Unlearning Incorrect Associations in Word Learning: Evidence From Eye-Tracking.

Area of Science:

• Cognitive Science
• Psycholinguistics
• Neuroscience

Background:

• Computational and animal models highlight the importance of unlearning incorrect word meanings for effective word acquisition.
• The precise mechanisms of this pruning process, particularly the roles of supervised versus unsupervised learning, remain incompletely understood.

Purpose of the Study:

• To investigate how incorrect word-object associations are pruned during the word learning process.
• To differentiate the contributions of supervised and unsupervised learning to the unlearning of semantic representations.
• To examine how the initial strength of incorrect mappings influences their subsequent pruning.

Main Methods:

• Two experiments involving adult participants learning novel word-object mappings.
• Utilized eye-tracking during training and testing to measure the activation of alternative word meanings.
• Manipulated the frequency of incorrect pairings during pretraining to disentangle supervised and unsupervised pruning effects.

Main Results:

• Weak incorrect associations were rapidly pruned, whereas stronger associations persisted even after successful word learning.
• Evidence for unsupervised pruning was observed, indicating that unlearning can occur without direct corrective feedback.
• The persistence of some incorrect associations suggests they may remain active in the lexicon post-learning.

Conclusions:

• The degree of unlearning incorrect word meanings is dependent on the initial strength of their association.
• Both supervised and unsupervised mechanisms contribute to the pruning of erroneous semantic representations.
• Lingering incorrect associations may influence subsequent language processing, such as word recognition.