Dividing up the vowel space

each listener. The response buttons contain well-known and meaningful monosyllabic (real) English words written in English orthography in a fixed /CVl/ carrier covering the 11 pure vowels of AE. In the carrier, C refers to a consonant phoneme in the syllable onset position, V refers to the synthesized vowel (long or short in duration), and a fixed coda /l/. Response alternatives were the following words: feel /i/, fill /ɪ/, sale /e/, tell /ɛ/, shall /æ/, null /ʌ/, doll /ɑ/, call /ɔ/, whole /o/, full /ʊ/, and fool /u/ (see Figure 4.2). The participants were asked to click on the response button with the word that contained the vowel they had just heard. Participants were instructed to always click a button when they heard a stimulus and to gamble if they could not choose between alternatives. This renders the task an eleven-alternative forced-choice (11-AFC) identification task.

Then, participants were required to provide a judgment on a 3-point Likert scale of how good an exemplar of the selected vowel they considered the vowel they had just heard was (typicality judgment), where 3 = Good. The process was repeated until all 86 tokens were covered.

Figure 4.2. User interface for the vowel identification test for native AE speakers presented in English (left panel) and for L1 PA participants with instructions translated into Arabic (right panel).

Presentation of instructions and stimuli as well as the collection of the responses were performed by a Praat MFC script (see Appendix 4.2). The choices the participant made, the typicality judgments and the time it took the respondent to enter the typicality judgment (in milliseconds from the offset of the stimulus) were logged and stored for offline data analysis.

bolded phonetic symbol is entered in the figure. Agreement between 25% and 50% of the responses is indicated by a smaller print (unbolded). When less than 25% agreement was obtained for a particular vowel category, the cell was left blank (this rarely occurred).

Therefore, the symbols in the cells of Figure 4.3 represent the modal response given to the synthesized vowel that is defined by the F1, F2, and duration coordinates defined by the cell position.³¹ Figure 4.3A shows the mapping of the English vowel space as entertained by native American listeners responding to the short vowel stimuli, while Figure 4.3B shows the results of the long stimuli for the same group of participants.

A. Short stimulus vowels B. Long stimulus vowels

x

: >50% agreement; x: 25-50% agreement

Figure 4.3. Modal responses by 20 American native listeners for 43 vowel stimuli differing in F1 (vertically) and F2 (horizontally) center frequencies. Vowel duration is either 200 ms (panel A) or 300 ms (panel B). For specifications of F1 and F2 steps, see Figure 4.1. Large bolded symbols denote a majority decision with 50% or more agreement. Small symbols indicate a modal response with an agreement between 25 and 50%. Cells with a

modal response < 25% agreement are left blank.

The native listeners have a rather straightforward division of their vowel space. The top left of the space is taken up by tense /i/, while the top center and top right areas are exclusively occupied by tense /u/. Then, going down along the front edge of the diagram, there is a well-demarcated area for /ɛ/ and a much smaller area for maximally open /æ/. The central portion of the vowel space is taken up by the other lax vowels /ɪ, ʊ, ʌ/. The open and half open back vowels /ɔ/ and /ɑ/ are not delineated from each other; in fact, open /ɑ/ is seen as an island surrounded by /ɔ/ responses. Note that all tense vowel categories – with the exception of the semi diphthong /e/ – are a preferred response for at least two sample points despite their short duration. This is an indication that duration is unlikely to be the primary cue in the tense-lax contrast in (American) English, in contradistinction to what was claimed by Wells (1982: 120).

31 Occasionally, the distribution of responses to a particular vowel type was bimodal or even multimodal. This happened with 5 out of the 86 vowel stimuli for the native listeners, but never for the PA listeners. In such cases the modal response was assigned such that contiguous areas in Figure 4.3 were maximized. Appendix 4.3.A contains a complete breakdown of response category by stimulus type for the native participants.

1. 2. 3. 4. 5. 6. 7. 8. 9.

i i

^u

u u u u u u i i

^{ɪ ɪ}

u u u u u

ɛ ɛ

^{ɪ ʊ ʊ o}

o ɛ ɛ

^{ɛ ʌ ʌ ɑ ɑ}

æ

^æ

æ

æ

ɔ ɑ

æ æ æ æ

^{æ æ}

1. 2. 3. 4. 5. 6. 7. 8. 9.

1.

i i

^u

u u u u u u

2.

i i

^{ɪ ɪ}

u u u u u

3. ɪ ɪ ɪ ʊ

ʊ

^o

o

4. æ

ɛ ɛ ɛ

^ʌ

ʌ

^{ʌ ɑ}

5.

ɛ

^{ɛ ʌ ʌ ɔ}

6.

æ æ

^ɔ

7.

æ æ

^ɔ

Figure 4.3B shows the results in terms of the preferred responses of the American listeners for the 43 vowel types with long duration (300 ms). It would appear that category boundaries between pairs of vowels that differ in height are very regular and are almost exclusively based on F1 frequency. The tense vowel categories have expanded their size somewhat (from 26 to 31 sample points) since the vowel duration of 300 ms fits the specification of this type of category. By the same token, the area taken up by lax vowels has shrunk (from 17 to 12 sample points). The effect of longer duration is surprisingly large for /æ/, which vowel expands its area from 4 sample points to 9. This would be yet another indication that /æ/ in American English is phonetically tense and long. The semi-diphthong /e/ is never perceived with at least 25%

agreement, not even when the duration matches its internal (long) specification. This would indicate that the closing gesture (diphthongal trajectory defined by gradual lowering of F1 frequency) is indispensable for /e/ but not necessarily so for /o/.

Now turning to the nonnative results, we may ask how PA learners, with their small L1 vowel inventory, map out the AE vowel space. The vowel labeling by the PA learners of English is shown in Figure 4.4A for short stimuli and 4.4B for long stimuli.

Figure 4.4. Modal responses by Palestinian listeners (20 boys, 20 girls) for 43 vowel stimuli differing in F1 (vertically) and F2 (horizontally) center frequencies. Vowel duration is either 200 ms (panel A) or 300 ms (panel

B). For specifications of F1 and F2 steps, see Figure 4.1. Large bold symbols denote a majority decision with more than 50% agreement. Small symbols indicate a modal response with an agreement between 25 and 50%.

Cells with a modal response < 25% agreement are left blank.

What is immediately noted for the nonnative PA listeners is that when the stimulus is short, the responses are mainly short/lax English vowels (see Figure 4.4A). Only 14 out of 43 stimuli were identified as long/tense AE vowels. Conversely, when the stimuli are long (300 ms), the responses are mainly tense vowels (see Figure 4.4B). Only 7 of the 43 stimuli were identified as short/lax vowels (exclusively /ʌ/).

The results for the short stimuli show mainly short/lax AE responses and therefore differ substantially from the native AE results on these stimuli (compare with Figure 4.3A). Lax /ɪ, ʊ, ʌ/ vowels obtain a majority vote only if the vowel duration is short (200 ms); they never reach a majority when the stimulus vowel is long (300 ms). Lax /ɛ/ never reaches 25% agreement, which would indicate that the mid vowels are indeed a special problem for PA learners of English. Duration, on the other hand, would seem to be a strong component of the (incorrect) perceptual representation of the AE lax-tense (short-long) contrast.

The top left area in Figure 4.4A is filled by lax /ɪ/, which overlaps and masks /ɛ/ downward, /i/ upward, and fills even part of the /æ/ area in the native AE listeners’ results in Figure 4.3A.

The top-right and top-central areas are fuzzily demarcated between a few /u/ tokens centered in the extreme upper right corner and an overgeneralized dispersion of /ʊ/ without any results

>50% while occupying large chunks of what should be /u/ tokens. Similarly, /ɑ/ tokens spread horizontally across the triangle, which happens by shifting the tongue position from back to front but is still considered by the majority (25%-50%) of the participants as good representatives of AE /ɑ/; vowel fronting seems to be compatible with the perceptual representation of /ɑ/ on the part of the PA learners. The bottom part of the triangle is occupied solely by /ʌ/ tokens, while the other AE low vowels /æ, ɑ/ never reach the 25% agreement criterion. As shown in Figure 4.4A above, lax /ɛ/ is never a response category; it is always crowded out by /ɪ/, which means that PA learners of English do not have a proper separation of /ɪ, ɛ/ vowels. Because Arabic has no mid-vowel phonemes, AE short (lax) /ɛ/ is not conceived of as distinct from PA short /i/, which leads to the provisional conclusion that short front-mid vowels are considered allophonic tokens of the PA short high front vowel type.

It is possible to conclude from the results in Figure 4.4A that short stimuli mainly yielded lax responses, except for tense /u, ɑ/, which occupy undisputed areas along the back edge of the triangle. The rest of the tense vowels, especially /e, o/, never made it as a preferred (‘modal’) response category for the short stimuli, which indicates that duration is more likely to be the primary cue in the tense-lax contrast in (American) English in the perception of PA learners.

The mid-central lax /ʌ/ is perceived too far low; low-mid lax /ɛ/ and the mid tense vowels never reach 25% agreement, which illustrates the problem with the mid-section vowels for L1 Arabic learners.

For the long stimuli, the results for the PA learners (Figure 4.4B) show that only /u/ and /i/

are well represented compared to other vowel tokens, but they are expanded substantially (relative to the native AE representation) when the stimulus duration is long (300 ms). All other long tokens are compromised when compared to the AE native results. The areas of /ʌ/ and /æ/

are about right – although /ʌ/ extends too far low and takes over sample points that should be /æ/, which in turn has shrunk to two sample points, and even these are too far back compared to the native norm.

It should be highlighted here that the longer stimuli yielded almost exclusively tense vowels as response categories. Accordingly, it is possible to say that duration is the most important cue for the PA EFL learners to differentiate between AE tense and lax high vowels. Duration, therefore, is given too much weight in the perceptual representation entertained by PA learners of American English. What confirms this claim in light of the AE tense/lax feature is that the presence of the AE lax counterpart ousted the tense counterparts and vice versa based on the stimulus duration, especially for the high vowels. Responses for the short stimuli filled 11 sample points with lax /ɪ/ in the high front area of the triangle, which covered the area that was supposed to be demarcated for tense /i/ and even /e/.

The opposite is true for the long stimuli where tense /i/ is fairly represented with the exact same 4 sample points as in the native results, and /e/ is never disputed with /i/. The same situation is applicable for AE /ʊ/ and /u/, but with better representation for the latter in the short stimuli than the former in the long stimuli. Additionally, /e/ and /o/ are represented as tense/long monophthongs, and diphthongization does not seem to be part of the perceptual representation of the tense English mid vowels in the mind of the Arab learners. Categories /ɑ/ and /ɔ/ are conflated, as they are in the vowel system of the Californian (and many other) native listeners.

The difference from the native AE representation is that the AE vowels are long and tense; they never extend into the central area of the vowel space. In the perceptual representation of the PA listeners, the undifferentiated /ɑ, ɔ/ vowel pair is the short counterpart of tense/long /o/; both mid-central and mid-back vowel qualities are deemed adequate for /ɑ/ and /o/. All open vowels are perceived as /ʌ/, unless the vowel is long and very front, in which case /æ/ is perceived.

Tense mid vowels /e/ and /o/ take up large areas in the vowel space when the stimulus duration is long. Apparently, the EFL learners accept these vowels on the basis of length and do not mind the absence of diphthongization. The AE vowel /æ/, which is long and generally considered phonetically tense in AE (Celce-Murcia et al. 2010; Strange et al., 2004; Wang &

Van Heuven, 2006), appeared as the response for long stimuli because it resembles (near-equivalent) low front long /a:/ in PA.

To conclude this section, it seems as if PA learners of EFL have some notion of mid vowels but only when they are long. The short mid vowels are unified with their short high counterparts.

The long mid vowels /e, o/ are entertained as viable response alternatives, even if they are

unacceptable to native AE listeners because the PA learners do not know that the AE tense mid vowels should diphthongize.

In document The Perception and Production of American English Sounds by Palestinian Arabic Adolescents (Pldal 105-110)