Semantic Language Model for Tunisian Dialect

Proceedings of Recent Advances in Natural Language Processing, pages 720–729, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_084 720 Semantic Language Model for Tunisian Dialect Abir Masmoudi MIRACL Laboratory University of Sfax masmoudiabir@gmail.com Rim Laatar MIRACL Laboratory University of Sfax Mariem Ellouze MIRACL Laboratory University of Sfax Mariem.Ellouze@planet.tn Lamia Belguith MIRACL Laboratory University of Sfax Abstract In this paper, we describe the process of creating a statistical Language Model (LM) for the Tunisian Dialect. Indeed, this work is part of the realization of Au- tomatic Speech Recognition (ASR) sys- tem for the Tunisian Railway Transport Network. Since our field of work has been limited, there are several words with similar behaviors (semantic for example) but they do not have the same appear- ance probability; their class groupings will therefore be possible. For these reasons, we propose to build an n-class LM that is based mainly on the integration of purely semantic data. Indeed, each class repre- sents an abstraction of similar labels. In order to improve the sequence labeling task, we proposed to use a discriminative algorithm based on the Conditional Ran- dom Field (CRF) model. To better judge our choice of creating an n-class word model, we compared the created model with the 3-gram type model on the same test corpus of evaluation. Additionally, to assess the impact of using the CRF model to perform the semantic labelling task in order to construct semantic classes, we compared the n-class created model with using the CRF in the semantic labelling task and the n-class model without using the CRF in the semantic labelling task. The drawn comparison of the predictive power of the n-class model obtained by ap- plying the CRF model in the semantic la- belling is that it is better than the other two models presenting the highest value of its perplexity. 1 Introduction Generally, the development of such ASR system for a specific language requires first and foremost the construction of a large speech corpus. This corpus must be based on both an orthographic and a phonetic transcription. In addition, tex- tual data for learning the LM of the system are also required. Nevertheless, these re- sources are not available directly for Ara- bic dialects. As a result, the ASR sys- tem development for Arabic dialects is fraught with many different kinds of dif- ficulties that it faces. In this perspective, this work is integrated in the field of de- veloping the Tunisian Dialect ASR system for the Tunisian Railway Transport Net- work. More precisely, we are interested in this paper in presenting our method for constructing a LM; one among the essen- tial components of the ASR system. This model proposes to define the probability distribution on sets of word sequences. Due to the lack of learning data of the Tunisian dialect, it is necessary to find a method that maximizes the amount of in- formation. This corresponds to the appear- ance of the n-class LM. The main idea of this model is to classify vocabulary words into …

Content truncated to 500 words