automatic content extraction

Mark Liberman had a post yesterday over on Language Log about Automatic Content Extraction. I spent the past two summers doing ACE annotation, so I know all about the metonymy issues he discusses.

What's worthy to note, however, is that, as an annotation group, we (I include myself, since I still occasionally go into the LDC and work a few hours) have been moving away from the explicit tagging of metonymy, largely due to the argument of Christopher Walker that it doesn't occur nearly as often as previously guidelines have suggested. Instead, we've been moving in the direction of tagging solely the "intended" referent, rather than the "literal" one, on the grounds that it's the former that actually matters.