October 30, 2003

Metonymy notes from all over

Could philosophers of language beat their analytical scalpels into data-mining shovels? Are language engineers discovering analytical distinctions that philosophers have missed?

Well, in any case, it's fall and the smell of metonymy is in the air.

Geoff Nunberg, who's been working in the metonymy trade since 1978, has a new piece entitled "Indexical Descriptions and Descriptive Indexicals". It's due to appear in a forthcoming OUP collection which is blurbed as "brand-new essays on important topics at the intersection of philosophy and linguistics."

Meanwhile, committees of engineers involved in the DARPA ACE (Automatic Content Extraction) project have extended their annotation task guidelines to cover Arabic and Chinese. The general approach is summarized by the slogan "Facts = entities + relations". A key issue in the ACE "entity" arena is the problem of metonymy, which the ACE guidelines divide into two types: "composite metonymy" or "role assignment", which is viewed as focusing attention on one of the intrinsic aspects or attributes of a "geo-political entity" (such as its territory or its government or its population), and "classic metonymy", which is viewed as using a GPE's name to refer to an entirely different thing (such as a sports team).

I have no doubt that Geoff and other philosophico-linguists could help the engineers sort this stuff out -- these seem like deep ontological waters -- but at the same time, it looks like the information-extraction community (it's not quite yet an industry) has some issues and ideas to offer that don't seem to be treated in the philosophical literature. For instance, is there really a qualitative distinction between "composite metonymy" and "classic metonymy"? What's the role of convention in this area -- do metonymic norms differ across languages or across genres, and does this matter?

Geoff has worked on automatic genre classification but not (I think) on automatic information extraction, and (although I've know him since he was thinking about metonymy while driving a cab as a graduate student at CCNY), I don't know if he thinks that his work on metonymic aspects of reference has practical applications in data mining, or what he things of the distinctions that the ACE crowd is making these days. Geoff?

Posted by Mark Liberman at October 30, 2003 09:35 AM