New evidence for chunk-based models in word segmentation.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Linguistics

Background:

• Infants leverage statistical cues for language acquisition, particularly word segmentation.
• Debates persist on whether word extraction relies on transitional probabilities or chunk-based processes.
• Dissociating these models has been challenging due to overlapping predictions.

Purpose of the Study:

• To propose a novel procedure to differentiate between transitional probability and chunk-based word segmentation models.
• To investigate the influence of a first language (L1) on second language (L2) segmentation.
• To test predictions using a chunk-based model (PARSER) and experimental data.

Main Methods:

• Simulations using PARSER (a chunk-based model) to predict L2 segmentation under L1 interference.
• Comparing predictions of chunk-based and transitional probability models.
• Experimental validation using a two-alternative forced choice test for L2 word segmentation.

Main Results:

• PARSER predicted depleted L2 learning when L1 words became L2 transitions, falling below chance performance.
• Transitional probability models predicted above-chance performance regardless of exposure duration.
• Experimental data confirmed PARSER's predictions, showing significantly below-chance performance in the interference condition.

Conclusions:

• Chunk-based models provide a more accurate account of L2 word segmentation under L1 interference.
• The findings challenge transitional probability as the sole mechanism for statistical learning in word segmentation.
• Language learning involves complex interactions between existing linguistic knowledge and new language input.