Download presentation
Presentation is loading. Please wait.
1
Acoustic variability in the speech of L2 learners of American English as a function of accentedness
Bruce L. Smith1 and Rachel Hayes-Harb2 1Department of Communication Sciences and Disorders, University of Utah 2Department of Linguistics, University of Utah BACKGROUND METHODS (cont.) RESULTS AND DISCUSSION (cont.) Considerable research has examined acoustic properties of L2-accented speech, but less is known about the relationship between acoustic properties shown by L2 speakers and their accentedness (e.g., Munro,1993). Other research has examined acoustic variability of L2 speakers’ productions and how this affects native English speakers’ perception of L2-accented English (e.g. Wade, Jongman & Sereno, 2007). However, there is little information pertaining to what relationships may exist between L2 speakers’ accentedness and the variability they manifest in their speech. The primary purpose of the present study was thus to determine whether L2 speakers of English exhibit more intra-speaker (i.e., token-to-token) spectral and/or temporal variability when producing various speech segments as compared to native, adult English speakers, and also whether L2 talkers with stronger accents are more variable in their productions than L2 talkers judged to be less-accented. • Formant and duration means, standard deviation, and coefficient of variation (CoV) were computed for the various vowel and consonant measurements; coefficient of variation was calculated using the formula: (S.D./mean) x 100. • 13 native English listeners rated the accentedness of each “Stella” passage on a nine-point Likert scale (1 = does not sound at all like a native speaker of English, and 9 = sounds like a native speaker of English); each judge rated each speaker once, and ratings were averaged across raters to provide a mean accentedness rating for each speaker. Because the 20 NNSs’ Accent Ratings reflected essentially a continuum, we also tried comparing two, more discrete subgroups by utilizing only the 7 most- and the 7 least-accented Ss’ values for several analyses. In this instance, for example, not only was VOT for /ptk/ significantly correlated with Accent Rating (rho = .75, p < .01), but as seen in Figure 4, Accent Rating also correlated with CoV for VOT of /ptk/. Once again, however, average CoV for the 16 acoustic measures did not correlate with the Accent Rating (rho = -0.18, ns). Figure 4 RESULTS AND DISCUSSION As shown in Table 1, the 20 NN talkers were judged by the 13 NE listeners to have accents ranging from 1.3 (#339) to 8.5 (#362), i.e., most- to least-accented. Table 1 Figure 1 Figure 1 shows a comparison of Coefficient of Variation (CoV) for the 20 individual NE speakers (blue bars) and the 20 individual NN English speakers, averaging across 16 different temporal and spectral measures of their speech. (The values within each speaker group are rank ordered, left to right, from most to least variable, but there is no relationship between the pairs other than their rank within their specific group.) As a group, the NE talkers show significantly less variability than the NN talkers (t = 4.021, p < .001). However, it is important to note that only 3 of the individual NN talkers (red #1, #2, #3) show CoV values greater than any of the NE talkers. Thus, although the NN talkers’ CoV values are more similar to the CoV values of the more variable NE talkers, 17 of the 20 NN Talkers (85%) fall within the range of the NE talkers’ CoV values. Figure 2 shows there was a statistically-significant correlation (rho = .53, p < .05) for the 20 NNs’ Accent Ratings compared with their VOT values for /ptk/. However, as seen in Figure 3, the Coefficient of Variation (CoV) for /ptk/ versus Accent Rating comparison approached, but did not reach significance (rho = -0.41, p < .10). A number of other comparisons between CoV and various acoustic measures were also calculated, but none of them reached statistical significance either. In addition, when CoV was averaged across 16 different acoustic measures and compared to the Accent Rating, there was not a significant correlation (rho = -0.13, ns). Thus, in general, it appears that the degree of NN speakers’ accents has little if any relationship to the amount of intra-subject (i.e., token-to-token) variability they manifest in their speech. Figure Figure 3 Talker 339 344 360 381 202 347 350 354 335 361 301 338 363 337 356 345 348 353 342 362 Mean 1.3 1.6 3.1 3.3 3.5 3.7 3.8 3.9 4.2 5.2 6.1 6.3 6.5 7.6 8.1 8.5 St. Dev. 0.6 0.9 1.5 2 1.8 1.1 1.9 0.8 rho = -0.66, p < .05 METHODS Subjects Production Task • 20 native English (NE) speakers (10 M, 10 F); mean age = 24 years • 20 non-native (NN) speakers of English (10 M, 10 F); mean age = 22 years - L1 backgrounds of L2 subjects: Korean (6), Chinese (3), Japanese (3), Spanish (3), Malay (1), Rwandan (1), Swahili (1), Tongan (1), and Vietnamese (1) - L2 subjects’ average age of arrival in U.S. = 18 yrs.,(6 – 22 yrs.) - mean length of U.S. residencewhen recorded = 4 yrs. (1 mo yrs.) - average age began learning English = 11 years, range = 5 – 20 yrs. - average length of formal study of English =7 yrs., range = 0 – 12 yrs. Listening Task • 13 native English listeners (?? M, ??F); mean age = ?? Years • Is there more relevant information that should be added? Stimuli • The NE and NN subjects were recorded producing 49 CVC words repeated in random order 4 times each in the carrier phrase, “I like to say ____ some of the time.” • Stimuli: bade, bead, beat, bed, bet, bid, bit, cab, cap, coat, cob, cooed, cop, could, dab, deed, did, dip, dock, dog, duck, fade, fed, gap, goat, good, heed, hid, hood, pad, pat, peck, peg, pet, Pete, pit, pod, pot, shade, shed, shooed, should, tab, tack, tag, tap, tip, tuck, who’d • Each NN talker also produced the “Stella” passage (Weinberger, 2013) three times; the second repetition was submitted to accentedness judgments by the 13 NE listeners Measurements • 66 different acoustic measurements: F1, F2, F3; durations of three tense/lax vowel pairs (/ i -I/, / e - ɛ /, and / u - ʊ /; initial and final consonant closure durations of / p, t, k, b, d, g /; voice onset time of / p, t, k, b, d, g /; voicing during word-final consonant closure; release burst durations of final / p, t, k, b, d, g /. CoV Accent Rating SUMMARY AND CONCLUSIONS The present study addressed the issue of whether NN speakers learning English as a second language show more intra-speaker variability when they are judged to have stronger versus weaker accents. Thirteen NE listeners rated the accentedness of twenty the NN speakers of English who came from a variety of language backgrounds and had varying amounts of experience and abilities in English. A variety of acoustic measures of their speech and that of 20 NE talkers were obtained. Although many differences were observed when comparing the primary acoustic measures of the NN versus NE speakers, few if any intra-speaker differences in variability were observed relative to the NE talkers. Similarly, there was very limited evidence of variability being associated with the NN’s degree of accentedness. rho = .53, p < .05 rho = -0.41, p < .10 REFERENCES Munro, M. J. (1993). Productions of English vowels by native speakers of Arabic: Acoustic measurements and accentedness ratings. Language and Speech, 36, Wade, T., Jongman, A, & Sereno, J. (2007). Effects of acoustic variability in the perceptual learning of non-native-accented speech sounds. Phonetica, 64, Weinberger, S. (2013). Speech Accent Archive. George Mason University. Retrieved from VOT CoV ACKNOWLEDGMENTS We would like to thank Hannah Allen, Micah Foster, and Cate Showalter for their help in collecting and analyzing the data. Accent Rating Accent Rating
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.