Annotating Attribution Relations in Arabic

Annotating Attribution Relations in Arabic Amal Alsaif, Tasniem Alyahya, Madawi Alotaibi, Huda almuzaini, Abeer Algahtani Al-Imam Mohammad Ibn Saud Islamic University College of Computer Sciences and Information {asmalsaif, tnalyahya, msalotaibi, hamozeani}@imamu.edu.sa;aakalqahtani@sm.imamu.edu.sa} Abstract We present a first empirical effort in annotating attribution in Modern Standard Arabic (MSA). Identifying attributed arguments to the source is applied successfully in diverse systems such as authorship identification, information retrieval, and opinion mining. Current studies focus on using lexical terms in long texts to verify, for example, the author identity. While attribution identification in short texts is still unexplored completely due to the lack of resources such as annotated corpora and tools especially in Arabic on one hand, and the limited coverage of different attribution usages in Arabic literature, on other hand. The paper presents our guidelines for annotating attribution elements: cue, source, and the content with required syntactical and semantic features in Arabic news (Arabic TreeBank - ATB) insight of earlier studies for other languages with all required adaptation. We also develop a new annotation tool for attribution in Arabic to ensure that all instances of attribution are reliably annotated. The results of a pilot annotation are discussed in addition to the inter-annotators agreement studies towards creating the first gold standard attribution corpus for Arabic. Keywords: attribution, annotation tool, NLP, Arabic discourse, annotation guidelines, ATB, inter-annotator agreement. 1. Introduction Textual information is one of the huge significant data available in the World Wide Web (WWW), which rapidly spread globally. This information is versatile and reflects different opinions as well as behaviours. For example, most of news is reporting people’s speech and opinions about particular events, with additional explanation and analysis by the writer to tend people into desired understanding and background. Referring attributed arguments to the source and distinguishing different opinions than the author/writer opinion are not straightforward processes (Pareti 2012). Attribution could be direct speech (quotations) with no influence by the author on the content, or indirect speech when the abstract object is presented in a different way than the exact speech or it is not clear who state the argument. Usually quotation tools are used in direct speech such as punctuations (: and “ ), speech acts such say/qAl/لاق or comment/Akbar/ربخأ and some particles such as that/An/نأ, as in Example 1. On other hand, indirect speech may miss one or more of these tools which leads to ambiguity on determining the exact source and content boundaries, Examples 2 and 3. This makes identifying quotations and opinions of other people require advanced analysis and tools in addition to the basic lexical-syntactic analysis in the literature (Pareti 2012). (1) لاق وينيروم دتيانويرتسشنام بردم ]ةارابملا صوصخب:[" ةصح تناك دج ةديج ةيبيردتا". Manchester United coach Mourinho said [about the match]: "it was a very good training share" (2) قلع لوؤسم ردصم موكحلا يفة }همسا ركذ ضفر{ ] حيرصت ىلع ا راعسا عفر نمضتملا سيئرلالورتبل[ تعزعز هنأش نم رارقلا اذه نأ .رسلال ماعلا لخدلا An official source in the government{refused to mention his name} commented [on the president's …

Content truncated to 500 words