Markov Model for French-Wolof Text Analysis

Sentiment analysis is an essential task in natural language processing. It aims to automatically evaluate the sentiment polarity. Several methods and tools have been developed to address this issue which is more difficult in a bilingual context, such as the case of Senegal where the official French language is often mixed with Wolof the national language. [1] has been recently proposed as effective solutions for overcoming the complexities of this bilingual sentiment analysis. The lemmatizer-methods used there are based respectively on word-level similarity and the word-level trigrams (list of consecutive three letters) to convert inflectional forms and derivationally related forms of a word to a common base form. This morphological analysis approach is very complex and has some limitations for normalizing some words because French words are morphologically complex. Furthermore, the lemmatization of Wolof words is very difficult to implement because they are written using the French alphabet. In addition, these previous approaches use Bayes independence assumption to determine the sentiment of a sentence. This is another limitation when considering syntactic rules in a linguistic expression. To overcome these drawbacks and address the challenges and opportunities of bilingual sentiment analysis, we propose: (i) a French-Wolof dictionary and dictionary-based lemmatizer for better deal with verbs’ conjugation or some words’ declination in both languages, and (ii) a method for determining context dependent sentiment words based on Markov Model. Experimental conducted on a more larger data [2], shows that it successfully predicts sentiment polarity with an accuracy of 0.94 (implying an increase of 21% compared to the previous work).