[Ltg] LTG Seminar [Maxim Khalilov, Monday, 27.09.2007, E6A357, 11am]

[Rolf Schwitter]

 ----
 LTG Seminar
  - see: http://www.clt.mq.edu.au/Events/Seminars.html

  Monday, 24th September, 2007, 11am
  Macquarie University, E6A, Room 357
 ----

Title: N-gram based statistical machine translation
Speaker: Maxim Khalilov
Affiliation: UPC - Polytechnic University of Catalonia (Barcelona, Spain)

The talk will be dedicated to the UPC ngram-based Statistical Machine Translation (SMT) system. Generally, SMT deals with statistical models generated based on the analysis of bilingual parallel text (the collection of text and their reference translations). Like the major of the modern SMT systems, the UPC system is based on the so-called noisy channel approach, supplemented with log-linear model, which allows including different features combined in a log-linear way. The main difference between ngram and state-of-the-art phrase-based systems lies in the distinct representation of the bilingual units. Unlike phrase-based systems the UPC system operates with the only possible text segmentation. The optimization procedure will be briefly discussed, despite the BLEU metric is being criticized it is still "de-facto" standard measure of translation quality evaluation. We try to improve both translation accuracy and fluency using other metrics as optimization criteria.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.ics.mq.edu.au/pipermail/ltg/attachments/20070920/575dc873/attachment.html