Chunking or not chunking? How do we find words in artificial language learning?

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Neuroscience

Background:

• Implicit statistical learning enables identification of word boundaries in continuous speech.
• Two primary mechanisms proposed: sequence chunking and sensitivity to transitional probabilities.
• Differentiating these mechanisms is challenging due to similar predictions in standard experiments.

Purpose of the Study:

• To contrast chunking and transitional probability accounts of implicit statistical learning.
• To investigate how learning conditions influence the nature of acquired representations.
• To determine if explicit cues affect learning strategies.

Main Methods:

• Two experiments using a serial reaction time task to learn artificial languages (L1 and L2).
• Manipulated the relationship between L1 and L2 transitions (unrelated vs. partly related).
• Varied the presence of cues to facilitate sequence parsing.

Main Results:

• When parsing cues were present, participants successfully learned artificial language words (chunking).
• Without explicit cues, learning was primarily driven by transitional probabilities.
• The relationship between language structures (L1/L2) influenced the learning strategy.

Conclusions:

• The nature of representations in implicit statistical learning is context-dependent.
• Explicit cues promote chunking strategies, while their absence favors transitional probability processing.
• Findings highlight the flexibility of statistical learning mechanisms in language acquisition.