Abstract
This paper proposes a new method of characterizing sentence meanings in terms of semantic features. In the first stage of this study, occurrences of the word osou were extracted from a corpus and manually coded, based on the methods used in FrameNet (Fillmore, Johnson, & Petruck, 2003). Fifteen senses were identified corresponding to comprehensible “situations.” In the second stage, sufficient features were specified to be able to effectively distinguish between these senses. In the third stage, two psychological experiments were conducted to investigate the validity of the 15 senses. In a feature-rating task, participants were asked to rate the applicability of properties, such as “The victim is animate,” to the situations described by 45 sentences of the form “Y was attacked by X.” In a card-sorting task, a different group of participants was asked to freely sort the same materials. Multivariate analysis showed that the feature-rating task predicted both the corpus-based analysis and the card sorting by normal Japanese participants, indicating that the proposed method is sufficiently powerful to investigate the conceptual structure of sentence meaning.
