Abstract :
The disambiguation of contextual multiple meanings has long been difficult, and it remains one of the leading issue to study for exploiting better the potentiality of information sources. In order to resolve this issue, a system is presented in this paper. The latter is able: to classify any Japanese sentence into one of the following four kinds of sentence patterns, which are, "verbal clause / adjective clause / nominal clause / clause containing emoticons (including a sigh)"; to achieve contextual analysis by identifying the coherence of sequential concepts and phenomena, and their connections, related to absolute expressions, that is to say expressions having an unchangeable meaning or an unchanging polarity, within the predicate. For the determination of the evaluation polarity, in a corpus composed of 2500 customer reviews, the percentage or correct answer, for all natures of CEEs except emoticons, was 75.0%, and the percentage for emoticons was 66.7%.