Towards an Automatic Text Comprehension for the Arabic Question-Answering: Semantic and Logical Representation of Texts

Towards an Automatic Text Comprehension for the Arabic Question- Answering: Semantic and Logical Representation of Texts Abstract Automatic text comprehension is an arduous task of automatic natural language processing. This task has been addressed in several NLP applications, namely, question-answering, information retrieval, etc. In some cases, these applications require a minimal understanding of the text. The use of this task in question- answering requires that the result must be the accurate answer to a question, instead of a number of references to documents that contain the answer. It makes it possible to construct a coherent representation of the content of a text to answer a question in natural language. In this paper, we propose to construct a semantic and logical representation of Arabic texts (the question and the passages of texts). Therefore, we propose a new approach for the Arabic question-answering. This approach is based on the automatic understanding of Arabic texts (question or passages of texts) to transform them into semantic and logical representations. The experiments results of our approach achieved an accuracy of 74%. 1 Introduction With the growth of electronic media, including the Web, which has become the main source of information for everyone, the amount of information has grown exponentially. In the existence of this large-scale digitization, finding high-precision information is a challenge. Thus, the question-answering research is developing every day; Research in the field of Arabic question-answering began in the 1990s. To our knowledge, there are many systems that have been developed for many languages of the world (i.e. English, French, Chinese, Japanese, etc.). However, there are few systems that have been developed for the Arabic language. These systems provide answers in the form of short passages, extracts of document collections. Therefore, the performance of these systems is limited by the difficulty of the language processing and the considerable lack of effective NLP tools that take into consideration the Arabic language [Alagha & Abu-Taha, 2015], [Al-Khalifa & Al-Wabil, 2007]. Arabic is an inflected and derivational Semitic language where from one root, several forms could be derived. This language poses challenges for the tasks of the NLP, as opposed to other languages, because of its morphological richness, relatively free words order and diglossic nature (where standard and dialects mix in most genres data) [Almarwani and Diab, 2017]. Moreover, with the absence of the capitalization in the Arabic language, we cannot distinguish the verbs from the nouns. The absence of voyellation in the Arabic texts is also a great source of ambiguity for morphological, syntactic and semantic analysis. Thus, the Arabic language still lacks large-scale computing resources that are very useful for English as the WordNet [Miller, 1995], or a resource, such as, VerbOcean [Chklovski and Pantel, 2004]. In this paper, we focus on a few dimensions whose development seems more particularly to the automatic understanding of Arabic texts. Specifically, we introduce a real analysis of text passages which may contain an answer to a question. Indeed, the automatic understanding of a text is one of the …

Content truncated to 500 words