AUDITORY MEMORY AND PROFICIENCY OF SECOND LANGUAGE SPEAKING: A LATENT VARIABLE ANALYSIS APPROACH'·2
Summary.-Previous studies of second language aptitude have mainly used verbal stimuli in memory tasks. Memory for musical stimuli has not been used in aptitude studies although music and language have structural similarity. In this slUdy, 30 Japa· nese university students who speak English as a second language (19 men, M = 21.3 yr., SD = 1.8) panicipated in the experiment as volunteers. They performed verba! memory tasks, musical memory tasks, and English pronunciation tasks. Factor analysis indicated that verbal and musical memory abilities are beller represented as a unitary factor rather than two independent factors. Further, a path analysis supponed the hy. pothesis th at the memory for both verbal and musical tasks affects proficiency of sec· ond language pronunciation, including prosodic features such as stress in word or in· tonation through a couple of sentences. The memory factor was interpreted as reflect· ing the performance of "auditory working memory."
It is weIl known that individual differences in second language proficiency are very large. Many researchers have conducted studies and tried to explain individual differences in second language proficiency (e.g., Harrington & Sawyer, 1992; Miyake & Friedman, 1998). But, how do the individual differences occur? Cognitive abilities of domains other than the languagespecific module might help adults to learn second language. Therefore, some cognitive ability might explain the individual differences in second language proficiency. One such candidate is memory span. Were memory span related to learning a second-language, there would be positive correlation between memory span and second-language achievement. ActuaIly, Miyake and Friedman (1998) reported that the individual differences in second language proficiency are correlated with memory span. Individual differences in second language proficiency can be found in several subdomains, e.g., grammar, vocabulary, pronunciation, listening, writing, ete. It is obvious that these individual differences have a different cognitive basis. In th is study, proficiencies in pronunciation are the focus. Previous studies of individual differences in pronunciation proficiency have focused mainly on segmental features of speech. On the other hand, suprasegmental features or prosody, i.e., pitch, loudness and duration in speech, have lWe thank Tomoyuki Nemoto for his assistance in data collection. 'Address correspondence to Akihiro Tanaka, Research Institute of Electrical Communication, Tohoku University, 2-1-1 Katahira, Aoba-ku, Sendai 980-8577, Japan or e-mail (
[email protected]. tohoku.ac.jp) .
been less examined. Prosody, as weil as segmental features, is an important part of speech. Prosody conveys state/question distinction, attitude, and emotional states of speakers, etc. When the prosodic features in an utterance of a less proficient second-Ianguage speaker of English are exchanged with those of a native speaker, the utterance was rated as more native-like (Waibel, 1988). Therefore, pronunciation proficiency measures should include a prosodic measure. Prosody in speech shares structural similarity with music. For example, intonation of a sentence is similar to melodic contour of music in that both have temporal change in fundamental frequency and rhythmic structure. Although similarity of structures does not necessarily require similarity of cognitive processes, it is possible that language processing and music cognition share the same cognitive processes (Patel, Peretz, Tramo, & Labreque, 1998). Then, what cognitive processes might language and music share? One possibility is that language and music share the short-term memory system or working memory because processing of both requires temporal storage of relevant information. Working memory temporarily stores and manipulates information required during cognitive tasks (Baddeley, 1986). For example, sentence comprehension requires the phonemes and words which constitute the sen ten ce to be held in the memory while semantic representations of the sentence are formed. Similarly, melodic processing requires each tone which constitutes the melody, to be held in the memory while melodic representation is formed. Baddeley's working memory model does not have a component that temporarily stores and manipulates musical information. However, Pechmann and Mohr (1992) showed that pitch information in musical tones is stored in atonalloop. Although the relationship between Baddeley's model and Pechmann and Mohr's model remains unclear, at least, working memory seems to hold musical information. Thus, it might be meaningful to include musical tasks in addition to linguistic tasks in memory measures in studies of second language aptitude. The possibility that language and music share the memory system suggests a hypothesis that verbal and musical memory abilities have a certain effect on second language pronunciation proficiency, including prosodic features such as stress or intonation. There are some studies that support th is hypothesis. Patel, el al. (1998) investigated an "amusic" patient with brain damage in left primary cortex and right prefrontal cortex. She had difficulty with both linguistic and musical tasks. Level of performance was similar across linguistic and musical domains. The linguistic tasks were intonational, accentual, and temporal discrimination. All of the linguistic tasks required short-term retention of speech prosody. The musical task required retention of tone pitch. Patel, el al. (1998) concluded that prosodic and musical processing share auditory working memory resources.
Another possibility is th at there are distinet memory systems for language and musie. Salame and Badde1ey (1982) showed that music does not interfere with verbal short-term memory. This suggests another two hypotheses. One is that verbal and musical memory are divided into two distinct factors, and one of them has a certain effect on proficiency in a second language pronunciation. There are many studies which showed the contribution of verbal memory for second language learning (for a review, Badde1ey, Gathercole, & Papagno, 1998). The other is that verbal and musical memory abilities are divided into two distinet factors, and both of them have respective effe cts on proficiency of second language pronunciation. The purpose of this study is to test the hypotheses with latent variabIe analysis, a statistical method to investigate the nature of latent variables that were assumed to explain the complex phenomenon by analyzing the multiple measures simultaneously. More precise1y, latent variabIe analysis deals not only with the corre1ations among the observed measures themse1ves but also with the latent structure that would produce these corre1ations. Thus, hypotheses about the latent structure of multiple measures can be tested. In this regard, the latent variabIe analysis approach is superior to the use of zero-order corre1ations or multiple regressions (Miyake, Friedman, Rettinger, Shah, & Hegarty, 2001). For memory measures, predictions from the hypotheses are as follows: If language and music share the memory system, verbal and musical memory measures would be better represented as a unitary factor than two dis tin ct factors. In contrast, if there are distinct memory systems for language and music, verbal and musical memory measures would be better represented as two distinct factors than as a unitary factor. For measures of second language proficiency, predictions are as follows: If all of the three measures reflect the same ability, the segmental, stress, and fluency scores would be better represented as a unitary factor than distinct factors. In contrast, if the th ree measures reflect different abilities, the segmental, stress, and fluency scores would be better represented as multiple factors than as a unitary factor. Finally, on the re1ationship between measures of memory and second language proficiency, the prediction from the hypotheses is as follows: If memory factor(s) have certain effects on proficiency of second language pronunciation, path coefficient from memory factor(s) to second language pronunciation proficiency factor(s) would be significant. Participants The participants were 30 undergraduate students at the University of Tokyo (19 men, M =21.3 yr., SD= 1.8). All were native speakers of ]apanese and learned English in junior and senior years of high school. Years of
learning English ranged between 9 and 16 years (M = 9.7, 5D = 1.7). None of the students majored in a music-related field. Musical experience of the participants ranged between 0 and 20 yr. (M=9.2, 5D=6.l). Materials and Procedure Measures of verbal memory.-Two measures we re used to estimate verbal memory ability: reading span and letter span. Reading span assesses working memory capacity, specifically in the trade-off between its processing and storage functions during language processing (Daneman & Carpenter, 1980). The Japanese version of th is task (Osaka & Osaka, 1994) was used because it is widely applied to measure the verbal working memory. In this task, participants read out aloud a set of unrelated sentences (each displayed on CRT) one at a time and recalled a specified word of each sen ten ce at the end of the set. Sen ten ces were presented in sets of increasing size, starting from two to five sen ten ces per set. Simple memory span for spoken letters was also measured. The sequences of letters were randomly generated. Sequences for measuring letter span were constructed from the set of seven consonants (B, J, H, Q, L, R, and Sj. Three sequences were used for each length (shortest leng th was 3, maximum length was 10). Participants were instructed to memorize the letters and repeat aloud the sequence after the experimenter finished reading aloud a sequence. The experimenter read the sequences one letter per second. The test continued until the participant made two errors. The mean number of the last three sequences repeated aloud correctly was recorded as letter span. This procedure is of ten used to measure capacity of verbal short-term memory. Musical memory measures.-Two measures we re used to estimate musical memory, melody discrimination and rhythm discrimination. Although "musical memory" corresponds to a wider concept than melody discrimination and rhythmic pattern discrimination, these two measures were chosen because structurally there is similarity between the linguistic and musical domains. Both language and music have "contour" and "rhythm." In these tests, percentages of correct responses were recorded. In linguistic tasks, normal individuals can usually report orally what they maintain in the shortterm memory. In contrast, in musical tasks, they cannot necessarily report orally what is stored in the short-term memory even if they really hold the musical stimuli correctly. For this reason, in this study, the two musical measures were recognition tasks while two linguistic measures were recall tasks. Melody discrimination was a measure of short-term memory for melody. Participants listened to the norm melody through headphones and then compared the melody to the comparison melodies, which were presented successively. Each melody stimulus was a pure tone sequence (number of tones ranged from 5 to 9). Duration of each tone was 500 msec. There were no
silences between the neighboring tones. Thus, the actual duration of the melodic sequences ranged from 2500 to 4500 msec. The duration of the interval between norm and comparison melody was 1000 msec. The pitches of the tones used were C4, D4, E4, F4, G4, A4, B4, and C5. The first and the last tones were identical in standard melodies. Other tones were randomly sequenced. Thus, the melodies followed a diatonic scale. Rhythm discrimination was a measure of short-term memory for rhythm patterns. Each rhythmic pattern consisted of 7 to 11 slots. Duration of each slot was 500 msec. Each slot was filled with either pure tone (pitch=A4) or silence. The first and the last slots were filled with A4 tone. In other slots, two, three, or four slots were randomly filled with silences whereas the remaining slots were filled with A4 tones. Participants listened to the norm rhythmic patterns and th en compared the pattern to others which were presented successively. The procedure was the same as that for the melody discrimination. English pronunciation proficiency measures-.Three measures were used to estimate proficiency of English pronunciation: Segmental Score, Stress Score, and Fluency Score. On the Word reading test, participants read 40 English words aloud successively. The words were selected from a standard text for Test of English as a Foreign Language (Matthiesen, 1999). The number of syllables within a word ranged from 1 to 4. Their utterances we re recorded by a minidisk recorder. The recorded utterances were rated later for accuracy of segmental features of each syllable (segmental score) and accuracy of the position of stress in each word (stress score) by a native speaker of English who teaches English at a ]apanese college. For segmental scores, the pronunciation within each syllable was rated either as correct (l point) or incorrect (0 point). For single-syllabic words, the above score was used. For multisyllabic words, the sum of scores for each syllable was divided by the number of syllables within the word. Thus, the possible maximal score was 40, and the minimum was 0 for segmental score. For stress scores, the positions of stress within each word were rated either as correct (1 point) or incorrect (0 point). This score was not rated for single-syllabic words. Thus, the possible maximal score was 30, and the minimum was 0 for stress score. Stress is one of the suprasegmental features in speech. Thus, stress scores represent a part of prosodie proficiency. For a single rater, the intrarater reliabilities of segmental and stress scores were .88 and .91, respectively. In a Paragraph reading test, partieipants read th ree English paragraphs aloud. The paragraphs were selected from a standard text for Test of English as a Foreign Language (Matthiesen, 1999). Each paragraph was read twice: first without any practice and secondly after plenty of practice. Their utterances were recorded on a minidisk recorder. The recorded utterances
were rated later by a native speaker in terms of global "fluency." Special care was paid that ratings included not only segmental features but also prosodie features such as intonation and rhythm (Fluency Score). The rating ranged from 1 (corresponding to "completely different from native speaker") to 7 (corresponding to "the same as native speaker"). Rating was conducted for six recordings per participant (three English paragraphs, twice each). Thus, the possible maximal score was 42, and the minimum was 6. Because special care was paid that rating included not only segmental features but also prosodie features, fluency scores can be regarded as representing a part of prosodie proficiency. The intrarater reliability of fluency scores was not calculated because the scores for each person were obtained only six items. RESULTS
The mean percentage of correct responses and standard deviations (in addition, their range) are shown in Table 1. Pearson product-moment correlation coefficients among aH the variables are also presented in Table 1. T 0 test the th ree hypotheses, two analyses were performed; (1) a series of factor analyses to examine the structure of the latent variables and th en (2) the latent variable analysis to test the hypotheses on the relationship between memory and second language factors. AH analyses were done using EQS program (Bentler, 1995). TABLE 1 MEANs,
STANDARD DEVIATIONS,
AND PEARSON CORRELATION
Target Measure
M
5D
Range
Melody discrimination Rhythm discrimination Letter span Reading span Stress score Segmental score FJuency score
87.14 85.49
7.64 8.09 124 0.86 3.29 1.79
100-76 97-70 9--4
MATRIX POR ALL MEAsuRES
2 1. 2. 3. 4. 5. 6. 7.
5.97 2.92 23.93 36.17 30.26
3.50
5-2 29-14 39-32 36-22
.42" .55t .42'"
.41"
.33 .07 -.09 .44'" .20 .42'" .23
JO .11 -.19 .42'" .23
J3
-.05
.72t J4
.45"
;'p< .05. tp< .01.
Factor Analysis A series of factor analyses were performed to examine the structure of the latent variables. First, an exploratory factor analysis was performed on the seven measures with maximum-likelihood procedure and rotated to final solution with a promax rotation, which is an oblique correlated solution. On the basis of consideration of the number of dependent variables, one-, two-, and three-factor solutions were computed, and th en these three solutions were evaluated by examining a combination of the fit indexes and the infor-
mation criteria. The fit indexes included goodness-of-fit index (GFI) and adjusted GFI (AGFI). These indexes can range from 0 to 1, and the high scores of these indexes indicate the good-fit model. The information criteria included Akaike's Information Criterion (AIC) and consistent AIC (CAIC). The model is chosen to give the lower scores of AIC and CAle. The fit indexes and information criteria of the th ree models are shown in Table 2. From this table, the two-factor solution seems best in terms of the fit indexes and the information criteria. The two-factor solution is superior to the other two solutions in AIC and the fit indexes. Especially, the value of GFI is over .90, suggesting that the two-factor solution fits data very weil. In contrast, although- for the one-factor solution CAIC is superior to estimates in the other two solutions (-17.69; two-factor solution= -11.71, three-factor solution= 13.61), the GFI and AGFI estimates are very low (GFI=.75, AGFI=.46). Information criteria of the three-factor solution are greatly inferior to those in the other two solutions, so the two-factor solution was adopted. TABLE 2 FIT INDICES
Akaike's Information Criterion I -Factor 2-Factor 3-Factor
I3.52 -2.11 6.40
OF EXPLORATORY
FACTOR ANALYSIS MODEL
Fit Indexes Consistent Goodness-of-fit Akaike's Index Information Criterion -17.69 -11.71 13.61
"'Goodness-of-fit Index and Adjusted Goodness-of-fit computed because the degree of freedom is zero.
0.75 0.95 Index
in 3-factor
Adjusted Goodness-of-fit Index 0.46 0.64 ~': model could not be
Based on the results of exploratory factor analyses, a series of confirmatory factor analyses were computed to ex amine whether the two-factor solution was better than the other solutions. The two-factor and three-factor models were computed for the confirmatory factor analysis. Both models are shown in Fig. 1. These models we re evaluated with the fit indexes, and the information criteria used in the exploratory factor analyses. The difference between the two-factor model and three-factor model is whether they explained the verbal memory measures and the auditory memory measures with one or two factors. The two-factor model explained both the verbal memory measures and the musical memory measures with only one factor; the three-factor model explained these measures with two factors, one corresponding to the verbal memory measures and the other correspon ding to the musical memory measures.
FIG. 1. Path models evaluated in the confirmatory factor analyses. la shows the two-factor model and 1b shows the three-factor model. MDT = melody discrimination test, RDT = rhythm discrimination lest, LST = letter span test, RST = reading span test, STR= stress score, SEG = segmental score, and FS = fluency score. Factors in the three-faclOr model were not interpreted because the two-factor model was chosen.
The results of the fit indexes and the information criteria are shown in Table 3. The two-factor model was superior to the three-factor model in the fit indexes and the information criteria. It was concluded that the results of the confirmatory factor analyses also supported the two-factor model. The two-factor model is shown in Fig. la (also, three-factor model in TABLE 3 FIT INDICES
Akaike's Information Criterion 2 Factor 3 Factor
-2.31 3.28
OF CONFIRMATORY
FACTOR ANALYSIS
Fit Indexes Consistent Goodness-of-fit Akaike's Index Information Criterion -28.74 -15.93
0.86 0.86
Adjusted Goodness-of-fit Index
Fig. lb). The factor shown on the left side is composed of the verbal memory measures and the musical memory measures. This factor was interpreted as "Auditory Memory" because both kinds of measures reflect the process of listening to and maintaining auditory information. The other factor was composed of English pronunciation proficiency measures. This factor was interpreted as "Second Language Pronunciation Proficiency" because these measures reflect the process of English pronunciation. Latent Variabie Analyst!; Based on the factor analysis, path analysis of latent variables was performed to examine whether Auditory Memory could be modeled as affecting Second Language Pronunciation Proficiency. As aresult, path coefficient from Auditory Memory to Second Language Pronunciation Proficiency was significant (.64, p < .Ol). These results supported the hypothesis that verbal and musical memory affects second language pronunciation proficiency. DISCUSSION
The results of latent variabIe analysis were consistent with the hypothesis that verbal and musical memory have a certain effect on proficiency of second language pronunciation, including prosodie features such as stress or intonation. The factor analysis supported a two-factor solution, and the path analysis showed th at auditory memory was related to proficiency of second language pronunciation. These results indicate that auditory memory was shared by both verbal and musical materiaIs. The results also indicate that the auditory memory has a certain effect on proficiency of second language pronunciation, including prosodie features. The exploratory factor analysis showed that the four memory measures, two for verbal and two for musical stimuli, can be interpreted as consisting of one common factor. It is noteworthy that two musical measures were recognition tasks, while two linguistic measures were recall tasks. There were methodological differences between musical and linguistic measures. Ir is possible that the methodological difference divides the four measures into two independent factors, one reflects recognition process and the other recall process. Nonetheless, the fit indices we re higher for the model that explains these four variables with only one factor. This further confirms a common memory process shared by linguistic and musical domains. As all the four varia bles include the process of listening to and maintaining auditory information, this factor was interpreted as auditory memory. As the Reading span was not developed to measure the auditory memory capacity selectively, one might argue that this test is not related to auditory modality. However, the Reading span includes some auditory modality-specific processes: the process of listening to the auditory information the participant produces when reading aloud.
The exploratory factor analysis also showed that the three pronunciation measures can be interpreted as one common factor. As ail th ree variab les include pronunciation of English, the factor was interpreted as proficiency of second language pronunciation. Of course, complete proficiency of second language pronunciation could not be represented only by the th ree scores obtained by this study. However, these measures include both segmental and suprasegmental features which are two major phonetic attributes. In addition, these measures cover from local (word) through global (paragraph) levels. For these reasons, it was considered that the Second Language Pronunciation Proficiency factor weil represents a general proficiency of second language pronunciation. The path analysis indicated that learning to speak a second language is based on auditory memory. Baddeley (1986) assumed that the phonological loop, a verbal component of working memory, maintains only verbal stimuli. The phonological loop model focuses only on segmental features of verbal stimuli. However, second language learners cannot imitate prosodie features without using a kind of memory system that maintains prosodie features. In fact, path analysis suggested that the range of stimuli the auditory memory factor reflected is wider than the phonologicalloop model (Baddeley, 1986) assumed. Thus, this study may be regarded as documenting another example of the theoreticallimitation in Baddeley's model. What kind of memory processes does the factor of auditory memory reflect? This factor includes musical memory so the auditory memory factor must reflect wider memory processes than the phonological loop model. One possibility is the auditory memory factor reflects the function of the phonologicalloop and the tonalloop; see Fig. 2. Prosodie features are maintained within the tonalloop. The current results are consistent with the above possibility. There is some evidence that prosodie features of speech are maintained in the same mechanism as musical tones (Patel, el al., 1998; Tanaka & Takano, 2004). Several studies have shown that pitch of musical tones is actively rehearsed in the tonalloop (Pechmann & Mohr, 1992; Tanaka & Takano, 2002). Also passive storage for auditory materials is unitary whether it is for speech or for musical tone (Semal, Demany, & Ueda, 1996). J akobson, Cuddy, and Kilgour (2003) recently proposed another model for auditory memory. They proposed a mechanism to account for the association between formal music training and proficiency of verbal recall. Their assumption was that musical and verbal abilities are mediated by auditory processing of temporal order. In the present study, the four memory measures required recognition (musical measures) or recail (verbal measures) of items (tones or words) rather than the temporal order of the items. For example, the Reading Span does not require a correct report on the order of
Coordinating Mechanism
=p"~=~=arsaIM=eChani=.m
1/
the target words. In melody discrimination, one tone in a compared melody was three degrees higher or lower than that in a standard melody. The participants could detect the difference between two melodies. without holding any information about order. Thus, the present study clarified other processes of auditory memory than J akobson, et al. (2003). In the correlations among the auditory memory measures, melodie discrimination scores correlated with the rhythm discrimination scores, the letter span scores, and the reading span scores. In contrast, scores on rhythm discrimination correlated only with letter span. As melodie discrimination scores correlated with those of aH the auditory memory measures, it is inferred that melodie discrimination includes a common process with aH kinds of auditory memory tasks. Among the measures of second language pronunciation, the stress score did not correlate with fluency scores while the segmental score correlated with fluency scores. At first sight, th is result might seem odd because both scores on stress and fluency need suprasegmental accuracy. However, these tasks require different parts of suprasegmental processing. The stress score reflects local suprasegmental processing, while the fluency score reflects global suprasegmental processing. The lack of correlation between scores for stress and fluency might indicate that local suprasegmental processing differs from global suprasegmental processing. Zero-order correlations between memory and second language measures also suggest some interesting interpretations. For example, scores for melody discrimination correlated with fluency scores, supporting the idea th at the fluency score reflects global prosodie processing. When we discriminate
melodies, melodie contour plays an important wIe (Dowling, 1978). Melodie contour is global information about melody, particularly on sequence of directions of pitch change. A sentence in a second language has structure like a melodie contour. The correlation between the melody discrimination scores and fluency scores suggests that these structurally similar sequences may be processed by the same mechanism. It remains unanswered which part of musical skill affects which domain of language acquisition. For example, Mandarin Chinese is a tonallanguage in which tone, expressed by pitch change, is a phonological feature. The hypothesis prediets th at good musicians are good learners of Mandarin Chinese, especially the ton a! features. In later research, investigation using languages with various prosodie features is expected. REFERENCES BADDELEY, A. D. (1986) Working memory. New York: Oxford Univer. Press. BADDELEY,A. D., GATHERCOLE,S. E., & PAPAGNO,c. (1998) The phonological loop as a Janguage learning device. Psycbological Review, 105, 158-173. BENTLER,P. M. (1995) EQS structural equations program manual. Encino, CA: Multivariate Software.
DANEMAN,M., & CARPENTER,P. A. (1980) Individual difference in working memory and reading. Journalof Verbal Learning and Verbal Bebavior, 19,450-466. DOWLING,W ]. (1978) Scale and contour: two components of a theory of memory for melodies. Psycbological Review, 85, 342-354. HARRINGTO , M., &SAWYER,M. (1992) L2 working memory capacity and L2 reading skill. Studies in Second Language Acquzsition, 14,25-38. ]AKOBSON,L. S., CUDDY, L. L., & KILGOUR,A. R. (2003) Time tagging: a key to musicians' superior memory. Music Perception, 20, 307-313. MATIHIESEN,S. J. (1999) Essential words for tbe TOEFL. (2nd ed.) Hauppange, NY: Barron. MIYAKE,A., & FRIEDMAN, P. (1998) Individual differences in second language proficiency: working memory as language aptitude. In A. Healy & 1. Bourne (Eds.), Forezgn language learning: psycbolinguistic studies on training and retention. Mahwah, NJ: Erlbaum. Pp. 339-365. MIYAKE,A., FRIEDMAN,N. P., RETTINGER,D. A., SHAH, P., & HEGARTY,M. (2001) How are visuospatial workinlS memory, executive functioning, and spatial abilities related? A latent variable analysis. journalof Experimental Psycbology: General, 130, 621-640. OSAKA, M., & OSAKA, N. (1994) [Reading and working memory span: measurement by Japanese version of the reading span test]. [Japanese Journal oj Psycbology], 65, 339-345. [in J apanese] PATEL,A. D., PERETZ, 1., TRAMO, M., & LABREQUE,R. (1998) Processing prosodie and musical patterns: a neuropsychological investigation. Brain and Language, 61, 123-144. PECHMANN,T., & MOHR, G. (1992) Interference in memory for tonal pitch: implications for a working-memory model. Memory & Cognition, 20, 314-320. SALAME,P., & BADDELEY,A. D. (1982) Disruption of short-term memory by irrelevant speech: implications for the structure of working memory. Journalof Verbal Learning and Verbal Bebavior, 21, 150-164. SEMAL,c., DEMANY,L., & UEDA, K. (1996) Speech versus nonspeech in pitch memory. Journal of tbe Acoust/cal Society of America, 100, 1132-1140. TANAKA,A., & TAKANO,Y. (2002) [Mechanism of active rehearsal for pitch information: an examination by dual-task paradigm]. [Journalof Music Perception and Cognition], 8, 81-91. [in J apanese with English abstract] TANAKA,A., & TAKANO,Y. (2004) Rehearsal mechanisms for pitch information in spoken language and musical tone. Paper presented at the 2nd International Conference on Working Memory (2004, August, Kyoto, Japan). WAlBEL,A. (1988) Prosody and speecb recognition. London: Pitman.
