Integrating bilingual lexicons in a probabilistic translation assistant

Integrating Bilingual Lexicons in a Probabilistic Translation Assistant Philippe Langlais, George Foster, Guy Lapalme RALI / DIRO Universit´e de Montr´eal C.P. 6128, succursale Centre-ville Montr´eal (Qu´ebec) Canada, H3C 3J7 www-rali.iro.umontreal.ca Abstract In this paper, we present a way to integrate bilingual lexicons into an operational probabilistic translation assistant (TransType). These lexicons could be any resource available to the translator (e.g. terminological lexicons) or any resource statistically derived from training material. We describe a bilingual lexicon acquisition process that we developped and we evaluate from a theoretical point of view its benefits to a translation completion task. Keywords Translator assistant, probabilistic translation, bilingual lexicon acquisition, bilingual phrasal alignment Introduction Translation needs are growing faster than machine transla- tion (MT) technology improves. Therefore, there are more and more situations where MT is just not an acceptable so- lution, especially when high quality translation is required. This statement is encouraging for (computational) linguists since people realize that despite good translation programs available on the web, more effort must still be invested in MT research. In the meantime, we believe that turning to alternatives to fully automatic translation is a challenging but promising approach. Among the tools that may make the translator’s difficult task a little easier, Foster et al.(1997) have developed TRANSTYPE, a system in which a translation emerges from a series of alternating contributions by the human and the machine. The machine’s contributions are basically pro- posals for parts of the target text, while the translator’s can take many forms, including pieces of target text, corrections to a previous machine contribution, hints about the nature of the desired translation, etc. In all cases, the translator remains fully in control of the process: the machine must work within the constraints implicit in the user’s contribu- tions, and he or she is free to accept, modify, or completely ignore its proposals. TRANSTYPE takes the form of a specialized text editor (see Figure 1). Embedded within this editor is a non-intrusive machine translation engine which can provide, at any point of the translation, a ranked list of units (words or sequences of words) that the translator is likely to type. The editor allows for the easy insertion of anyone of these units at a keystroke. These completions are computed according to a translation model and a language model that both take into account the translation already typed. Within such a scenario, we have investigated the possibility of integrating bilingual lexicons and report on our results in this paper. These lexicons could be any resource available to the trans- lator (e.g. terminological lexicons) or any resource statisti- cally derived from training material. In the next section, we give a brief overview of the TRANSTYPE system and its evaluation by translators. In section three, we describe the strategy we devised to au- tomatically extract bilingual lexicons from training ma- terial. In the fourth section, we describe the way we integrated these automatically acquired lexicons within TRANSTYPE’s completion mechanism, followed by a per- formance evaluation of this integration. …

Content truncated to 500 words