Pattern based pruning of morphological alternatives of Bengali wordforms

Multiple morphological interpretations of wordforms is a bottleneck for different level of syntactico-semantic analyses of Natural Language (NL) sentences. Common vocabulary words of morphologically rich languages typically have more than one morphological analyses. However, if a word contains multiple morphological alternatives, one of them is appropriate with respect to the context where the word is used. Different text processing tasks including rule-based Machine Translation (MT) require morphological features of each surface word-form of the sentence. Therefore, given a context, an efficient procedure is needed to choose the appropriate morphological analysis of each word having multiple alternatives. In this paper, we propose a method to identify correct morphological analysis of each Bengali word-form by cancelling (or pruning) all such analyses which show context incompatibility.