Computational Models of Real World Phonological Change

Language change refers to the phenomenon of alterations in the linguistic structures of the language used by a community over time, which may subsequently give rise to one or more languages or dialects. Phonological change is a special type of language change that involves alteration in the pronunciation patterns. The aim of this thesis is to develop computational models for cases of phonological changes that have been observed for real languages and validate the same against real data. Several theories have been proposed to explain the cause and nature of phonological change, which, nevertheless, cannot be validated and established firmly due to the lack of sufficient linguistic data and scope of experimentation. Computational models not only provide a formal description of the linguistic theories, but also serve as excellent virtual laboratories. The inherent complexities of our cognitive and social structures, however, make it hard or impossible to replicate reality within a computational framework. Consequently, most of the present computational models are developed for toy languages and are rarely compared with real linguistic data – an obvious weakness that has attracted criticisms from linguists and computer scientists alike. In this thesis, we present computational models pertaining to the different explanations of phonological change for two cases of real world language change: schwa deletion in Hindi (SDH) and change affecting the verb inflections of the dialects of Bengali. Hindi and Bengali are two major languages of the IndoAryan family that are spoken in the Indian sub-continent. The models are developed in an incremental manner, where the findings of a more detailed, but computationally intensive model are used to abstract out the details in the sub-