Title Vowels of Hong Kong English: from an acoustic perspective

The present study investigated how the vowel system of the first language (Cantonese) affects the acquisition of vowels of a second language. Forty native Cantonese-speaking adults (20 males and 20 females), with ages between 19 years 4 months and 26 years 10 months were recruited. Data from the first and second formant frequencies indicated that, for both female and male speakers, production of American English vowels was influenced by the Cantonese vowel system. This is also true even for those English vowels that were found in Cantonese. The perceptual ability in identifying the English vowels was also carried out to account for the deviated production of American English. It is found that perceptual and production abilities are related. Vowels of Hong 3 Vowels of Hong Kong English: from an acoustic perspective Cantonese is a dialect of Chinese. It is the first language spoken by the people residing mainly in the southern China including Hong Kong. In Hong Kong most of the native speakers of Cantonese speak English as their second language. Although most children in Hong Kong start to learn to speak English as early as in the kindergarten, English spoken by native Cantonese speaking Hong Kong people is usually judged to be heavily accented. Accent is defined as non-native like pronunciation (e.g. Markham, 1997; Major, 2001; Flege, Bohn, & Jang, 1997). Inaccurate production of English vowels is one of the major sources of accent (Markham, 1997). Vowel Systems in Cantonese and English In English, there are 11 monophthongs /ɪ, i, e, æ , ɑ, ɔ, ʊ, u, ɛ, ʌ, o/ (Chen, Robb, Gilbert, & Lerman, 2001). They are classified by the place of articulation in terms of tongue height and anteriorposterior constriction position (Peter, 2001). Tense-lax vowels are considered as distinct vowels in English. Peter (2001) claimed that tense and lax vowels are similar in vowel quality, but „lax vowel is shorter, lower, and slightly more centralized than the corresponding tense vowel‟ (p.81). In English, there are suggested three pairs of tense-lax vowels /ɪ, i/, /e, æ /, and /ʊ, u/. However, in Cantonese, tense and lax vowel are not distinctive and only considered as allophones of each other (Bauer & Benedict, 1997). Bauer and Benedict (1997) carried out an in-depth study of Cantonese phonology. They concluded that Cantonese comprises of eight monophthongs /i, y, ɛ, œ, a, ɐ, u, ɔ/ and 13 allophones /i:, ɪ, y:, ɛ:, e, œ:, ø , ɐ, a:, u:, ʊ, ɔ:, o/. Similarly, Cantonese vowels are classified in terms of tongue height and front-back constriction position. Acoustic Theory of Vowel Production The Source-Filter Theory Vowels of Hong 4 Research on second language acquisition often uses phonetic symbols such as International Phonetic Alphabet (IPA), or the acoustic measurement of formant frequencies, to compare the phonetic inventories of first language (L1) and second language (L2). Though International Phonetic Alphabet (IPA) is commonly used worldwide as a phonetic symbol system to represent pronunciation, the same phonetic symbol may actually represent different configurations of the articulators in the production of different phonemes in different languages (Flege, 1987). For example, Flege (1987) claimed that although /u/ is present in both the phonetic inventories of French and English, the tongue position in producing the French /u/ is in a more posterior position of the oral cavity than that in producing English /u/. Therefore, it may not be valid to represent actual vocal tract configuration during vowel production. The same IPA symbol may indicate different articulatory configurations. Acoustic measurements yield more valid and objective information. Proposed by researchers (e.g. Fant,1970; Stevens & House, 1961), the source-filter theory can be used to understand vowel production. According to Fant (1970), vowel is produced as a product of the energy source from the larynx and the effect of vocal tract resonators. The laryngeal source determines various aspects of a speech sound including the loudness, voice quality, and pitch. Pitch is closely correlated with fundamental frequency. Fundamental frequency refers to the rate at which the vocal folds vibrate when driven by an outward flow of air stream from the lungs. The source energy is modified by the filter, the configuration of the vocal tract. Resonance is thus resulted, and formants are created. Formants are the frequencies at which energy peaks are found, and they are labeled as F1, F2, F3, etc. following the order they appear in the frequency spectrum (Fant, 1970). Each vowel can be identified by its first three formant frequencies (Kent & Read, 2002). However, usually first and second formant frequencies alone are adequate for identifying most vowels in English (Kent & Read, 2002). Vowels of Hong 5 Tube Models of Vowel Production Fant (1970) suggested that independence between the energy source and the resonator. Formant frequency is only affected by articulatory configuration. In producing vowels, the vocal tract can be viewed as a tube, or a combination of tubes for resonance, depending on the tongue position of the vowel (Johnson, 2003). For example, in producing schwa, the vocal tract is a tube of uniform cross-sectional area. In producing the English vowel /ɑ/, the vocal tract is regarded as two tubes, a back tube with smaller cross-sectional area, and a front tube with larger cross-sectional area. In producing /i/, the vocal tract can be viewed as two tubes with similar cross-sectional areas which are separated by a constriction. Regardless of the articulatory gesture, the formant frequency of tube(s) is inversely proportional to the length of the tube (Johnson, 2003). The shorter is the tube, the higher is the formant frequency. Children have shorter vocal tract than adults, and women have shorter vocal tract than men. Therefore, vowels produced by children should have higher formant frequencies than adults, and females may have higher formant frequencies than males (Fant, 1970). Besides the length, the ratio between the cross sectional area of the front tube and back tube also influences resonance frequency (Fant, 1970). Fant suggested that an increase of the cross-sectional area of the front cavity would lead to an increase of F1. Kent and Read (2002) summarized Fant‟s argument and stated that the two lowest formant frequencies relate to articulation of vowels. They stated that F1 of vowel is related to tongue height. The higher is the tongue position, the lower is the F1 value. Meanwhile, F2 is related to the anterior-posterior position of the tongue during vowel production. The more posterior is the tongue, the lower is the F2 value. Pickett (1999) claimed that F1 was also influenced by place constriction during articulation. While constriction in the front cavity increased F1, constriction at larynx reduced F1. The formant frequencies associated with a vowel therefore indicate the positioning of the tongue inside the vocal tract during the production of that vowel. The discrepancy in F1 of a vowel indicates Vowels of Hong 6 the tongue height and place of constriction, while deviation in F2 of reflects the difference in anterior-posterior position in producing the vowel. Therefore, the variance of F1 and/or F2 between a vowel in L1 and L2 should indicate the errors in production. Accent Theory Major (2001) reviewed early researches about accent when speaking a second language. Using Contrastive Analysis (CA), the main reason of having an accent in L2 is argued to be the transfer of phonological system of the L1 to that of L2. When acquiring L2, people discover the similarity in the phonological system between L1 and L2. The learner substitutes a new phoneme of L2 by the similar phoneme in L1 (e.g., Lado, 1957 as cited in Major, 2001). In a more recent study, Flege (1992) introduced Speech Learning Model (SLM) and the concept of „equivalence classification‟. In SLM, phonemes in L2 are classified into three types: „identical‟, „similar‟, or „new‟. Flege argued that when learning L2, a similar phoneme is poorly acquired, as the learner would substitute it with a phoneme in L1 system; while acquisition of „new‟ phoneme will be native like eventually. This theory is supported by Flege‟s earlier study in 1987. Flege studied the production of „new‟ and „similar‟ phonemes of French vowels by English speaking individuals (Flege, 1987). He found that the new phoneme /y/ produced by the English speakers were not significantly different from that by native French speakers, regardless of French speaking experience. However, for the similar phoneme /u/, none of the English speaking subject, despite the extensive experience of speaking the L2 (an average of 11.7 years), attained native like /u/ in French, though the more experienced subjects were able to produce /u/ more similarly to the native speaker than the less experienced subjects. Later studies have also supported the SLM. Chen et al. (2001) examined the familiar and unfamiliar English vowels produced by Mandarin adult speakers. They defined familiar Vowels of Hong 7 vowels of their Mandarin subjects, who spoke English as L2, as common vowels found in the phonological systems of both English and Mandarin. These vowels may be viewed as „same‟ and „similar‟ vowels as suggested by Fledge (1992). Those English vowels that do not exist in Mandarin were defined as unfamiliar vowels, and „new‟ vowels according to Fledge (1992). In Chen et al.‟s study, however, the formant frequencies of the vowels in these two language systems were not compared. They noticed significant differences in the first (F1) and second (F2) formant frequencies of the familiar English vowels produced by both male and female Mandarin speakers. However, significant differences were also found in the unfamiliar vowels. Chen et al.‟s findings of unfamiliar vowels opposed the SLM. They argued that this may be due to inability in perceiving the difference in acoustic features, and inability in the control of articulators. Chen et al. suggested an absence of perceptual evaluation was one of the limitations of their study. Therefore, it was hard to conclude that if the difference in the vowel production found in Chen et al.‟s study was due to the lack of perceptual acuity in differentiating different vowels in the two phonological systems of the speakers or purely speech motor control. Flege, Bohn, and Jang (1997) observed a relationship between the perception and production ability of L2 learners. They studied the production and perception ability of English vowels by four groups of speakers – German, Spanish, Mandarin and Korean, and found that subjects who were able to perceive English vowels more accurately tended to perform better in production of the corresponding English vowels. They also found that the production of L2 (English) was influenced by the vowel inventory of the subject‟s L1. As the subjects identify a particular English vowel as a vowel in L1 inventory („similar‟ vowel), they produced less native like vowel than those who identified the vowel as a new phoneme („new‟ vowel). This finding supports Flege‟s notion of SLM (Flege, 1992). It can be concluded that the acquisition of the phonological system of L2 is influenced Vowels of Hong 8 by the phonological system of L1, and by an ability to distinguish vowel quality, such as F1 and F2, between the vowels in L1 and L2. The present study is to investigate the influence of L1 on acquisition of L2, and the effect of perceptual ability on acquisition of the vowel system of F2. Perhaps, some other factors, such as age of acquisition, also play an important role in acquiring L2 (Markham, 1997). However, it is not the purpose of the present study. Purpose of the Present Study Previous researches n L2 acquisition have focused on various languages such as French spoken by English speakers (Flege, 1987), English spoken by French, Spanish, Chinese, and German (Flege et al., 1997; Chen et al., 2001). Hung (2000) carried out a study of the phonological system of Hong Kong English. He found that the English tense and lax vowels produced by his Cantonese speaking subjects were not acoustically significantly different in terms of vowel durations, and F1 and F2 values. However, the English vowels produced by the Cantonese-speaking subjects were not compared with those produced by native English speakers, or with the vowels in Cantonese. A systematic acoustical analysis of the English vowels spoken by Hong Kong people is lacking. In order to account for the accent produced in Hong Kong English, the ability in perceiving English vowels should be evaluated. Furthermore, the vowel systems between Hong Kong English and Cantonese, and that between Hong Kong English and American English should be compared. The present study attempted to determine: (1) how the F1 and F2 of English vowels spoken by native Cantonese speakers resided in Hong Kong (HKE) are different from those of native American English (AE) speakers; (2) how vowels in HKE (L2) is affected by the vowel system of Cantonese (L1); and (3) if there is relationship between perception and production of English vowels in learning English as an L2. The hypotheses of the present study are (1) the production of „familiar‟ vowels will be more deviated than the „unfamiliar‟ vowels from the corresponding vowel of AE, and will be Vowels of Hong 9 similar with a vowel in Cantonese; (2) the production of „unfamiliar‟ vowels will be native like; (3) perception ability and production ability is correlated. Method Participants Forty native Cantonese speakers (20 males and 20 females) participated in the present study. A 29-year-old male speaker of native American English from Los Angeles was also recruited. The Cantonese-speaking female subjects were between 19.5 and 23 years of age, with a mean of 21.7 years. The male subjects were between 19.3 and 26.8 years of age with a mean age of 21.5 years. All of them obtained a grade C or above in oral English in Hong Kong Advanced Level Examination (HKALE). All participants passed the hearing screening at 500 Hz, 1000 Hz, 2000 Hz, and 4000 Hz at 20 dB in a sound-treated room. Materials The speech materials included the 11 English vowels (/ɪ, i, e, æ , ɑ, ɔ, ʊ, u, ɛ, ʌ, o/) produced in an /hVd/ context; and eight Cantonese vowels (/i, y, ɛ, œ, a, ɐ, u, ɔ/) produced in either /kVn/ or /kV/ context. Tables 1 and 2 show the IPA symbols and the English words and Cantonese words with meanings, respectively. Table 1. Eleven American English vowels and the corresponding words used in the present study Vowel Word Vowel Word Vowel Word Vowel Word i heed ɪ hid ɛ head æ had u who‟d ʊ hood ɑ hod ɔ hawed e hayed ʌ hud o hoed Vowels of Hong 10 Table 2. Eight Cantonese vowels and the corresponding words used in the present study Vowel Word Vowel Word Vowel Word Vowel Word i 堅 /kin/ ɛ 驚 /kɛŋ/ y 捐 /kyn/ æ 薑 /kæ ŋ/ a 奸 /kan/ ɐ 根 /kɐn/ u 官 /kun/ ɔ 肝 /kɔn/ Data Collection Native American English (AE) speaker Native American English (AE) vowels were elicited from the American English (AE) speaker. To obtain the speech samples, the speaker was instructed to produce the 11 English words listed in Table 1 at a comfortable level. Similar to the study reported by Peterson and Barney (1952), the /hVd/ syllable was embedded in a carrier phrase of „Say ___ again‟. All speech samples were recorded by using a high quality recorder (M-Audio Mircotrack Professional 2-channel Mobile Digital Recorder), via a high quality microphone (M-Audio Aries Professional Condenser Vocal Microphone) in a sound-treated room. The productions made by the AE speaker were later used in the perception task. Cantonese-Speaking Subjects The Cantonese-speaking participants completed three experimental tasks: (1) production of English vowels (HKE), (2) production of Cantonese vowels, and (3) identification of English vowels. The entire procedure was carried out in a sound-treated room. The procedure used to obtain the productions of English and Cantonese vowels by the Cantonese speakers was similar to the recording of native American English vowels stated previously. production of Hong Kong English (HKE) vowels The Cantonese-speaking subjects were instructed to produce the 11 English words Vowels of Hong 11 listed in Table 1 in a carrier phrase of „Say ___ again‟. Each phrase was produced by each subject thrice in a randomized order. Thirty-three English phrases were obtained from each subject. A practice session was provided to each subject before recording in order to allow the subject to get familiar with the chosen English words. The /hVd/ syllable produced by the AE speaker was separated from the carrier phrase and used as materials in the practice session for the Cantonese-speaking subjects. Two experimenters were present throughout the entire experiment. If the syllable was judged to have misproduced the words by the two experimenters, the subject was required to listen to the word produced by the AE speaker, and produce the phrase again. The subject was allowed to repeat the phrase four times at most for each syllable. A total of 33 English phrases were produced by each subject. production of Cantonese vowels All Cantonese-speaking subjects also produced the eight Cantonese vowels listed in Table 2. The /kVn/ or /kVŋ/ syllable was embedded in a carrier phrase of „我要讀___俾你聽‟ („I want to read ____ to you‟). Similar to the production of English vowels, each carrier phrase was spoken by each subject thrice in randomized order. For each subject, a total of 24 Cantonese phrases were produced. identification of American English (AE) vowels The phrases produced by the AE speaker were used in the identification task. Each of the 11 phrases was presented to the subject thrice in a randomized order at a comfortable loudness level via high quality headphones. Each subject was required to identify the vowel which they perceived by forced choice of the 11 English words in Table 1. The subject was allowed to listen to the phrase as many times as he/she desired in each trial in order to obtain the best answer. Data Analysis The English and Cantonese vowels produced by each subject were acoustically Vowels of Hong 12 analyzed. In order to acoustically describe how the vowels were produced, formant frequencies were used. A signal analysis software, Praat, was used to obtain F1 and F2 values. To avoid initiation and termination effects, only the medial 80% of the vowel was used for analysis. The F1 and F2 values of the vowels were evaluated by using linear predictive coding (LPC) analysis. The built-in LPC algorithm was used to superimpose the spectral peaks on the spectrogram. The first two spectral peaks of each frame were then calculated. These values were averaged to represent the mean F1 and F2 values. For the identification task, the percent correct identification of each vowel was calculated and the pattern of mis-identification was also noted. Statistical Analysis The mean F1 and F2 values obtained from the English vowels produced by male and female Cantonese subjects were compared against the normative data reported by Kent and Read (2002) and Chen et al. (2001) (see Tables A1 and A2 in Appendix A). One-way ANOVA was carried out to determine if there is statistically significant difference in F1 and F2 between the 11 HKE vowels. To compare each vowel pair, Tukey HSD test of multiple comparisons was carried out. Results Reliability Measurements Five percent of the entire data corpus (114 out of 2,280 speech samples) was randomly selected from the English and Cantonese vowels produced by the 40 Cantonese-speaking subjects for inter-rater and intra-rater reliability measurements. The selected speech samples were analyzed a second time by the primary investigator and another investigator. The first and second measurements made by the first investigator were used to calculate intra-rater reliability, and the measurements made by the first and second investigators were used to calculate for inter-rater reliability. Vowels of Hong 13 For intra-rater reliability, the average absolute error of F1 and F2 obtained from the first and second measurements made by the first investigator were 9.81 Hz and 20.66 Hz, respectively. The Pearson product-moment correlation coefficients (r) for F1 and F2 values were 0.994 and 0.995 (p < 0.01) respectively. For inter-rater reliability, the average absolute error of F1 and F2 obtained from the measurements made by the first and second investigators were 12.98 Hz and 25.13 Hz, respectively. The Pearson product-moment correlation coefficients (r) for F1 and F2 values were 0.990 and 0.994 (p < 0.01) respectively. Both the average absolute error and Pearson product-moment correlation coefficients show that measurements obtained by the first investigator were reliable and consistent. Acoustic Findings Cantonese Vowels The F1 and F2 values associated with the eight Cantonese vowels (/i, y, ɛ, œ, a, ɐ, u, ɔ/) produced by female and male Cantonese speakers are shown in Tables 3. Comparison of Cantonese and American English (AE) Vowel Systems The vowel spaces corresponding to the corner vowels /ɑ, i, u/ produced by male and female Cantonese speakers are displayed in Figures 1 and 2 respectively. Vowel spaces were developed based on the F1 and F2 values associated with the vowels. As indicated in Figures 1 and 2, for both Cantonese and English, the F2 of the front vowel /i/ and central vowel /ɑ/ for both female and male Cantonese speakers was slightly higher than the corresponding vowels in AE, and the back vowels /u/ are lower than the corresponding vowels in AE. Hong Kong English (HKE) Vowels and American English (AE) Vowels The F1 and F2 values of the vowels produced by the Cantonese subjects were compared by using Tukey HSD test of multiple comparisons. The results indicated that the vowels /ɑ/ and i i Formatted: Font: 8 pt Formatted: Font: 8 pt Vowels of Hong 14 Table 3. Mean, standard deviation and range (in Hz) of F1 and F2 values of eight Cantonese vowels Vowel F1 (Hz) F2 (Hz) Mean SD Range Mean SD Range Female a 874.48 148.35 582.70 – 1343.4 1659.17 130.01 1409.70 – 1963.60 ɐ 767.89 88.46 567.93 – 1007.5 1679.97 99.55 1413.90 – 1895.00 ɛ 688.59 73.15 536.44 – 865.67 2284.76 176.89 1704.90 – 2848.60 œ 654.64 66.63 490.51 – 804.38 1601.01 128.63 1155.70 – 1859.30 i 379.71 54.54 276.75 – 473.71 2860.05 159.01 2514.50 – 3281.70 ɔ 643.20 81.85 330.12 – 822.55 1068.88 94.47 866.93 – 1270.30 u 435.94 31.46 378.63 – 504.39 885.03 116.05 606.75 – 1147.00 y 413.95 35.3 298.61 – 495.58 2002.32 143.09 1735.7 – 2949.2 Male a 727.20 852.22 556.51 – 912.98 1426.71 107.16 1209.10 – 1626.80 ɐ 653.30 57.74 521.18794.22 1442.55 113.8