<p>Anaphora resolution, a key task in natural language processing (NLP), involves identifying relationships between pronouns and their referents in text. In morphologically rich languages like Arabic, resolving third-person pronouns attached as suffixes Haa (e.g., "ـه", "ه") presents a unique challenge. This paper proposes a novel rule-based approach to distinguish whether the suffix "Haa" functions as a referential pronoun or an integral word part. The method extracts linguistic features and applies a series of heuristic rules, achieving a linear time complexity of O(n), where n is the number of tokens in the dataset. We used a composite dataset compiled from five Arabic news sources: Saba News Agency, Echorouk Online, AlmasryAlyoum, Almustaqbal, and Alittihad. In total, the datasets consist of 1232 sentences and 32,947 words. Annotation was done using the CAMeL Tools morphological analysis, which offers detailed linguistic features specifically designed for Arabic. To evaluate the performance of our model, we applied standard evaluation metrics: precision, recall, accuracy, and F1-score. Evaluation of the AlmasryAlyoum dataset, which consists of 143 sentences and 5025 words, yielded a precision of 96%, a recall of 94.07%, and an accuracy of 92%, demonstrating the approach’s effectiveness for Arabic anaphora resolution tasks in applications such as question answering and machine translation.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

A Rule-Based Method for Identifying Third-Person Arabic Attached Pronouns in Texts

  • Alaa Sayed Omar,
  • Soheir M. Khamis,
  • Azza A. Taha

摘要

Anaphora resolution, a key task in natural language processing (NLP), involves identifying relationships between pronouns and their referents in text. In morphologically rich languages like Arabic, resolving third-person pronouns attached as suffixes Haa (e.g., "ـه", "ه") presents a unique challenge. This paper proposes a novel rule-based approach to distinguish whether the suffix "Haa" functions as a referential pronoun or an integral word part. The method extracts linguistic features and applies a series of heuristic rules, achieving a linear time complexity of O(n), where n is the number of tokens in the dataset. We used a composite dataset compiled from five Arabic news sources: Saba News Agency, Echorouk Online, AlmasryAlyoum, Almustaqbal, and Alittihad. In total, the datasets consist of 1232 sentences and 32,947 words. Annotation was done using the CAMeL Tools morphological analysis, which offers detailed linguistic features specifically designed for Arabic. To evaluate the performance of our model, we applied standard evaluation metrics: precision, recall, accuracy, and F1-score. Evaluation of the AlmasryAlyoum dataset, which consists of 143 sentences and 5025 words, yielded a precision of 96%, a recall of 94.07%, and an accuracy of 92%, demonstrating the approach’s effectiveness for Arabic anaphora resolution tasks in applications such as question answering and machine translation.