The influence of familiarity on the ability to form associations: A connectionist account

Examination of numerous past results, such as word frequency effects, suggests that there might be an influence of stimulus familiarity on the ability to form associations with that stimulus. Specifically, that familiarity facilitates association formation. However, to date there has been no direct experimental evidence of this effect. To demonstrate familiarity facilitation, as well as to examine other variables thought to modulate the effect of familiarity facilitation, two experiments were conducted. Evidence in favor of familiarity facilitation, modulated by stimulus complexity and amount/quality of prior exposure, was found. Additionally, it is noted that this behavior appears to resemble that of an autoencoder network, suggesting that autoencoder-like processing might be in part the source of these effects. Application of these results is also briefly discussed. INFLUENCE OF FAMILIARITY ON ASSOCIATION FORMATION INFLUENCE OF FAMILIARITY ON ASSOCIATION FORMATION How humans learn, and, more specifically, which variables influence successful learning, has been a topic of enormous interest to psychologists and educators alike for well over a century, if not longer (Ebbinghaus, 1885). Accordingly, much research, both past and present, has been dedicated to discriminating the effective variables from those of no consequence to learning. This research has yielded a number of well-established variables known to affect learning: general practice and rehearsal, distributed practice, encoding variability, effortful encoding, and use of retrieval cues to access information, to name a few (Cepeda, Pashler, Vul, Wixted, & Rohrer, 2006; Wickelgren, 1964). Though we have undoubtedly uncovered much about learning, memory, and the interactions between the two, there remain a number of related puzzles yet to be fully explained. Among these puzzles are the strange patterns of results that word frequency memory tests consistently produce. Furthermore, while many of these patterns are perplexing in themselves, when taken as a whole, word frequency effects are even more provocative. Perhaps the most well-known and well-replicated of these patterns is that, for lists containing both high and low frequency words, recognition is better for low frequency than high frequency words, but recall is better for high frequency than low frequency words (Mandler, Goodman, & Wilkesgibbs, 1982). Accordingly, a number of explanations have been provided for this effect. Less well-known, recognition memory for high frequency word pairs is better than that for low frequency word pairs (Clark, 1992). Additionally, for lists of purely high and purely low frequency words, high frequency words perform better at recognition tests than low frequency words (MacLeod & Kampe, 1996; Watkins, LeCompte, & Kim, 2000). Although a number of explanations have been provided for subsets of these and other word frequency effects, no present explanation can intuitively account for the entirety of these effects. This paper seeks to provide the necessary explanation by highlighting a new variable proposed to modulate the interaction between learning and memory. More specifically, this paper puts forward the following hypothesis: the familiarity of a stimulus interacts with the ability to form associations with that stimulus. In particular, the higher the familiarity of a stimulus, the easier it is to form an association with said stimulus. The first evidence for this familiarity facilitation hypothesis, as simply a behavioral phenomenon, comes from its ability to intuitively account for all of the various strange patterns of word frequency results without exception or limitation. Assuming that recollection requires an association of a stimulus with context, while recognition requires at minimum a feeling of familiarity with the stimulus, the most popular pattern can be explained. More specifically, when the list is mixed, people will attend to the low frequency words more that high frequency words (Rayner & Duffy, 1986); low frequency words will undergo the largest change in familiarity, and people will thus be more accurate at judging low frequency words by recognition. Conversely, because the association of high frequency words to the experimental context will be facilitated by their higher familiarity relative to low frequency words, people will be better able to recollect high frequency words. Similarly, when tested on recognition of word pairs, high frequency word pairs will perform better than low frequency word pairs as, for the more familiar high frequency word pairs, familiarity will aid the association of the pair itself, while, for low frequency word pairs, there will be no such facilitation. Finally, for pure lists containing only high or only low frequency words, one is unable to differentially attend more closely to low frequency words at the expense of high frequency words. After attending to each INFLUENCE OF FAMILIARITY ON ASSOCIATION FORMATION