LAUSR

In this paper, we propose a new approach for the Arabic question-answering system. This approach is based on the automatic understanding of Arabic texts to convert them into semantic and logical representations. Our approach is designed to determine the relation of the textual entailment between the logical representations of the question and the text passage in order to select the text passage that contains an answer to the question and find the precise answer. For a logical representation, we referred to a semantic representation. The idea is to convert the Arabic texts into conceptual graphs that allow the modelling of textual information by concepts and relations. From these graphs, we proposed an algorithm that transforms each conceptual graph to a logical representation. Finally, we extracted three features and combined them to determine the textual entailment relation between two logical representations of the question and its passage answer. Our approach has been validated through a question-answering system, called NArQAS. For the evaluation, we used questions and text passages from our corpus of questions-texts, AQA-WebCorp. The performance of our approach has reached an accuracy of 74%.