broadsheets from the empirical underground

I’ve been corresponding with Zoltán Varjú, an enthusiastic proponent of what he describes as “rationalist” linguistics, and Melody Dye, who I would describe as a strongly “empirical” linguist — or rather, psychologist of language.  Also chiming in on that conversation has been Asad Sayeed, an old colleague of mine at the 2005 2003 JHU CLSP Summer School, chiming in in favor of rationalist approaches.

Or, in the form of a tag team match:

Team East: Rationalist linguists: Zoltán and Asad
Team West: Empiricist linguists: Jeremy and Melody

There’s not a lot at stake in these arguments — we each have places that we’re able to do the kind of linguistic research that we think is important, for the most part. But many words have been written. When a twitter-flood got overwhelming late last week, I asked Zoltán to take it to a longer-form medium, and I received a 1000-word info-bomb two hours later.  He didn’t want to blog it, so I won’t quote it here, but I’ll put instead an opening to a new conversation, inspired by a fragment of Zoltán’s info-bomb:

Zoltan writes:

I do not think that the two movements are talking about the same thing. Empiricism is similar to physics, and rationalism plays the role of logic and maths.

It occurred to me that this analogy is extraordinarily seductive, and yet — alas! — it tends to lead to vitriol and mutual contempt between the rationalist and the empiricist camps.  I think I’ve figured out part of why, too: there are actually two analogies packed in here.

Analogy 1: What I mean when I say “I know…”

Rationalist and empiricist linguists actually are talking about two different kinds of knowledge.  Let’s first look at empiricist linguists (and physicists!).  These scientists:

  • try to account for observable data in the universe
  • report their results using statistical measures of certainty (e.g., p values against the Null Hypothesis)
  • accept that there are sources of noise (or signal) beyond the control of the observer (they do not believe, or they have given up waiting for, Laplace’s Demon); these sources are treated as “random” in the Bayesian sense

By contrast, rationalist linguists (and mathematicians!):

  • try to account for philosophical properties of The Universe or The Mind
  • report their results using syllogistic argumentation (e.g.modus ponens and modus tollens)
  • accept that there are sources of noise (or variation) beyond the philosophical bounds of their argumentation; these sources are treated as “e-language” (as opposed to “i-language”) or — in mathematics — proof that a solution must exist is considered sufficient

I think the analogy works out okay here, but I’m not so sure that there’s a meaningful debate to be had: these two views simply talk past each other (over and over and over).

Analogy 2: how do we value our teachers?

However, there’s another domain packed into that simple analogy: the domain of how our culture of science has its valuation expressed in a community of learning (and outside that community).

Among the empirical linguists (and the physicists!), most:

  • work with machinery that costs a lot of money
  • are friendly with engineers, including industry contractors
  • get paid relatively well
  • find work outside the academy easily
  • are best rewarded when their research “just works

Many of the rationalist linguists, on the other hand:

  • work with blackboards and LaTeX
  • have their offices within Humanities colleges
  • get paid like literature professors
  • have difficulty finding work closely related to their studies outside of the academy
  • are best rewarded when their research publishes often

Morlock, trying to cut back on Eloi snacks

Let’s be clear: there’s a big difference in these two parallels. The first parallelism is distinguishing between what we mean when we say we know something; the second parallelism is really about how do we value our teachers and ourselves, and it’s often very easy to conflate the two.

I’m a little dismayed to find this second split, though, because it’s the split that hurts feelings and pride on both sides. I benefit tremendously from it, of course, because I work in the well-paid, grungy, engineering-oriented (“Morlock“) half, rather than the poorly-paid, pretty, theoretically-oriented (“Eloi“) half.

But that personal benefit doesn’t outweigh the risks of really alienating ourselves — all of us who care about learning about natural language — from each other. Intellectually, I’d be much more interested in whether the division in the first analogy can be resolved than in tackling the second one — resolving the second one seems like a far larger question than simple epistemology, since it cuts to the heart of what we think a liberal education should include.


Posted

in

,

by

Tags:

Comments

10 responses to “broadsheets from the empirical underground”

  1. Zoltán Varjú Avatar

    I’m not a rationalist… I’m a methodological anarchist 😀 I do think that anything goes as long as you play the game according to your field’s rules. While we see accepted ‘standard’ paradigms in sciences, I think the ‘multi-paradigmatic’ state of the social sciences and the humanities is a value that we have to preserve. (Mill’s On Liberty applied to science!)

    I like the analogies, and yes this describes the situation more or less exactly. But I think what Jeremy calls engineering is electrical engineering. I know nothing about pay scales, but I guess there must be ‘rationalists’ in those areas since the theory of computation is based on the armchair theories of logicians and mathematicians. In academia, you can find rationalist at maths, informatics, and other departments.

    It is a common misconception that rationalist do not use computational models – but yes, they do use them, this is the 21th century. Just one example, Jan van Eijck’s homepage shows what those guys like to do with computers http://homepages.cwi.nl/~jve/

    Outside academia, I guess hybrid methods are the best when you want something that works. Maybe my own experience and preconceptions say this, but I found similar thoughts on Stuart Robinson’s blog too (http://prospero.bluescarf.net/stuart/2009/01/symbolic_versus_statistical_ap.html)

    1. Jeremy Avatar
      Jeremy

      Zoltán, “anything goes as long as you obey your field’s rules” is a fine approach, but the empiricist/rationalist divide is precisely a divide over what the norms (“rules”) of linguistic research should be.

      You say “electrical engineer” as a distinction among engineer, and I suppose you’re right — I would perhaps argue “information engineer” is the term I’m looking for. Theories of computation have been very helpful for compiler and programming-formalism design over the last forty years, but the engineers I’m talking about (yes, mostly in “electrical engineering” by accident of history) draw their strength from the noisy channel model, cryptography, and — most fundamentally — information theory.

      I have no objection to hybrid models — on the statistical side, the power of information-oriented approaches; on the “knowledge-driven” side, a tremendous wealth of information, encoded in a highly-compact form. These approaches are compatible. But approaches that ignore information-driven approaches do so at the peril of … well, not working very well.

      1. Zoltán Varjú Avatar

        No only compilers and programming formalism, but the whole idea of computing is relying on logic and computation!

        I love electrical engineers (I know many of them, and they are the best source when somebody wants to learn about information theory) and some of them are real ‘information engineers’. But they know that making a computer requires logic design…

        The rationalist/empiricist debate suffers from that both sides wants to be universalist. I think both can be true. For me, it seems that empiricists are following the Humean tradition, they are ‘positivists’ and belive that observation is everything. Rationalist think they know big things, like logic, and they are constructing consistent theories – hoping once reality will fit into their model.

        When I’m less critical, I use the physics, mathematics analogy. But I can put it into other words: empiricists are doing science, rationalists are doing philosophy (where philosophy means analytic here, a kind of meta-theory).

        But don’t forget that Newton’s empirical system worked well for two hundred years. It was based on observations, it used experiments and it was highly mathematical. It described the word using Euclidean geometry. Before Einstein, Bólyai and Lobachevsky developed the system of non-Euclidean geometry (as Bolyai wrote “I made a new world from nothing, using pure negation”) and after relativity theory and quantum mechanics came and found non-Euclidean geometry very useful. Does relativity theory refuted Euclid’s system?

        Information is important, and this is acknowledged by the development of (dynamic) epistemic logic which found to be useful in various fields. Why? Because sometimes you should be able to reasoning about information and knowledge.

        1. Jeremy Avatar
          Jeremy

          See, Z: you can write just fine in public. We could raise the thread (and profile) of this conversation if you just blog your end with trackbacks!

          But anyway: I agree that the empiricist (“neo-empiricist”, pace Asad) linguists are doing science and that most of the rationalists are doing some kind of philosophy. I’m not convinced that the variety of philosophy as practiced within the C of C (again, thanks Asad for the term) — that that variety of philosophy has much to say about natural language “as she is spoke“.

          Along those lines: High Chomskyanism has had absolutely revolutionary effects on mathematical logic and compilers — it’s not for nothing that the Chomsky Hierarchy is better understood by Haskell programmers than most linguists — but I’m not sure that it has had correspondingly revolutionary effects in the scientific understanding of how human language actually happens.

          Chomsky’s contributions to formal language happened just around the time that information theory was beginning to creep out from Shannon’s experiments into the rest of the sciences and engineering disciplines. But it’s only quite recently (since 1990?) that empirical sciences (psychology and linguistics) have begun to cope with the implications of information theory in its ability to reflect partial knowledge; most of linguistics (especially the C of C) still proves things with singleton counterexamples — an awful lot of it is modus tollens with ∀ symbols in the first premise.

          1. Asad Sayeed Avatar

            Re rationalist linguists as philosophers: I kind of alluded to that thought in a blog post I wrote a few weeks ago about the previous debate that Melody started.

            I never got around to writing the sequels.

            I consider myself one of the small intersection of people who

            1. doesn’t think that these magisteria are quite non-overlapping.

            2. does believe that the experiment can be ill-posed while the question and reasoning well-posed and that this is basically the case for the attempt to connect subtle syntactic phenomena with claims for indirect negative evidence.

            But that would require reams of blogposts and maybe a postdoc and some money to discuss properly.

  2. Anonymoose Avatar
    Anonymoose

    What does rationalist linguistics buy us? Sentence trees which seem to have no empirical basis for existing? Models which are only expanded a couple sentences at a time? Endless hand-waving and rhetorical flourishes? Any applications at all? I’ll stick to experiments.

    1. Asad Sayeed Avatar

      Why does the knowledge have to “buy” you anything?

    2. Zoltán Varjú Avatar

      If you are keen on applications, have a look at Mozilla Labs Ubiquity project, its parser is based on the minimalist program.

      My point is we can NOT say that one approach is better than the other! Empiricism is very important and you’re right it gives us a lot. But even empiricism is driven by our preconceptions and it can NOT answer every question. Think about rationalism as a meta-theory, a way to speak about the ‘big picture’. Arguing against rationalism (or empiricism) is a categorical mistake since they are talking about different things! Think of Saussure or the Prague circle or the old grammarians’ school – there’s a big difference between studying actual language and speaking about the language faculty.

      1. Jeremy Avatar
        Jeremy

        Neat, Zoltán! It’s nice to see that some components were actually released. On the other hand, I see that Ubiquity was put on hiatus: an experimental branch that was scrapped is that’s not particularly good evidence for “application”.

        More to the point: I think every time you’re saying “rationalism” you’re lumping together everything from logic and philosophy to Chomskyan linguistics (the minimalist program included). I have no objection to logic or mathematics; I’m just unconvinced that the further extensions (doing “reasoning about the language faculty”) have much to do with how language actually is used by human beings.

        So that ends up looking like you’re defending the Minimalist Program by claiming that anyone who finds it unconvincing must not like logic or deductive proof.

        My point is more that I find deductive proof an ineffective tool in dealing with the properties of language: exceptions abound. Inductive reasoning (empiricism!) is much more worthwhile, and rejecting information theory’s ability to quantify one’s own ignorance isn’t really fair to oneself, let alone the empiricists.

  3. […] evidence, as I suggested before, that Chomsky — and his acolytes — really don’t mean the same thing that […]

Leave a Reply to Jeremy Cancel reply

Your email address will not be published. Required fields are marked *