The study of memory for texts has had an long tradition of re-search in psychology. According to most general accounts oftext memory, the recognition or recall of items in a text is basedon querying a memory representation that is built up on the ba-sis of background knowledge. The objective of this paper is todescribe and thoroughly test a Bayesian model of this generalaccount. In particular, we develop a model that describes howwe use our background knowledge to form memories as a pro-cess of Bayesian inference of the statistical patterns that areinherent in a text, followed by posterior predictive inference ofthe words that are typical of those inferred patterns. This pro-vides us with precise predictions about what words will be re-membered, whether veridically or erroneously, from any giventext. We then test these predictions using data from a memoryexperiment using a relatively large sample of randomly chosentexts from a representative corpus of British English.

Statistical learning paradigms traditionally use transitional probabilities as a measure of statistical distribution within a language. The current study suggests that alternative metrics may exist that can account for differences in language processing ability. Two primed lexical decision tasks are used to examine the effects of bigram frequency and diversity on speed and accuracy of word recognition. It is demonstrated that both frequency and diversity contribute to word recognition performance; findings and theoretical implications are discussed.