Japanese linguistics conference

Posted August 4, 2005

Last week, Harvard hosted a workshop on Japanese linguistics as part of the summer events of the Linguistic Society of America. Although this looked to be a little bit out of my league, being neither a professional linguist nor a speaker of Japanese, I couldn't pass up the opportunity to attend.

The pleasant surprise was that my lack of Japanese skill wasn't much of a hindrance: all examples were generally provided in romaji with full translations and grammatical markup. The unpleasant surprise was that my lack of linguistic background was a bit of a hindrance: the talks were peppered with jargon that was rarely defined, so if you didn't already know what "DP ellipsis" or "pied piping" was, you were out of luck.

Still, it was interesting. A great deal of effort is directed towards the problem of explaining linguistic "data", namely what sentences are seen by native speakers as "fine", "a little odd but okay", or "wrong". The subtleties are fascinating. For instance, English can have indirect objects in several ways: "I sent a package to John," versus "I sent John a package." But if you keep the same form while changing the semantic content a little, you end up with "I sent a package to London," versus "I sent London a package." While the first is fine, the second is a bit odd... the only way to make it okay is to interpret "London" as something like "the London office". Human language is full of these sorts of subtle rules that few of us ever consciously consider, but linguists have to.

That being said, as a naive outside observer, I'm not a big fan of the flavor of the theories that linguists are using the deal with these issues. There is lot of work expended on debating how to construct grammatical tree structures to properly break down a sentence into component parts while preserving the inter-relations between them. But these models seemed to me to be stuck on a path towards epicycles: more and more complexity and special cases, without really capturing the essence of how human brains process the language. I hate to criticize a field without really studying it, but I would be a lot more content if linguists would close the loop and start predicting things with their models: program computers with the grammatical structures they are inventing and see what kind of sentences they end up creating, for instance, as opposed to being stuck in a purely explanatory mode.