中国高级英语学习者对反义关系 的习得 山东大学外国语学院 王勇 摘要 • 根据国外研究者对反义关系典型性的研究,本文作 者设计了一套76对反义词对,对山东某大学英语专业三 年级学生和美国某高校的本族语者进行了反义关系典型 性判断的测试,并将测试结果进行对比,以期找出中国 高级英语学习者在反义关系习得方面的规律,并对结果 做出解释。实验结果表明,中国高级英语学习者在部分 反义词对的判断上与英语本族语者有着显著差异,这些 显著差异在典型性反义词和一般性反义词中均有反映, 其原因较为复杂,包括但不限于反义词对的共现频率、 语义范围以及概念对立性;其中反义词对的共现频率及 MI值对于中国高级英语学习者及英语本族语者的反义词 对典型性判断测试具有最强烈的影响。同时,对某些反 义词对进行了案例研究,深入探讨了上述因素对于反义 词对典型性判断的影响。 Canonicity of antonymous pairs 反义词对的典型性 • “Language users can intuitively sort ‘good’ (or prototypical) antonyms from not-so-good ones and downright bad ones” (Murphy, 2003: 11). This is often referred to as the ‘clang phenomenon’ – a term used to describe the reaction to those pairs that intuitively strike the hearer as being good ‘opposites’ (Charles and Miller, 1989; Muehleisen, 1997). The following working definition of canonical antonyms is adopted in this thesis: Canonicity of antonymous pairs 反义词对的典型性 • Canonical antonyms are pairs of words in binary semantic opposition associated by convention as well as by semantic relatedness (e.g. wide/narrow). The notion of canonical antonymy is different from semantic opposition in which the meanings are incompatible, but the words are not necessarily conventionally paired (e.g. cold/scorching, calm/nervous). antonymous pairs 反义词对的选择 • The antonymous pairs used in the canonicity judgment task are mainly taken from two sources: that of Deese (1964) and Sabourin (1998). • Deese (1964: 347-57) picked from the data from the psycholinguistic elicitation tests forty word pairs which he considered among the most important in English. Justeson and Katz used these antonyms in their research and regarded them as “historically important” (1991: 142). However, since Deese’s work was conducted before access to corpora was possible, it was based entirely on the results of word association tests. • Deese chose 278 adjectives and used them to elicit response from 100 informants. When a pair of contrast words successfully elicited one another more than any other word, they were added to the list of antonymous pairs, which ultimately numbered forty. Antonym canonicity judgment test 反义词对典型性测试 • • • • • • • • • • • • • Antonymous pairs list To what extent can the following word pairs be regarded as “perfect” antonyms? Please rate them according to your intuition, circling a number on the 10-point scale following each pair. (10 stands for perfect antonym, 1 means not antonym at all. For example, 10—⑨—8—7—6—5—4—3—2—1). active – passive 10—9—8—7—6—5—4—3—2—1 active – inactive 10—9—8—7—6—5—4—3—2—1 agitated – quiet 10—9—8—7—6—5—4—3—2—1 agitated – calm 10—9—8—7—6—5—4—3—2—1 alive – inanimate 10—9—8—7—6—5—4—3—2—1 alive – dead 10—9—8—7—6—5—4—3—2—1 bad – good 10—9—8—7—6—5—4—3—2—1 bad – evil 10—9—8—7—6—5—4—3—2—1 big – small 10—9—8—7—6—5—4—3—2—1 big – little 10—9—8—7—6—5—4—3—2—1 The difference between NSs’ and NNSs’ results of judgments of the canonicity of the antonymous pairs 学习者与本族语者在反义词对典型性判断测试中的差异 • In order to examine any possible difference between NSs’ and NNSs’ results of judgments of the canonicity of the antonymous pairs, a Oneway ANOVA is adopted to achieve this end. • As can be seen from the table in Appendix VIII, among the 76 antonymous pairs, 28 of them demonstrate significant difference between NSs’ and NNSs’ results, accounting for 37% of the whole set of antonymous pairs. 学习者与本族语者在反义词对典型 性判断测试中差异显著的词对 • These pairs are: agitated / quiet, alive / dead, bad / good, bad / evil, big / small, big / little, expensive / cheap, expensive / inexpensive, good / bad, good / evil, illegal / legitimate, permanent / transient, poor / affluent, right / incorrect, safe / dangerous, simple / complicated, simple / sophisticated, bend / stretch, lose / win, addition / subtraction, fact / fiction, fact / rumor, failure / success, generosity / greed, generosity / miserliness, guilt / punishment, strength / weakness, strength / vulnerability. Possible factors influencing the scores on antonym canonicity judgment test 影响反义词对典型性判断测试分数的因素 • Is there any relation between the scores on the judgment of canonicity of antonymous pairs on the one hand, and MI score, the frequency of stimulus word (first member of the antonymous pairs) in the corpus, the frequency of response word (second member of the antonymous pairs) in the corpus, or the frequency of co-occurrence of the two members in the antonymous pairs within a certain span in the corpus, on the other hand? • In order to answer this question, a bivariate Correlations Test is performed to see whether there is any relation. • As can be seen from the above two tables, both NSs’ and NNSs’ judgment scores are significantly correlated with the co-occurrence frequency of the two members of the antonymous pairs (all ps < .01), but not significantly correlated with other factors mentioned earlier. • That attests the assumption that native speakers have an intuition about the cooccurrence of the canonical antonymous pairs. And the result of the NNSs’ judgment scores also demonstrate the tendency of advanced Chinese EFL learners to become familiar with this kind of information due to the intensive teaching and extensive reading requested by English majors. The relation between NNSs’ discrepancy from NSs’ antonym canonicity judgment scores and NNSs’ overall L2 proficiency and specific lexical proficiency 学习者与本族语者在反义词对典型性判断测试中的分数差异与二语水平 以及词汇水平的关系 • From above discussion, we know that NNSs’ performance on antonym canonicity judgment test deviates from that of NSs’. We wonder if we can use this discrepancy as an indicator of NNS subjects’ overall L2 proficiency and specific lexical proficiency. In order to do that, we calculate the discrepancy using the following formula: • discrepancy = √Σ (SNNSi-SNSmeani)2 • in which SNNSi means the score of an NNS individual on ith antonymous pair (i ranges from 1 to 76), while SNSmeani is the mean score of NS on ith antonymous pair. By (SNNSi-SNSmean)2 we get the square of the difference between NNS individual’s score on ith antonymous pair and that of NNSs’ mean score on this pair. After we get the sum of all the 76 pairs’ square differences, we may get NNS individual’s discrepancy by deriving the square root of this sum. • After we get each NNS individual’s discrepancy score, we use SPSS to detect if it is correlation with his/her general linguistic proficiency (as indicated by TEM4 score) or with his/her specific lexical proficiency (as indicated by the several lexical mentioned above). Table 5.18 shows the results of correlation test. Discrepancy vs. TEM4 • From the above table, we can see that NNS individuals’ discrepancy score has a negative correlation with their TEM4 scores (r = -.225, p = .024). This result conforms to our expectation; it means the less the discrepancy score, the closer the NNS individual’s performance with that of NSs, the higher the NNS individuals’ performance on general linguistic proficiency. However, their discrepancy scores show no relationship with their performance on synonym differentiation test (score20) and specific lexical test, i.e., Word Associates Test (WAT), Productive Levels Test (PLT), and Levels Test of Vocabulary – Recognition (LTV-R). • The reason might be that antonymy is indeed a complicated construct and thus need further investigation before we can tease out its nature. Besides this, the NNS individuals’ discrepancy scores also have a negative correlation with their self-rating scores on their writing ability (r = -.206, p = .031). This is also reasonable, because writing could be considered the most complicated process among the four linguistic skills, and acquisition of antonymy will occur at an advanced stage of lexical competence. An extension of the present study • In order to picture the route followed by advanced Chinese EFL learners when acquiring the L2 lexical semantic relations, the synonym differentiation test and the antonym canonicity judgment test were also administered to a group of freshmen of English major. For fear that the unfamiliar word would frustrate their understanding of the sentence meaning and influence their judgment, the subjects are required to mark them. The results showed that, the synonym differentiation test has too many unfamiliar words for a large part of the subjects. Therefore, the test results were set aside. • The results of the antonym canonicity judgment test nonetheless could serve the purpose, since we may delete antonymous pairs with unfamiliar words marked. The procedures are as follows: first, the results are typed into the dataset; then, a simple count of the marked words is carried out; third, the results with marking are treated as missing value, and if these marked results were more than 15% of the number of the subjects (that is, it is higher than 20), then the concerned antonymous pairs will be deleted; only the antonymous pairs with less than 15% markings are reserved for further consideration. We can follow the rule of thumb to treat them as missing value and thus replace them with series mean of the concerned item. • A One-Way ANOVA is administered to show whether there is any significant difference between the results by freshmen Chinese English majors, junior Chinese English majors, and native speakers (for significant results, see Appendix VII). At the same time, the means plots of the One-Way ANOVA procedure are examined in order to find any significant results. On examining the results, we find an interesting phenomenon: to the native speakers, the negative prefix is a strong indicator of antonymous status. If the antonymous pairs consist of one word and another word with a negative prefix, the native speakers usually give higher marks to them than Chinese EFL learners. Negative Prefix Hypothesis 反义前缀假设 • In the remaining antonymous list, there are 9 sets of such pairs: active-inactive, correct-incorrect, expensiveinexpensive, happy-unhappy, healthy-unhealthy, illegallegal, married-unmarried, safe-unsafe, and disprove-prove. Except for happy-unhappy, the other 8 sets are invariably shows a sharp rise on the native speakers’ judgment scores. We could cautiously call this phenomenon as Negative Prefix Hypothesis: native speakers of English rely heavily on negative prefix to judge the antonymous status of a word pairs in a context-free setting. We could in the future include more such pairs in our test and extend this test to EFL learners with other language background to see its reliability. • In order to look at the comparison between each groups of subjects to make a more detailed study of the antonymous pairs, an Independent-Samples t test is administered (1) between freshmen Chinese English majors and junior Chinese English majors, (2) between freshmen Chinese English majors and native speakers, and (3) between junior Chinese English majors and native speakers. The results show that: • The scores on 11 antonymous pairs are significantly different between freshmen Chinese English majors and junior Chinese English majors. These pairs include bad-evil, difficult-simple, false-real, good-evil, poor-rich, poor-affluent, confirm-deny, fail-fulfill, addition-reduction, failure-achievement, and generosity-greed. There are still two antonymous pairs with a marginally significant difference between the above two groups. They are right-incorrect and shortenlengthen. • The scores on 22 antonymous pairs are significantly different between freshmen Chinese English majors and native speakers. These pairs include alive-dead, bad-good, big-little, correct-incorrect, difficult-simple, expensive-cheap, expensive-inexpensive, good-bad, good-evil, illegal-legitimate, right-wrong, rightincorrect, begin-end, bend-stretch, confirm-deny, failfulfill, lose-win, shorten-lengthen, fact-fiction, failureachievement, generosity-greed, and guilt-punishment. There are still 6 antonymous pairs with a marginally significant difference between the above two groups. They are active-inactive, big-small, correct-wrong, married-single, simple-complicated, and lower-raise. • The scores on 20 antonymous pairs are significantly different between junior Chinese English majors and native speakers. These pairs include alive-dead, biglittle, big-small, correct-incorrect, expensive-cheap, expensive-inexpensive, good-evil, healthy-unhealthy, illegal-legitimate, poor-affluent, right-incorrect, safedangerous, simple-complicated, begin-end, bendstretch, lose-win, fact-fiction, generosity-greed, guiltpunishment, and strength-weakness. There are still 6 antonymous pairs with a marginally significant difference between the above two groups. They are bad-evil, good-bad, safe-unsafe, lose-gain, lower-raise, and addition-reduction. • What factors caused these? First, we performed a Multiple Regression Analysis to see which of the factors mentioned earlier (MI score, the frequency of stimulus word in the corpus, the frequency of response word in the corpus, or the frequency of co-occurrence of the two members in the antonymous pairs within a certain span in the corpus) contribute most to the judgment result. The results of Multiple Regression Analysis show that there are differences between freshmen of English majors on the one hand and junior of English majors and native speakers on the other. • Comparing the results, we find that the factor of frequency of co-occurrence contribute most (R2 = .121) to the scores by freshmen of Chinese English majors, which reflects a medium effect. Apart from this factor, no other factors have a significant role in it. • On the other hand, the factor of frequency of cooccurrence also features as the most contributing factor to the scores by juniors of Chinese English majors as well as by native speakers. Apart from this, there is another factor, Mutual Information of F1*F2 co-occurrence, also contributes to their scores. They combined together will account for 9.7% and 8.4% of the variance of scores by Chinese juniors and native speakers respectively. • This result shows that the intermediate EFL learners rely on frequency of co-occurrence to judge the antonym canonicity. As their proficiency increases, they also turn to MI value for help. The addition of this further source could be understood as the increasing exposure to the L2 in their learning settings. • These results seem promising, since they can explain 10% or so variance of the scores, which amounts to a medium-sized effect. It also suggests the judgment of antonym canonicity is an issue much more complex than we first think. More experiments are needed to reveal its nature.