Sunday, December 16, 2012

Physics Envy

Yes, I suffer, badly. Not only that, but I distrust those that don’t. A good dose of Physics Envy is the sign of good theoretical taste.  That’s why I prize fMRI snaps of frontal lobes awash in green whenever ‘CERN,’ ‘Higgs boson,’ ‘renormalizable,’ ‘Standard Theory,’ ‘SU-3 symmetry,’ etc. are flashed to visual and auditory fields.  It is beyond dispute: Physics is THE science. It provides THE models for explanation.  Its methodological precepts should be stolen and copied at every opportunity. What physics has is what every right-minded science should aspire to. The fact that we are likely to fall short (maybe very short) is no excuse.

I mention this because I had an acute spike in my physics-enviometer recently as I read some old essays by Steven Weinberg (collected here).  Here are four observations he made with suggestions of how these may be relevant to linguistics.

1. Weinberg, aficionados of Chomsky’s philosophical writings might recall, is a strong proponent of the Galilean style (GS), which consists in ”making abstract models of the universe to which the physicists, at least, give a higher degree of reality than they accord the ordinary world of sensations.” GS values a theory’s explanatory reach as well as its empirical coverage. Indeed, at some times, explanatory power outweighs (and should outweigh) empirical coverage and empirical anomalies and (apparently) contradictory evidence is (and should be) set aside for the sake of further theoretical development.   

The relevance of this methodological attitude to linguistics is pretty evident. Linguistics is very data rich. Experiments (acceptability judgments) are cheap and easy. The problem is not finding another “fact” that resists explanation, but putting together a theory that has even a hint of non-trivial deductive explanatory structure. GS values this enterprise and reminds us that theories can be worth pursuing (and discarding) for reasons other than how many data points they (appear) to cover (or miss).

2. Weinberg remarks on what we want from our theories:

…there are explanations and explanations.  We should not be satisfied with a theory that explains the Standard Model in terms of something complicated an arbitrary…To qualify as an explanation, a fundamental theory has to be simple- not necessarily a few short equations, but equations that are based on a simple physical principle…And the theory has to be compelling- it has to give us the feeling that it could scarcely be different from what it is (ch 1).

These desiderata are reminiscent of those endorsed by the Minimalist Program, which, as Chomsky has repeatedly observed, are the in tune with the standard methodological tenets widespread in the successful sciences, viz. physics. Note Weinberg’s gloss on “simple” and “compelling.” Simple theories are based on natural principles, ones that make sense. This is also what makes them compelling. Within linguistics, the Minimalist Program endorses a similar mindset.  What makes for a simple and compelling theory of FL?  What are natural principles of UG?  These are very abstract very hard questions. But, as Weinberg observes for physical theories, they are the questions that any science with explanatory ambitions must confront.  Indeed, we might do worse than evaluate the success of a research program with how well it manages to operationalize these very important concerns.  For what it’s worth, I believe that one of the successes of the Minimalist Program is that it has tried (somewhat successfully in my view) to grapple with these issues. The central motif is that FL/UG is a computational system. We should explore its computational design features and look for natural ones.  Principles like Extension/No-tampering, Inclusiveness, Minimality, and Locality make computational sense. The program is to explore exactly how such notions operate and exactly what computational virtues they reflect.  Natural, simple principles, with great explanatory potential, that’s what we ought to be looking for!

3. Weinberg notes that what basic theory finds is likely to be irrelevant for many concerns:

The discovery of a final theory is not going to help us cure cancer or understand consciousness…We probably already know all the fundamental physics we need for these tasks (Ch 1).

There are many questions within linguistics where the results of minimalist theorizing will likely be irrelevant.  So far as I can tell, if you are interested in how kids acquire their grammatical competence in real time or how people parse sentences in real time or in how exotic language ‘E’ forms questions or relative clauses or how it expresses anaphoric dependencies then GB (and for many questions the Aspects Standard Model) is more than adequate to your task.  This is especially the case if, as I believe, a desirable boundary condition on adequate Minimalist theory (note: theory, not program) is that it preserve the generalizations embodied in these earlier descriptions of FL/UG. At any rate, much of what minimalism aims to accomplish will likely be of little relevance to these other investigations for roughly the same reasons that Weinberg notes above.

4. Weinberg describes the explananda of physics as follows:

…physicists are interested in the explanation of regularities, of physical principles, rather than individual events…Biologists, meteorologists, historians and so on are concerned with the causes of individual events…while a physicist only becomes interested in an event…when the event reveals a regularity of nature…(Ch2).

Another way of putting this is that physicists aren’t interested in what happens but in what could happen.  Linguists, at least those interested in competence, are very similar.  They are not interested in performances (what so and so said at this and this time) but the capacity that underlies these performances (which, to confess, I for one believe we will never be able to explain).  But more than this, linguists of the generative variety are interested in regularities, and high level ones at that. We often describe these as “effects,” island effects, principle C effects, Weak Cross Over effects etc.  These are highly stylized regularities and work in syntax aims to discover such effects and explain why they hold. Gathering linguistic data is worthwhile to the degree that these sorts of effects/generalizations are discovered.  Why? Because understanding the etiology of these effects is the key to unraveling the general properties of FL.  Linguists tend to underappreciate this point.  Effects (and anomalies, which are systematic counter-examples to effects) are what drive theory in the serious sciences.[1] They should do so in linguistics as well. Moreover, just as in physics, the aim should be to deduce these regularities from more and more general principles. As Weinberg puts it:

…we explain a physical principle when we show how it can be deduced from a more fundamental physical principle.

Weinberg has a very interesting discussion of what ‘fundamental,’ ‘deduced,’ and ‘principle,’ mean in the context of physics, which I urge you to look at.[2] In the context of contemporary linguistics, it is equally important to get a bead on the interpretation of these terms, especially given minimalist aspirations.  I have discussed my views of this elsewhere (here) (and I have some papers discussing this that are on the way out that I will link to when they come out) but for now, let’s just note that the ambitions are partially reductive, the aim being to deduce the laws of GB from more general principles.  Reduction does not imply elimination, rather the converse. And finding general accounts for the structure of FL need not imply that the principles of UG are domain general rather than domain specific (see here).  However, looking to deduce principles of UG from more general considerations, most likely of a computational sort, is what physics enviers should be aspiring to and Weinebrg’s examples illustrate the subtleties of this enterprise in a domain far more successful than our own.

There is a lot more in these Weinberg’s essays that I found thought provoking. Physics envy is a great motivator.  If you want to fuel your methodological mirror neurons, Weinberg’s popular writings are not a bad place to go.

[1] I will discuss effects and anomalies and their role in linguistics in a future post.
[2] Weinberg notes that “physicists try to explain just those things that are not dependent on accidents, but in the real world most of what we try to understand does depend on accidents.” Compare this to Norvig’s conception of scientific understanding (discussed here).

1 comment:

  1. "The central motif is that FL/UG is a computational system. We should explore its computational design features and look for natural ones. Principles like Extension/No-tampering, Inclusiveness, Minimality, and Locality make computational sense. The program is to explore exactly how such notions operate and exactly what computational virtues they reflect."

    There is already a very well developed science of computation -- it's called computer science (nothing to do with computers though!). Could you explain why Minimalism rejects CS?
    What is the relation between the NTC (for example) and any concept in CS, or between efficiency in your sense, and in the technical senses it is used in CS.

    And more generally why Minimalist theories are not mathematically formalised? (with the exception of Stabler).
    One of the defining characteristics of modern physics is its highly mathematical character -- viz Wigner's paper, the unreasonable effectiveness of mathematics.