Associations and manipulations in the mental lexicon: A model of word-stem completion

• There are no published computational models of word stem completion.
• Data suggest frequency and word length are important contributors.
• Initial model based on Estes SST accounts for major effects.
• The model is unsatisfactory for constrained stem completion.
• Response-manipulation strategy helps account for limitations of the base model.

Although word stem completion is commonly used to understand verbal fluency, lexical access, and incidental learning, no formal computational model of the process has been previously developed. We describe an association-based model based on Estes’s (1950) Stimulus Sampling Theory (SST). The model proposes that when words are experienced, associations form between word elements and wholes, via associative mechanisms described by SST. The association-based model provides a reasonable fit to unconstrained-length stems, but was inadequate for constrained-length stems. To address this, we hypothesize that a string manipulation strategy is used, editing previous responses to create new possibilities. Results of the new model showed substantially-improved fits to data, even without assuming a length-based encoding advantage for short words. We conclude that both associative mechanisms and manipulative strategies are used to solve stem completion tasks, and that these manipulation strategies are as important as the hypothesized form of the word generation process.