Discourse Constraint in Computer Manuals

This paper describes the result of our analysis of computer manuals to evaluate the strength of discourse constraint, due to which morphologically identical words tend to be used in the same way in the same context. According to our analysis, more than 90% of all the content words in a single chapter of a computer manual have the same word sense. For a manual concerning a narrow domain, such as a manual for experts, this constraint is stronger, but it becomes weaker with in a larger scope such as that of the whole manual, where 92.6% of the verbs have the same word sense, as opposed to 96.3% within a single chapter. This discourse constraint can be used to solve various problems in natural language processing, such as lexical and structural disambiguation and pronoun resolution. The results of this analysis offer a good prospect for the possibility of developing practical method of context processing that improves the accuracy of sentence analysis. However, generation of a good translation in a target language also involves the issue of free translation. In our analysis, free translation was detected in 35.5% of verbs in a manual professionally translated into Japanese.