Language habits, acoustic confusability, and immediate memory for redundant letter sequences

Abstract

Immediate memory was tested for sequences of 7, 8, 9, or 10 auditorily presented letters which comprised either words or zero-, first-, second-, or third-order approximations to English words. At all lengths, recall probability correlated highly with letter sequence predictability (.58-.78) but was unrelated to acoustic confusabilitv. It is suggested that coding was still phonemic but involved speech sounds comprising 'several letters rather than letter names.Short-term memory for letter sequences is influenced by two major variables, predictability and acoustic confusability. Predictability is a measure of the closeness with which a letter sequence approximates to the structure of English, and has been shown to correlate highly with probability of recalling approximations to English words (Baddeley, 1964a), and somewhat less highly with the recall of sequentially presented consonant sequences (Baddeley, Conrad, & Hull, 1965). A series of studies have shown a high correlation between probability of recalling a visually presented consonant sequence and its acoustic confusability, as measured by the probability of mishearing the constituent letters when presented in noise (Conrad, 1962;Conrad & Hull, 1964). A later study by Conrad, Freeman, & Hull (1965) directly compared the contributions of predictability and acoustic confusability to the recall of sequences of six random consonants and concluded that the main determinant of recall probability in this situation was acoustic confusability, with predictability playing a relatively minor role. Kintsch (1970, p.191) has subsequently concluded, on the basis of this one very limited study, that "Language f actors appear to be ineffective in short-term memory, or at least are overridden in importance by acoustic confusions." Such a conclusion is clearly at variance with the results of Baddeley (1964a , 1970). The following experiment, therefore, uses the teclmique that seems most likely to favor acoustic coding, sequential auditory presentation, and studies the relationship between acoustic confusability, predictability, and probability of recall of letter sequences of varying approximation to English. MATERIAL The technique described by Miller, Bruner, & Postman (1954) was used to generate eight sequences of 7, 8,9, and 10 letters of zero-, first-, second-, and third-order approximation to English. These were based on the tables of letter frequency in written English compiled by Baddeley, Conrad, & Thomson (1960). In addition, four sets of eight words (one set each of length 7,8,9, and 10 letters) were selected at random from the Thorndike-Lorge list, making a total of 40 sequences of each length. PROCEDURE A separate group of 20 enlisted men was tested at each of the four sequence lengths. Letters were read out at a rate of one per second, and Ss were allowed 2 sec per letter to recall as much of the sequence as possible. Ss were instructed to write each letter in the appropriate serial pOSition on a prepared answer sheet. They were familiarized with the material t...

Bound feature combinations in visual short-term memory are fragile but influence long-term learning

“…This is analogous to the well established finding that verbal short-term memory for letter sequences is better for sequences that resemble real words represented in long-term memory (e.g. Baddeley, 1971). Returning to our own data, the finding that bound objects constrain both VSTM and long-term learning further suggests that VSTM and visual long-term memory share a common representational format.…”

Optimal forgetting: Semantic compression of episodic memories

“…For instance, the effect of domain expertise has widespread support in the memory literature not just in the domain of chess but also memory for sports trivia, software code, medical images, and other games [ 48 ]. In addition to the effect of varying levels of expertise on recall accuracy, a similar effect arises if the congruence of stimuli to the statistical structure of the domain is varied along a spectrum, for example the order to which letter statistics of words conform to that of the English language [ 49 ]. The encoding of the observation into a posterior over latent variables can be understood as compressing sensory experience into sufficient statistics for the latents, which in addition to the gist-based distortion experiments analysed here explains seemingly paradoxical results that discrimination performance of sound textures decreases with increasing stimulus duration when stimulus samples come from the same texture family [ 50 ].…”

Working memory and binding in sentence recall

“…Furthermore, memory for sequences of unrelated consonants is strongly influenced by phonological similarity between the names of the letters, suggesting that sequences are remembered in terms of individual subvocalised letters. However, when order of approximation increases and vowels are included, the correlation with phonological similarity becomes insignificant (r = .05) while letter-based predictability becomes an important factor (r = .66) (Baddeley, 1971). There is therefore abundant evidence that sequential redundancy within language has a powerful impact on immediate recall, and that it operates through the process of chunking.…”