T RANSCRIPTION ANALYSIS

4.1 Transcription analysis

Nine speech errors were found from the transcription data: (1) Non-retroflex produced as retroflex, (2) Retroflex produced as non-retroflex (3) Substitution: Mandarin /ʐ/

misarticulated as /l/, /n/, or deretroflexed, (4) Nasal confusion: velar nasal produced as alveolar nasal or final nasal deletion. (5) /h/ substitution: /h/ produced as /f/ or deleted.

(6) /tɕ^h, tɕ, ɕ/ confusion: these three consonants are randomly substituted by one another, (7) Vowel confusion: diphthong produced as monophthong, or certain vowels were substituted by one another (8) Tone confusion: the most common error: T3 produced as T2, T4 as T1, and (9) Others. These errors are grouped and discussed at the segmental level: consonant and vowel as well as suprasegmental levels below.

Consonant

Regarding voicing, Bernhardt and Stemberger (1998) indicated that voiced segments are acquired before voiceless ones. However, voicing is not contrastive in Mandarin, so the discussion focused on the other contrastive laryngeal feature- aspiration. In general, unaspirated consonants are usually acquired before aspirated ones (Bernhardt and Stemberger, 1998). More solid evidence supporting this claim can be found in prevailing cases of /tɕ^h/ realized as [tɕ]. This type of substitution was not observed in aspiration-contrasted stops which might be the result of stops being one of the earlier acquired MOA, hence facilitating the acquisition of aspiration unlike affricates which are acquired later and may hinder the acquisition process.

In terms of MOA, the agreement among scholars is that the stops are acquired first, followed by the nasals, fricatives, and finally the affricates. The liquids are constantly avoided in the early stages (Goodluck, 1991). The results of present study show a universal development in manner of articulation which indicate that affricates were the most frequent errors, followed by fricatives, then nasals, and stops were the last.

In terms of POA, Bernhardt and Stemberger (1998) suggested that labials are one of the earliest place of articulation to be acquired followed by coronals and then by dorsals. Surprisingly, the error rates these POA present in the present study are bilabials (4%), coronals (94%), and dorsals (3%). That is, these children have mastered bilabials and dorsals but have extreme difficulties with coronals.

Furthermore, children produced the most errors for coronal affricates (47%), coronal fricatives (42%), coronal nasals (6%) and coronal stops (5%) (cf. Fig. 18). This result is postulated to be caused by the large inventory of coronal affricates in Mandarin.

Interestingly, these Shanghainese children were found to master dorsals quite well. It postulated to be the result of the greater dorsal inventory in Shanghainese - /ʔm/, /ʔn/, /k/, /kʰ/, /g/, /h/, /ɦ/, /ʔ/, /ʔɲ/, /ŋ/, /ʔŋ/.

Given that most retroflexes in Mandarin are affricates, which are one of the most difficult POA for children to acquire, many errors were found in our data. Examples such as /ʈʂʰ/ realized as unretroflexed /tsʰ/ while /ʈʂ/ were realized as [ts]. The result

children's learning process. It also echoes the finding of Jeng (2009) which indicated that children do not acquire retroflex until they are six years old.

Vowel

The vowel development of Shanghainese children in the present study is ranked by error percentage: /u/ (27%), /a/ (26%), /e/ (19%), /o/ (15%), and /i/ (12%). Although previous research showed different patterns in vowel development, some studies indicate that the low vowels [æ ], [ɑ], [a], mid vowels [ʌ], [ɛ] and high vowel [i] are early mastered, while back rounded vowel [u] is a late mastery (Bernhardt &

Stemberger, 1998; Ito, 1990; Fee, 1991) (see 1.2.1 for detailed review). The high frequcy of /u/, and /a/ misarticulation agree with the universal development; in general, the rounded vowel [u] is mastered late and it is assumed that [a] has more features to be acquired later (Bernhardt & Stemberger, 1998).

Interestingly, in most of the misarticulated cases, the above mentioned vowels were overwhelmly substituted by /a/ (38% of the cases), followed by /e/ (21.73%), /u/

(16.3%), /i/ (10.86%), /o/ (9.78%), and /y/ (3.26%). Although [a] is mastered late, it appeared frequently in early stages (Ito, 1990; Fee, 1991). Jakobson (1968) suggested that the openness of vowels is the earliest contrasted feature for children. Children in the present study might use a vowel which occurs most frequently for substituting other vowels. The vowels might be substituted by another easy distinguished vowel, i.e., openness: narrow[i] versus wide [a]. Thus, children choose [a] to substitute other vowels, due to the fact that [a] occurs most frequently and is the most mouth-opened vowel.

Syllable structure

In terms of the syllable structure, monophthong V was best controlled by children, and VV was the worst; which echoes the finding of Chen (2004). Chen suggested that children acquire monophthongs first, followed by diphthong, the triphthong is the last.

As the examiner predicted, monophthongs are easier for children to produce, but the unexpected pattern, VVN, was mastered better than VV. It postulated that the nasal coda might not cause difficulties for children to produce, the most difficult for

Tones

There are total of 209 tone errors found. Tone 3 was misarticulated the most, followed by Tone 1 and 4, Tone 2 was the last. The delay T3 master is the same as the results of Li and Thompson (1977) and Su (1985). Generally T2 is acquired later than T1 and T4, however results in the present study show different patterns. According to Shen's (1989) results, American learners who learn Mandarin tones acquired T2 earliest, followed by T3, T1 then T4. It is thus assumed that the tone acquisition order in L1 is different from that of L2, and the L1 language background would influence L2 tone acquisition order. Furthermore, there is no actual T1 and T4 in Shanghainese so it might cause the difficulties in master Mandarin T1 and T4 for Shanghainese children.

For T1 errors, the most frequent error was T1 as T2, T1 as T4, and T1 as T3.

Children produce T1 as T2 most, which might be due to the fact that children master T2 better than T1 as mentioned in the above. For T 2 errors, the most frequent error was T2 as T4, followed by T2 as T1, and T2 as T3 is the last. For the T3 errors, T3 as T2 is the most frequent error, followed by T3 as T4, and T3 as T1 is the last. These patterns are correlated to the tone acquisition order in that children use the best mastered tone to substitute a tone they cannot control well.

For T4 errors, subjects misarticulated T4 as T1 most frequently, T4 as T2 next, and T4 as T3 last. This pattern (children did not misarticulate T4 as T2 more than T1) is unexpected. It postulated that Mandarin T1-T4 contrast is more difficult than T2-T4 contrast for Shanghainese children. Children master T2 better than T1 and T3, that's why they could clearly identify the differences between T4 and T2. On the other hand, children could not yet master T1 and T4 well, so they might get confused between these two tones. When misarticulation occurs, children seem to choose the tone which is the easiest one for them to master.

The tone acquisition of Shanghainese subjects indicated that they master T2 best, followed by T1 and 4, and T3 is the most difficult for them. The results of Li and Thompson (1977) indicated T1 and T4 were acquired earlier than T2 and T3, and they concluded that rising tones (T2 and T3) are acquired later than the falling tone (T4).

however T3 is the most difficult tone for children to master in both studies.

4.2 Acoustic analysis Retroflex

One of the major contributions for the present study is an in depth acoustic investigation of retroflexes. Previous studies measured F3 in the consonant parts of retroflexes; however, the retroflexes in those studies are voiced retroflex consonants. In the present study, the retroflex consonants in Mandarin include voiced and voiceless retroflex consonants. The formants of voiceless retroflex consonants cannot be measured, which is the reason we measured the F3 of the following vowels. Although there is no powerful support research for this measurement, we aim to provide a reference value for future studies. In the present study, we separated retroflexes into voiced and voiceless. There are three measurements in the retroflex section: a) F3 of the following vowel of all retroflex consonants, b) F3 of voiced retroflex consonant itself, and c) M1-M4 values of voiceless retroflex consonants.

The result of F3 in voiced retroflex consonants indicated that F3 dipping was found in correct pronunciation, but not in incorrect pronunciation for three different rhymes (VVN, VV, and V). In order to make sure that the following vowels will affect the F3 dipping patterns, the experimenter further compared F3 values in different following vowels (section 3.2.1). Results indicated that F3 dipping was found in correct pronunciation, but not in incorrect pronunciation for retroflexes preceding V (/i/, /e/, /u/) and VN (/en/) rhymes. However, F3 dipping was found in correct as well as in incorrect pronunciation of retroflex preceding VV (/ua/, /ou/) and VVN (/uaŋ/) rhymes.

The result of retroflex in transcription and acoustic analysis show an asymmetry pattern. Even experimenter hear a deretroflex sound in transcription, the result of acoustic analysis show that the F3 dipped in the deretroflex sound. It is postulated that children are in the progress in acquiring retroflex, and produce an retroflex acoustic cue (F3 dipping) in both retroflex and deretroflex sounds. It is also postulated that the unexpected F3 dipping in both correct and incorrect pronunciation is due to the complexity of rhyme or the difficulties in producing an affricate. According to the results above, the monophthong (retroflex + V) rhyme for children is the best

These results show a universal development in the syllable structure acquisition.

Furthermore, there is no retroflex consonant in Shanghainese; therefore it might be the reason that Shanghainese children misarticulated the most error for retroflex.

For the result of M1-M4, M1 and M2 value of correct pronunciation (retroflex) is significantly lower than that of incorrect pronunciation (nonretroflex). M3 and M4 of correct pronunciation are higher than that of incorrect pronunciation. The results of the present study showed different patterns to the result of Jeng (2009), in which M1 and M4 of retroflexes are lower than that of deretroflexes, and M2 and M3 of retroflexes are higher than that of deretroflexes. In other words, Jeng (2006) suggested that retroflexes in Mandarin have lower central gravity frequency, low but weak energy and wide energy distribution (section 1.2.1). It is postulated that the different results between the two studies is due to the fact that the subjects in the present study were children from Shanghai, which means and the production of retroflex might be influenced by their L1.

However the subjects from Jeng (2009) were 19-25 years old adults. The production of retroflex by children and adults might show different patterns. According to the literature review, different researches show different patterns. Forrest et al., (1988) investigated ten English adult speakers; results indicated that M3 is the most efficient parameter to distinguish /s/ and /ʃ/ which is different to Jeng’s study. Flipen et al., (1999) investigated /s/ produced by 26 teenagers and found out that M1 and M3 are the most salient characteristics to represent /s/. It is postulated that the different results shown in the present study might be affected by the language background and age of the subjects.

Vowel

Comparisons among Shanghainese speaking children, Mandarin speaking and Shanghainese speaking female adults yield three sets of results of vowel production: (a) the vowel space of Shanghainese children was narrower than that of Mandarin adults, (b) the vowel space of Shanghainese adults (producing Shanghainese) was found narrower than that of Mandarin speaking adults. (c) The vowel space of Shanghainese children was found narrower than that of Shanghainese adults and Mandarin-speaking adults. Generally, children's vowel space shows three different patterns (section 1.2.1 vowels). Results from the present research show that the vowel space of children is

universal development. However, since Shanghainese-speaking adults have a narrower vowel space than Mandarin-speaking adults (c.f. fig 62), it is thus postulated that the dense vowel space of children might be influenced by their L1, Shanghainese.

Tones

The F0 contours and duration pattern of tones produced by the children resemble adult norms. The four tones are significantly different: T3 > T2 > T1 > T4 (Fig. 67). This result suggests that they have managed to utilize F0 and duration cue as adult speakers.

This result agrees with the result of Li and Thompson (1977), in which they argued that tone acquisition is accomplished within a short period of time, and tones are mastered before segmentals. Moreover, the rising and dipping tones are substituted for each other. However, the acquisition order was different to the result of Li and Thompson (1977). Interestingly, in the tone production of children, the T4 pitch range is not as wide as the adult norm (cf. Fig. 64-65), which is posited to be the effect of their L1 (Shanghainese) (cf. Fig.66) which does not have a high falling tone as Mandarin T4.

Conclusion

In conclusion, Transcription and acoustic analysis from normal speaking children provided evidence of the effect of universal development and that of L1 background.

For consonants, bilabials acquired earlier than coronals were in line with the universal development, but dorsals mastered better than coronals might be due to the L1 background, in which there are eleven dorsals in Shanghainese but only three dorsals in Mandarin. Stops are acquired the earliest, followed by nasals, fricatives, and affricates are the last, which show a universal development trend. The misarticulation of retroflexes might be affected by the universal development trend--- retroflexes were generally acquired after the age of six and influenced by their L1---as well as the fact that there are no retroflexes in Shanghainese. The high frequcy of /u/, /a/

misarticulation agree with the universal development. The delayed master of T3 agrees with the universal development. However, the early mastery of T2 might be the L1 effect.

studies show different vowel space acquisition patterns, this finding is assumed to be the effect from their L1. Finally, the children perform similarly to adults in tone production, which corresponds to the general development trend---suprasegment acquired before segment. On the other hand, the narrower T4 pitch range may be caused by their L1.