IV. DATA ANALYSIS
4.2 True Classifiers
4.2.2 An Experiment on Classifier Identifications
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
chu1 齣 jia4 駕 ren4 任 zhan4 站 chu4 處 jian1 間 shan4 扇 zhao1 招 chuang2 床 jie4 介 sheng1 聲 zheng4 幀 chuang2 幢 jie2 節 c suo1 梭 zhi1 只 dang3 檔 jing1 莖 suo3 所 zhi1 支 c dao4 道 ju4 句 tai2 台 c zhi3 紙 dian3 點 c ju4 具 tang4 趟 zhou2 軸 ding4 錠 juan3 卷 ti2 題 zhu1 株 dong4 棟 kou3 口 c ting3 挺 zhu4 柱 dong4 洞 long3 壟 tong1 通 zhu4 炷 du3 堵 lü3 縷 c wan1 彎 zhuang1 樁 duo3 朵 lun2 輪 c wan1 灣 zong1 宗 c fa1 發 mei2 枚 wan2 丸 zun1 尊 fang1 方 men2 門 c wei3 尾 zuo4 座 fen4 分 c mian4 面 xi2 席
fen4 份 c ming2 名 xi2 襲
According to identical norms to re-classify classifiers proposed by these representative studies, I find that these twenty-two core classifiers are shared by representative studies. Thus, these twenty-two core classifiers are definitely true classifiers. Non-core classifiers are not shared by representative studies, so these non-core classifiers are only classifiers, not true classifiers. However, I think that there is the possibility for these non-core classifiers to become true classifiers. Thus, through an obejective questionnaire experiment to double check the possibility for these non-core classifiers to become definitely true classifiers.
4.2.2 An Experiment on Classifier Identifications
In this section, a questionnaire experiment is adopted to investigate the degree of possibility by which the non-core classifiers could become definitely true classifiers.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
The methodology, results and discussion are provided below.
Methodology
Our subjects in this questionnaire experiment limit to have a linguistic knowledge because subjects need linguistic knowledge when they use linguistic-based tests to differentiate classifiers and measure words. In this experiment, subjects are twenty-six linguistics graduate school students in National Chengchi University.
They are all Taiwnese and Mandarin speakers and all have received formal linguistic training. The questionnaire experiment comprises two parts. Part One includes a brief introduction of using yi-multiplier, de-insertion and ge-substitution to distinguish classifiers from measure words and a pre-test of classifier identifications. Becasuse the numeral / adjectival stacking is an optional test, it is not included in the Part 1.
The pre-test contains twenty-six test items with eighteen core classifiers from Table 22 and eight measure words whose multipliers are definitely not 1. (Please refer to Appendix A for the details of the pre-test). Subjects were asked to do the pre-test after reading the introduction. No time limit was set for the introduction and the pre-test. In the pre-test, three options were offered for each test item. If subjects view the test item as a classifier, they are to circle option C which represents classifiers. If subjects view the test item as a measure word, they are to circle option M which represents measure words. Or, if it is possible for the test item to be both a classifier and a measure word, subjects are to circle option O which represents classifiers and
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
measure words. If subjects circle option O, they are asked to write down their interpretations of the meaning of the classifiers and measure words. If the correctness in the pre-test has 92 % to 100 %, subjects will be requested to do the formal test.
Totoally twenty subjects are requested to do the formal test. Part Two is composed of ninety non-core classifiers that are obtained from Table 29. The details of the formal test are shown in Appendix B. There is also no time limit during the formal test and three options are also offered for each test item. Subjects are asked to carry out the formal test in the same way as the pre-test.
Results and Discussion
In the data analysis, I use percentages to represent the statistics. If a subject circles
C (classifiers) once, then the classifiers will be calculated once. If a subject circles M
(measure words) once, then the measure words will be calculated once. Also, if a subject circles O, O will be counted once. The percentage of three options in each test item adds to one hundred percent. Table 30 shows the statistics of the results from the formal test. In the following, two aspects from Table 30 will be discussed.
First, to discuss the relations between these three options from option O’s point of view: the option O is chosen twenty-four times. This result shows that the subjects have experienced some degree of confusion over these twenty-four items when differentiating these twenty-four items as classifies or measure words. The higher the percentage of the choice of the option O, the more uncertainty the subjects feel. For
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
example, the percentage of option O for bi3筆 is 20 %. And the percentages of the occurrence of bi3筆
as a classifier and of bi3
筆as a measure word are 40 % and
40%, respectively. We can thus infer that if the percentage of the occurrence of option O is high, then the percentages of the occurrence of one word as a classifier and of one word as a measure word are likely to be concordant. However, this inference is not absolute because there are some counter-examples. For example, the percentage of the occurrence of option O of chu4處 is 10% and the percentage of the occurrence of chu4處 as a classifier and of chu4處 as a measure word are 80%and 10%, respectively, and the percentage of the occurrence of option O of dang3檔 is 5% and the percentage of the occurrence of dang3檔 as a classifier and of dang3 檔 as a measure word are 65% and 30%, respectively. Although the percentage of the occurrence of option O of chu4處
is higher than that of dang3
檔, the difference in the percentages of the occurrence of chu4處as a classifier and of chu4
處as a
measure word is larger than that for dang3檔.Second, the relations among percentages of one word as a classifier and as a measure word and as option O show the following three phenomena. One phenomenon is that the percentage of one word as a measure word is over than as a classifier, such as bu4部
c, lü3
縷c, pan2
盤, tang4
趟and xing1星. Most subjects circled option M for bu4部c, lü3
縷c rather than option C. This implies that the
properties of bu4部c and lü3
縷c for being a measure word are more prominent than
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
for being a classifier for most subjects. Next, pan2盤
and tang4
趟are strongly recognized as measure words by the subjects because the percentages of pan2盤and
tang4
趟as a measure word is 50% more than that of pan2
盤and tang4
趟as a
classifier. This implies that pan2盤
and tang4
趟 are measure words for most of subjects. Last is xing1星. Although xing1星 is rare in Taiwan Mandarin, most the subjects regard xing1星 as a measure word. This implies that the subjects tend to regard a new word as measure word. This further supports that measure words are an open set and are acceptable to innovations as proposed by Her and Hsieh (2010).The second phenomenon is that the percentages of one word as a measure word and as a classifier are equal, such as qi2 畦. Unless the number of subjects is increased, it will be difficult to show if qi2畦 is a classifier or a measure word.
The last phenomenon is the percentages of one word as a measure word and as a classifier are quite similar, such as jia4駕
, long3
壟and wan1
灣. Jia4駕, long3
壟 and wan1灣 may not so common in subjects’ daily lives, or jia4駕, long3
壟and
wan1
灣, may be metaphorical usages in literature, so subjects may have difficulties in differentiating these words. Thus, the percentages of these words as a classifier and as a measure word are quite close.Table 30: Statistics of Non-core Classifiers
No. Test Items Percentage of Test Items as a classifier (%)
Percentage of Test Item as a measure word (%)
Percentage of Test Item as both a classifier and a measure word (%)
‧
13
C mark is used to symbolize this ambiguous classifier functioning as a classifier rather than as a
measure word. C mark does not appear in the pre-test or the formal test.
‧
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
75
ze2 則 100 0 0
76
zhan3 盞 100 0 0
77
zhan4 站 60 40 0
78
zhao1 招 90 10 0
79
zheng4 幀 100 0 0
80
zhi1 只 100 0 0
81
zhi1 支 c 100 0 0
82
zhi3 紙 95 0 5
83
zhou2 軸 90 5 5
84
zhu1 株 90 10 0
85
zhu4 柱 95 5 0
86
zhu4 炷 95 5 0
87
zhuang1 樁 90 10 0
88
zong1 宗 100 0 0
89
zun1 尊 100 0 0
90
zuo4 座 100 0 0
After discussing the relations between the three options, Table 31 shows the percentage of test items as a classifier from high to low. In Table 31, I strictly stipulate that only test items with a 100% identification as a classifier are true classifiers because these test items are not only classifiers that mentioned in Section 4.2.1 but also objectively reconfirmed as classifiers by twenty subjects. The above dual certifications support that test items with a 100% identification as a classifier are true classifiers. There are a total of thirty-nine test items with a 100%
identification as a classifier in Table 31. These thirty-nine test items are thus definitely true classifiers under my stipulation. The remaining test items are classifiers, but they are not true classifiers because they violate my stipulation that only test items with a 100% identification as a classifier are true classifiers. Merely,
‧
the remaining test items are closer to true classifiers if the percentage of the test items as a classifier is higher. For example, test items with a score of 95% are closer to true classifiers than those with one of 90%. Then, test items with a score of 90%
are closer to true classifiers than those with one of 85% again.
Table 31: Percentage of Test Items as a Classifier in Non-core Classifiers
No. Test Item Percentage of Test Items as a classifier (%)
No. Test Item Percentage of Test Items as a classifier (%)
1
ban4 瓣 c 100
46men2 門 c 95
‧
Because the twenty-two core classifiers in Section 4.2.1 and the thirty-nine non-core classifiers in this section are definite true classifiers, both of them compose true classifiers which are a group of definitely true classifiers. Sixty-one true classifiers are shown in Table 32.
Table 32: 61 True Classifiers