A role for the developing lexicon in phonetic category acquisition.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Linguistics

Background:

• Infants learn to segment words from continuous speech concurrently with acquiring phonetic categories.
• Existing models of phonetic category acquisition often overlook the influence of word-level information.
• The interplay between word segmentation and phonetic learning in infants is not fully understood.

Purpose of the Study:

• To investigate how word segmentation might constrain and inform phonetic category learning in infants.
• To model the role of lexical feedback in resolving ambiguities in phonetic perception.
• To explore the computational mechanisms underlying early language acquisition.

Main Methods:

• Developed a Bayesian computational model to simulate phonetic category learning.
• Incorporated feedback mechanisms where segmented words influence phonetic perception.
• Tested the model's ability to disambiguate overlapping phonetic categories using simulated data.

Main Results:

• Word-level information significantly aids in disambiguating overlapping phonetic categories, such as English vowels.
• The model's learning patterns closely replicate human behavior observed in artificial language learning experiments.
• Demonstrated that top-down lexical information can constrain phonetic learning effectively.

Conclusions:

• The developing lexicon plays a crucial role in the acquisition of phonetic categories during infancy.
• A framework is proposed for integrating top-down constraints, derived from word knowledge, into models of category learning.
• Highlights the interconnectedness of word segmentation and phonetic perception in early language development.