The Affects of Inflectional and Derivational Morphology in Bengali Phonology and its Implications in Speech Technology

Development of a text-to-speech synthesizer in a language calls for either a pronouncing dictionary or a grapheme-to-phoneme (G2P) converter. For agglutinative and morphologically rich languages it is not possible to build an exhaustive pronouncing dictionary. Moreover, since phonology is overtly dependent on morphological structures, morphological analysis becomes important even for building an accurate G2P converter. In this paper we discuss the case of Bengali phonology, which is influenced by its derivational and inflectional morphologies to a large extent. We describe the different phenomena and show that although inflectional morphology seems to be absolutely necessary for a reasonably good G2P converter, derivational morphology is computationally intensive and therefore, it might be overkill for the G2P converter. Rather, simple heuristics coupled with a list of common exceptions can improve the accuracy of the G2P.