A Sequence-to-Sequence based Approach For the double Transliteration of Tunisian Dialect

Abstract

Transliteration consists of automatically transforming a grapheme’s transcription from one writing system to another, while preserving its pronunciation. It is usually used in the context of machine translation and cross language information retrieval, mainly to deal with the issue of named entities and technical terms. In the case of some Arabic dialects, which are used on the social web in both Latin and Arabic scripts and which are still low-resource languages, transliteration is of great benefit for the automatic generation of various linguistic resources (parallel corpora and lexica), useful for their automatic processing. In this work, we focus on the Tunisian dialect transliteration. We propose a deep learning based Sequence-to-Sequence approach to perform a word-level transliteration of the user generated Tunisian dialect on the social web, in both Latin to Arabic and Arabic to Latin senses.