Paper: Incremental Topic-Based Translation Model Adaptation for Conversational Spoken Language Translation

ACL ID P13-2122
Title Incremental Topic-Based Translation Model Adaptation for Conversational Spoken Language Translation
Venue Annual Meeting of the Association of Computational Linguistics
Session Short Paper
Year 2013
Authors

We describe a translation model adapta- tion approach for conversational spoken language translation (CSLT), which en- courages the use of contextually appropri- ate translation options from relevant train- ing conversations. Our approach employs a monolingual LDA topic model to de- rive a similarity measure between the test conversation and the set of training con- versations, which is used to bias trans- lation choices towards the current con- text. A significant novelty of our adap- tation technique is its incremental nature; we continuously update the topic distribu- tion on the evolving test conversation as new utterances become available. Thus, our approach is well-suited to the causal constraint of spoken conversations. On an English-to-Iraqi CSLT task, the pro- posed approach gives...