Psychological Review 1998, Vol. 105. No. 1, 158-173 Copyright 1998 by the American Psychological Association, Inc. 0033-295X/98/J3.00 The Phonological Loop as a Language Learning Device Alan Baddeley and Susan Gathercole University of Bristol Cpstanza Papagno Universita Degli Studi di Milano A relatively simple model of the phonological loop (A. D. Baddeley, 1986), a component of working memory, has proved capable of accommodating a great deal of experimental evidence from normal adult participants, children, and neuropsychological patients. Until recently, however, the role of this subsystem in everyday cognitive activities was unclear. In this article the authors review studies of word learning by normal adults and children, neuropsychological patients, and special developmental populations, which provide evidence that the phonological loop plays a crucial role in learning the novel phonological forms of new words. The authors propose that the primary purpose for which the phonological loop evolved is to store unfamiliar sound patterns while more permanent memory records are being constructed. Its use in retaining sequences of familiar words is, it is argued, secondary. Baddeley and Hitch (1974) considered the possibility that short-term memory (STM) may serve as a general working memory designed to support complex cognitive activities. This suggestion led to the development of a specific multicomponent model of working memory and has subsequentlycontributed to an enduring interest in the specific cognitive functions that are fulfilled by the separate subcomponents of working memory. The aspect of working memory for which the fullest theoretical account is now available is the phonological loop (Baddeley, 1986). The loop is specialized for the retention of verbal infor- mation over short periods of time it comprises both a phonologi- cal store, which holds information in phonological form, and a rehearsal process, which serves to maintaindecaying representa- tions in the phonological store. This relatively simple model has proved capable of accommodating a great deal of experimental evidence from normal adult participants, children, and neuro- psychoiogical patients (see Baddeley, 1997, and Gathercole & Baddeley, 1993, for reviews). Although the evidence for the existence of such a short-term system is strong, it is not obvious why the phonological loop should be a feature of human cognition at all. People have a remarkable capacity to repeat what they hear, a capacity that Alan Baddeley and Susan Gathercole, Department of Psychology, Centre for Study of Memory and Learning, University of Bristol, Bristol, England Costanza Papagno, Clinica Neurologica III, Universita Degli Studi di Milano, Milano, Italy. Much of the work reported in this article was supported by the Medi- cal Research Council (MRC) of Great Britain and the Economic and Social Research Council. The support of MRC Grant G9423916 during the writing of the final draft is gratefully acknowledged. Finally, we are very grateful to Dorothy Bishop of the MRC Applied Psychology Unit Cambridge for stimulating discussion and making many constructive suggestions. Correspondence concerning this article should be addressed to Alan Baddeley, Department of Psychology, Centre for Study of Memory and Learning, University of Bristol, 8 Woodland Road, Bristol BS8 ITN, England. Electronic mail may be sent via Internet to alan.baddeley@ bristol.ac.uk. has extensively been investigated by using lists of digits or unrelated words. When looking for a function that this capacity serves, Baddeley and Hitch (1974) concentrated on asking why it should be useful for people to remember sequences of words, and this led them to study comprehension and verbal reasoning. However, the evidence of a major role for the phonological loop was far from compelling (see Baddeley, 1986, for review). Indeed, much of the neuropsychological evidence that has led to the development of the current model of the phonological loop (e.g., Vallar & Baddeley, 1984) itself raises questionsabout its function. Many individuals with specific deficits in short- term phonological memory appear to have few problems in coping with everyday cognition: Despite dramatic reductionsin the capacity of the phonological loop, such individuals typically have normal abilities to produce spontaneous speech (Shal- lice & Butterworth, 1977) and encounter few significant diffi- culties in language comprehension (Vallar & Shallice, 1990). Does this mean that the loop is of little practical significance and that at least this aspect of STM does not serve as a working memory? Some authors have argued that this is indeed the case (Butterworth, Campbell, & Howard, 1986). The purpose of the present article is to propose that the phono- logical loop does indeed have a very important function to fulfill, but that it is one that is not readily uncovered by experimental studies of adult participants. We suggest that the function of the phonological loop is not to remember familiar words but to help learn new words. According to this view, the ability to repeat a string of digits is simply a beneficiary of a more fundamental human capacity to generate a longer lasting representation of a brief and novel speech event���a new word. For anexperimental psychologist working exclusively with adults, this might at first seem a singularly arcane and useless skill for humans to possess. For a developmentalist, though, the point of such a skill is all too evident because the task of forming long-term representations of novel phonological material is a key component of language development. At a conservative estimate, the average 5-year-old child will have learned more than 2,000 words (Smith, 1926) and will learn up to 3,000 more per year in the coming school years (Nagy & Herman, 1987). Indeed, successful vocabulary

160 BADDELEY, GATHERCOLE, AND PAPAGNO tion to both STM performance and vocabulary knowledge of a general intelligence factor. It should be noted that the STM- vocabulary association is particularly high for measures of non- word repetition, in which coefficients typically fall in the range .4- .6 for digit span, the coefficients are consistently lower, in the range .25- .45. Possible reasons for this especially close link between vocabulary and nonword repetition are discussed later. Of course, correlation does not imply causation. It is in princi- ple as plausible that good vocabulary knowledge supports accu- rate nonword repetition as the reverse. One way of collecting further evidence on the direction of causality is to carry out a cross-lagged correlational analysis of longitudinal data. Such an analysis compares the correlation between two measures across a particular time period in the two possible causal directions (i.e., the correlations are calculated and compared between early x and later y and early y and later x). According to the logic of cross-lagged correlations (e.g., Crano & Mellon, 1978), the correlation should be stronger in the causal than in the noncausal direction. In other words, if verbal STM ability is the causal factor in the developmental relationship between nonword repe- tition and vocabulary, one would expect a stronger prediction from nonword repetition at the first assessment to vocabulary 1 year later than the reverse pattern. Gathercole, Willis, Emslie, and Baddeley (1992) applied a cross-lagged correlational analy- sis to data obtained in a longitudinal study of 80 children tested on three occasions between 4 and 8 years of age (see also Gathercole & Baddeley, 1989) and yielded results that were consistent with the phonological loop hypothesis. Nonword rep- etition at age 4 was found to be significantly associated with vocabulary test scores 1 year later (partial r = .38, p .001, with variance associated with age and nonverbal ability con- trolled), whereas the vocabulary measure at age 4 was not a significant predictor of nonword repetition scores at age 5 (par- tial r= .14, p .05). Although such a pattern does notprovide watertight evidence for causation, it certainly lends further sup- port to the view that ability to repeat nonwords influences learn- ing of new words. Research carried out by Service and colleagues on Finnish children learning English at school has extended the link be- tween ability to repeat nonwords and word learning to the acqui- sition of foreign language vocabulary. The original group of children studied by Service (1992) started to learn English at school at 9 or 10years. Before commencing the English course, the children were given a series of cognitive tests, one of which involved repeating pseudo-English nonwords. The measure of nonword repetition accuracy proved to a very strong predictor of English language learning when it was tested 2 years later. Further longitudinal analysis of Finnish children learning En- glish as a foreign language has provided more direct evidence that the children's later success at acquiring English is princi- pally mediated by a direct link between repetition ability and vocabulary acquisition (Service & Kohonen, 1995). Similar results have recently been reported in a study of 12-year-old children learning English as a second language (Cheung, 1996). The data reviewed so far have established a close, natural association between children's phonological loop abilities and their knowledge of native vocabulary. Correlational studies of this kind are inevitably prey, however, to a number of important limitations. One such limitation is that studies of naturalvocabu- lary learning do not permit close control of the word-learning opportunities of individual children. Could it therefore be the case that individuals with both good phonological loop skillsand vocabulary knowledge are simply exposed to richer linguistic environments at home and that the greater variety of linguistic forms experienced will boost any language-related ability? Another limitation concerns the specificity of the hypothe- sized causal relationship between the phonological loop and word learning. Although we have assumed that the loop serves to support the immediate retention and eventual learning of the novel phonological form of new words, the data from studies of children reviewed so far merely establish a link between loop function and ability to demonstrate knowledge of the meaning of a spoken word. Is it really the case that learning the sound of a new word taxes the phonological loop, or is it linked with all aspects of word learning, phonological and nonphonological? If so, the theoretical account of the relationship, which is that the temporary representation of the novel phonological form provided by the phonological loop provides the basis for the construction of a more enduring phonological specification, would clearly require substantial modification. At least some of these concerns have been laid to rest by experimental studies of word learning in children. In an initial study, Gathercole and Baddeley (1990a) tested the abilities of 5-year-old children of either high or low nonword repetition ability (matched on a measure of nonverbal ability) to learn new names of toy animals. Across 15 trials, the experimenter named four toys and tested the children's memory for these names. The toys were either given familiar names such as Peter and Michael or phonologically unfamiliar names such as Pye- mass and Meeton (constructed from the same phonological pool as the familiar names). The findings were clear: The children with the low nonword repetition scores were significantly poorer at learning the phonologically unfamiliar names than the high- repetition children. In contrast, there was no reliable difference in the rates at which the two groups of children learned the familiar names. These results provide some reassurance that new-word learn- ing is indeed linked to phonological memory skills, even when environmental exposure to new words is controlled across sub- jects. Furthermore, the specific pattern of findings, in which phonological loop function is significantly related to children's abilities to learn nonwords but not words, has turned out to be highly characteristic of studies of STM and long-term learning and is the signature of many of the studies discussed elsewhere in this article. Similar findings were obtained in a recent study of phonological memory and word learning in 65 5-year-old children (Gathercole, Hitch, Service, & Martin, 1997). The principal concern in this study was to investigate the specificity of the association between phonological loop function and the learning of the phonological form of new words. The children were tested, in separate sessions, on their abilities to learn either pairs of familiar words such as table-rabbit or word���nonword pairs such asfairy -bleximus. The main finding was that phono- logical loop ability in this sample of children, as indexed by their scores on nonword repetition and digit span tasks, was highly associated with rate of learning the word���nonword pairs (r = .63, p .001) but not with word pair learning (r = .23, p .05). Even after variance attributableto differences in age,

PHONOLOGICAL LOOP AND LANGUAGE LEARNING 161 nonverbal ability, and vocabulary knowledge were taken into account, the partial correlation between phonological memory and word-nonword learning remained strong (partial r = .49, p .001) the corresponding partial correlation between mem- ory and word pair learning diminished further to .07, p .05. Thus, ability to learn to associate pairs of familiar words was quite independent of phonological loop function. In contrast, the children's ease of learning new words was strongly constrained by their phonological loop capacity. A similar finding emerged from a recent study of experimental word learning by Michas and Henry (1994), in which young children were taught the names of three new words, such as gondola, platypus, and minstrel. An important degree of speci- ficity to the memory���vocabulary association established by Mi- chas and Henry was that it was independent of spatial memory skill. Further explorations of the developmental relationship be- tween nonword repetition in particular and vocabulary acquisi- tion indicate that it is oversimplistic to claim that the phonologi- cal loop mediates long-term phonological learning in a unidirec- tional manner. Instead, vocabulary knowledge, phonological loop capacity, and nonword learning share a highly interactive relationship. There is accumulating evidence that, for at least some nonwords, the task of nonword repetition taps both the phonological loop and knowledge about the structure of the native language. This fact is demonstrated most simply by the finding that children are reliably more accurate at repeating nonwords that are high in degree of rated wordlikeness (Gath- ercole, 1995a: Gathercole, Willis, Emslie, & Baddeley, 1991). So even though the stimuli are by definition nonlexical, it ap- pears that children are drawing on their knowledge of either specific familiar words in the language or generalized knowl- edge of the statistical properties of the language to support the repetition of the novel sound pattern. The sensitivity of nonword repetition to word likeness pro- vides an important clue as to the relationship between phonolog- ical loop function and vocabulary knowledge. It explains why nonword repetition is more highly correlated with vocabulary knowledge than digit span (see Table 1) the reason is that the repetition task itself draws to some degree on the child's vocabulary knowledge and on reflecting phonological loop con- straints. For wordlike nonwords, the contribution of long-term knowledge will probably reduce the phonological loop contribu- tion to repetition and hence the sensitivity of the task to phono- logical loop constraints. Other evidence also points to a highly interactive relationship between the phonological loop, language knowledge, and long- term learning of the sounds of new words. In the study discussed earlier by Gathercole, Hitch, Service, and Martin (1997), in which experimental word-learning tasks were used as a means of assessing the cognitive components in vocabulary acquisition, speed of learning was correlated to a highly significant extent with children's vocabulary knowledge, even after shared vari- ance with phonological STM had been partialed out. Thus, learning the sounds of new words appears to be mediated by both the phonological loop and long-term knowledge of the native language. The combination of these two types of learning support yields a highly flexible word-learning system in which, where possible, the capacity constraints of the phonological loop are offset by the use of stored knowledge about the language (Gathercole & Martin, 1996). In summary, evidence from studies of children indicates that the phonological loop mediates the long-term phonological learning involved in acquiring new vocabulary items. This role appears to be particularly significant when the novel phonologi- cal forms to be learned have highly unfamiliar sound structures. Experimental Word Learning and the Phonological Loop in Adults If the phonological loop is important for acquiring new vocab- ulary, it should be possible to hinder such acquisition by interfer- ing with the operation of the loop. Any given manipulation may of course be regarded as affecting several underlying variables, and it could be one of these rather than phonological storage that plays a crucial role in vocabulary acquisition. It is at this point that a coherent model of the phonological loop, which is tied to well-explored experimental phenomena, becomes partic- ularly valuable. Several quite distinct variables share a known impact on the phonological loop if each of these has a corre- sponding influence on the learning of unfamiliar phonologically novel vocabulary items, it becomes much harder to provide an alternative account of the results. Accordingly, the three sets of experiments described later study the influence on vocabulary acquisition of variables that are known to influence the operation of the phonological loop in clearly specified ways. Detailed accounts of these empirical phenomena in terms of the phono- logical loop model are provided elsewhere (e.g., Baddeley, 1986, 1997). Briefly, the effective capacity of the phonological loop is diminished when list items have long names rather than short names, the word length effect (Baddeley, Thomson, & Buchanan, 1975), have names that are phonologically similar to one another, the phonological similarity effect (Conrad & Hull, 1964), and when participants are required to engage in irrelevant articulation during presentation of the memory list, the articulatory suppression effect (Murray, 1967). Although the word length and articulatory suppression effects appear to be located in the rehearsal process, the source of the phonological similarity effect is believed to be the phonological store (e.g., Baddeley, 1986). The important question for the hypothesis that the phonological loop mediates the long-term learning of the sounds of new words is the following: Do these three variables also influence phonological learning? In an initial series of experiments, Papagno, Valentine, and Baddeley (1991) studied the effect of articulatory suppression on the acquisition rate of pairs of familiar words and items of a foreign language vocabulary by normal participants. It is important to note here that articulatory suppression places mini- mal demands on executive processes but has a precise effect on the capacity for phonologically encoding visually presented material and for actively maintaining it by rehearsal. It thus has no effect on performance for patients such as P.V. who do not use this mode of encoding (Baddeley, Papagno, & Vallar, 1988), but it does remove the phonological similarity effect with visual though not auditory presentation, in line with the phonological loop model (Murray, 1967). The participants in the initial exper- iments reported by Papagno et al. were native Italian speakers, and they were asked to learn two types of material. The first