Statistical word segmentation in spontaneous child-directed speech of Korean

Abstract

The present study demonstrates advantages of child-directed speech (CDS) over adult-directed speech (ADS) in statistical word segmentation of spontaneous Korean. We derived phonetic input from phonemic corpus by applying a set of phonological rules. For modeling the statistical word segmentation based on transitional probability (TP), we used two syllable-based algorithms (i.e., Absolute and Relative) in two directions (i.e., Forward TP and Backward TP). Results show that (i) segmentation accuracy is greater with phonetic input than phonemic, (ii) The model performs better when trained on CDS than ADS, and (iii) segmentation accuracy improves with child age.

Tasks

Reproductions