Method.

Participants. Participants in Experiment 1 were 60 native speakers (NSs) of English with no previous formal instruction in Spanish. All of the participants were undergraduate psychology students at a private Midwestern university in the United States. Participants received credit related to requirements of their psychology class for their participation.

Experimental words. The experimental words used were 24 Spanish concrete nouns, divided into 3 groups. Each word group had a total of 20 to 21 syllables. Words 1 through 8 were hongos “mushrooms,” oso “bear,” reloj “clock,” tijeras “scissors,” sandía “watermelon,” cuaderno “notebook,” ardilla “squirrel,” and cinta “tape.” Words 9 through 16 were fresas “strawberries,” tiza “chalk,” caballo “horse,” manzana “apple,” cebolla “onion,” elote “(ear of) corn,” lechuga “lettuce,” and grapadora “stapler.” Words 17 through 24 were loro “parrot,” pato “duck,” lápiz “pencil,” conejo “rabbit,” gato “cat,” naranja “orange,” basurero “garbage can,” and pez “fish.” The experimental words were the same as those used by Barcroft (2001). In the present study, however, the words were grouped so that semantic categories (animals, fruits and vegetables, classroom items) were mixed and so that the average number of syllables in each word group was approximately the same. The average number of syllables was 2.5 for words 1 through 8, 2.63 for words 9 through 16, and 2.63 for words 17 through 24.

All of the target words had been previously recorded by a male NS of Mexican Spanish in a soundproof room. Stimuli were recorded using a sampling rate of 44.1 kHz and 16-bit resolution. The NS was instructed to read the list of the 24 target words in 4 different voice types: neutral voice, excited voice, whispered voice, and nasal voice. To obtain 2 additional voice types for each target word, all of the 24 words produced in neutral voice were subsequently digitally altered using commercially available sound-editing software to produce high-pitched and elongated varieties of each word. For the high-pitched modification, the fundamental frequency of each neutral production was increased 53% without altering any other temporal parameters of the speech signal. In the case of elongated words, duration was increased 53% without altering the fundamental frequency. The RMS amplitude of all productions was equated using digital signal processing software. This process yielded six different voice types for all of the words (neutral, nasal, elongated, whispered, excited, and pitch-shifted), in which the RMS amplitude was equated across voice type.

A separate test was conducted to ensure that the productions were highly intelligible to Spanish NSs. Four NSs of Spanish heard all 144 productions (24 items × 6 voice types) presented over headphones at approximately 75 dB sound pressure level and were asked to write down the word that was presented. Presentation order was randomized for all participants. Scoring was based on exact phonetic matches (i.e., adding, deleting, or substituting a single phoneme was counted as an incorrect response). Mean identification accuracy was 99.3% (SD = 0.4), indicating that all of the productions were highly intelligible to NSs of Spanish.

We conducted two additional pilot studies to determine whether naïve L2 listeners perceived six distinct voice types and to assess the quality of each production as an example of a given voice type. In the first study, all of the words (24 tokens for each of 6 voice types) were presented in random order to 4 speech-language pathologists. The four participants had a minimum of 11 years experience as licensed speech-language pathologists, and none had previous formal instruction in Spanish. Each token was presented individually. Participants were first asked to assign the production to one of the six voice-type categories. After this decision, they were asked to rate “how good an example” the token was of their selected voice type, using a seven-point scale. Correct categorization (categorizing consistent with the voice type designated at production) exceeded 95% for all four participants. Mean category goodness ratings (7 indicating an ideal example; 1 indicating a poor example) were as follows: nasal = 6.7, whispered = 6.9, excited = 6.5, neutral = 6.7, elongated = 6.9, and pitch-shifted = 6.6. In the second study, we examined whether results with the speech-language pathologists would generalize to a group of participants similar to those who would participate in the proposed experiments. Ten young adults from the student population at a private Midwestern university were asked to perform the voice-type categorization task. Mean correct categorization for the group of students exceeded 96%, with three participants obtaining perfect scores. Taken together, the results of these two pilot studies indicate that the six voice types were easily distinguished and that the productions represented good examples of each voice type.

Procedures. Participants were tested individually in a quiet room according to the following procedures. In the learning phase, each participant sat in front of the computer screen and viewed six repetitions of 24 pictures while hearing the spoken form of the Spanish word for each picture. Each picture appeared on the screen for 5 seconds. Each target Spanish word was spoken 750 milliseconds (ms) after the picture appeared. The picture remained visible for the entire 5 seconds. Each participant learned eight words in each of the three learning conditions: (a) no variability—or six repetitions of neutral voice; (b) moderate variability—or two repetitions each of neutral, nasal, and elongated voices; and (c) high variability—or one repetition each of neutral, nasal, elongated, whispered, excited, and pitch-shifted voices. Participants were informed that all of the words would be produced by a single talker although sometimes in different voice types. Word groups and presentation orders were counterbalanced across learning conditions so that 20 participants heard each of the word groups in 1 of the 3 variability conditions. Within a word group, order of presentation was constant across the six repetitions (e.g., in the no-variability condition, participants heard all six repetitions of each of the eight words repeated in the same order). In the moderate- and high-variability conditions, participants always heard the words in the same order, but the voice type associated with that item was selected randomly without replacement. We presented six repetitions of the eight words within each condition in light of prior reports (Barcroft, 2001) that indicated performance levels that did not approach either floor or ceiling using this method.

The picture-to-L2 recall posttest was administered immediately after the learning phase. All participants performed the cued recall task prior to the translation task in order to avoid additional exposure to the spoken L2 word forms (cues for the L2-to-L1 recall task were the spoken L2 word forms in nasal and neutral voices). Whereas the inclusion of L2-to-L1 recall provided a new measure for assessing the effects of acoustic variability on L2 vocabulary learning, any differences in overall performance between picture-to-L2 and L2-to-L1 recall should be interpreted in light of the order in which these two tasks were administered, given that the picture-to-L2 task made available a form of additional practice that could affect the L2-to-L1 translation task. For the L2 cued recall posttest, the participants attempted to produce the Spanish word for each picture. The picture for each target word appeared on the screen, with order determined by structured randomization. The first 3 pictures were selected at random from each of the 3 word groups in the learning phase; the second 3 words were selected at random from each of the 3 word groups based on the remaining words; the third set of 3 words was selected at random from each of the 3 word groups based on the remaining words, and so forth until all 24 words had been selected. After each set of three words was selected at random from each of the three word groups, the order in which the three words in question would appear on the posttest also was selected at random. A voice key was used to measure response latencies, which extended from the onset of the picture cue to the participant's initial vocalization of the word. The voice key was calibrated to minimize unintentional triggering while maintaining high sensitivity. Each picture appeared on the screen for a maximum of 10 seconds or until the participant provided a spoken response.

The L2-to-L1 recall posttest was administered after the cued L2 recall. For this posttest, participants heard Spanish words and were asked to provide the English translation of each word. The order of the Spanish words was determined using the same type of structured randomization used to determine the order of the picture-to-L2 recall task. Participants were given a maximum of 10 seconds to provide the appropriate English translation for each word they heard. Half the words used as cues in the translation task were presented in neutral voice. The other half of the words used as cues were presented in nasal voice. The use of neutral versus nasal voice cues was determined as follows: Within each word group (1–8, 9–16, 17–24), four words were selected at random for the neutral cue and four words for the nasal cue. Half of the participants received cues presented in this manner. For the other half of the participants, the cue types were reversed.

Scoring. The criterion used to score the picture-to-L2 recall and L2-to-L1 recall awarded 1 point for a completely correct production, 0.5 points for missing one or more of the target phonemes or using one or more incorrect phonemes within a single syllable, and 0 points for all other responses. In the case of L2-to-L1 recall, the 0.5 score was never used because participants did not make any phonetic errors in their L1.

Analyses.

All accuracy scores and RTs for correct responses were submitted to repeated-measures analyses of variance (ANOVAs). When analyzing the RT data, latencies for trials receiving a score of 0 on the accuracy measure were excluded from the analyses (i.e., only fully or partially correct trials were included in the RT analyses). We also conducted the analyses using only RTs from trials that were fully correct (i.e., that received a score of 1), and the pattern of results was unchanged from that obtained when both fully and partially correct trials were included. Latencies exceeding 2.5 standard deviations (SDs) of the mean of individual conditions were also excluded from the analyses as outliers. Additionally, as part of a questionnaire administered at the end of the experiment, participants were asked if they knew any of the Spanish words prior to the experiment. Responses to items reported as being known previously were excluded from all analyses. Percentages in these cases were based only on the other items. In Experiment 1, two participants indicated that they knew the Spanish word for “cat” was gato.

In each ANOVA, condition (no variability, moderate, high) and recall type (picture-to-L2, L2-to-L1) were treated as within-subjects variables. Accuracy and RT were the dependent variables. To examine the effect of voice type used for cueing during L2-to-L1 recall, accuracy and RT scores for L2-to-L1 recall were submitted to additional repeated-measures ANOVAs. For this analysis, condition (no variability, moderate, high) and cue type (neutral, nasal) were treated as within-subjects independent variables, and accuracy and RT were the dependent variables.

Results.

Means and SDs for accuracy and RT based on condition and recall type appear in Table 1. Results of the ANOVA for accuracy based on condition and recall type revealed that performance was significantly worse for picture-to-L2 recall than for L2-to-L1 recall, F(1, 59) = 117.63, p < .001, η² = .667. No other significant main effects or interactions were observed for the accuracy measure. The effects of acoustic variability were not significant for either accuracy of picture-to-L2 recall, F(2, 118) = 1.58, p = .21, η² = .026, or L2-to-L1 recall, F(2, 118) = 1.7, p = .17, η² = .029. Results of the ANOVA for RT revealed that listeners were significantly faster recalling from L2-to-L1 than from picture-to-L2, F(1, 59) = 13.11, p < .001, η² = .198. No other significant main effects or interactions were observed. As with the accuracy results, no significant effect of variability was observed for either the picture-to-L2 recall, F(2, 118) = 1.2, p = .28, η² = .021, or L2-to-L1 recall, F(2, 118) = 0.73, p = .49, η² = .012.

Means for accuracy and reaction time based on condition by recall type (Experiment 1)

In examining whether the effects of acoustic variability differed for the two different cue types (nasal vs. neutral) in the L2-to-L1 translation task, several participants were incorrect for all trials with a particular cue type and, therefore, the sample size was reduced for this analysis. Results of the ANOVA for accuracy revealed a significant main effect for cue type, F(1, 54) = 37.1, p < .001, η² = .407, reflecting significantly higher performance using neutral voice as compared to nasal voice as a cue for L2-to-L1 recall. Overall means were 0.73, SD = 0.20, for neutral voice and 0.61, SD = 0.20, for nasal voice. No other significant main effects or interactions were observed. Results of the ANOVA for RT revealed a significant main effect for cue type, F(1, 51) = 6.8, p < .05, η² = .117, reflecting significantly longer RTs using nasal voice as compared to neutral voice as a cue. Overall RT means were 2243, SD = 947, for neutral voice and 2726, SD = 899, for nasal voice. No other significant main effects or interactions were observed.

Discussion.

The results of Experiment 1 can be summarized as follows. As compared to neutral voice, voice-type variability did not affect L2 vocabulary learning using accuracy and latency of picture-to-L2 recall or L2-to-L1 translation as dependent measures. Although accuracy and latency of L2-to-L1 recall were not affected by acoustic variability using both neutral and nasal voice types as cues, L2-to-L1 translation performance was better when cued by a neutral voice type as compared to a nasal voice type. These results are consistent with those of Barcroft (2001) and with both the independent modulation and strong versus robust connectivity hypotheses. According to independent modulation, acoustic normalization produces L2 lexical representations that are abstract in nature (i.e., they contain lexical but not indexical information). According to the strong versus robust connectivity hypothesis, increased acoustic variation increases the number of connections between forms of a word and its referent, which facilitates L2 vocabulary learning; however, each of these connections is weaker as compared to the connection derived from multiple repetitions of a word in a single voice, which impairs L2 vocabulary learning. The combined effects of these two opposing processes result in no observable effect of acoustic variability on L2 vocabulary learning. Although the null effects for acoustic variability in Experiment 1 and the earlier Barcroft study are consistent with these two hypotheses, other types of acoustic variability warrant investigation as well. Among these other types of acoustic variability is between-talker acoustic variability, which was the focus of Experiment 2.

Method.

Participants. Participants in Experiment 2 were 60 NSs of English with no previous formal instruction in Spanish and were recruited from the same participant pool as Experiment 1. Participants received credit related to requirements of their psychology class for their participation. None of the participants from Experiment 1 took part in Experiment 2.

Experimental words. The experimental words and word groups used in Experiment 2 were the same as those used in Experiment 1. All of the target words had been previously recorded in a soundproof room by three female and three male NSs of Spanish. The female speakers were from central Spain, Chile, and Mexico. The male speakers were from northern Spain, Chile, and Mexico. Each of the NSs was instructed to read the list of the 24 target words in their normal voice. As in Experiment 1, pilot testing was conducted to determine if all of the words would be perceived correctly by NSs of Spanish. Four NSs of Spanish heard all 144 productions (24 words × 6 speakers) and were asked to write down the word that was produced. Three of the pilot participants obtained perfect scores on the test and the other missed a single item (99.3%).

Procedures. Participants were tested individually in a quiet room based on the same procedures as those used in Experiment 1 with the following exceptions. First, each participant learned eight words in each of the following three learning conditions: (a) no variability—six repetitions of each word in the voice of one speaker; (b) moderate variability—two repetitions of each word in the voice of three different speakers; and (c) high variability—one repetition of each word in the voice of six different speakers. Thus, in all three conditions, each participant heard six repetitions of each target word. Second, participants were informed that some of the words would be produced by different talkers. Next, to ensure that equal numbers of participants heard each speaker in the no-variability and moderate-variability conditions, the 60 participants were divided into 6 groups of 10 participants. In the no-variability condition, each group was assigned to the voice of only one of the six different speakers (Speakers 1 through 6). Speakers 1, 3, and 5 were men; Speakers 2, 4, and 6 were women. The assignment of speakers to the moderate-variability conditions across the six groups was as follows: Group 1 = Speakers 1, 2, 3; Group 2 = Speakers 2, 3, 4; Group 3 = Speakers 3, 4, 5; Group 4 = Speakers 4, 5, 6; Group 5 = Speakers 5, 6, 1; and Group 6 = Speakers 6, 1, 2. Note that for each group, the speaker in the no-variability condition was also included in the moderate-variability condition and that a given speaker was included in the moderate-variability condition for three of the groups. Finally, for the L2-to-L1 recall task, half of the words used as cues in the translation task were spoken by a male voice, and the other half were presented in a female voice.

Scoring. The scoring procedure was the same as in Experiment 1. For the words manzana and lápiz, however, either of two phonemes (θ or s) was accepted as correct in the place of the letter “z” in light of the regional dialects of the NSs to which the participants had been exposed.

Analyses. The analyses conducted for Experiment 2 paralleled those of Experiment 1, including deletion of outliers and inclusion of only correct or partially correct responses in the analysis of RTs. None of the participants in Experiment 2 indicated on the postexperiment questionnaire that they were familiar with any of the Spanish words. To examine the effect of speaker's voice used for cueing during L2-to-L1 recall, an additional ANOVA was conducted with condition (no variability, moderate, high) and cue type (male voice, female voice) as within-subjects variables, and accuracy and RT as the dependent variables.

Results.

Means for accuracy and RT based on condition and recall type appear in Table 2 and are depicted graphically in Figures 1 and 2. Results of the ANOVA for accuracy based on condition and recall type revealed that accuracy differed significantly across the three conditions, F(2, 118) = 108.08, p < .001, η² = .647, and for the two different recall types, F(1, 59) = 114.21, p < .001, η² = .659. Additionally, a significant Condition × Recall Type interaction was observed, F(2, 118) = 6.98, p < .001, η² = .106. To examine the source of this interaction, we analyzed the main effects of condition separately for the two types of recall. The effects of acoustic variability were significant for both picture-to-L2, F(2, 118) = 62.5, p < .001, η² = .51, and L2-to-L1 translation, F(2, 118) = 49.1, p < .001, η² = .45. Separate Tukey HSD post hoc comparisons using a Bonferroni correction indicated significant differences among all levels of acoustic variability for both types of recall (picture-to-L2 and L2-to-L1). For the picture-to-L2 recall task, differences between the no-variability and high-variability and between the medium- and high-variability conditions were both significant at p < .001. The difference between the no-variability and medium-variability condition was also significant, p < .01. For the L2-to-L1 task, differences between the no-variability and medium-variability and between the no-variability and high-variability conditions were significant at p < .001. The difference between the moderate- and high-variability conditions was also significant, p < .01.

Means for accuracy and reaction time based on condition by recall type (Experiment 2)

Effects of talker variability on accuracy of picture-to-L2 and L2-to-L1 recall (Experiment 2). Error bars represent standard deviations of the mean.

Effects of talker variability on latency of picture-to-L2 and L2-to-L1 recall (Experiment 2). Error bars represent standard deviations of the mean.

Results of the ANOVA for RT based on condition and recall type revealed significant differences in reaction times across the three conditions, F(2, 118) = 12.92, p < .001, η² = .185. No other significant main effects or interactions were revealed. Pairwise Tukey HSD analyses with a Bonferroni correction for multiple comparisons indicated the following: (a) lower RTs in the high-variability condition as compared to the no-variability condition, p < .001, (b) lower RTs in the high-variability condition as compared to the moderate-variability condition, p < .05, and (c) marginally lower RTs in the moderate-variability condition as compared to the no-variability condition, p = .06.

Results of the ANOVA for accuracy based on condition and cue type (male or female voice) in the L2-to-L1 task revealed a significant main effect for condition, F(2, 94) = 56.14, p < .001, η² = .548, but no other significant main effects or interactions. Overall means were 0.67, SD = 0.22, for male voice and 0.66, SD = 0.19 for female voice. Results of the ANOVA for RT based on condition and cue type revealed a significant main effect for condition, F(2, 94) = 23.87, p < .001, η² = .331, but no other significant main effects or interactions. Overall RT means were 2054, SD = 809, for male voice and 2108, SD = 1091, for female voice.

Discussion.

The findings from Experiment 2 can be summarized as follows. Talker variability resulted in higher accuracy and lower RTs for both picture-to-L2 and L2-to-L1 recall. The effect of variability was not moderated by the speaker's voice (male voice, female voice) used to cue the translation. To our knowledge, the results of Experiment 2 represent the first demonstration of a positive effect of acoustic variability on L2 vocabulary learning. As such, they have both important theoretical and practical implications. From a theoretical standpoint, the results are consistent with the elaborative processing hypothesis but weigh against predictions of independent modulation. Specifically, the finding that between-talker variability can improve L2 vocabulary learning argues against the independent modulation hypothesis, as this model would predict that sources of variability are normalized prior to encoding and, therefore, have no effect on vocabulary learning. Thus, the independent modulation hypothesis is not consistent with the improved L2 vocabulary learning observed with the high-variability compared with the low-variability condition of Experiment 2. Furthermore, as predicted by the elaborative processing hypothesis, we found systematic improvement in vocabulary learning moving from the no-variability to the moderate-variability to the high-variability conditions.

Thus, the results of Experiment 2 are consistent with the elaborative processing hypothesis, which maintains that talker variability can promote more robust lexical representations of new L2 vocabulary. According to this position, acoustically varied instances of each new lexical item in the input combine to form a representation that is more robust than would have been obtained by an equivalent number of acoustically consistent instances of the same item (i.e., multiple presentations of the same speech signal). The increased robustness might be due to the presence of more indexical information in each representation. Thus, although further research is warranted, the findings of Experiment 2 can be reconciled more easily with a theoretical framework in which—at least for the initial stages of learning new word forms—indexical information is retained as part of lexical representations and can facilitate access to those representations.

The differences between the findings of Experiments 1 and 2 also raise the question of why voice-type variability failed to improve L2 vocabulary acquisition, whereas talker variability produced systematic increases in vocabulary learning. One possible reason for the differences between voice-type and talker variability observed in Experiments 1 and 2 concerns the lack of rotation of voice types across participants in the moderate-variability and no-variability conditions. In Experiment 1 (as in Barcroft, 2001), all target words were presented to all participants in neutral voice in the no-variability condition and in three selected voice types in the moderate-variability condition. A potential consequence of this procedure is that the null effects of voice-type variability might reflect differences related to these specific voice types as opposed to differences between voice-type variability versus voice-type consistency in general. For example, if one or more of the voice types used in the high-variability condition were significantly less intelligible than the neutral voice type and this decrease in intelligibility resulted in worse vocabulary learning, then the potential benefits of voice-type variability might have been offset by this reduced intelligibility. One finding consistent with reduced intelligibility of specific voice types is that performance for the L2-to-L1 translation task in Experiment 1 was significantly reduced when the nasal voice was used as a cue as compared to when the normal voice was used as a cue. If this differential voice-type intelligibility contributed to the null findings in Experiment 1, then positive effects for voice-type variability might emerge using a rotation procedure similar to the one used in Experiment 2. Experiment 3 was conducted to examine this issue directly.

Method.

Participants. Participants in the study were 36 NSs of English from the same participant pool as Experiments 1 and 2. They had no previous formal instruction in Spanish. All of the participants were undergraduate psychology students at a private Midwestern university in the United States. Participants received credit related to requirements of their psychology class for their participation. None of the participants from Experiments 1 and 2 took part in Experiment 3.

Experimental words. The experimental words, word recordings, and word groups in Experiment 3 were the same as those in Experiment 1.

Procedures. Procedures for Experiment 3 followed those of Experiment 1 except for the inclusion of the rotation procedure. In Experiment 3, to ensure that equal numbers of participants heard each voice type in the no-variability and moderate-variability conditions, the 36 participants were divided into 6 groups of 6 participants. The assignment of voice types (voice type 1 = normal; 2 = nasal; 3 = elongated; 4 = high pitch; 5 = whispered; 6 = excited) to the moderate-variability conditions across the six groups was as follows: Group 1 = Voice Types 1, 2, 3; Group 2 = Voice Types 2, 3, 4; Group 3 = Voice Types 3, 4, 5; Group 4 = Voice Types 4, 5, 6; Group 5 = Voice Types 5, 6, 1; and Group 6 = Voice Types 6, 1, 2.

Scoring. Scoring was identical to Experiment 1.

Analyses. The analyses conducted for Experiment 3 paralleled those in Experiment 1. On the postexperiment questionnaire, one participant in Experiment 3 indicated having known the Spanish word for “cat,” gato, and one indicated having known the Spanish word for “orange,” naranja. These two items were excluded from all analyses based on the same procedures as in Experiment 1. All accuracy scores and RTs for correct responses were submitted to two repeated-measures ANOVAs. In both ANOVAs, condition (no variability, moderate, high) and recall type (picture-to-L2, L2-to-L1) were treated as within-subjects independent variables. Accuracy was the dependent variable in the first ANOVA. Reaction time was the dependent variable in the second ANOVA. To examine the effect of the voice type used for cueing during L2-to-L1 recall, accuracy and RT scores for L2-to-L1 recall were submitted to additional repeated-measures ANOVAs. For these analyses, condition and cue type (neutral, nasal) were treated as within-subjects variables, and accuracy and RT were the dependent variables.

Results.

Means for accuracy and RT based on condition and recall type appear in Table 3 and are depicted graphically in Figures 3 and 4. Results of the ANOVA for accuracy based on condition and recall type revealed that cued recall scores differed significantly across the three conditions, F(2, 70) = 32.24, p < .001, η² = .481, and that accuracy was higher for L2-to-L1 recall than for picture-to-L2 recall, F(1, 35) = 31.04, p < .001, η² = .472. No other significant main effects or interactions were observed. Tukey HSD post hoc pairwise comparisons with a Bonferroni correction for multiple comparisons indicated (a) higher accuracy in the high-variability condition as compared to the no-variability condition, p < .001; (b) higher accuracy in the high-variability condition as compared to the moderate-variability condition, p < .001; and (c) higher accuracy in the moderate-variability condition as compared to the no-variability condition, p < .001. Results for accuracy based on condition and measure appear in Figure 3.

Means for accuracy and reaction time based on condition by recall type (Experiment 3)

Effects of voice type variability on accuracy of picture-to-L2 and L2-to-L1 recall (Experiment 3). Error bars represent standard deviations of the mean.

Effects of voice type variability on latency of picture-to-L2 and L2-to-L1 recall (Experiment 3). Error bars represent standard deviations of the mean.

Results of the ANOVA for RT based on condition and recall type revealed that latency differed significantly across the three variability conditions, F(2, 70) = 57.53, p < .001, η² = .622, and recall type, F(1, 35) = 53.91, p < .001, η² = .603. No other significant main effects or interactions were observed. Tukey HSD post hoc pairwise analyses with a Bonferroni correction for multiple comparisons indicated that RTs were (a) lower in the high-variability condition as compared to the no-variability condition, p < .001; (b) lower in the high-variability condition as compared to the moderate-variability condition, p < .01; and (c) lower in the moderate-variability condition as compared to the no-variability condition, p < .001. Results for RTs based on condition and recall type appear in Figure 4.

Results of the ANOVA for accuracy based on condition and cue type (neutral, nasal) revealed a significant main effect for condition, F(2, 54) = 5.8, p < .01, η² = .179, and cue type, F(1,27) = 10.2, p < .01, η² = .27, but no significant interaction between condition and cue type. Results of the ANOVA for RT also revealed a significant main effect for condition, F(2, 54) = 52.2, p < .001, η² = .658, and for cue type, F(1, 27) = 9.4, p < .001, η² = .27, but no significant interaction between condition and cue type.

Discussion.

The findings from Experiment 3 can be summarized as follows. Voice-type variability resulted in higher L2 cued recall scores and lower RTs for picture-to-L2 recall as well as for L2-to-L1 translation. On the L2-to-L1 translation task, accuracy was higher and RTs shorter for neutral-voice cues as compared to nasal-voice cues. These results help to reconcile the pattern of results in the study of Barcroft (2001) and Experiment 1 with those of Experiment 2. Specifically, the results of Experiment 3 suggest that the null effects of voice-type variability in Experiment 1 and in Barcroft (2001) are specific to comparisons of voice-type variability with conditions of less variability that maintain (without rotation) specific voice types, such as neutral voice in the no-variability condition and three specific voices in the moderate-variability condition. In contrast to Experiment 1 and Barcroft's study, when voice types were rotated across participants in Experiment 3, a clear positive and additive effect for voice-type variability was obtained. As such, in addition to reconciling the previous mixed findings, the results of Experiment 3 provide a second demonstration of a positive effect of acoustic-phonetic variability on L2 vocabulary learning. Whereas Experiment 2 demonstrated this effect for talker variability, a type of between-talker acoustic variability, Experiment 3 demonstrated this effect using voice type, a type of within-speaker acoustic variability.