Of Minds and Language (36 page)

I think that the contrast between Aristotelian and Galilean laws is very relevant to the study of language because there are various ways of approaching language universals. One of the ways in which you could approach them is like what Joseph Greenberg did with his various arguments on universals. That is not the kind that I am interested in, and it is not the kind of universals that generative grammar really is interested in. The kind of typological universals that Greenberg discovered might be interesting for discovering the type of hidden universals that generative grammar is interested in, but they are not the end of the enterprise. It is worth noting that Greenberg's universals are really surfacing properties of language that typically can be explained in functionalist terms and allow for a variety of exceptions. That is, they are basically tendencies of various sorts, but that is not the kind of thing that generative grammarians have focused on in the past fifty years.

In fact generativists conceived of universals as basically properties of universal grammar (UG). This is the most general definition of universals that I could give, if you ask me what a language universal or linguistic universal (LU) is for a generative grammarian. But that definition actually depends on the specific understanding of UG, and that has been changing for the past 30–35 years. I should say though that no matter how you characterize UG, its content is defined along Galilean lines. We cannot expect universals to be necessarily found on the surface in all languages. That probably is not the case. Conversely, all languages might have a word for yes and no. (I haven't checked, but say it's true.) I don't think we would include this as part of UG, even though it is in all languages. So the understanding of universals that we have as generative grammarians is based on a theory of language that has, as I said, been changing for the past 30–35 years in many ways that do not, I think, make some people very happy as consumers because, to anticipate the conclusion that I will be reaching, the list of universals that we will reach as syntacticians or grammarians will be very refined and abstract, and not directly useful to, for example, the study of language acquisition. We should not be discouraged by that fact. This is a natural result of pursuing a naturalistic approach to language.

What I would like to stress first of all is that the study of syntactic or linguistic universals has run through various stages in generative grammar. In particular, one of the first advances that we were able to make in the understanding of linguistic universals was the distinction that Chomsky (1986b) introduced between I-language and E-language. As soon as you make that distinction, you really have the distinction between I-universals and E-universals. E-universals
are the type of thing that for instance Greenberg universals could be. I-universals would be something like, for example, some deep computational principles of a very abstract sort that are only manifested in very refined and rarified phenomena. It is not something that you can observe by just walking around with a tape recorder or anything of the sort. In fact I think the study of I-universals in this sense started with “Conditions on Transformations” (Chomsky 1973), or if you want, with the discovery of the A-over-A principle – that is, an attempt to try to factor out what the abstract computational principles are, based on a fairly refined empirical view of language. It is true that “Conditions on Transformations” wouldn't have been possible before Ross's (1967) investigation of islands. It was only once you reached that very detailed empirical picture that you could try to extract from it this very abstract rule, so Galilean in nature. And so it will be, I think, with other universals.

I think that the stage of the principles and parameters (P&P) approach constitutes a serious attempt to come up with more of those universals, once you have a very good empirical map. That is, once you have attained very good descriptive adequacy, you can try to find and formulate those abstract universals. Things changed, I think, with the advance of the minimalist program, and in particular more recently with the distinction that Hauser, Chomsky, and Fitch (2002) have introduced between the narrow faculty of language (FLN) and the broad faculty of language (FLB). This further distinction basically narrows down the domain of what we take to be language, to be specifically linguistic, and that of course has a direct influence on what we take LU to be. That is, if by LU we mean specific universals for language, then we are going to be looking at a very narrow field, a very narrow set, that is FLN. And there, what we expect to find will be basically abstract general principles such as minimal search, or various refinements of relativized minimality, cyclicity, etc.

Once we reached that stage, then people began to see that perhaps those universals are not specifically linguistic, but might be generic or general principles of efficient computations belonging to third-factor properties, for example. But these would be the kind of LU that may actually be at the core of FLN. Remember that, as Chomsky has discussed recently,
¹
there are basically two ways of approaching UG – from above, or from below. And these two approaches will conspire, ideally, in yielding the sources of LU, but for a while we will get a very different picture depending on which perspective we take. Notice, by the way, that if some of these LU are part of third-factor properties, then they may not be genetically encoded, for example. They may be part of general physics or chemical properties, not directly encoded in the
genome. In this case, the study of LU dissociates itself from genetic nativism (the most common way of understanding the “innateness hypothesis”).

The refinements that we have seen in the study of language and LU will force us to reconsider the nature of variation. In this sense, one very good and productive way of studying universals is actually studying variation.
²
Here again, recent advances in the minimalist program have been quite significant because the notion of parameter that we have currently is very different from the notion of parameter that we had, say, in the 1980s. In the 1980s we had a very rich understanding of parameters, including a fair amount of so-called macroparameters of the type that Mark Baker (2001) discussed in his
Atoms of Language
. We no longer have those macroparameters in the theory, simply because we don't have the principles on which those macroparameters were defined. However, we still have the effects of macroparameters. For example, there is something like a polysynthetic language, but I don't think we have a polysynthetic parameter, or rather I don't think we have the room for a poly-synthetic macroparameter in FLN. How to accommodate macroparametric effects in a minimalist view of grammar is a challenge for the near future. But it is a positive challenge. That is, maybe this new view of grammar is actually a good one, as I'll attempt to illustrate through just one example. Take headedness as a parameter. We used to have a very rich structure for P&P, and one of those parameters was basically one that took care of whether complements were to the left or to the right of their heads in a given language. Now the minimalist take on UG no longer has room for such a parameter, but instead tells us that if you have a simple operation like Merge that combines alpha and beta, there are basically two ways in which you can linearize that group (either alpha comes before beta, or after). You must linearize A-B, due to the physical constraints imposed on speech, and there are two ways of doing it. Notice that there you have an effect, since you have a choice between two possibilities depending on the language, but it is no longer the case that we have to look for a parameter in the theory that encodes that. It may just be that by virtue of the physics of speech, once you combine alpha and beta, you have to linearize that set by going one way (alpha before beta) or the other way. I think that this offers new perspectives for studying parameters because LUs are different depending on your theory of language.

Now let me briefly conclude by saying that in a sense, the linguistic progress that we have seen over the past thirty years has taken us closer to a study of LU that is truly Galilean in nature. But that actually should raise a couple of flags, if language is just part of our biological world, and linguistics therefore part of
biology, because biologists are typically, and by tradition, not very interested in universals in the Galilean sense; they are more interested in the Aristotelian kind of universals and tendencies. Gould, Lewontin, and others were fond of noticing two facts about biologists. First, they love details, they love diversity, the same way philologists love details. I certainly don't like diversity for its own sake. I am interested in general principles and only use the details to the extent that they can inform the study of general principles. Secondly, biologists don't usually think that there are biological laws of the kind that you find in physics, just because the world of biology is much messier than physics. But here I think linguistics has an advantage, because in a very short history (roughly fifty years) we have been able to isolate invariance amidst diversity, and this is what I was thinking of when discussing I-language vs. E-language, or FLN vs. FLB. One of the things that we have been able to do is make the study of language the study of very simple systems. By narrowing down and deepening our understanding of language we can actually exclude things that belong to details and focus on things where we can discover very deep and comprehensive principles that will be just like what you can find in Galilean laws. That is, they will be exceptionless, abstract, invariant, and hidden.

For me, being asked to talk for ten minutes about universals is a bit like being asked to talk for ten minutes on the economy of northern Minnesota in 1890. That is to say, I don't know much about Minnesota and I don't know many universals either. But that's fine, because it allows me to take a very selfish perspective on the subject. I am a psycholinguist and as such it's not my job to discover linguistic universals, but to
consume
them.
³
I work on language acquisition, and it is very important when we are trying to understand language acquisition to assess how much children already know when they begin the task of acquiring their target language from their linguistic input. So what matters to me is not just that something is universal, but the idea that if it is universal, it can be innate. And in fact it probably is – how else did it get to be universal? So I will assume here that universals are innate, that they are there at the beginning of the acquisition process,
⁴
and that they can guide acquisition, increasing its accuracy and its efficiency. Language acquisition is very difficult and needs all
the guidance UG can give it.
⁵
What I will do here is to highlight universals in relation to syntax acquisition. I am going to be walking into the universals store with my shopping bag, and explaining what I would like to buy for my language acquisition model, and why.

A very important point that is often overlooked is that universals (embodied in innate knowledge) play a role not only when learners are trying to find a grammar to fit the sentences they have heard, but at the very moment they perceive an input sentence and assign a mental representation to it. They have to represent it to themselves in some way or other, and it had better be the right way, because if they don't represent it correctly there is no chance that they will arrive at the correct grammar. So innate knowledge has its first impact on the acquisition process in guiding how children
perceive
the sentences in the sample of the language they are exposed to. They have to be able to recognize nouns and verbs and phrases and the heads of phrases; they have to know when a constituent has been moved; they have to be able to detect empty categories, even though empty categories (phonologically null elements) are not audible; and so forth. And that is why they need a lot of help, even before they begin constructing a grammar or setting parameters. I want to emphasize that this is so even if acquisition consists in setting parameters. In the P&P model we like to think that an input sentence (a trigger) just switches the relevant parameter to the appropriate value. But for someone who doesn't know what the linguistic composition and structure of that sentence is, it won't set any parameters, or it won't set them right. So if children get their representations right, that's a very good first step, because it will greatly limit the range of grammars that they need to contemplate as candidates for licensing the input they receive.

Learners need to know what sorts of phenomena to expect – what sorts of elements and patterns they are likely to encounter out there in this language world that is all around them. As one example, consider clitics. Children have to be alert to the possibility that they might bump into a clitic. Imagine a child who has begun to recognize that certain noises compose sentences that contain verbs and objects, and that objects consist of a noun with a possible determiner and that they normally follow (let's say) the verb, and so on. This child shouldn't be too amazed if, instead of the direct object she was expecting at the usual place in a sentence, she finds a little morpheme that seems to be attached to the beginning of the verb – in other words, a clitic. Infants need to be pre-prepared for clitics, because if they weren't it could take them a long time to catch on to what those little morphemes are and how they work. You could imagine a
world of natural languages that didn't have any clitics, but our world of natural languages does, and infants pick them up extremely early: they are among the earliest things that they get right (Blasco Aznar 2002). So it seems that somehow they are pretuned to clitics, and to the ways in which a clitic might behave. Sometimes a clitic can co-occur with a full DP object (usually it doesn't, but it can); and there can be indirect object clitics, and locative clitics and reflexive clitics and partitive clitics; and sometimes multiple clitics have to come in a certain order before the verb, and learners should watch out for whether that order is determined by an array of properties that includes person as well as case. None of these differences from phrasal arguments seem to take children by surprise.