A talk with Neil Smith

Links & Letters 1, 1994 109-121 A talk with Neil Smith Mireia Llinis i Grau Universitat Autbnoma de Barcelona Q: Your work and interests in linguistics have been very varied, and they have touched upon many dzfferent areas, couldyou te11 us a little bit about the development ofyour research interests in linguistics? N.S: It started when 1 was a student at Cambridge and 1 was bored with modern languages, which was mainly romance philology and the history of German literature before 1500. So 1 did a paper in linguistics and got hooked on traditional (pre-Chomskyan) linguistics so hooked that 1 then came to London to do a PhD. 1 was caught by the idea of doing field work on an unwritten language so 1 read books on anthropology ti11 1 found a nice one, namely the Nupe in Nigeria. Then 1 hitch-hiked out to Nigeria and spent a year sitting in a mud hut working on the language. And then 1 wrote out my thesis which was a grammar of Nupe. Although it was on Nupe it was a thesis in general linguistics, influenced mainly by Michael Halliday and his work - Halliday was in London at that time. As a result of that, 1 was given a job at the School of Oriental and African Studies (SOAS), where 1 was converted to generative grammar by one of my colleagues in the Department of Indian Studies. As a result of that 1 decided to go to MIT and spend a couple of years there, from 1966 to 1968, with the idea of working on generative grammar of West African languages. In particular, 1 wanted to look at the syntax of Ewe, but when 1 got there Chomsky was on sabbatical and 1 was adopted by Morris Halle, who was very kind to me and inspired me with an interest in phonology which 1 had not had before, at least generative phonology. As a result of that, 1 spent most of my time at MIT, at least for the first half, working on phonology. Then Chomsky came back and 1 worked on syntax. 1 also spent some time in California (at UCLA) working on African languages and linguistics, but my first really decent work as opposed to the dabblings in West African languages arose from that, because while 1 was in the States my first son was born and 1 started work on language acquisition in about 1968, when he started to speak. 110 Links & Letters 1, 1994 Mireia Llinis Grau Because 1 had spent so much time working on phonology it turned out to be a book on the acquisition of phonology rather than the acquisition of syntax. It was as a result of Halle's influence and the fact that 1 had been able to work at MIT that 1 could do that. But 1 was only there for two years and when 1 got back 1 was working mainly on syntax and general theory, but 1 was doing research on phonology. And that left me interested in phonology for many years. It was probably another ten years before 1 gave up phonology and 1 haven't done much in that since. After that I suppose it was a continuation of my conversion to theory. When 1 started as a student 1 was only interested in facts, and description, and it was really at MIT that 1 began to be interested in explanations and theory. That developed with my work on the acquisition of phonology, and was encouraged when 1 got back mainly by Deirdre Wilson with whom 1 worked very closely. We wrote a book togetherland that and other things 1 did on the philosophy of language was a development from description to explanation. Q: Wbat didyou JZnd in tbeo y that was more motivating than continuing with description? N.S.: 1 suppose the real reason was that every time you describe something you just add another little pebble on top of your heap, following the old fashioned idea that science is cumulative; you add a brick to the wall, everything is factual and so on. 1 think the problem with that view is extremely wellexpressed by Popper in his Conjectures and Refitations: in order to be intellectually honest and in order to be able to make a serious contribution, you need to state the conditions under which you will give up your beliefs rather than simply add to them. That's a position that later became known as naive falsificationism, but for a long time 1 found it very appealing and 1 decided that, although 1 was quite good at manipulating data, it wasn't really the description that turned me on as much as what might consitute an explanation. It gradually dawned on me, much later than it should on most people and much later than it does on students these days, that if one wants an explanation one has to have a theory. And the only coherent theory that 1 came across was Chomskyan in one sense or another. At the same time, 1 got interested in pragmatics because of Deirdre Wilson and her work with Dan Sperber2. So, 1 set up a conference on mutual knowledge that resulted in a book that 1 edited, which sort of generalized my interest away from core linguistics to problems of communication. Again the idea was to see how on earth one could explain how communication took place. Although the book was called Mutual Know- 1. Srnith, Neil and Deirdre Wilson. 1979. Modern Ltnguwtrcs. The Results of Chomsky? Revolut2on. London: Pcnguin. 2. Sperber, Dan and Deirdre Wison. 1986. Releuance. Oxford: Basil Blackwell. - A talk with Neil Smith Links & Letters 1. 1794 11 1 ledge, there wasn't really a consensus view but the conclusion 1 drew from it was that mutual knowledge was a waste of time. Q: What does mutual knowledge standfor? N. S. : "Mutual knowledge" was a locution devised by Schiffer in the 1960's and the idea was that in order to communicate successfully you needed to establish mutual knowledge between speaker and hearer. Sperber and Wilson have essentially demolished that position and replaced it by a situation in which you optimize relevance rather than dealing in terms of knowledge which presupposes factuality or truth, and factuality is not necessarily important. That again was a diversification into greater interest in theory. And 1 suppose al1 the time what 1 try to do is unite as many different strands in linguistics as 1 can, so that 1 have an overall conspectus of the discipline, which includes phonology, syntax, and a little bit of semantics and pragmatics, as well as detailed analyses of any one part of it. 1 have always been a jack of al1 trades and tried to get my fingers into as many pies as possible and draw everything together, so that there is an overall theory that will account for everything that is going on. Q: You've mentioned so many dzfferent areas in the answer to thejrst question that Iguess I will now ask you to expand on some of the dzfferent things thatyou've talked about. One of thejrst things you've mentioned is language acquisition and howyou became interested in it. What is thepresentstate ofjrst language acquisition studies and what has been achieved? N.S.: It's sort of fun at the moment because so much seems to be going on simultaneously. There's Radford's recent book3 and that work has been influential, but 1 think the real turning point came with the arrival of the principies andparameters framework4 at the end of the 1970s and beginning of the 1980s in Chomsky's theory where at last it began to seem possible to provide some sort of explanation of how it was that children could learn their first language. In Chomsky's early work he had postulated a language acquisition device which presupposed some kind of evaluation measure. There were al1 sorts ofproblems with the evaluation measure: how on earth the child could choose among competing descriptively adequate grammars, and if you had a descriptively adequate grammar why would you want to go and look at seventy more of them and then start choosing between them by whatever kind of metric. The details of the metric were never particularly well worked out, anyway. 3. Radford, Andrew. 1770. Syntactic Tbeory and tbe Acquisition of Englisb Syntax. Oxford: Basil Blackwell.See the review section of this issue for a review of this book. 4. See the review section in this issue. The following books reviewed there are within this framework: Ouhalla (1771), Radford (1770), Botha (1787), Speas. 112 Links & Letters 1, 1994 Mireia Llinis Grau The developments in the principles and parameters framework seemed to reiterate the kind of work in biology especially in immunology. It is described quite nicely in an article by Piattelli- Palmarini5 where he describes how work in immunology underwent a change from being what one might cal1 instructive to being selective. That is, it used to be believed that the system had to be instructed by impinging external stimuli in order to react appropriately and develop the correct kind of antigen. Then gradually al1 those theories were replaced by a selective theory that presupposed that the organism had available right from the beginning al1 conceivable antigens and therefore al1 it had to do when some external stimulus impinged was select the correct antidote from an antecedently available reservoir. If you think about it, that is exactly similar to what happened with having an evaluation measure where you had to generate a particular reponse to the incoming stimuli, where you constructed a grammar and then when you got lots of them you chose among them. In principles and parameters you dont really have to go through that procedure, there are no rules of grammar anymore, what you have is universal principles, a number of which are parametrized and, of course, you have lexical entries, and opinions differ as to where the parametrization lies. In the case of language then what happens is that the child selects from an antecedently available set of grammars the actual grammar that he happens to be exposed to. That is an interesting transition in that it now makes it feasible to acquire language if there is essentially one supergrammar specified by Universal Grammar (UG) and the idiosyncratic differences across languages can then be selected from that available set perfectly straightforwardly. 1 think that development especially in syntax and gradually in phonology has had a beneficial influence on language acquisition research because people can begin to see how they can answer the questions as to how children can possibly solve particular problems - of negative data, problems of retreat, problems of tadpoles becoming frogs and so on. 1 think Chomsky's influence in that respect has been for the second time extremely beneficial to language acquisition research. Q: Could we say that it has given it a methodology? N.S.: Right, it's given it a methodology and it's given it a framework which makes it look like a feasible enterprise rather than a miracle. In some of his earlier work, Chomsky drew the distinction benveen problems and mysteries. Problems were those things which one might rationally address and solve. Mysteries were really fascinating but phenomena which one had really no hope ever of understanding, for instance, the mystery of free will. 1 think the principles and parameters framework has probably made language acquisition 5. Piattelli-Palmarini, M. 1989. "Evolution, Selection and Cognition: from "Learning" to Parameter- Setting in Biology and in the Study of Language". Cognition 31. 1-44. -- A talk with Neil Smith Links & Letters 1, 1994 113 study change from being some kind of mystery to being a problem that one might conceivably solve. And both in syntax and phonology there is a huge amount of work now attempting to put into practice the program that Chomsky devised in the late 1970s and early 1980s. There are further - maybe more technical, but quite fascinating - developments in first language acquisition arising from the details of syntactic theory. In particular, if you think of what's happened in post-Barriers work where X-bar syntax has been generalized to functional categories. Radford's work and a lot of other work has concentrated on the development of functional categories and there's the interesting hypothesis that what characterizes children's speech is the absence of functional categories, and that simple claim, given that we now have a suitably complex and well articulated theory of what the functional categories are and how they might even constitute some sort of natural class, makes empirical predictions across a huge range of constructions. For instance, if there are no functional categories and if functional categories are responsible for things like directionality constraints then it would follow that little children have no directionality consiraints and they ought to have basically freer word order. Now, again and again one looks at child language data and one sees that there is much more free word order than there is in the adult language - unless, of course the adult language has totally free word order. But if you look at French and English and Spanish and Irish, lots of languages, children' S language always seems to display greater freedom of word order than the adult language, and now we have an automatic ex~lanantion for that without having to make up some other kind of story. Ianthi Tsimpli's recentl~ completed London PhD thesis6 is an elegant demonstration of these points. Q: You are saying that research in jrst language acquisitionfiom a generative perspective (pecz$cally within the principies and parameters frdmework) is valid and desirable. Ii other work in jrst language acquisition we$l? N.S.: 1 think we' re now back to the dichotomy we went into before with regard to theory and description. 1 think that what is lovely about current work in first language acquisition is that it really is imbued with a theory which, for independent reasons, seems reasonably good and has begun to make predictions across a wide enough range for people to: a) take it seriously as a syntactic theory or as a general theory of language; and b) to have implications across a really wide domain. Now, there's clearly need and room for al1 sorts of studies on first . ~ language acquisition in what one might think of as a theory-neutral framework. There is the childes database, for instance, and if you want to find out the 6. Tsimpli, Ianthi M. 1992. Functional Categories and Maturation: Tbe Prefinctional Stage ofLanguage Acquisition. PhD thesis. University College London. 114 Links & Letters 1, 1994 Mireia Llinis Grau incidence of free word order in Catalan, for example, you can look it up to see if there are any studies in the language. If so they can be encoded and that information can be accessed and used for further interpretation. But if one wants explanation then one has to go beyond this. 1 don't know of any other theory of first language acquisition that is remotely plausible or that has anything like these results. Now, ofcourse, that doesn't mean there is agreement, there was radical disagreement at the Boston Conference7 a few months ago. There were almost stand up fights between those who thought functional categories were available from the beginning and those who swore that functional categories matured later on. Nonetheless, there are empirical issues bearing on this, and it should be possible to come to a conclusion. But one can only ask these questions once one has a theory of functional categories. 1 mean, previously people would say things like "children's speech is vaguely telegraphic "o? they miss out the function words", but there wasn't a theory of function words and nobody knew what function words were. Q: Can second language acquisition be usefilfor the studies on jrst language acquisition? Are the two processes parallel? N.S.: 1 think they can be. And 1 think it is a matter of theoretical disagreement at the moment, as to whether the parallels are interesting . The position that 1. Tsimpli and 1 developed in the WorkingPaperf recently implies that the parallels are not nearly as great as most people think they are. What we argued for in that paper was that parameter resetting in L2 is not a possible strategy and we did it for various reasons, largely, for instance, because of the discontinuities between L1 acquisition and L2 acquisition; that is, L1 acquisition appears to be deterministic whereas L2 acquisition clearly isn't , you have to work awfully hard and al1 sort of special inputs have to be taken into consideration. The "end state", the "steady state" that one ends up with after L1 acquisition is pretty uniform, it is close to identical across al1 the members of the speech community, indeed across the species, but the "steady state" one ends up with in L2 acquisition is clearly not uniform in the same way and is not usually remotely as perfect. That suggested to us that it would be an interesting hypothesis to suggest that, whereas L1 acquisition is correctly characterised in terms of parameter setting - where parameters in the version we're interested in are associated with lexical items rather than with principles of UG as they used to be for Chomsky - with L2 acquisition, parameter setting is no longer an option; that is, you can't reset parameters, al1 you can do is still 7. The Sixteenth Annuai Boston University Conference on Language Development. October 18-20, 1991. 8. Tsimpli, Ianthi and Neil Smith ,1991. "Second Language Learning: Evidence From a Polyglot Savant". University College Working Papers in Linguistics 3. 171 -1 83. A talk with Neil Smith Links & Letters 1, 1994 115 have access to principles of UG but only nonparametrized ones. Then you have various other possibilities like errors due to general learning strategies - if there are general learning strategies , transfer errors from your first language, maybe inductive errors of one kind or another ,etc. Now, if that position is right - and there's a huge literature saying that what goes on in L2 acquisition is precisely parameter resetting -, but if they're al1 wrong and if we are right, then L2 acquisition can't say nearly as much about L1 acquisition as it could in the other system, simply because by hypothesis now - and hopefully by argument as wellthe parallels are not nearly as great as they were supposed to be. Q: Which data do you work on to test your hypotheses? N.S.: We've looked at various bits of data. The nicest data that we've looked at recently is work with an " idiot savant", a man called Christopher (C) . This is a man who is institutionalized because he is mentally retarded such that on non-verbal IQ tests, he scores between 40 and 75. He's unable to cut his nails, he cannot indulge in normal social interaction, he's awkward, he loses his way when he goes from place to place. When he came and stayed in my home for two days, even at the end he was still puzzled as to how to get upstairs. But he can speak 20 languages, with varying degrees of fluency. He knows English, French, German, Spanish, Italian, Dutch, Modern Greek, and some Hindi reasonably well, and then another dozen languages like Welsh, Turkish, Finnish, Norwegian, Swedish, Danish, Polish and Russian not quite so well. As part of our investigation of this man, we've been teaching him two new languages ( Berber and an invented language) under controlled conditions, so that we can control the input. The original idea when I thought of doing this was : wouldnt it be nice if he were really learning these languages as though they were first languages? And that when you therefore set a parameter you might get some kind of parametric cascade and you could see what parameters were related to what particular sets of phenomena, and we could have answers to al1 these questions such as "1s the pro-drop parameter" one parameter or two parameters, because of inversion problems and so on? Q: So when you started working on this, you thought that there were more parallels? N.S.: At that time, when we were setting this up, 1 thought there were far more parallels than there turned out to be, but then when we started teaching C two languages it transpired that the kind of mistakes he was making in these - and in fact in other languages especially in Modern Greek, which we've looked at in great detail, but also French, German, Spanish and Italian - were typic+ly characteristic of L2 learners. They were not the kind of mistakes typically characteristic of L1 learners, so you get transfer errors al1 the time, you get 116 Links & Letters 1, 1994 Mireia Llinhs Grau general learning strategies, you get analogies of a kind you dont find in L1 acquisition. That gradually persuaded us - maybe the others didn't need persuading, 1 needed persuading - that L2 acquisition was different in kind to L1 acquisition, at least for the majority of people - there's a residue of very , . skillfui expert L2 learners who may retain sime;hing of their L1 ability, but 1 doubt it, 1 mean, Christopher is extremely good at learning languages and we had an example this week. On Friday, Ianthi and 1 went to see Christopher, who lives in the North of England, with a Dutch woman who is representing Dutch television - because theywant us to talk on Dutch television next month. Now, Christopher knew a little bit of Dutch and we got him to translate a passage a long time ago, which indicated that he had some rudimentary knowledge, but it wasn't particularly good. On Wednesday of last week, the director of the institute where he lives gave him two days off work and a copy of Tedch YourselfDutch. He spent the intervening time just reading it and going through al1 the exercises and so when we arrived he was able to converse in reasoiably fluent Dutch. It had al1 sorts of mistakes, but it was remarkable a) that someone with his mental problems could concentrate to that extent, and b) that he could absorb the material so fast. He clearly does have a phenomenal ability to absorb vocabulary items. But when you look at the syntax, his syntactic errors are nearlyall transfer errors from L1, his English mother tongue, or they are general inductive generalizations of a kind that one would expect in any L2 learner, rather than the kinds of mistakes that little children make. For instance, in learning Berber he adopted a general strategy to put al1 clitics after the verb, whereas in fact in Berber clitics quite often are in second position, if there is a focus marker or a future maker or negation, and in any of those situations, the clitic is attracted and climbs to the superordinate functional category. Now, although he'd had lots of examples of these, he didn't induce that generalization, he induced a generalization which happened to be false, but which was very sensible: namely, "clitic goes after the verb". 1 don't know of any L1 acquisition data for Berber syntax but that is not a typical kind of child mistalte. So we think that there is evidence that L2 acquisition is different in kind to L1 acquisition and certainly there is evidence from Christopher. 1. Tsimpli and A. Roussou have got a paper in our Working Papers indicating the same conclusion from experiments carried out on adult speakers of Greek learning English as their second language. So, again, there is converging evidence that parameter resetting doesn' t take place. So, that's a long-winded way of saying that the parallels between the two are not nearly as close as we might like, but, obviously, ifyou are going to characterise the notion of transfer error even, then you need to have a theory of what the grammar is so that you know what a transfer error could be. So you can still get evidence for the theory 9. Tsimpli, Ianthi and Anna Roussou.1991. "Parameter Resetting in L2?" Uniuersity College Working Papers in Linguistics 3.149-170. A talk with Neil Smith Links & Letters 1, 1994 117 of grammar from L2 acquisition and you can still get evidence or useful insights into how you might structure L2 instruction. Q: Having talked about all these dzfferent areas, do yoti envisage ageneral theo y of language compatible with the generative approach to grammar on the basis of existing theories? N.S.: What 1 think is significant in a number of recent developments is the way that at least three different strands have come together. The three strands that appear to me as being most important are: the work in generative grammar - especially the syntax of Chomsky, and in fact one can think of the whole of generative grammar including phonology as being uniform in this respect, despite what people like Halle and BrombergerIo say; the second strand is the work ofsperber and Wilsonl'in pragmatics which is one of the most significant developments in areas peripheral to core grammar of the past ten years - . It represents the growth of a fully explicit and sophisticated theory of pragmatics, the theory of utterance interpretation, and 1 think Sperber and Wilson's Relevance theoy has now enabled us to redraw the lines of demarcation in a number of areas so that we no longer try falsely to give syntactic explanations for facts that are not syntactic but are semantic, as happened in the sixties. Now we've discovered that we can give explanations which are actually pragmatic rather even than semantic, and some of my own work has been devoted to demonstrating that you can indeed provide pragmatic explanations for phenomena which simplify the syntax or the semantics. One particular area that I've looked at with my son is on the analysis of c~nditionals'~. There are traditional problems in the semantics of conditionals, in particular in the so-called paradoxes of material implication. One of the things we tried to show was that those paradoxes dissolve once you have a pragmatic input rather than simply a semantic or syntactic one. Now, that pragmatic analysis is totally or largely . parasitic on the existente of a generative grammar of a Chomskyan kind, but what we want to do is say that certain phenomena of language ought not to be handled by a generative grarnmar, not even a generative grammar that includes a leve1 of LF where you have particular logical or quantificational statements, but that some phenomena of language will fa11 out from a theory of pragmatics, a theory of utterance interpretation rather than a theory of utterance meaning. 10. Bromberger, Sylvain and Morris Halle.1989. "Why Phonology is Different". Linguistic Inquiry 20. 51-70. -. 11. See note 2 and the article by Sperber and Wilson in this volume. 12. Smith, Neil and Amahl Smith ,1988. "A Relevance Theoretic Account of Conditionals". Hyman, L. & C. Li (eds.) Language, Speech andMind: Studies in Honourof VtctoriaA. Fromkin, pp. 322-352. London: Routledge.