Is there a bilingual disadvantage for word segmentation? A computational modeling approach

Area of Science:

• Cognitive Science
• Computational Linguistics
• Developmental Psychology

Background:

• Word segmentation, identifying word boundaries in continuous speech, is a fundamental challenge in language acquisition.
• This challenge is particularly pronounced due to the absence of clear pauses between words in spoken language.

Purpose of the Study:

• To computationally model and assess the difficulty of word segmentation for bilingual infants compared to monolingual infants.
• To investigate the influence of linguistic similarity and algorithmic approach on word segmentation performance in different language settings.

Main Methods:

• Utilized seven distinct computational algorithms representing various cognitive approaches to word segmentation.
• Applied these algorithms to naturalistic speech corpora, creating matched monolingual and bilingual datasets.
• Varied the phonological and lexical overlap between language pairs (e.g., Catalan/Spanish vs. English/Spanish) to simulate different bilingual exposure scenarios.

Main Results:

• The performance variation across segmentation algorithms was the most significant factor influencing results.
• Language differences also showed a substantial effect on segmentation performance.
• The impact of bilingualism on word segmentation difficulty was found to be smaller than the effects of both the algorithms used and the languages themselves.

Conclusions:

• Bilingualism does not inherently pose a substantially greater challenge to word segmentation than monolingualism, contrary to some assumptions.
• Computational modeling provides valuable insights into the factors affecting early language acquisition, highlighting the importance of algorithmic approaches and language characteristics.