A Hybrid Approach to Develop the First Stemmer in Maithili

This paper presents a hybrid approach to extract the stems from the Maithili words. Development of stemmer for various languages has gained adequate attention of the researchers for decades. Stemmers have also been developed for most of the major Indian languages. However, no work is found on the development of Maithili stemmer. In this first attempt, we use a hybrid model where rule-base acts as the core module. We study the inflections of the major parts-of-speech categories and define rules accordingly. The rule-based module is supported by neural word-embedding and suffix stripping. Suffix stripping is used to increase the coverage and word embedding is used to restrict erroneous stem generation. The system is evaluated using a test data collected from Maithili literature and news text. The final system achieves an accuracy of 84.6%.