Speech segmentation by statistical learning is supported by domain-general processes within working memory.

Area of Science:

• Cognitive Psychology
• Neuroscience
• Psycholinguistics

Background:

• Statistical learning (SL) is crucial for language acquisition, enabling word segmentation from continuous speech.
• The cognitive mechanisms supporting SL, particularly working memory involvement, remain under investigation.
• Previous research suggests processing speed influences SL, hinting at active maintenance mechanisms.

Purpose of the Study:

• To determine the extent to which working memory resources are engaged during statistical learning.
• To investigate whether SL relies on domain-specific (phonological) or domain-general working memory resources.

Main Methods:

• Participants identified novel words in an artificial speech stream using syllable transitional probabilities.
• Speech rate was manipulated (slow vs. fast) to assess its impact on segmentation performance.
• A dual-task paradigm involved a two-back task (rhyme vs. shape stimuli) during speech stream processing to measure working memory load.

Main Results:

• Slower speech rates significantly improved statistical learning performance, indicating the need for active processing.
• Both phonological (rhyme) and non-phonological (shape) two-back tasks equally impaired learning.
• Cognitive load from dual tasks eliminated the benefit of slower speech rates on segmentation.

Conclusions:

• Statistical learning is supported by working memory processes.
• These working memory resources appear to be domain-general, rather than exclusively domain-specific (e.g., phonological).
• The findings highlight the role of general cognitive resources in implicit language learning.