A comparison study on contextual modeling for estimating functional loads of phonological contrasts

Abstract

Functional load (FL) is the quantitative measure of the importance of phonological contrasts, which stand for the differentiation of communicative linguistic units. Correct estimate of FLs is useful for the studies of speech recognition, language evolution, language teaching and etc. Conventional approaches use phonological transcriptions and unigram probabilities for the estimation, hence weak in contextual modeling. Based on the measurement of mutual information (MI) between the text and its phonological transcription, we previously proposed a novel FL measurement which utilizes n-gram word probabilities, hence owing better context modeling power. In this study, we compare the effects of different context on the estimation of FL: syllable, word, n-gram word model, and open data. Experimental results show: the wider the context modeling, the smaller the FL; FL based on MI with the trigram model achieves the best performance in modeling the context in our experiments. Compared with FL based on entropy, FL based on MI showed smaller value and is applicable to open data.