Sunday, 19 April 2015

On Being a Generative Linguist

I am a Generative Linguist

A. Defining Myself as a Generativist:
My own commitments as a linguist are to the ideas that (i) the human language faculty must be described in terms of a system that generates infinite creative possibilities from finite symbolic means and that (ii) the ultimate object of inquiry is our internal knowledge of language, not some description of the data
produced by it. ( I think further that Chomsky introduced these ideas to linguistics, so in that sense I am a Chomskian Linguist.)  I am fascinated by the question of to what extent the ‘special’-ness of human language flows from innate (and undoubtedlly rather abstract) capacities. I take a rather open minded view of what those innate capacities are—representational vs. procedural vs. a product of the special ways in which we learn and generalize— though I lean more to the latter than the former. I do not particularly care whether those special things are unique to the language faculty or not, although I do think that language provides the richest and clearest instantiation of those properties. I certainly don’t believe that representational facts about language (whether we are talking about Principles of the Binding theory, or the Functional Sequence) are innate per se. But I am fascinated by the emergence of patterns in language that appear to be common to all instantiations of it, and I suspect that discovering these kinds of commonalities are a good clue to what is unique and special about the human mind. All of these things, I think I have in common with a broad swathe of linguists who are usually labelled ‘generative’, but it is a far cry from the simple way in which generative grammar is characterized as: “Massive amounts of representational innateness, loads of highly specific universals and a belief that Chomsky is always right”. I do not recognise this description of the field, and I am dismayed by the fake ideological battles fought in its name.

B. Improving the Nature of the Generative-internal Conversation
If my credo in (i) and (ii) above is really shared by most generativists then we should actually be a broader church than we appear to be from the outside. There are parts of ‘generative’ grammar that I do not feel a particularly strong part of . For example, I am not sympathetic to recent trends in Biolinguistics, which to my mind is guilty of extreme Overreach in attempting to connect linguistics to Biology. I think it gives the whole field a bad name. The granularity gap and the terminology gap (to put it in Poeppel’s terms) are still too great to sustain the specific kinds of proposals that are  being taken seriously in this sub-group. I also am dismayed by a culture in which much theorizing in a highly theory-internal way is done on the basis of a priori and conceptual arguments concerning simplicity and ‘perfect design’. I think we have no bloody idea what constitutes perfect design when it comes to the brain and I suspect that Jackendoff, for example, is right that redundancy and parallel architectures are the more realistic structures. I find personally that alternative architectures that involve non-derivational constraint based systems, or parallel architectures do not actually get taken seriously in our field, and that there is a privileged discourse within syntax that uses a rather specific set of analytical choices and terms (Merge, feature checking, DM-style architectures for morphology) which I cannot even begin to participate in because I do not share the assumptions. The privileged tool box exists even though the broad generative perspective that I sketched above is consistent with all of these alternative architectures and implementations. The same is true about the particular ways in which the interfaces are conceived, which have not actually changed in 50 years, even though the background philosophical rhetoric has changed massively. We need to find a way to keep the system open to technical innovation, and different ways of doing things, while still maintaining the ability to have a common conversation. Right now, there seems to be no way in terms of the current linguistic disagreements in popular media, of distinguishing between a Jackendoff (who I think is NOT crazy) and truly uninformed and purely ideology driven people. Our image in the world, as well as our ability to make progress on our core scientific questions are both dependent on us getting our house in order from the point of view of what is our scientific starting point and what we are committed to. (It would be nice in this regard to have a list of major results and improvement in understanding that generative linguistics (broadly construed) has contributed to the field during the past 50 years).

C. Improving the Nature of the Conversation between Generativists and the Others:
If we do this clearly, and make clear what our core is, distancing ourselves from minority positions and extremes in our own field, then we have a better chance of heading off the real big source of debate affecting our  enterprise, and setting itself up to undermine it, which is the trendy view that
everything can be solved with Big Data and Lots of Counting. This is the behaviourist mindset rebooted with shiny new computational tools. The reasons it exists are the reasons it has always existed: people feel comforted by  objectivity; it feels hardcore and scientific. But behaviourism does not work, and we must fight the simplistic versions of it which would seek to shut down classic generative linguistics.
Having said that, I wonder though, how many of those on the ‘other’ planet out there would be comfortable with my simplistic characterization of them as rebooted behaviourists. There is of course an interesting old question lurking here— the classic mind-body problem. People who emphasize connectionist architectures and emergent grammars are also in their own way fascinated by the idea that the illusion of order can be made to emerge out of chaos, that the illusion of mind can emerge out of matter, once things are complicated enough and incorporate enough ‘feedback loops’. Learning and ways of learning are vitally important questions for our field, and vitally important to understanding the human mind. We do not make progress on them if we simply mock the connectionists or the counters. Rather, simulations of this type are ideal for exploring the tension and interplay between categorization, levels of generalization, and the emergent rule systems. One of the differences between the earliest stages of generative grammar and today is that we understand much better how an innate tendency or ability that is very abstract can give rise to quite rich and specific solutions in the logical problem of language acquisition. Our differences with the connectionists and the counters are that we are pushing on different ends of the problem, and have a tendency to favour solutions with a different balance between the hard wired and the emergent. The rhetoric that underlies this conversation is what is most damaging to progress. The debate is always set up as a battleground where someone is right and someone is wrong. In point of fact, the connectionists conceded ages ago that something about the human mind must be innate (see Elman’s classic textbook on Rethinking Innateness where he makes this point in the introduction), and generativists have long backed off from the idea that Principle A could in any way be representationally innate in the human mind. The divisive rhetoric belies this essential convergence. The question of what in fact needs to be considered innate is an open one, and one which both kinds of linguists are interested in. I for one am also completely open to the idea that the magic innate thing or things is not unique to language, although I do think as I said above, that language is the richest and most accessible manifestation of that special thing. Moreover, we need to keep open the lines of communication across the two camps, with their different methodologies and different ‘gut feelings’, because that way we will learn from each other and actually find answers.

Incidentally, I do think that there is some positive action going on on the B and C fronts, but not enough.  

D Specific Big Picture Issues I am Passionate About:
  • Separating structure/hierarchy from linearization and rethinking our standard architectures concerning the interface relationship between the narrow computational system and Externalization
  • Rethinking the interface with the semantics: putting some structural semantics back into form; understanding the work done by the concep- tual primitives at the interface
  • Spanning and the formal status of ‘word’
  • Derivational vs. Constraint based and Parallel architecture models