Mendeley出力
＜紀要論文＞
言語情報の確実性アノテーションのための様相表現の分類
| 作成者 | |
|---|---|
| 本文言語 | |
| 出版者 | |
| 発行日 | |
| 収録物名 | |
| 巻 | |
| 開始ページ | |
| 終了ページ | |
| 出版タイプ | |
| アクセス権 | |
| JaLC DOI | |
| 概要 | Natural language texts contain pieces of information with several levels of certainty, namely, factual assertions, speculations, inferences and hypothetical thoughts. Recognition of certainty levels i...n textual information is crucial for efficient extraction of newly reported facts and rapid judgment of the reliability of the information. Some groups in natural language processing (e.g., Light et al. (2004), Medlock and Briscoe (2007), Szarvas et al. (2008), Kilicoglu and Bergler (2008), Eguchi et al. (2009)) have started to develop technologies for automatic certainty recognition. We are now constructing a corpus in which modal, negative, and conditional expressions and their scopes are semantically annotated, as a basis for dealing with certainty of information in Japanese texts. In order to achieve high-quality annotation, we have designed an annotation schema on the basis of the classification of key expressions according to the certainty levels of propositions they introduce. We classify modal expressions into 6 categories including factive, evidential and epistemic expressions, and sub-classify epistemic expressions into 8 categories based on the following observations: 1) co-occurrence with "100%" expressions (e.g., 'kanarazu', 'zettai'), 2) occurrence in a conjoined sentence where the same proposition is both affirmed and negated, and 3) co-occurrence with "0%" expressions (e.g., 'kanoosei-wa nai'). These empirical observations are incorporated into the annotation schema and can be used by human annotators to ensure consistent annotation.続きを見る |
| 目次 | 1. 目的 2. 確実性アノテーションプロジェクトの概要 3. 様相表現の分類 4. 結語 |
詳細
| レコードID | |
|---|---|
| 査読有無 | |
| 主題 | |
| ISSN | |
| NCID | |
| 登録日 | 2011.05.20 |
| 更新日 | 2023.11.17 |