Comments on: Predicate-Argument EXtractor (PAX)

By: celsocosta

celsocosta — Fri, 21 May 2010 09:45:36 +0000

Certainly, triples wont be the solution, but some alternative or combination of alternatives, will rise, sooner or later.

We will go from impossible, pass on useless implementations, step on serviceable and finally get something suited for the purpose of understanding human language.

Its all a question of time and money.

By: Oleg Kreymer

Oleg Kreymer — Thu, 20 May 2010 15:25:08 +0000

“Anaphora resolution” indeed.
I often get an impression that we are under illusion that true NLP is possible and even mandatory for Semantic Web to be realized.
Nobody speaks in “triples”.
Are we ready to accept the fact that AI will never be able to adequately operate with human language?

By: celsocosta

celsocosta — Thu, 20 May 2010 10:05:49 +0000

Without some special care the result wont be satisfactory.

Even by using the algorithm inspired in “Triplet extraction from sentences”, it only produces good result in simple, stereotypical sentences. Still, by looking at sentence parse tree, the results wont be that bad, going to run it later.

But to take care of sentences as complex as your Oleg, avoiding information loss, it needs algorithms to split coordination conjunctions, extract dependency clauses, propositions and adjective phrases, besides the nightmare of co-referencing and anaphora resolution.

By: fmalheiro

fmalheiro — Thu, 20 May 2010 09:34:23 +0000

Hi there,
I parsed it using PAX+Stanford Parser.
This was the result:
{obj=, sub=everything, verb=start}
{obj=problem, sub=i, verb=have}

By: Oleg Kreymer

Oleg Kreymer — Thu, 20 May 2010 02:02:21 +0000

Just curious, did you have a chance to parse the following sentence with PAX?
“By the way In the beginning I was having some configuration problems, but after some emails with project creators everything started working fine.”