Running Head Word Association Spaces 1 Word Association Spaces Semantic Spaces based on Free Association that Predict Memory Performance Send correspondence to
study by Brainerd and Reyna (1998), subjects gave two ratings for each of 100 test items. In one
rating, they were instructed to judge whether the item was old or new and were told to judge
semantically similar distractors as “new”. In another rating, they were instructed to rate (on a six
point scale) how semantically similar the item was to the studied items. We focused on the
semantic similarity ratings for the new items from this study. For each subject, the 72 new test
items were randomly selected from a larger pool of 144 words. An average of 44 (SD=4.87)
subjects rated the semantic similarity for each of the 144 words that might appear as new words
in the test list7. The semantic similarity ratings are theoretically interesting because they can be
used to test models of semantic similarity. Subjects merely have to remember how similar the
item was to the studied items without being forced to give old-new judgments that might be more
influenced by various strategic retrieval factors (such as word frequency or previous retrievals).
Many memory models assume that a recognition memory judgment is produced by
calculating the global familiarity involving the summed similarity between the test item and the
episodic traces in memory (e.g. Minerva II, Hintzman 1984, 1988; SAM, Gillund & Shiffrin,
1984). More recently, Shiffrin and Steyvers (1997, 1998) and McClelland & Chappell (1998)
have proposed recognition memory models that produce recognition judgments with Bayesian
decision processes. McClelland & Chappell (1998) proposed that the best match (i.e., maximum
similarity) between the test item and the episodic traces in memory forms the basis for the
recognition judgment. Shiffrin & Steyvers (1998) showed that in the Bayesian framework, a
maximum similarity process produced results very similar to a summed similarity process. In this
research, our aim is not to test these models specifically but to use and simplify the underlying
mechanisms to predict semantic similarity ratings.
热门文档