Detection of semantic errors from simple Bangla sentences

We describe a methodology to detect semantic errors from Bangla sentences. According to Bangla grammar, a single verb can have many forms depending on its tense and person of its subject. The subject of a sentence can be noun or pronoun, may indicate human, animal, or any non-living entity. There is a fixed semantic relation between every verb and subject and object of a sentence. For example, a non-living entity can never feel hungry but living entity feels. This semantic difference checking for its correctness in a language is very important for the purpose of machine learning study and intelligent agent development for human computer interaction. Semantic error detection for Bangla language is an important research problem because of the variety that Bangla language offers in its grammatical, structural and semantic diversity. In this paper, we have established the relationship between subject and verb as well as object and verb of Bangla sentence. Hence we have proposed an algorithm for semantic correctness of simple Bangle sentences. The algorithm can easily be extended for other forms such as complex and compound sentences.