Adaptive Language and Translation Models for Interactive Machine Translation

Adaptive Language and Translation Models for Interactive Machine Translation Laurent Nepveu, Guy Lapalme Philippe Langlais RALI/DIRO - Universit´e de Montr´eal, C.P. 6128, succursale Centre-ville Montr´eal, Qu´ebec, Canada H3C 3J7 {nepveul,lapalme,felipe} @iro.umontreal.ca George Foster Language Technologies Research Centre National Research Council Canada A-1330, 101 rue Saint-Jean Bosco, Gatineau, Qu´ebec, Canada K1A 0R6 George.Foster@nrc-cnrc.gc.ca Abstract We describe experiments carried out with adaptive language and translation models in the context of an interactive computer-assisted translation program. We developed cache-based language models which were then extended to the bilingual case for a cache- based translation model. We present the improve- ments we obtained in two contexts: in a theoretical setting, we achieved a drop in perplexity for the new models and, in a more practical situation simulat- ing a user working with the system, we showed that fewer keystrokes would be needed to enter a trans- lation. 1 Introduction Cache-based language models were introduced by Kuhn and de Mori (1990) for the dynamic adap- tation of speech language models. These models, inspired by the memory caches on modern com- puter architectures, are motivated by the principle of locality which states that a program tends to re- peatedly use memory cells that are physically close. Similarly, when speaking or writing, humans tend to use the same words and phrase constructs from paragraph to paragraph and from sentence to sen- tence. This leads us to believe that, when processing a document, the part of a document that is already processed (e.g. for speech recognition, translation or text prediction) gives us very useful information for future processing in the same document or in other related documents. A cache-based language model is a language model to which is added a smaller model trained only on the history of the document being pro- cessed. The history is usually the last N words or sentences seen in the document. Kuhn and de Mori (1990) obtained a drop in per- plexity of nearly 68% when adding an unigram POS (part-of-speech) cache on a 3g-gram model. Martin and al. (1997) obtained a drop of nearly 21% when adding a bigram cache to a trigram model. Clarkson and Robertson (1997) also obtained similar results with an exponentially decaying unigram cache. The major problem with these theoretical results is that they assume the correctness of the material entering the cache. In practice, this assumption does not always hold, and so a cache can sometimes do more harm than good. 1.1 Interactive translation context Over the last few years, an interactive machine translation (IMT) system (Foster et al., 2002) has been developed which, as the translator is typing, suggests word and phrase completions that the user can accept or ignore. The system uses a transla- tion engine to propose the words or phrases which it judges the most probable to be immediately typed. This engine includes a translation model (TM) and a language model (LM) used jointly to produce pro- posals that are appropriate translations of source words and plausible completions of the current text …

Content truncated to 500 words