Syntax-based statistical machine translation, and evaluation of machine translation systems

“Syntax-based statistical machine translation, and evaluation of machine translation systems” by W. Byrne, Cognition Institute Summer School: Bilingual Minds, Bilingual Machines. June 2013. Three lecture short course.

Abstract

Syntax-base statistical machine translation

The lecture will introduce statistical machine translation. Following an overview of the problem, the following topics will be presented: the use of parallel text for estimation of alignment models and extraction of phrase-based and synchronous translation grammars; the use monolingual text for the estimation and construction of target language models; and the construction of SMT `decoders' to perform actual translation. The presentation will be at a fairly high level so that students not familiar with statistical natural language processing should still find this an accessible introduction.

Evaluation of machine translation systems

Current approaches to SMT system development rely heavily on automatic measures of translation quality. Although existing measures are widely considered to be less than perfect, they serve as proxies for human judgements of translation quality, which are time consuming and expensive to gather. This lecture will continue material from the previous day, and will describe automatic metrics and their use in translation system development. Current research into improving these schemes will be also be described.

BibTeX entry:

@misc{byrneplymouth13,
   author = {W. Byrne},
   title = {Syntax-based statistical machine translation, and evaluation
	of machine translation systems},
   publisher = {Cognition Institute Summer School: Bilingual Minds,
	Bilingual Machines},
   month = jun,
   year = {2013},
   note = {Three lecture short course.}
}

Back to Bill Byrne publications.