|
Implicit linguistic entities and the problem of coreference resolution| old_uid | 10326 |
|---|
| title | Implicit linguistic entities and the problem of coreference resolution |
|---|
| start_date | 2011/10/28 |
|---|
| schedule | 11h-13h |
|---|
| online | no |
|---|
| summary | In my talk I will present current research work in our laboratory which
extends from deep linguistic processing, its application to Q/A,
treebanking and parsing of oral corpora.
At first I will introduce the problem of the recovery of implicit
linguistic entities in the context of coreference resolution in open
domain texts, which can be regarded as the most important task to cope
with in either IR/IE or Q/A. I will present attempts at solving it and
experiments reported in the literature of NLP and in particular
computational linguistics. As most researchers in the field, I also assume
that in order to improve performance of any system, knowledge of the
world, commonsense knowledge and precise linguistic information are all
important prerequisites. I will concentrate then on some specific cases,
using freely available corpora which I will use to test our system: in
particular I will show text entailment and coreference resolution
examples. As will appear, semantic representation should be made available
to the modules taking care of coreference resolution. In turn, semantic
representation requires to be properly treated for pronominal anaphora in
order to be complete.
I will then introduce the parallel topic of implicit arguments as they are
treated within FrameNet framework and show a solution to that.
I will make a short presentation of the Q/A system and show the
effectiveness of a hybrid architecture. |
|---|
| responsibles | Candito |
|---|
| |
|