Saturday, September 15, 2012

Where's the meeting that was cancelled?

When I was still at PARC -- 2008 seems an awful long time ago now -- I invited Patricia Amaral, then a recently graduated research student to come and visit us for a short while. She was applying to be a research assistant in Linguistics, Stanford and being in Palo Alto a little earlier was a good idea. It was lucky that the Gulbekian Foundation agreed too. We wanted to work on improving the system Bridge (about which I will write informally some other day, but an official description can be found in PARC's Bridge and Question Answering System and a personal preliminary take can be found here).

Being a latecomer to computational linguistics and perhaps a bit too optimistic about research, I always have dozens of ideas of things I want to get done, but little ability to predict the time that it takes to get those things done. In this case, however, everything worked well. Patricia got the hang of the system very quickly and soon enough made gigantic progress on the issue I wanted to deal with.

Despite our claims in Preventing Existence (and despite the beautiful algorithm that Rowan, Cleo and Lauri described in Computing Relative Polarity for Textual Inference) the Bridge system could, at that stage, only deal with logical contexts introduced by verbs with a full complement phrase. So something like Ed knew that Mary arrived  would be fine, a logical context would be created with the right polarity, but Ed knew of Mary's arrival  would not create a logical context. (The Bridge system works with full deep LFG parses of sentences and it easier to mark the complemented verbs.)

Thus my goal for the work with Patricia was the extension of the marking of verbs that introduce context, from the ones with full complement to ones that were simply directly transitive.  As usual the problem was much harder than I expected and Patricia, at the end of her stay, gave us a huge presentation with millions of issues and a possible classification that would be a first stab at solving the original problem.

This was written up in a paper  (by Patricia, myself, Cleo Condoravdi and Annie Zaenen) submitted to the workshop on contexts associated with ECAI2008. The paper accepted, but none of us could attend. So we withdraw the paper, hoping to improve it.

Very recently I took the initiative of re-submitting the paper to ONTOBRAS2012 in Recife, Brazil, as I had hoped to go visit some  friends there. But this was not possible, so the paper, again accepted,  had to be withdrawn. Again.

The newly named Where's the meeting that was cancelled? paper seems a bit unlucky, but the work is very cool. Patricia came up with this idea of treating verbs that change  existential commitments as pre and post conditions reminiscent of Hoare's logic. This connection is hinted at in the paper, but not pursued, as the need for classification large-scale was definitely more important then. Someone should work on this, though.

(Meanwhile together with Lottie Price and Tracy King, I worked a little on nouns that introduce logical contexts, cf. Contexts Inducing Nouns.)

No comments:

Post a Comment