Segment predictability as a cue in word segmentation

摘要: Several computational simulations of how children solve the word segmentation problem have been proposed, but most applied only to a limited number languages. One model with some experimental support uses distributional statistics sound sequence predictability (Saffran et al. 1996). However, design does not fully specify is best measured or modeled in simulation. Saffran (1996) assume transitional probability, Brent (1999a) claims mutual information (MI) more appropriate. Both locally, relative neighboring segment-pairs. This paper replicates Brent's mutual-information on corpus childdirected speech Modern Greek, and introduces variant using global threshold. finding regarding superiority MI confirmed; performance local comparisons thresholds depends evaluation metric.