Chapter 3 Methodology
3.2 Data analysis
3.2.4 Substitution pattern in tonal errors
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
46
many researchers (see Shriberg & Kwiatkowski, 1982; Sheiberg, et al., 1997)
Accuracy rate=
When calculating the accuracy rate, the denominator should be the number of targeted tone that the children intended to produce. And the numerator would be the number of correct tones which are determined both by children’s referential meanings and care-takers’ target tones. For example, given that the reduplication form of ‘a pen’ the children learned is [pi35 pi21], when the child uttered [pi55 pi21], the tone of the first syllable would be considered a tone error. Hua & Dodd (1995) provided a criterion that a tone was viewed stable when two-third (66.7%) of the tones were produced correctly.
With this criterion, we could precisely examine whether a tone is acquired or not by checking the accuracy rate. Zhu (2002) applied the 66.7% criterion of stabilization from Hua & Dodd (1995), and also applied a 90% criterion of stabilization to measure the advanced level of tone stabilization.
3.2.4 Substitution pattern in tonal errors
Meaningful tokens which were determined to be a tonal error would be further analyzed in this measure. Unstable and immature tones would sometimes be changed into other tones, and would be determined as a tonal error. Although the accuracy rate could pretty much depict the states for children’s tonal development, it could not tell us how
the number of correct tokens of a tone the number of targeted tokens of a tone
x 100%
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
47
children deal with immature tones. Therefore, the substitution pattern was used to demonstrate which tones were more likely to be realized in replacing the error tones.
The substitution pattern in tonal errors would be presented in a matrix that the row represents the target tones, and the column represents the actual realized tones. The example of the matrix is presented in Table 3.3.
Table 3.3 A sample matrix of substitution pattern in tonal errors
Target toneRealized tone
[55] [35] [21] [51] Total
[55]
24 19 17 60 50.8%[35]
3 13 10 26 22%[21]
4 7 11 22 18.6%[51]
0 5 5 10 8%Total
7 36 37 38 1185.9% 30.5% 31.3% 32.2%
The matrix could present the number of tokens and the frequency of substitution in each tone. If a tone should be produced in [55] but is realized as [35], it will be put into the first column and the second row. The percentages on the right column would tell us which tone is more frequently used in replacing others in tonal errors, and the percentages on the bottom row tell us which tone tends to make more tonal errors. In this sample in Table 3.3, there are 3 tonal error tokens that replace [55] to [35], and there are totally 26 tonal errors that are realized into [35], accounting for 22 % of the total errors. There are only 7 tonal errors whose target tones are [55], accounting for 5.9% of the total errors. With this
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
48
matrix, the substitution pattern in tonal errors could be displayed clearly.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
49
Chapter 4
Results and Analysis
This section will present the results and general analysis of the data, and the analysis will follow the method introduced in chapter 3. Section 4.1 will be the overview of the overall data including monosyllabic and disyllabic tokens. The results regarding the age of tone emergence will be presented in 4.1.1, and the tone frequency and accuracy rate will be shown in graphs in 4.1.2. In section 4.2, to obtain more specific results, the monosyllabic and disyllabic tokens will be analyzed separately. The monosyllabic tokens will be analyzed in 4.2.1, and the disyllabic tokens will be analyzed in 4.2.2.
We found that the unusual high frequencies of [21] in the first syllable and [35] in the second syllable in disyllabic tokens might result from the tone combination [21-35]
related to the reduplication of motherese, so the results of reduplication in motherese will be displayed particularly in 4.2.3. After we determined to exclude tokens produced in [21-35], the reanalysis of the modified disyllabic data will be shown in 4.2.4, and the reanalysis of the modified data combining both monosyllabic and disyllabic tokens will be put in 4.2.5. Then in section 4.3, the substitution patterns of tonal errors will be illustrated. Last but not least, the age of tone emergence and stabilization will be
‧
As mentioned in chapter 3, there were 16 children enrolled under Professor Wan’s NSC project, but only 6 children fitted in this study. Among the 97 hours of observational data collected from the 6 subjects, only 44 hours of the recordings were adopted. Table 4.1 summarized the subject information including the subject number, gender, the age range during the observation, and the total number of tokens uttered by each subject.
Table 4.1 Subject information
Subject Gender Age range Duration Total number of tokens
Three male and three female children were adopted in this study. The observation started from the age of 0;10 to 1;1 in the beginning and ended at the age of 1;5 to 1;6. The children enrolled were all between the babbling stage to one-word stage and their conceptual and lexical abilities were still under construction, so it was common for children to produce vocalization without recognizable meanings. However, the development of prosody has already started at the babbling stage, so it was applicable to collect tonal acquisition data from this stage. The observation lasted for 8 months, and
‧
there were originally 2286 tokens. Yet, the speech tokens which were spoken in Taiwanese or English, such as [ne55-ne1] 'milk' and [‘paj-paj] 'bye-bye’ were not supposed to include in this Mandarin tonal study. Plus, the utterances which were more than two syllables or utterances whose tones were not clear would also be inappropriate to this topic. After excluding these tokens, there were a total number of 2062 tokens which could be analyzed in this study. Children’s production had individual differences. As Table 4.1 shows, the most productive children were subject #1, #3, and #4 who uttered more than 400 tokens, while subject #5 and #6 produced the least number of tokens, which were under 100 tokens.
There were monosyllabic and disyllabic tokens in the data. Under these two big categories, there were two subgroups classified the tokens with meaning and tokens without meaning. The following table summarized the total number of tokens in different subcategories.
Table 4.2 Number of tokens in subcategories
Monosyllabic tokens Disyllabic tokens
Subject without
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
52
There were several findings derived from Table 4.2. First, it shows that most of the subjects, including #1, #3, #4, and #5, preferred producing disyllabic tokens than monosyllabic tokens. The disyllabic utterances outnumbered monosyllabic ones no matter in tokens with meanings or without meanings. There were 1474 disyllabic tokens, but only 588 in monosyllabic tokens. Second, tokens with clear meanings were more than those without clear meanings. The number of meaningful tokens was 1330, but that of the meaningless tokens was only 732. Third, the subjects who were more productive tended to produce more meaningful words. Subject #1, #3 and #4 were productive children that their numbers of tokens were much bigger than those of the other three subjects, and they all produced higher numbers of meaningful tokens than meaningless tokens. For example, subject #1 produced totally 396 meaningful tokens, but there were only 90 meaningless tokens among all his utterances. Yet, there were exceptions, such as subject #5 who produced the least number of tokens but had more meaningful tokens. The exception would be accounted for individual differences or sampling fluctuations.
4.1.1 Tone emergence ordering
Many researchers were curious about the ordering of tone emergence. The best timing for collecting this ordering was from the very beginning of children’s one-word stage. In children’s one-word stage, the referential meaning of the utterances became more clear, and children have learned to refer to a particular object by using a particular
‧
form. By tracing the meaningful words that children uttered, we were able to know which tone appeared first, and what was the ordering of tone emergence in early lexical development.
Table 4.3 showed the age-tracked data of tone emergence in six individual subjects. The criterion for determining whether a tone had emerged was to see if the tone was produced more than once by children (Vihman, 1996). Noted that the data collection of subject #1 started from the age of 1;1. By that time, all his tones had already emerged, so the exact age for each tone’s emergence was hard to detect from his data. The only thing we could make sure was that all his tones have emerged by the age of 1;1. Moreover, subject #6's lexical development was still before one-word stage that most his utterances were meaningless by the age of 1;6, so the analysis of tone emergence ordering would not be applicable in his utterances.
Most children uttered the high-level tone [55] in their first words, including subject
#2, #4, #5, and #6 between the age from 0;11 to 1;0. Only subject #3's [55] appeared the last and later than other tones. Later on, at the age of 1;1, falling tone [51] had emerged in
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
54
subject #2, #3, #4, and #5. Regarding the emergence of [35] and [21], these two tones seemed to emerged together after other lexical tones. Among all the subjects, except for subject #4 whose rising tone [35] and low-level [21] appeared together with [55] at 1;0, other subjects’ [35] and [21] developed later than [55] and [51] at the age between 1;2 to 1;3. As for the neutral tone, it appeared as early as other citation tones between 1;1 to 1;2 in most subjects, but subject #2 did not develop the tone until 1;6. The differences of the ordering might mostly accounted for individual differences. However, it should be noted that the data collection method had flaws that from children’s spontaneous speech, the opportunity for children to utter each tone was not equal, so there might be sampling fluctuation. For instance, a child may have already learned the rising tone [35] in 1;1, but we missed collecting her [35] at 1;1. Thus, the age of emergence of [35] would then be less accurate due to the sampling fluctuation.
Although the individual differences and the sampling fluctuation caused it hard to make conclusion, the overview of tone emergence still showed some similarities among subjects. First, [55] and [51] seemed to emerge earlier than [35] and [21] at their first-word stage. Table 4.3 indicated that both subject #2 and #5 uttered high-level tone [55] first and falling tone [51] the second by 1;1, and their [35] and [21] were appeared later by 1;2 and 1;3. Subject #3’s [35] and [21] also appeared later than [51] by 1;2, but her [55] appeared the last at 1;3. It may result from sampling fluctuation again because
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
55
her [55] also reached 90% of stabilization at 1;3. The data of stabilization will be presented later in the last section in this chapter. In subject #4’s data, the four contrastive tones appeared roughly at the same time, and the data in subject #6 could only show that [55] was the first appeared tone in his tonal development.
Second, once children have reached the one-word stage, the four contrastive tones would emerge one after another in a short period of time. Subjects #2, #3, #4, and #5 spent about one to three months to contrast all the citation tones, but it took more than five months for subject #6 to apply all tones in his lexicon. Third, the neutral tone tended to appeared last comparing to the four citation tones. Data showed that the neutral tone was the last emerged tone in subject #2, #4, and #5. Only subject #3 was the exception that her neutral tone was the second appeared tone which showed up later than [51].
4.1.2 Frequency and Accuracy rate
In the previous section, I presented the ordering and ages of tone emergence, and found that the high-level tone [55] and falling tone [51] appeared earlier than [35] and [21], and the neutral tone tended to appear last. Thus, in order to see whether the frequency and accuracy rate of tones would rank in the same sequence as the tone emergence, the number of occurrences and the correctness of each tone would be computed in the section. The tone frequency would inform which tone is more frequently used and which tone is less used. In addition to tone frequency, the accuracy rate is also
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
56
crucial for determining the degree of development. A tone with higher accuracy rate reflects a more mature and more stable degree of acquisition. By analyzing the tone frequency and accuracy rate, the detail of tone acquisition could be better described.
Among the 2062 tokens, there were 588 monosyllabic tokens and 1474 disyllabic tokens, and each of the disyllabic tokens had 2 syllables, so the 1474 disyllabic tokens would have 2948 syllables. Thus, there were totally 3536 syllables (588+1474x2=3536) in the data. As mentioned in chapter 3, the tokens without meaning could also apply frequency analysis. To see which tone was used more frequent, the 3536 syllables would all be included. The four tones were written in the tone values modified from Chao’s (1968) tone number, and the neutral tone was also included in the tone frequency chart with the label of ‘Neut.’
From the total of 3536 syllable tokens, the frequencies of all tones were calculated below and the number of tokens and frequencies were presented in Table 4.4 and Figure 4.1.
Table 4.4 Number of tokens and frequencies of tones in all syllabic tokens
[55] [35] [21] [51] Neut Total
Number of tokens 960 954 894 565 163 3536 Frequencies of tones 27.1% 27.0 % 25.3% 16.0% 4.6%‧
Figure 4.1 Frequencies of tones in all syllabic tokens
In Figure 4.1, the bar graph presented the frequency of each tone. The difference of the five percentages was statistically significant (χ2=18.2, p<.001). The highest three bars, [55], [35], and [21] were all higher than 25%. Falling tone [51] ranked as the fourth place with 16% of appearance, and neutral tone appeared last frequently, accounting for 4.6%.
Actually, it was unfair to compare the neutral tone with other tones in frequency, because in Mandarin, neutral tones only appear in weak stress that are usually in utterance-final position. If children acquire this phonological rule early in this stage, the neutral tone would scarcely be found in utterance-initial positions in monosyllabic and disyllabic tokens. In fact, Figure 4.1 showed that children preferred using [55], [35], and [21] the most, and the neutral tone appeared last frequently. The hypothesis that children learned the phonological rule of neutral tone in this stage might be approved. Then, to see whether the acquisition of tones could demonstrate the rule that ‘practice makes perfect,’
we need to compare the ranking with accuracy rate below.
The accuracy rate was calculated by taking meaningful tokens. Meaningful tokens
27.1% 27.0% 25.3%
‧
had clear targets which encoded with target tones, so the correctness could then be decided. Among the 3536 syllables, only 2354 tones were syllables that had target tones, so the 2354 syllables were included in the accuracy rate analysis. The accuracy rates were calculated by dividing the number of correct tokens by the number of targeted tones.
Table 4.5 presented the fraction of the accuracy rates in each tone, and the percentage of accuracy rates were shown in Figure 4.2.
Table 4.5 Number of tokens and accuracy rates of tones in all syllabic tokens
[55] [35] [21] [51] Neut
Number of correct tokens
/ number of targeted tones 468/529 586/642 598/703 298/356 91/111 Accuracy rates of tones 88.5% 90.3 % 85.1% 83.7% 82.0%
Figure 4.2 Accuracy rates of tones in all syllabic tokens
The result suggested that all the lexical tones including neutral tone showed high percentages in accuracy rate. The most stable tone was [35] that reached 90% of accuracy rate. The high-level tone [55] showed lower rate than [35], accounting for 88.5%. The third and fourth places were [21] and [51], and the neutral tone was the last. However, the numbers of accuracy rates in all tones were all high, and there was no significant
88.5% 90.3%
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
59
differences found among the five percentages (χ2=5.229, p=0.265). If the accuracy rates were not different statistically, we could only say that the degree of tonal development was similar during the age between 0;10 to 1;6.
Now that the accuracy rate in the overall data could not indicate which tone was more mature and which was not, the following analyses would separate the data into monosyllabic and disyllabic tokens and see whether there would be more specific results in the ordering of tonal acquisition.
4.2 Subgroup analyses in monosyllabic and disyllabic tokens
It is mentioned that the tokens collected in this study included monosyllables and disyllables. To see whether different types of tokens would affect the development in tonal acquisition, the two types of subgroups would be analyzed separately. During the observation, we found that younger children could only produce short utterances. There were few tokens that were longer than two syllables, so the tokens collected here were all one-to-two syllabic utterances. As mentioned in section 4.1, the number of tokens in disyllables was much higher than that in monosyllables. The table below provided more information about the number of tokens in different syllable positions in disyllabic tokens.
‧
Table 4.6 Number of tokens of each syllable in monosyllabic and disyllabic
tokensTone Monosyllable Disyllable 1 Disyllable 2 Total
[55]
162 445 353 960Table 4.6 listed the number of tokens in each tone that showed up in three different positions, including monosyllable, and the first and second syllable in disyllabic tokens.
There was a significant position effect found in number of tokens among tones (χ2=735, p<.001), which means that children tended to produced different number of tokens in different positions. Take the rising tone [35] for example, it appeared 127 times in monosyllabic tokens, but appeared 238 times in the first syllable position in disyllabic tokens, and the number increased dramatically in the second syllable of disyllabic tokens.
The circumstances were not significantly found in [35] and [21], and it would be discussed later in the further exploration.
4.2.1 Monosyllabic tokens
The number of tokens and frequencies of monosyllabic tokens were presented in Table 4.7. The bar graph in Figure 4.3 makes it easier to compare the differences among
frequencies of tones.
1 The neutral tone in Mandarin is not supposed to be in the utterance-initial position. The neutral tone here was produced by children imitating the utterance-final affix.
‧
Table 4.7 Number of tokens and frequencies of tones in monosyllabic tokens [55] [35] [21] [51] Neut Total
Number of tokens 162 127 127 171 1 588
Frequencies of tones 27.6% 21.6 % 21.6% 29.1% 0.2%
Figure 4.3 Frequencies of tones in monosyllabic tokens
Different from the frequencies presented in the overall data in Figure 4.1, the ranking of tone frequencies in monosyllabic tokens was [51]> [55]> [35], [21]> N, and their percentages were proved to have significant difference (χ2=26.9, p<.001). The falling tone [51] accounted for 29.1% of the total number of monosyllabic use, and high-level tone [55] accounted for 27.6%. The other two lexical tones [35] and [21] reached the same percentage, 21.6%, which appeared less frequent than [55] and [51]. The neutral tone had only 0.2% of occurrences, because the phonological constraint in Mandarin does not allow it to appear in the utterance-initial position. Compare to Figure 4.1, the ordering of frequencies in Figure 4.3 could better correspond to the sequence of tone emergence that the [55] and [51] appeared earlier and more frequently than [35] and [21]. It seemed that the earlier appeared tones were produced more frequently in monosyllabic tokens.
With regard to the accuracy rates in monosyllabic tokens, the results here were also
27.6%
‧
different from that of all syllabic data.
Table 4.8 Number of tokens and accuracy rates of tones in monosyllabic tokens
[55] [35] [21] [51]
Number of correct tokens
/ number of targeted tones 61/72 42/44 84/119 61/71 Accuracy rates of tones 84.7% 95.5 % 70.6% 85.9%
Figure 4.4 Accuracy rates of tones in monosyllabic tokens
The tone which obtained the highest accuracy rate was [35] that accounted for 95.5%, followed by [51] and [55] whose accuracy rates were 85.9% and 84.7%. The most immature tone in monosyllabic tokens was [21] whose degree of stabilization was at 70.6%. The accuracy rate in neutral tone was not applicable here in the monosyllabic tokens, because in Mandarin the neutral tone is only allowed to appear in the utterance-final position. The monosyllables were all in the utterance-initial positions, so it would be impossible for a neutral tone to produce correctly in the monosyllabic speech tokens. Therefore, the ranking of the accuracy rate of monosyllabic tokens was [35]>
[51]> [55]> [21]. The accuracy rates showed similar ranking with tone emergence that [55] and [51] was more stable and appeared earlier than [21] except for [35]. It was more
84.7%
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
63
questionable why the later appeared tone [35] gained the higher rate in accuracy.
According to the raw data, it was showed that children uttered only few [35] in monosyllables, but once they uttered [35], they produced it with high correctness. More interestingly, the monosyllables children uttered were sometimes the second syllable of the disyllabic tokens, such as [je35] in [je21-je35] ‘grandpa’ and [ma35] in [ma21-ma35]
‘mother.’ These examples though were part of the reduplications we still viewed them as correct tokens. Thus, the high accuracy rate of [35] might result from this situation.
4.2.2 Disyllabic tokens
When the monosyllabic tokens were analyzed separately in tone frequency and accuracy rate, we found that the ranking was different from that in the overall data. As for disyllabic tokens, would the results of frequencies and accuracy rates be similar to the results presented in overall data in section 4.1.2 or to the results in monosyllable data in section 4.2.1? In the following, the frequencies and accuracy rates of disyllabic tokens would be shown in separate syllable positions.
When the monosyllabic tokens were analyzed separately in tone frequency and accuracy rate, we found that the ranking was different from that in the overall data. As for disyllabic tokens, would the results of frequencies and accuracy rates be similar to the results presented in overall data in section 4.1.2 or to the results in monosyllable data in section 4.2.1? In the following, the frequencies and accuracy rates of disyllabic tokens would be shown in separate syllable positions.