sentence find them uninterpretable (the last I heard -- remember I have been retired for awhile). The general consensus has been that such sentences illustrate what is called the "A over A" Principle: if you have construction of type "A" (a noun phrase like "the boy who kissed some girl" which contains a construction of the same type, as is true of "some girl," which is also a noun phrase, then you cannot "extract" the subordinate instance of A from the superior instance of A. In

Which girl did John tell [the boy who kissed [e] ] to get out of town.

I have put in the "identity element "e" in the position from which one could say "which girl" was extracted. There have been many ways of trying to account for this phenomenon. Without perusing the recent literature I think I can say with confidence that no real consensus has developed. Nevertheless, the phenomenon is real and it seems we must employ our imagination to think up such a possible sentence and our intuition to judge it unacceptable.

A second reason Chomsky said we should use one's intuition is that real speech can be rather chaotic: full of false starts, pauses, pause fillers like "uh", and the like. In my book, Speech Acts and Conversational Interaction, I analyzed a conversation that was full of these and other messy things. This conversation can found at pages 109, 110, and 111,. Here we find inward and outward breathing (the .h's and h's), pauses, the lengths of which are indicated within parentheses, and numerous pause fillers, incomplete sentences, and the like. I had access to a copy of this conversation before going to a conference in Italy where Manny Schegloff, a Conversation Analyst, played the actual tape recording. As "defective" as this conversation is, it sounded perfectly natural, like those we engage in every day. The fact that I had read it before hearing it might make me into a problematic source but no one in the crowd said they found it to be nothing but word salad.

What made this conversation so interesting was that the incomplete sentences are easily filled out. When Donny says, "I don't know if its possible but I have to open up the bank" he means "I don't know if its possible for you to give me a ride but I need for you to do so because I have to open up the bank." In my book, I provide a theory of certain classes of conversations, inspired in large part by work I did with Terry Patten and Barbara Becker on computer simulation of conversation, as well as much reading in the speech act, conversational analytic, sociolinguistic, and artificial intelligence literatures, that makes perfect sense out of what Donny said as well as what Marsha said in the next turn of the conversation. What she says means "I know you want a ride and i would give you one but I can't because I've got to leave in about five minutes." Her shift from "but" to "except" is socially significant for "except" means "in every circumstance other than the current one" in this context. This is powerfully strong and is intended to mitigate the face threat associated with her rejection of his request. What is most remarkable about this conversation is that there is no direct reference to someone giving a ride to someone else; the request is never explicitly made; and it is never explicitly rejected. Yet the conversation was perfectly clear to the participants.

I have talked to linguists who say that they believe they have intuitions as to what are and are not well-formed conversations. To that, I say, and "4 million angels can dance on the head of a pin." The idea that we have intuitions as to the well-formedness of conversations is one of the silliest things any putative linguistic scientist could possibly say. Nevertheless, linguists and semanticians routinely work on "home made" conversations the way that syntacticians work on "home made" sentences. I'm not going to tell syntacticians that they shouldn't use intuition as they do even though doing so is clearly problematic. My reason is that I believe that what linguists should be working on is conversations like the Donny and Marsha ride request. Real data. It will force a radical revision in how linguists do theory construction. No longer would constructing a grammar Chomsky-style be a desideratum and to some degree some of the motivations for using intuition would vanish. A better idea would be to create what Michael Halliday called a systemic functional grammar in the form of a complex "munge" of pragmatic, sociolinguistic, semantic, morphological, and syntactic features, examples of which are in my book. That is the sort of grammar that Patten, Becker, and I¹ worked with in a computer simulation of conversation.

---

¹Terry Patten, Michael L. Geis, and Barbara D. Becker. Toward a theory of compilation for natural language generation. Computational Intelligence, 8(1):77--101, February 1992.

---

Tags: linguistics, conversation, intuition, linguistic data, theory construction