lcc 10
Word Segmentation and Language Learning: Summary Guide
1. Brent (1999) – Speech Segmentation and Word Discovery
This review categorizes computational models of speech segmentation into three primary strategies: (1) utterance-boundary (segmenting at end of utterances), (2) predictability (segmenting where unexpected transitions occur), and (3) word-recognition (hypothesizing and matching word-like units). The INCDROP model, based on word-recognition, shows the best performance. The review compares memory demands, processing type (on-line vs batch), and behavioral predictions of these models. The integration of phonotactic, rhythmic, and statistical cues is proposed as essential for comprehensive modeling.
2. LCC Lecture 10 – Language Learning and Segmentation Models
This lecture outlines computational and experimental methods for studying language learning. It emphasizes cognitive plausibility in models (e.g., realistic input, memory limits, incremental learning). The course highlights segmentation challenges like word boundary identification, the role of transitional probabilities, and phonological cues. It discusses grounded language learning and evaluates models using humanlike behaviors like error patterns and developmental stages. Resources like CHILDES and TalkBank are presented as data sources.