PhD Qualifying Examination


"Survey: Statistical Machine Translation and Example-Based
Machine Translation"

By Mr. Yihai Shen


Abstract:

We present, contrast and critique two major paradigms within corpus-based
machine translation (MT), namely statistical machine translation (SMT) and
example-based machine translation (EBMT). We first present the major
techniques utilized in translation models in SMT in the order of
word-based translation model, phrase-based model and syntax-based model.
We find that translation models in SMT have evolved in a way to
incorporate more and more linguistic information, especially structural
aspect of the language, which is very important, especially for language
pairs that differ much in structure, for speed and accuracy
considerations. We then present the EBMT paradigm, its three major steps
being matching, alignment and recombination. We contrast the major
techniques used in these three steps. We observe that EBMT works well for
cases that syntactic information (rules) serves badly. Therefore we think
it promising to combine the strength of syntax-based SMT systems and EBMT
under one framework in which EBMT does the work that is not suitable for
syntax-based SMT. Possible approaches include applying stochastic methods
to EBMT systems and using multi-engine systems.


Date:     		Tuesday, 10 February 2004

Time:                   9:00a.m.-11:00a.m.

Venue:                  Room 3401
			lifts 17-18

Committee Members:      Prof. Dekai Wu (Supervisor)
                        Prof. Brian Mak (Chairperson)
			Prof. Hongjun Lu
			Prof. Pascale Fung (ELEC)


**** ALL are Welcome ****