The Alignment between Language Properties and Computational Algorithms Enhances Statistical Word Segmentation:...

Area of Science:

• Linguistics
• Computational Linguistics
• Developmental Psychology

Background:

• Child-directed speech (CDS) is hypothesized to possess linguistic characteristics that aid language acquisition.
• Understanding how CDS facilitates early language processing, such as word segmentation, is crucial for developmental theories.

Purpose of the Study:

• To compare the word segmentability of Korean child-directed speech (CDS) versus adult-directed speech (ADS).
• To investigate the influence of specific linguistic properties of CDS and ADS on computational word segmentation performance.

Main Methods:

• A speaker-matched corpus of naturalistic Korean CDS and ADS was created.
• Computational algorithms were used to evaluate word segmentation performance on both speech registers.
• Statistical modeling was employed to identify linguistic features modulating segmentation performance.

Main Results:

• Korean CDS exhibited shorter utterances and words, lower lexical diversity, and reduced lexical ambiguity compared to ADS.
• Computational analysis showed significantly higher word segmentation F-scores for CDS than ADS.
• Linguistic properties unique to CDS were found to modulate its enhanced segmentability.

Conclusions:

• The linguistic adaptations present in Korean CDS enhance its segmentability for computational models.
• Specific features of CDS, such as simplified vocabulary and structure, facilitate word segmentation.
• The findings contribute to understanding the role of input characteristics in early language processing and computational language acquisition.