Generalizations and the stupidity of linguists
Let’s say you’re a person studying language, and you observe a word, or a construction, in some language, and think that it is interesting for some reason or another. After much study and hard work, you come to the conclusion that this single word or construction must be polysemous. That is to say, you decide that its distribution in the syntax, or its semantics, or its patterns of usage are such that there is no “it” per se, and you are actually looking at two (or seven) distinct forms. Or, say that you observe some pattern of sounds among the lexical items of a language, and notice that there is some correlation between (say) syllable-per-lexeme count and vowel harmony, or between certain stretches of segments and certain types of semantics (in the former case, perhaps you are on drugs and need to reexamine the data; in the latter case, perhaps you look at words in English that start with gl, or ideophones in Japanese that start with voiced velars). In either case, you think: wow, this is an amazing observation. Perhaps I can even call it a generalization! But then you say: shucks–this generalization doesn’t “follow” “logically” from any “principles” of the grammar (perhaps you are spitting upon uttering each of those words, or perhaps not).
You might have been Kay (1992) writing about at least, and its three distinct yet related meanings (exemplified briefly here: (1) The store has at least 3 copies left, (2) The professor talks slowly but at least she’s clear, (3) He’s home, or at least his car is in the driveway; (1) is about quantities, (2) is about a non-optimal but somewhat-okay situation, (3) is about a rhetorical back-off). You might have examined these sentences, and seen that they all use at least, but come to the conclusion that
The notional observations we can make relating the various usages of at least to each other belong to the ad hoc, non-productive family of polysemy relationships. By the criterion just suggested, [namely] that a grammar needs to contain everything, but only such things, as a speaker can’t figure out about his language from the other things he already knows about it, the apparent notional relations among the three uses of at least do not qualify as grammatical observations.
This was initially rather a disappointing thing to read. It means that a lot of generalizations that some linguists have made (say, regarding metaphor-driven polysemy, or even most cases of so-called polysemy) are not really generalizations, just post hoc descriptions of similarities. A speaker might eventually come to realize these similarities exist, but only after finding that out on a case-by-case basis. One example of this is the use of temperature words to describe attitudes. You can have a primary metaphor like affection is warmth, which you can (maybe) motivate from non-linguistic principles, but from that you still can’t necessarily know that a cold response is proper English, but a freezing response is marginal at best. And you might not get that a cool response is a bad thing, but a cool breeze is a good thing.
This seems to say that there are some patterns that a linguist might find are actually not “real” facts about the language (or the “grammar,” or perhaps the “competence”), but perhaps about the history of the language. One might be led to believe then that linguists are quite smart. They know things about the language that other people don’t. However, what they know, I think Paul is saying, is that linguists may know more about the language, but not about the grammar. That is to say, you don’t even have to unconsciously know anything about the possible relationships between temperature words and attitudes in order to use them correctly.
On the other hand, Kay also says that linguists (at least the type that don’t run psychological or neurological experiments) have to be stupid. Because there’s no way to tell without experimentation if some real generalization is actually made by speakers or if the facts are just recorded in the speaker’s mind. So you have to assume, for the purpose of developing a grammar, that whatever real facts linguists can find, other people can find (and may also exploit).
Yes, I certainly agree. Certainly, many generalizations that linguists have made aren’t really synchronic generalizations, but are just reflections of generalizations about diachronic change. So do you think that means that linguists should be in the job of doing “psychological or neurological experiments” to find out what sorts of generalizations people really make? Because that’s certainly the conclusion I’ve come to.
As one example, a guy named Alex del Giudice who is in my class at UCSD did some work as an undergad with Adele Goldberg, arguing that all the uses of subject-auxiliary inversion formed a natural class around a radial category. You can see the paper here. I mean, it’s certainly an interesting analysis as far as radial categories, etc., are concerned, but it really seems unlikely to me that people are actually making/using that generalization at all. I mean sure, if these senses are related, it might help in learning a little, but in what sense are they really all the same category?
Anyway, I’ve got a lot to say on this topic, and I’m glad to hear that Paul Kay agrees with me. But let me know what you think.
I’m not sure if the link to that paper worked. It’s in Linguistic Review 2005 (i think around december?) titled “Subject-auxiliary inversion: A natural category?”
Yes, experimentation should be the job of linguists. The only other barely-reasonable position would be: well, we can let other people do the experimentation — what we (linguists) will do is come up with -all- of the possible generalizations that could be made. It will then be revealed via experimentation -which- of those generalizations are the real ones. On a very abstract level, that might be okay. You just have to hope that your choice of framework allows you to make those generalizations which are possible. Something as constrained as Minimalism might not even allow you to make the right generalizations. Something as vague as cognitive grammar (not that I really know much/anything about it) will probably let you make as many generalizations as you want (even some that are ridiculous). OTOH, you’ve basically multiplied the amount of work necessary: you could have just done the experimentation and then cut down on a bunch of unnecessary pencil-and-paper theorizing. (However, it’s hard to interpret experimental results without some “pretheoretical” expectations, so there’s always that problem; but it’s not limited to linguistics).
About the SAI paper, it has been noticed before that SAI is “about” so-called irrealis statements, or about non-veridical utterances (like neg-polarity items). The paper puts disparate cases together into a radial category, so I suppose it’s not as vague as just saying “irrealis.” Again, this is always the problem with motivation: you can come up with a story that fits with some basic principles, but there’s no way to be sure that some -other- story wouldn’t have done the trick just as well. They hint at this with the exclamatives. They say that the SAI is a relic of the cxn’s earlier status as some sort of rhetorical question but it’s still motivated because it’s related in speakers’ minds to questions (and then they provide corpus evidence). One would like a theory of rhetorical questions before one became convinced that these exclamatives are related to them. And while they’re at it, I’m still not convinced about what they mean by prototypical/aprototypical sentences (dependence, polarity, assertiveness, etc.). They have the cluster of dependent/non-positive/non-assertive/etc. as their non-prototypical sentence. But I’d like to see evidence from psychological, or even conversation/discourse analysis, that these are actually “non-prototypical.” Seems like we might be carrying over some misguided ideas from derivational theories about “basicness” of sentences.
Finally, a thought: say that you claim that you cannot unite all cases of SAI in English. That is, you demonstrate that speakers do not use the existence of SAI in cxn-1 to learn/motivate/coin a cxn-2 that has SAI. Maybe there are three or four types, or even a separate type for each use. You might still want to ask: was it -ever- a single type? Why the heck -is- there SAI in the first place? It’s been argued that SAI as for questions is in fact motivated (I’ll dig up the reference), at least historically. It’s hard to know where to draw the line between historical and synchronic explanations. Some people think it’s al diachrony: we’ve inherited certain traits from our ancestors, and beings looking back at us from some distant point in the future might not be able to tell how exactly our behavior is related to theirs. It’s like, at some level we’re still amoeba, or something (okay, that was just a strange thought I had when I was about to fall asleep several weeks ago).
Probably if I keep typing I’ll start making less sense. I’ll think it over more and then write more.
The note in the paper about historical motivation for SAI reminds me of Blevin’s central premise of “evolutionary phonology”, which is that a plausible and/or documented diachronic account for why something is the way it is should be preferred to a synchronic one unless there is strong evidence (i.e., experiments) that a synchronic account should be preferred. I think it’s a great principle, and can do wonders for phonology. But, it’s interesting that you could apply it to syntax as well, which is something I’m very interested in trying to do maybe someday. Anyway, I think it applies here. Corpus evidence just won’t be strong enough to make the synchronic account override a perfectly good diachonic explanation.
As for the question of whether there was one type of SAI in the first place, I would assume there had to be… I mean, I don’t know the history of the construction, but I would assume that the history of any word or construction begins with a single kind of use, which is then extended to other uses. The only other possibility would be kindof an analogy to homophones, such that maybe subject-auxiliary inversion developed twice, and just happens to look like each other. In the case of the construction, though, I would think this would be hard to do, since after it developed once, then people would interpret another use of SAI to have the meaning of the first development.
As a side-note, I think you had an interesting sort of blend in your comment.
Maybe this was intentional, (or perhaps you’re just running low on sleep), but it’s certainly interesting.
Oh, and while you’re at it, I’m afraid I don’t really understand the amoeba thing. (and isn’t the plural amoebae?)
Yeah, my question that always lurks in the background is, given some situation in the modern language that originally arose from some change in behavior on the part of some speakers (like using SAI for questions, or counterfactual conditionals, or not using comparative -er on some adjectives), how far away from that initial impetus do you have to get before you wouldn’t reasonably want to attribute the synchronic pattern to the diachronic facts? Is there, in fact, never a boundary? How do we evaluate SPE, for instance? That’s where the amoeba thing came from, when I was thinking about the fact that the living things alive now are continuations from the original source(s) of life, in turn (possibly) from replicating molecules, and so on. And so we are just “elaborations” on that, just as (say) modern English is just us trying to do the exact same thing as our language-developing ancestors from long ago, except …well, just different. It’s mostly just a naive opinion, since I don’t know much about that sort of thing.
I thought recently of another analogy: consider trying to do a synchronic analysis of social action regarding Christmas. In fact, you might end up with an “underlying” conception, /christmas/, which undergoes certain rules in certain “action-lects” of Americans, such that some observe public and/or private religious rites, others exchange gifts, others get do nothing, etc., and a mix of any of the above. But this might instead be a recapitualation of historical change whereby an originally purely religious ceremony developed in different communities in different ways (pretend that Christmas was at some point a purely Christian and religious affair, which of course is false). You might then contend that a synchronic explanation of social beharior wrt Christmas crucially misses the diachronic aspect. But what if you wanted to extend your explanation to extend to the pre-Christian era and its analogue of modern Christmas. You might then say that Chrismas is a sort of borrowing, or in some action-lects, Chrismas was inherited from a pagan holiday with some radical changes. But certainly this has no place in a synchronic explanation, although certainly people are willing to posit some degree of historical information stuck inside people’s synchronic heads. How does it work for phonology?
Another analogy is superstition. Like not walking under a ladder, or not breathing in a tunnel. These actions may, for some people, be totally unmotivated (or at some point in life unmotivated, and later given a sort of post hoc motivation), but they do them anyway. So, is language basically a superstition? Is the semi-productivity of comparative -er (*funner; originally the suffix was fully productive) a “superstition,” or must people develop some motivation for it? How about the SAI construction(s)? Are all instances of SAI originally “superstitions,” which motivations for are later attached? Or does history recapitulate itself in the acquisition process, whereby some SAI constructions motivate others which motivate others — and later the practice is made into a superstition for each particular person. I’m sure that’s not clear at all. Oh well. =)