The Roles of Semantic Role Labeling in Statistical Machine Translation

PhD Qualifying Examination


Title: "The Roles of Semantic Role Labeling in Statistical Machine Translation"

by

Miss Chi-Kiu Lo


Abstract:

In this survey, we review semantic role labeling (SRL) and statistical machine 
translation (SMT) literature with the recent rise of work in adapting SRL to 
improve SMT.

For years, heavy effort has been put in designing and evaluating SMT models. 
Translation quality achieved by SMT systems has proven to be improving 
continuously as measured by lexical and n-gram precision based 
metrics. However, today's SMT systems still often make many glaring errors 
caused by confusion of semantic roles. Semantic role confusions often result in 
serious misunderstandings of the essential meaning of the source input 
sentences.

SRL, the task of identifying basic event structures (i.e. who did what to whom, 
when, where, why and how) of the source input sentence to capture the essential 
meanings in it and preserve them in the translation output is one of the major 
challenges faced by SMT. With the increasing availability of semantically 
annotated corpora (e.g. Propbank, FrameNet), there was a proliferation of 
automatic SRL systems with promising performance in the last seven years. 
However, not until recently has there been a rise of work incorporating SRL 
in SMT to improve translation accuracy.

In this survey, we will present how typical SMT models fail to address 
translation accuracy, and show that recent studies have illuminated the 
possible research directions of using SRL for MT evaluation and semantic SMT. 
Finally, we will discuss the main challenges in using SRL for semantic SMT.


Date:                   Thursday, 12 August 2010

Time:                   2:30pm - 4:30pm

Venue:                  Room 4472
                         lifts 25/26

Committee Members:      Dr. Dekai Wu (Supervisor)
                         Dr. Raymond Wong (Chairperson)
                         Dr. Pascale Fung (ECE)
 			Prof. Bertram Shi (ECE)


**** ALL are Welcome ****