Forget me, forget me not: unlearning incorrect associations in word learning

People are estimated to acquire tens of thousands of words by the time they reach adulthood. To do this, they have to form associations between words and objects to build a vocabulary. Data from an animal model of word learning as well as data from computational modeling suggest that people not only strengthen associations between words and objects but also prune incorrect associations. This means that people learn not only that the word “ball” maps onto the object “ball” but also that the word “ball” does not map onto the objects “ball” or “apple”.

As there currently is no behavioral evidence for such pruning in humans, I conducted six experiments in adults in my dissertation. In all experiments, participants were first trained to link two objects to each word, so that each word was associated with two meanings. Subsequently, they learned that each word only mapped onto one object. However, during training, the amount of experience with objects that were available as choices was manipulated, offering the possibility to prune some incorrect associations but not others. Throughout training, participants received feedback as to whether they selected the target object. I measured association strength between words and objects using eye-movements, a post-training accuracy assessment and a yes/no task. Across experiments, I manipulated whether symbols were auditory words, orthographic words or non-linguistic symbols.

Participants showed rapid pruning of incorrect associations, and this process was accelerated when words were auditory. Thus, speed of pruning may be related to ease of acquiring the mappings. Moreover, there was evidence that small associations between symbols and objects remained, and that they could influence subsequent word learning.