The influence of semantic predictability on transposition effects in Chinese and English

2.1 Subjects

Thirty third-year English majors from a university in Jiangsu Province participated in the experiment. They were all native speakers of Chinese, with an average age of 21.67 (SD = 1.35). All subjects were right-handed, with normal or corrected-to-normal vision. They had all passed TEM-4, a nationwide proficiency test for English majors in China, and therefore were regarded as L2 learners of intermediate level.

2.2 Materials

Before Experiment 1, 36 students were asked to rank the familiarity scale of 120 target words used in Luke and Christianson’s (2012) experiment (“1” means very unfamiliar and “5” means very familiar). Based on the ranking, we maintained 60 words with a familiarity scale above 4.0 (Mean = 4.33). Since the materials used by Luke and Christianson (2012) only include sentences with high and low semantic predictability, we adjusted the word order of these sentences accordingly so that sentences with preceding and following contexts are also included in the materials. After each trial, a true-or-false tag question was added to the sentence to ensure that the subjects would read the sentences carefully. In addition, their answers to the questions would be used as a criterion in data selection: Only correct response times would be included in final data analysis. Examples of sentences and tag questions with 4 types of context are shown in Table 1.

Each target word has 8 prime conditions: Identity (the prime is the same as the target word); TL Initial (initial letters transposed); SUB Initial (initial letters substituted); TL Internal (internal letters transposed); SUB Internal (internal letters substituted); TL Final (Final letters transposed); SUB Final (final letters substituted); and URL (unrelated word prime). See Table 2 for details.

2.3 Procedures

A within-subject design of 2 (semantic predictability: high vs. low) × 2 (context position: preceding context vs. following context) × 8 (prime conditions) was adopted in Experiment 1. The experiment was run using E-Prime Professional software (2.0 version). The subjects were about 75 cm in front of a 19-in. monitor with the refresh rate set to 1000 Hz.

All materials were presented in the middle of the screen, with black words on a white background. The steps of the SPaM paradigm ran as follows. First, a red fixation cross (“+”) appeared on the left side of the screen, indicating the beginning of the experiment. When the subjects pressed the space bar, the “+” sign disappeared and the first word of the sentence would appear on the screen. When the participants finished reading it and pressed the space bar, the second word of the sentence would appear on the screen (If there was no press of the space bar in 3,000 ms, the first word would disappear automatically and the second word would appear on the screen). If the word was the target word, then there would be a prime word (any one of the 8 prime conditions) before it, and the prime would be displayed for 57 ms. After the subjects read the sentence in this way, they would see a true-or-false question based on the sentence they had just read, and they had to make a judgment whether the statement was correct or false. The time limit for the judgment was 3,000 ms. If the subject did not respond within 3,000 ms, the software would automatically determine that the answer was wrong. The purpose of these tag questions was to make sure the subjects read the sentences carefully.

Before the experiment, there were 12 practice trials. If the accuracy rate of practice was 90 % or above, the subjects would go directly into the formal experiment. If their accuracy rate was below 90 %, they would have to repeat the practice trials until their accuracy reached the desired rate.

2.4 Results and discussion

Reaction times were measured from the onset of the target word. The accuracy rate of answers to the tag question was 95.3 %, showing that the subjects carefully read the sentence and the subsequent tag questions. Reaction times with incorrect response were excluded from data analysis. In addition, reaction times below 150 ms, above 1500 ms or 3 times above the standard deviations were excluded, accounting for 1.7 % of the total data. Therefore, a total of 6.4 % of the data was excluded. The mean reaction times under the four contextual conditions are displayed in Table 3.

As shown in Table 3, the mean response times of L2 subjects under the four contextual conditions were 625, 628, 542 and 629 ms, respectively, while the mean response times of L1 subjects in Luke and Christianson’s (2012) experiment were 239 and 321 ms, respectively. Obviously, the processing speed of L2 learners was far slower than that of their L1 counterparts. The reason for this difference lies in the language proficiency level of the subjects: The subjects in Experiment 1 of the present study were third-grade English majors, who were mostly L2 learners with an intermediate proficiency level in English, while the subjects of Luke and Christianson (2012) were L1 college students with an advanced proficiency level. This shows that the subjects’ language proficiency level had an impact on transposition effect.

It can also be seen from Table 3 that the reaction times under the high-predictability preceding context was significantly shorter than those under the other three contextual conditions.

In order to learn about the influence of context type and priming type on reaction times, SPSS (Version 26.0) was applied to analyze these data. The results showed that the main effect of semantic predictability was significant, F(1, 238) = 24.2381, p < 0.001. The main effect of contextual position was not significant, F(1, 238) = 0.7635, p > 0.05. The main effect of priming type was significant, F(7, 554) = 19.1987; p < 0.001. The interaction effect between the three factors was significant, F(14, 1108) = 12.6371, p < 0.01. The results of post hoc analysis showed that there was no significant difference in response times under the three contextual conditions of low-predictability preceding context, low-predictability following context, and high-predictability following context, ps > 0.05. However, there were significant differences between the above three contexts and the high-predictability preceding context, ps < 0.001.

To explore the influence of different transposition positions on transposition effects, we analyzed relevant data by paired-sample T-test. Results are shown in Table 4.

It can be seen from Table 4 that under the three conditions of low-predictability preceding, low-predictability following, and high-predictability following contexts, there was a significant priming effect in internal and final transpositions, but no significant priming effect was observed in final transposition. In high-predictability preceding context, there was also a significant priming effect in internal and final transpositions. Although there was no significant priming effect in initial transposition, the p value was 0.06, close to the significant level.

In order to know the difference of priming magnitude in three different positions, we used the reaction times of unrelated priming as the baseline, and the difference between this baseline and the reaction time of a certain transposed-letter prime was the priming magnitude. The results are shown in Figure 1.

Figure 1:

Magnitude differences of three transposition types.

As shown in Figure 1, under the four contextual conditions, internal transposition produced the greatest priming magnitude, followed by final and initial transpositions. This means initial transposition is the most disruptive in word processing and recognition in English, followed by final and internal transpositions. In other words, initial letters play the most important role in word processing and recognition.

Luke and Christianson’s (2012) experiment observed a significant priming effect in internal transposition, but not in final transposition under the low-predictability context. Under the high-predictability context, no transposition effect was observed in internal or final transpositions. Their experiment did not involve two priming conditions: initial transposition and initial substitution. Experiment 1 of the present study added those two priming conditions. The results of Experiment 1 show that there is no priming effect in initial transposition, but there is a priming effect in internal and final transpositions. Why are there differences between the two studies? Since the two studies used similar materials and research paradigms, it is reasonable to contribute the differences in experiment results to the different subjects. Castles et al. (2007) explored the influence of individual differences on transposed-letter effects in English, and their subjects included adults, third-graders and fifth-graders. The results show that the three kinds of subjects display different patterns in transposition effect: the adults do not show any form of transposition effect, fifth-graders show significant transposition effect, while third-graders show priming effect in both transposition and substitution effects. Castles et al. (2007) maintain that the adults are familiar with vocabulary, so they are highly accurate in word recognition, while the word recognition system of 3rd- and 5th-graders is highly tolerant to the mismatch between letter position and identity and base word. Therefore, the differences in language proficiency of the subjects may be one of the reasons for the differences between the results of this study and those of Luke and Christianson’s (2012). In addition, the research results of Yang et al.’s (2021) study show that the L1 word processing habits of L2 learners have an impact on L2 word processing. Their study investigated the effects of different language backgrounds on the backward transposition effect of English words (such as naelc-CLEAR). They found that Chinese–English bilinguals showed a clear backward priming effect in the lexical decision task, but no such effect was observed in English monolinguals, Spanish–English or Arabic–English bilinguals.

3.1 Subjects

Forty third-year English majors from a university in Jiangsu participated in the experiment. They were native speakers of Chinese, with an average age of 21.5 (SD = 1.65). All subjects were right-handed, with normal or corrected-to-normal vision. None of them participated in Experiment 1.

3.2 Materials

Xu and Sui’s (2018) study employed two-character Chinese words as materials. However, their experiments failed to take into account the positions of character transposition (i.e., initial, internal, and final transpositions). To make up for this limitation, Experiment 2 of the present study chose four-character Chinese words as materials. One hundred high-frequency four-character Chinese words were selected from The Dictionary of Word Frequency of Common Words in Modern Chinese (in Phonological Order) (Liu et al. 1990). Based on these 100 words, we selected some sentences with target words close to the end of sentences (an average of 4 sentences per word) from the BCC Corpus of Beijing Language and Culture University (Xun et al. 2016) and the CCL Corpus of Peking University (Zhan et al. 2019). Next, the target words were removed from these sentences. Thirty-six non-English majors (these students did not participate in any part of Experiment 1 or Experiment 2) were asked to fill in the blanks with four-character words. Based on the results of blank filling, the sentences with a frequency of above 70 % are high-predictability context and those below 25 % are low-predictability context. Altogether, a total of 120 sentences with 60 high-predictability context and 60 low-predictability context were selected. Then, we rewrote the sentences and adjusted the target word to the front of the sentences, trying not to change the meaning of the original sentences. Finally, we slightly adapted the sentences of the four context types to ensure that they are roughly the same length (± one or two words). After that, we designed a tag question based on the content of each sentence to ensure the subjects would read the sentences carefully. Only correct response times would be included in the final data analysis. Examples of sentence materials and supporting questions in four contexts are shown in Table 5.

Each target has 8 prime conditions: Identity (the prime is the same as the target); TL Initial (initial characters transposed); SUB Initial (initial characters substituted); TL Internal (internal characters transposed); SUB Internal (internal characters substituted); TL Final (Final characters transposed); SUB Final (final characters substituted); and URL (unrelated prime). See Table 6 for details.

3.3 Procedures

The procedure of Experiment 2 is exactly the same as that of Experiment 1.

3.4 Results and discussion

Reaction times were measured from the onset of the target word. The accuracy rate of the answers to the tag question was 96.5 %, showing that the subjects had carefully read the sentence and the subsequent tag questions. Reaction times with incorrect response were excluded from data analysis. In addition, reaction times below 150 ms and above 1500 ms or 3 times above the standard deviations were excluded, accounting for 1.2 % of the total data. Therefore, a total of 4.7 % of the data was excluded. The mean reaction times under the four contextual conditions are displayed in Table 7.

As shown in Table 7, under the high-predictability preceding context, the mean reaction time of the target (751 ms) was significantly shorter than those under the other three contextual conditions (897, 899, and 894 ms, respectively). This shows that compared with the other three contextual types, high-predictability preceding context could better facilitate the processing and recognition of Chinese words, which is consistent with the results of Experiment 1.

In order to explore the influence of contextual and prime types on reaction time, we used SPSS software to analyze the data. The results showed that the main effect of semantic predictability was significant, F(1, 238) = 21.362, p < 0.001. The main effect of contextual position was not significant, F(1, 238) = 0.981, p > 0.05. The main effect of prime type was significant, F(7, 552) = 20.157; p < 0.001. The interaction effect between the three factors was significant, F(14, 1104) = 13.578, p < 0.01. The results of post hoc analysis failed to reveal any significant difference among low-predictability preceding context, low-predictability following context, and high-predictability following context, ps > 0.05. However, there were significant differences between the above three contextual types and high-predictability preceding context, ps < 0.001.

In order to understand the effect of different positions in four contexts, we analyzed the relevant data by paired-sample T-test. Results are shown in Table 8.

As shown in Table 8, under the low-predictability preceding, low-predictability following, and high-predictability following contexts, there was a significant priming effect in internal and final transpositions, but no significant transposition effect was observed in initial transposition. Under high-predictability preceding context, there was no significant difference between the reaction times of internal, final, and initial transpositions and their corresponding control conditions. This means no significant transposition effect was observed.

In order to explore the reasons for the disappearance of transposition effect in the high-predictability preceding context, we further analyzed the data. When paired-sample T-test analysis was conducted between these reaction times and those of the identity priming conditions, no significant difference was observed between these six priming conditions and the identity prime condition (ps > 0.05). This indicates that these six priming conditions can facilitate the processing and recognition of target words as much as the identity priming condition.

In order to reveal the influence of different transposition positions on transposition effect in Chinese, we further analyzed the priming magnitude of the three transpositions under four different contextual types. The results are shown in Figure 2.

Figure 2:

Magnitude differences of three transposition types.

As displayed in Figure 2, under the four contextual conditions, the priming magnitude of final transposition was the largest, followed by internal and initial transpositions. This shows that in the process of Chinese word processing and recognition, the transposition of initial characters was the most disruptive, followed by the internal and final transpositions.

On the one hand, the results of the present study are different from those of Xu and Sui (2018), which investigated the influence of semantic predictability on the transposition effect of two-character Chinese words by using eye movement paradigm. Each sentence in their experiment included one of the four conditions: the original word (OR), the transposed non-word (TN), the initial substituted non-word (FS), and the final substituted non-word (ES). Results show that the TN condition is not significant with FS and ES condition under the high-predictability context, but it is the opposite under the low-predictability context. Therefore, they concluded that high-predictability contexts may facilitate the encoding of the Chinese two-character words. However, in the present study, under the low-predictability preceding, low-predictability following, and high-predictability following contexts, there was a significant priming effect in internal and final transpositions, but no significant transposition effect was observed in initial transposition. Under high-predictability preceding context, there was no significant difference between the reaction times of internal, final, and initial transpositions and their corresponding control conditions. This means no significant transposition effect was observed. The difference in the results of the two studies may be attributed to the different experiment paradigms (Blythe et al. 2014).

On the other hand, the results of both studies have shown that contextual type (high- and low-predictability contexts) may have an impact on the encoding and processing of words in the Chinese language. This is consistent with results of previous studies on the processing of Chinese words. Rayner et al. (2005) adopted the eye movement paradigm to study the phenomenon of word skipping in the Chinese language by using two-character Chinese words with different predictability levels. The results showed that words with low predictability had the longest fixation time and were less likely to be skipped, while high-predictability words had the shortest fixation time and higher probability of skipping. Li (2000) used the same paradigm to study sentence reading in the Chinese language, and found that under the condition of preceding context, the lexical decision time was shorter than that in the following context condition, and the difference was significant. All the studies above have shown the impact of context predictability on Chinese word recognition and processing.