An explicit statistical model of learning lexical segmentation using multiple cues

Abstract

This paper presents an unsupervised and incremental model of learning segmentation that combines multiple cues whose use by children and adults were attested by experimental studies. The cues we exploit in this study are predictability statistics, phonotactics, lexical stress and partial lexical information. The performance of the model presented in this paper is competitive with the state-of-the-art segmentation models in the literature, while following the child language acquisition more faithfully. Besides the performance improvements over the similar models in the literature, the cues are combined in an explicit manner, allowing easier interpretation of what the model learns.