Abstract :
The architecture of a system able to interpret multiple knowledge representations throughout context, by capturing coherence of sequential foci arising in contextual sentences, disambiguating lexical meaning and specifying anaphor/cataphora/antecedent, is presented in this paper. The system can detect cohesion of structure of complicated contents, thanks to the construction of four novel sentence patterns1-any Japanese sentence being classifiable into one of theseiquestthat clarify role relations between words related with an absolute expression called CEE within the predicate. For instance, having a ldquobig heartrdquo, a tolerant mind, and the expression "pretty girl", are both usually understood as positives; whereas the expressions ldquobig words" and "pretty boy", or effeminate boy, have rather a pejorative connotation. To resolve such ambiguity, the system achieves an accurate contextual analysis, by identifying role relations between words. Our system evaluated correctly ldquopositiverdquo or ldquonegativerdquo nuance for 75.0 % of those.
Keywords :
classification; computational linguistics; knowledge representation; natural language processing; pattern classification; Japanese sentence pattern; anaphor-cataphora-antecedent specification; contextual analysis method; knowledge representation; lexical meaning; sequential foci coherence; Brightness; Heart; Internet; Knowledge representation; Natural languages; Pattern analysis; interpreting coherence of multiple semantic; overall role relation between words based on CEE; representations in contextual sentences;