An Arabic Semantic Parser and Meaning Analyzer

Abstract

Arabic language is very rich in derivations, vocabulary, and grammatical structures. The problem of determining the correct meaning of a word in a non-vowelized Arabic sentence is not a trivial task since Arabic is very rich in the polysemy phenomena. This paper attempts to reveal the word sense ambiguity, by building a semantic parser powered by a statistical semantic analyzer, which may aid in the improvement of machine translation, question answering and other Arabic NLP systems. Building the parser was done in three steps. The first step was to acquire the grammatical rules for Arabic that was covered in an Arabic grammar textbook, and develop constraints that aided in revealing part of the parsing ambiguity. The grammar and the constraints were then written in an XML format to make them readable and available for future uses. The second step was to build the semantic parser that assigns grammatical structure onto input sentence. The final step was to impose a semantically statistical technique on the resulting grammatical structures to determine the most accurate structure, the one that result in resolving the word sense ambiguity, and determining the most accurate meaning of the word.