A distributional perspective on the gavagai problem in early word learning

Area of Science:

• Developmental Psychology
• Linguistics
• Cognitive Science

Background:

• Word learning involves mapping sounds to meanings and grammatical categories.
• The 'gavagai problem' highlights the challenge of referential ambiguity in word acquisition.
• Infants must overcome ambiguity to map words to intended referents.

Purpose of the Study:

• To analyze distributional information in infant-directed speech regarding object and property words.
• To investigate how the ratio of object versus property labels affects word learning robustness in adults.
• To understand the origins of the whole-object bias in early word learning.

Main Methods:

• Corpus analysis of parent-directed speech to infants (6-20 months).
• Two experiments with adults using a cross-situational statistical learning paradigm.
• Manipulation of label distributions favoring objects or properties.

Main Results:

• Infant-directed language contains frequent object and property words, but they rarely co-occur.
• Adult word learning experiments demonstrated a whole-object bias, even when property labels were more frequent.
• Property labels did not generalize to novel objects, irrespective of label distribution.

Conclusions:

• The whole-object bias in early word learning may stem from infants' limited ability to encode object details during rapid naming.
• Mapping words to whole objects is inherently easier than to properties, simplifying the gavagai problem.
• Immature encoding and memory abilities in infants further contribute to this bias and word learning process.