Construction of Adverb Dictionary that Relates to Speaker Attitudes and Evaluation of Its Effectiveness

Construction of Adverb Dictionary that Relates to Speaker Attitudes and Evaluation of Its Effectiveness Toshiyuki Kanamaru1, Masaki Murata1, and Hitoshi Isahara1 National Institute of Information and Communications Technology (NICT) 3-5 Hikaridai, Seikacho, Sorakugun Kyoto, 619-0289, Japan {kanamaru, murata, isahara}@nict.go.jp Abstract. Adverbs can express a speaker’s attitude in a given situation on a specific matter. We constructed an adverb dictionary in which the attitude of the speaker is described. We also looked into whether or not adverbs could effectively be used as basic data to analyze reputations. We conducted three kinds of experiments to verify how effective and precise our dictionary was. First, we calculated the coverage ratio of the dictionary by comparing the ratios of appearances of all adverbs to the ratios of appearances of our dictionary items. Next, we attached a tag to 988 adverbs, and found that the coverage ratio of the tagged adverbs was 97.76% in the open data. Finally, we classified whether sentences were positively or negatively represented using the adverb dictionary and calculated that the accuracy of the classification was 86.5%. 1 Introduction Japanese adverbs are mainly classified into two kinds. The first is adverbs that indicate the state and mode of motion. The second is adverbs that indicate the psychological attitude or feeling of the speaker. [1, 2]. Reputation analysis1, which excludes adverbs from the objects being pro- cessed, has been used in previous research to handle sensitive information such as the extraction of evaluative expressions. Therefore, little research has been done to investigate the importance of adverbs in language processing. Adverbs are mostly excluded during information retrieval as stop words, because they are not part of the propositional content of sentences. Adverbs are not used in language processing such as reputation analysis or opinion extraction, because it involves the same process as information retrieval. In addition, when doing such processing, either adverb forms are not defined, or their relative frequency is not sufficiently high. Few adverbs have been extracted as important words to determine reputations in reputation analysis. We need to define adverbs manually in advance in a dictionary for reputation analysis 1 “Reputation” for our purposes refers to whether speakers appear to evaluate a given subject as positive, negative, or neutral. 295 2 since it is different from other evaluations of expressions in that it does not use automatic acquisition or automatic classification. We therefore constructed an adverb dictionary that describes speakers’ atti- tudes. We found that the dictionary provides effective data for analyzing repu- tations. We verified how effective the dictionary we constructed was. First, we ex- amined its coverage ratio using a Web text. Next, we verified the accuracy of the classifications using randomly selected dictionary data from the Web text. Finally, we assessed how effective the adverb dictionary was. 2 Construction of the adverb dictionary As was mentioned above, we need to define adverbs manually in advance in a dictionary as reputation analysis is different from other evaluations of expres- sions in that it does not use automatic acquisition …

Content truncated to 500 words