evan_tech

Previous Entry Share Next Entry
12:42 pm, 3 Dec 03

linguistic side effects

This is the sort of thing I wish I was writing about:

Linguistic Side Effects.
Abstract. Making a linguistic theory is like specifying a programming language: one typically devises a type system to delineate the acceptable utterances and a denotational semantics to explain their operational behavior. Via this connection, programming and natural language research can inform each other; in particular, computational side effects are intimately related to referential opacity in natural languages. To illustrate this link, I use continuations and composable contexts, concepts from the study of side effects in programming languages, to analyze quantification, a natural language phenomenon.


The meat of the paper is a detailed description of how quantification ("every [x]") could work using a sort of exceptions for raising (instead of positing something like logical form--though it's interesting to see how similar the result is). The analysis includes some nondeterminism to handle the ambiguity of multiply-quantified sentences like "Everyone loves someone", and even claims:
[...] For example, we predict correctly that the following sentence, with three quantifiers, is five-way (not six-way) ambiguous.
Every representative of some company saw most samples.
Wow. (But why is that? Something to do with the monotonicity of the quantifiers? Help me out, trochee. :P)

It's written from a CS perspective, so you don't need to be knowledgeable about linguistics to follow, but it's probably more interesting to linguists who appreciate how hard some of these linguistic problems are. (Something like pronoun reference [binding theory] seems kinda obvious until you look at it really hard. Natural language is really hard to appreciate in that respect: because it comes so naturally to us, it appears really intuitive.) Though I understood most of it, the CS aspect is pretty intense. And Section 6.1 especially: I have a vague understanding of functors but still need to learn what type-lifting really means.

I now need to go find all of this guy's other stuff. He briefly mentions another paper that discusses natural language evaluation order...