Departmental Bulletin Paper テキストからの対象物認識に有用な記述内容 : 動物を例に
Encyclopaedic Descriptions That are Useful for Identifying Entities: A Case Study of Descriptions of Animals

加藤, 祥  ,  Sachi, KATO

(9)  , pp.23 - 50 , 2015-07 , 国立国語研究所
ISSN:2186-134x print/2186-1358 online
This paper reports what features of encyclopaedic descriptions are useful for recognising entities based on the results of three experiments on target object identification from texts. We used gloss descriptions of animals compiled from dictionaries (Experiment 1), texts with additional information not found in the dictionaries (Experiment 2), and usages acquired from corpora (Balanced Corpus of Contemporary Written Japanese, Google Japanese N-grams; Experiment 3). In all three experiments, the percentage of the entities which could be identified correctly from texts was only about half. Therefore, we conclude that it is difficult to recognise an entity based on its description in a text. The results of the three experiments suggest that the following information is important for the identification of the target: specific features selected based on participants' personal experiences and information that would distinguish the target animals from other members in the ad hoc categories were important for identification of the target. When readers have knowledge about the target entities, individual or general experience and knowledge are easily obtained from corpora, as the actual texts are useful for readers' recognition. In contrast, when readers have insufficient knowledge about the target entities, a description of the features that distinguish the target from the prototype of the superordinate category (ad hoc category) is useful.

Number of accesses :  

Other information