Target-Side Context for Discriminative Models in Statistical MT

Speaker:
Aleš Tamchyna
Abstract:
Discriminative translation models utilizing source context have been shown to help statistical machine translation performance. In this talk, I will describe a novel extension of this work using target context information. I will show how this model can be efficiently integrated directly in the decoding process of a phrase-based translation system. I will present results which validate that our approach scales to large training data sizes and results in consistent improvements in translation quality on four language pairs. I will also provide an analysis comparing the strengths of the baseline source-context model with our extended source-context and target-context model.
Length:
01:39:39
Date:
07/11/2016
views: 1277

Images:
Attachments: (video, slides, etc.)
137 MB
843 downloads
1.2 GB
1278 downloads
168 MB
849 downloads
242 MB
823 downloads
507 MB
772 downloads