The selection of a parsing strategy for an on-line machine translation system in a sublanguage domain. A new practical comparison

The paper reports the results o f a practical comparison of different parsing strategies. The research was carried out in the context of a larger project for the development o f a machine translation (MT) system for translating avalanche forecast bulletins from German to French. The design of the MT system requires controlled input and no post-editing o f the translated texts. The parsing experiment had as a goal to select the most suitable parsing strategy for a parser that allows the composition of the sentences in on-line fashion with mouse and windowing2. In order to guarantee correct translation, the input system accepts only words and sentences that are known by their grammar and dictionary and it refuses wrong input. To minimize input errors, the user can select the possible next words with the mouse from different windows, which display the choices at a particular point in the sentence. The sentences are parsed word by word from left to right so that wrong input is detected immediately. After each word, the input device has to predict, with the help of the parser, all the words that can possibly continue the sentence that is being made. For our type of on-line parser, time is critical. The interface window has to be refreshed immediately after each word chosen by the user.