Chapter Three Methodology
In this chapter the methodology of the longitudinal and the cross-sectional study is
outlined in three major sections: the general design, a pretest and the revised
methodology from the pretest. In the section of general design there are descriptions
about the background of the target subjects, the speech stimuli for the subjects to read
out, the equipment for speech data analysis and the procedure of data collection. The
following section presents an account of how the administration of the pretest was
conducted to evaluate the appropriateness of the general design in terms of speech
stimuli and the sampling procedure. The final section is the formal execution of the
experiment modified from the experience of conducting the pretest.
3.1 General Design
General design of the methodology was intended to control the means of
collecting and analyzing data from every subject in identical procedure. It is believed
that better controlled procedure would to some extent prevent unexpected variables
from appearing during the experiment on children. Moreover that may help avoid
differences in VOT values due to the process of data collection (Cho and Ladefoged
1999, Romaine 1984, Smith and Kenney 1999).
3.1.1 Subjects
In the longitudinal study, there were two girls aged 1;8 and 1;6 respectively when
they were seen in the beginning of the study. Both of them live in Taipei and speak only
Mandarin Chinese. They have no problems regarding linguistic development and
hearing ability according to parents’ report. The two children’s family members were
able to use Mandarin and Taiwanese but the majority of their domestic conversations
was carried on by Mandarin. Given that both Mandarin and Taiwanese are spoken by a
large population in Taiwan, some contact with Taiwanese is unavoidable. However the
influence of Taiwanese on the two children’s Mandarin is presumed to be minimal
because Mandarin is the major language in communication used by the two families. In
addition children’s parents did not teach them any foreign language at home nor did
they send the two girls to learn one at school. Therefore both of the two children were
assumed to speak Mandarin Chinese as their first language.
On the other hand, there were five groups of children in the cross-sectional study.
The children were two to six years old. Each age group consisted of at least twenty
people and almost equal number for both genders. They were recruited from
kindergartens and nursery schools in Taipei and were native speakers of Mandarin
Chinese. It is assumed that the children might be exposed to Taiwanese to some extent
because it is one of the local dialects. To minimize the effect of Taiwanese, we
consulted the teachers of the schools and that allowed for a preliminary screening of
prospective subjects who only used Mandarin as long as they are at the kindergarten or
nursery school. It is thus expected that the subject’s production of Mandarin was not
crucially affected by Taiwanese. Moreover teacher’s report was employed to exclude
the children who spoke any foreign language from participating in this experiment.
Before we are allowed to collect children’s speech, an official letter and parental
consent form was sent to the schools and parents (see Appendix).
Besides, there is an adult group which serves as the norms of speech in the
comparison with the results of children’s VOT values. This group is composed of
sixteen native speakers of Mandarin, seven males and nine females, aged from 18;11 to
43;03 (mean age = 28;4). They are students and faculty members of a college located in
Taipei. Some of them can speak Taiwanese but they use Mandarin most of the time in
their daily life. All of the adults have studied English in middle school and college but
none of them has ever spent more than one month in English-speaking country.
3.1.2 Stimulus
This study investigates the VOT values of word-initial stops, i.e. /p/, /ph/, /t/, /th/,
/k/, and /kh/ from one- to six-year-old children. Some previous studies suggested that
word-initial and word-medial contexts might affect VOT values so that the present
study placed all of the target stop consonants in word-initial position to avoid the
influence (Cho and Ladefoged 1999, Davis 1995, Smith and Kenney 1999). The tokens
for each stop in question were listed in Table 1. As can be seen from the data in Table 1,
each type of stop consonant was represented by six tokens. The prearranged stimulus
tokens can help assemble equal amount of instances across the three pairs of cognates
and prevent young subjects from avoiding saying certain types of sound in
spontaneous speech (Menn 1983). The stimulus words were depicted in picture,
Chinese character, and Chinese phonetic transcription in the same flesh card for the
subjects to name them. They were randomly arranged to avoid subjects’ detection of
the aim of the recording. The first word of the six tokens of /p/ was level tone, of /ph/
and /th/ was second tone, of /k/ was third tone, and of /t/ and /kh/ was fourth tone. As
sources of young children’s language can primarily be found at home (Luinge et al.
2006, Macken 1980, Smith and Kenney 1999), the tokens were intentionally selected
on the basis of children’s daily-life experience including names of cartoon characters
(e.g. Picachu), snacks (e.g. ice cream), animals (e.g. a dog), household appliances (e.g.
a refrigerator), and so on.
Table 1 List of Tokens in Pretest Stop consonant Tokens
/p/ /pa/ “eight”, /pei pau/ “a bag”, /pei ts/ “a cup”, /pi iang/ “a refrigerator”, /pü li i/ “glass shoes”, /pi thi lin/ “ice cream”
/ph/ /phi kha thiou/ “Picachu”, /pha lou thi/“to climb the stairs”,
Stop consonant Tokens
/pha i/ “a crab”, /phu thau/ “grapes”, /phi kuo/ “an apple”, /phai tuei/ “to line up”
/t/ /tin xua/ “a telephone”, /tan kau/ “a cake”, /ta ia/ “an elephant”, /tin / “a TV”, /tu ts/ “the belly”, /tin t/ “a lamp”
/th/ /th i t V ta/ “the parking lot”, /thin/ “the farmland”, /tha kuo/ “candy”, /thou fa/ “hair”, /thuo niau/ “an ostrich”, /thau ts/ “a peach”
/k/ /ku/ “a drum”, /kuei/ “a ghost”, /ku thou/ “a bone”, /kuai ta/
“a stick”, /kuo t/ “juice’, /kou/ “a dog”
/kh/ /khV thi/ “a living room”, /khV rn/ “customers”, /khu ts/
“pants”, /khou ts/ “a button”, /khuai ts/ “chopsticks”, /khV pn/ “a textbook”
The tokens were then used in the pretest for the purpose of evaluating children’s
familiarity degree to the words. Since subjects were all preschool children whose age
ranged from one to six years old, it was assumed that they could not yet read Chinese
words. Thus the words which were familiar and able to be drawn were considered the
best stimulus words for illiterate children. However the requirement for stimulus words
may be at the cost of some familiar words which could not be adequately represented in
picture. It is acknowledged that the familiarity degree of words could not be totally
maintained to an identical degree to every child.
3.1.3 Equipment
The subjects’ production was recorded on a digital recorder, Samsung Voice Yapp
BR-1640, connected to a microphone at a distance of approximately five to ten cm from
the subject’ mouth. The files of voice recordings were subsequently measured on Praat
which is a computer software of speech sound analysis commonly used by phoneticians
(Boersma and Weenink 2006). Then the computer program produces spectrograms and
sound waves of the subject’ pronunciation. Afterwards the cursor was placed at the
onset of the release of the plosive and of vocal fold vibration so that VOT of stops can
be measured and calculated. If there are multiple bursts, the measurement started from
the first burst.
The measurement of VOT will be illustrated by two examples. The first one is
the demonstration of VOT observed after the release of stop consonant, the typical
fashion of uttering either an aspirated or unaspirated stop consonant. The second
exemplifies the VOT occurring before the release of stop which exists in some
productions of the youngest subjects who have not yet developed the desired
articulatory ability to control aspiration of stops.
As can be seen from Figure 2, the lag time of the aspirated phoneme /th/ in /thu ts/
“a rabbit” was measured. In the Praat window the sound waves are placed in the
topmost space. Below the sound waves is the spectrogram indicating the formants of
/thu ts/. The following three columns serve as note pad for the users to write down
notes for a marked area. When the word-initial /th/ is uttered, the sound waves and wide
range of frequencies are presented in their given places. When a sonorant like /u/ is
produced, the sound waves and voicing bars exhibit in the wideband spectrogram. The
voicing bars can be observed from the straight lines, representing glottal pulses, in the
topmost space and by clear dark formants in the second space. To measure the VOT
value, one has to mark the area between the onset of /th/ and /u/. The marked area is then
turned pink and the VOT value 0.077469 is shown right above the area.
Figure 2 VOT measurement with Praat illustrated by /th ts/ “a rabbit”.
As shown in Figure 3, the voiceless unaspirated /p/ in /pi iang/ “a refrigerator”
exemplified the measurement of voicing lead. A child in the experiment pronounced /p/
in /pi iang/ “a refrigerator” as prevoiced sound. In the middle column the stop release
of /p/ is indicated by the burst of frication noise. Voicing preceding the stop release can
be observed by the straight lines in the top column and the low energy dark bands
without the appearance of formant in the middle column. Since voicing was detected
before the burst of the stop, here /p/ was produced as a voiced sound.
Instead of a voiceless unaspirated /p/, which should have a very short positive
VOT, she pronounced it as a voiced unaspirated one, making the VOT a negative one
(with voice bars preceding the release burst).
Figure 3 VOT measurement with Praat illustrated by /pi iang/ “a refrigerator”.
3.1.4 Data collection
Both adults’ and children’s speech data were collected from the task of picture
naming. For the longitudinal study, the recording sessions were conducted
approximately every two weeks for a four- and five-month period at the subject’s home.
The recording began when the two subjects were about 1;8 and 1;6 respectively and
ended after they turned 1;11. Each session lasted between twenty and thirty minutes.
Every time the children’s mothers presented the pictures and asked the child what it was.
If the child could not identify the object in the picture, the mother would tell her and
then ask her to say the word again. The whole process of interaction was monitored by
the experimenter and recorded by a digital recorder with a microphone which was held
in front of the child.
The recording session of the cross-sectional study proceeded either in an empty
classroom on campus or at home. The subjects were instructed to name the pictures
which were arranged in random order and speak to the microphone connected to the
digital recorder about five to ten centimeters in front of their mouth. The child was
accompanied by a teacher or a parent during the recording session. If the child failed to
name the picture, they were suggested to repeat the experimenter’s pronunciation.
When they finished naming the pictures, they would receive cookies for rewards.
The adults were also recorded in an empty classroom on campus or at home. They
also have the chance to preview all of the pictures before naming them. The only
instruction for them was to read out the words printed in the flesh card in which
Chinese characters, phonetic transcription and pictures were shown together. They
were not particularly suggested how to name the pictures but just reminded to speak to
the microphone.
After collecting their speech output, subjects’ sound files were transferred from
the digital recorder to a computer. Later they were analyzed with the software Praat and
the experimenter took down the VOT values. According to the convention of assigning
time values to stop consonants, VOT measured after the stop release was taken down as
positive numbers, otherwise as negative numbers (Gandour et al. 1986, Macken and
Barton 1980, Öğüt et al. 2006). Sometimes VOT values could not be obtained because
of extraneous noises (e.g. the school bell), voice overlay (where the teacher’s or
parent’s voice superimposed on the child’s), no burst, the following voiceless vowel,
and continuous voicing (where voicing went on from the previous voiced segment to
the target sound). All of the remaining VOT values went through statistic analyses
running on statistical package for social sciences 13.0 (SPSS 13.0): ANOVA and
one-tailed t-test of significance of the differences between the mean VOT values for the
aspirated and unaspirated stops at three places of articulation. The statistical significant
level of p<0.05 was used for all tests. Furthermore the frequency distributions of each
of the six stop consonants were drawn and the mean, standard deviation and range were
calculated.
3.2 Pretest
The administration of the pretest encountered a difficulty in reaching
kindergartens and nursery schools which did not teach English to children. The pretest
was intended to be conducted on the children of no foreign language background.
However it was not easy to find an English-free kindergarten or nursery school in
Taipei except for some public ones. While the linguistic background of children could
not be totally controlled in private schools, doing pre-test in public institution would
need some time to follow the official procedures and prepare some document. Due to
the limited time for this study, the pretest would have to be completed on schedule in
order to move on to the next stage. A reconciliatory solution was to do the pretest at
English-teaching private schools. The goal of this pretest was thus restricted to
evaluate the effectiveness of the stimulus tokens and procedure.
Before officially carrying out the experiment, we ran a pre-test on 75 children
recruited from one kindergarten and three day nurseries in Taipei County. Teachers
reported that no children had speech, language or hearing impairments. The ages of the
children ranged from two to six years old. The children were divided into five groups
by their ages. The two-year-old group consisted of four subjects, two boys and two girls,
ranging in age from 2;1 to 2;8 (mean age = 2;3); the three-year-old group consisted of
fourteen subjects, six boys and eight girls, ranging in age from 3;1 to 3;11 (mean age =
3;4); the four-year-old group consisted of thirty one subjects, nineteen boys and twelve
girls, ranging in age from 4;0 to 4;11 (mean age = 4;6); the five-year-old consisted of
seventeen subjects, and nine boy and eight girls, ranging in age from 5;0 to 5;11 (mean
age = 5;6); the six-year-old consisted of nine subjects, six boys and three girls, ranging
in age from 6;0 to 6;6 (mean age = 6;2).
All children who participated in this pretest were native speakers of Mandarin
Chinese with limited Taiwanese input mostly from grandparents and TV. They could
speak and comprehend only few isolated words in Taiwanese like body parts, kinship
terms and greetings. In addition they had English lessons one to two hours every school
day. We were concerned over the influence of English on children’s pronunciation on
Mandarin and thus decided not to measure and analyze their VOT values. We just
considered whether the procedure of data collection and the stimuli listed in Table 1
were adequate enough to elicit responses from children.
All of the thirty-six stimulus words were arranged randomly before the recording
session began. All recordings were carried out in an empty classroom on campus or at
home. If recording was conducted on campus, after the child came to the classroom,
they were instructed to name the pictures and speak to the microphone attached to a
digital recorder. If the child failed to name the pictures, they were told to repeat after the
experimenter. On the other hand when the child was recorded at home, their parent
would randomly present the pictures and encouraged the child to speak the name the
picture to a microphone. The experimenter was responsible for monitoring the process
and recording the child’s speech output. Once the naming process was finished, the
child was rewarded with some cookies.
During the process of the experiment we took notes of children’s general reaction
to the recording procedure and the stimulus words. According to this record a few
modifications were made to refine the administration of the investigation. First of all,
the subjects should not learn any foreign language such as English. Second, some of the
tokens which seemed too difficult for half of the children to name will be replaced with
other words which were considered more familiar to them. Third, the tone of the tokens
would be adjusted to the same one between the consonants of the same place of
articulation such as /p/ and /ph/. The adjustment was intended to reduce the possible
effect from different tones between cognate pairs. Also this improvement would also
require some replacement of the tokens. Furthermore some children appeared
uncooperative or reluctant to follow the procedure during the recording due to facing
strange surroundings so that attempts were to be made to prevent the child from feeling
unpleasant during the recording. For example, two children would accompany each
other during the recording to soothe their sense of insecurity from staying alone with a
stranger (i.e. the experimenter). Though teachers may be the better candidate for
children’s company, they had to look after other students while the experiment proceeds.
Consequently the alternative was to arrange two children at a time to record their
responses if the teachers give their consent to our practice.
In addition the children would have a chance to preview all of the pictures to
reduce their anxiety resulting either from the experience of encountering an unfamiliar
picture, or from the experience of speaking to a microphone. During the preview, we
would read the pictures, along with some explanatory remarks, to all of the subjects
even if they could identify the picture (Davis 1995). Previewing may also help younger
subjects to identify more pictures. Since we presented the same stimuli to elicit
responses from every age group, it was common to notice that younger children’s
receptive and expressive skills were more limited and had more difficulty naming the
object in the picture (Luinge et al. 2006). It is believed that when children saw the
same picture again, they would be more capable of naming the pictures than their first
encounter. Besides, preview was employed to pacify children’s intense emotion which
was found to be aroused by some of their favorite pictures. In the process of pretest
some children would suddenly scream as soon as they saw their preferable objects
depicted in the picture, such as ice cream. Such abnormal exclamatory utterances may
generate emphatic stress pitch, prolonged VOT and so on (Smith and Kenney 1999). To
keep children’s mood and attitude as normal as possible, we provided them with a
chance to preview the pictures. The abovementioned points were going to be put into
practice in the formal experiment.
On the other hand we learned that the teachers were concerned over the time the
children were out of their sight when we recorded children’s speech production in
another empty classroom on campus. Accordingly the teachers suggested sending the
children back to their classroom within about ten to fifteen minutes. Due to the time
limit on recording, the number of speech tokens used in the formal experiment might
be thus restricted. It usually took three to four minutes for a child to read the thirty-six
speech tokens, six tokens for a type of stop consonant, in the pretest. If we wanted to
increase the number of the stimulus token, it should be limited to the number that
could be finished within ten minutes.
3.3. Revised Methodology 3.3.1 Subjects
The subjects of the longitudinal study and the adult group were the same and their
background is not going to be repeated here. Thus the following description only points
out the young subjects in the cross-sectional study.
There were one hundred and two children participating in this study. They were
two to six years old coming from a kindergarten and three nursery schools in Taipei
County and Taipei City. They were classified into five groups according to their ages
when they were recorded. The first is the two-year-old group which consisted of twenty
one children, ten boys and eleven girls, aged from 2;2.18 to 2;11.08 (mean age = 2;7.14).
The second is the three-year-old group, with eleven boys and ten girls, aged from 3;1.11
to 3;11.26 (mean age = 3;6.16). The third is the four-year-old children consisted of
twenty children, ten boys and ten girls aged from 4;0.3 to 4;11.16 (mean age = 4;5.27).
Next is five-year-old group with ten boys and ten girls aged from 5;1.1 to 5;10.12
(mean age = 5;5.15). The six-year-old group is made up of twenty one children, ten
boys and eleven girls, aged from 6;0.9 to 6;10.10 (mean age = 6;4.18).
The children are native speakers of Mandarin Chinese. Some children have
limited exposure to Taiwanese and can comprehend several isolated Taiwanese words
like greetings, body parts and kinship terms. However teachers reported that they had
never heard Taiwanese uttered by these children. As for English, none but one
six-year-old girl just began learning English after school at the time of recording. Her
VOT values of the stop consonants were among the common ranges found in the
six-year-old group. Therefore her data were included in this experiment.
3.3.2 Stimulus
This study investigates the VOT values of word-initial stops. A total of 56
stimulus words some of which were revised from the pretest were used to elicit speech
data from adults and children. The number of the tokens was determined because of
the time limit on the recording session. To obtain the maximum quantity of speech
data from a child within ten minutes, fifty-six words were considered the largest
number of stimulus words based on the experience from the pretest. Every stimulus
word began with a stop consonant in question each of which had at least nine tokens as
shown in Table 2. For a quantitative study, it is required that every type of stop
consonant has at least fifteen tokens. The failure to prepare fro the prerequisite
number of words can be ascribed to limitations on time of recording and the
possibility of portraying the words. There are three pairs of stops which differ from one
another by place of articulation: /p/and /ph/ are labial cognates, /t/and /th/ are dental
stops and /k/and /kh/ are velars. The first word of the tokens of labials and velars is level
tone and of dentals fourth tone. Every token was represented in pictures, Chinese
characters and Chinese phonetic transcription in the same flash card. Some of the new
tokens replaced the old ones used in the pretest due to the consideration of familiarity
degree to children and maintenance of identical tone in homorganic stops. Though
some of the tokens were loan words, their syllable structure and phonotactic rules
coincide with those in original Mandarin words. Take /phi sa/ “pizza” for example. The combination of /ph/ and /i/ can be illustrated by /phi phi/ “criticism” and /phi ua/
“arsenic”. Otherwise the form of loan words was adapted to conform to Mandarin
phonological system, such as /pa pi ua ua/ “Barbie” where r in the first syllable of
“Barbie” was deleted in Mandarin pronunciation. These stimuli were to be spoken by
both children and adults in the picture naming process.
Table 2 List of Tokens in Formal Experiment Stop consonant Tokens
/p/ /pi thi lin/ “ice cream”, /pei ts/ “a cup”, /pa pi ua ua/
“Barbie”, /pi ia/ “a refrigerator”, /pan pi/ “Bambi”, /pau ts/ “bao zi”, /pan ma/ “a zebra”, /pa pu/ “Bob”, /pi khuai/ “an ice cube”, /pa/ “eight”
/ph/ /phi sa/ “pizza”, /phin thu/ “puzzles”, /phu khV phai/ “poker”, /phn uei th/ “a fountain”, /phai ou/ “to clap hands”, /phai tau/ “to shoot a photo”, /pha pha u/ “Tarepanda”, /phi pha thou/ “pingpong”, /phi li mau/ “Thunder Cats”,
/t/ /tin / “a TV”, /tan kau/ “a cake”, /ta ia/ “an elephant”, /tin xua/ “a telephone”, /tan/ “an egg”, /tin nau/ “a computer”, /tou tou ian / “Mr. Bean”, /tai u/ “a kangaroo”, /ta thou thian/ “swings”
/th/ /thu ts/ “a rabbit”, /thai ia/ “the sun”, /thiau u/ “to dance”, /thiau thiau xu/ “Tigger”, /thai an/ “Tarzan”, /thai khu tuan/
“a spaceship”, /tha tian/ “a mat”, /thiau / “to jump the rope”, /thai la/ “(Hamu)taro”
/k/ /ku tu/ “a princess”, /kau kn i/ “high-heeled shoes”, /ka
thin/ “a piano”, /kau fei/ “Goofy”, /kua niou/ “a snail”, /kuai kuai/ “Guai-Guai”, /ku ti/ “a rooster”, /kV ts/ “a pigeon”, /kuo ts/ “a pot”
/kh/ /khai tV/ “to drive a car”, /khV nan/ “Conan”, /khu/ “to cry”,
Stop consonant Tokens
/kha kha/ “Kang-Kang”, /kha fei/ “coffee”, /khV tou/
“tadpoles”, /khu lou thou/ “a skeleton”, /khu tia/ “a suit of armor”, /khun tu/ “an insect”
3.3.3 Data collection
The means of data collection for the longitudinal study were mentioned in section
3.1.4 and thus would not be repeated here. In principle the procedure and environment
in collecting data from the two girls was to remain as consistent as possible. For
example they were always recorded at the living room and their speech was elicited by
their mothers. Accordingly both girls’ mothers were responsible for the elicitation from
beginning to end. Both of the two children were always recorded at home. The time of
recording generally lasted between twenty and thirty minutes for young children
couldn’t sit still long. After a period of time they often looked uncooperative and
distracted by toys, snacks or other things within sight. When they were not able to focus
on picture naming, the session was called off whether they finished naming or not.
The procedural protocol of the recording session in the cross-sectional study was
partially adjusted as below to make the children feel more at ease. The child was always
accompanied by a teacher, a parent or a classmate during the recording session. Before
they named the pictures, they would have a chance to hear the teacher, parent or
experimenter read the pictures for them. When the recording was conducted on campus,
the teachers were concerned over the time children would miss the class so that we were
advised that every child finish recording within ten to fifteen minutes. The other
practice not mentioned here remained the same as described in section 3.1.4. As for the
adult subjects, the procedure was also untouched.
The subsequent procedure of analyses of the speech data were reported in section
3.1.4.
