PhD Thesis Proposal Defence


"Word sense disambiguation vs. statistical machine translation"

By Miss Marine Carpuat


Abstract:

We propose to empirically demonstrate that dedicated word sense
disambiguation (WSD) systems are useful to statistical machine
translation (SMT), and directly investigate the related issues raised
by the WSD vs. SMT debate.

WSD, the task of resolving sense ambiguity to identify the right
translation of a word is one of the major challenges faced by language
translation systems. If the English word "drug" translates into French
as either "drogue" (used as a narcotic) or "medicament" (used as a
medicine), then an English-French machine translation system needs to
disambiguate every use of "drug" in order to make the correct
translations.

Heavy effort has been put in designing and evaluating dedicated WSD
models, in particular with the Senseval series of workshops. This is
partly motivated by the often unstated assumption that any full
translation system, to achieve full performance, will sooner or later
have to incorporate individual WSD components. However, in most
machine translation architectures, in particular SMT, the WSD problem
is typically not explicitly addressed. This paradoxical situation
encouraged speculation that SMT models are already very good at WSD
and that current WSD systems have nothing to offer to state-of-the-art
SMT.

We propose to directly address these issues by conducting an empirical
investigation of the WSD vs. SMT debate. A critical survey of both WSD
and SMT literature shows that current SMT systems already benefit from
some WSD insights. But it is still unclear whether the new
state-of-the-art WSD models can actually help improve translation
quality.

We will first introduce the HKUST WSD system, which achieves the best
known performance on the Senseval-3 Chinese lexical sample task, among
other desirable properties for our study. Then we will present
empirical results suggesting that while typical SMT models cannot
disambiguate word translations as well as dedicated WSD systems,
simple methods for incorporating WSD predictions do not help
translation quality. Based on error analysis, we will suggest new
directions to incorporate WSD predictions in SMT.


Date:     		Monday, 21 November 2005

Time:                   1:00p.m.-3:00p.m.

Venue:                  Room 4480
			lifts 25-26

Committee Members:      Dr. Dekai Wu (Supervisor)
                        Dr. Dit-Yan Yeung (Chairperson)
			Dr. Brian Mak
			Dr. Pascale Fung (ELEC)


**** ALL are Welcome ****