行政院國家科學委員會專題研究計畫 成果報告
詞內位置機率對中文句子閱讀的斷詞與詞彙辨識歷程的影
響
研究成果報告(精簡版)
計 畫 類 別 ： 個別型 計 畫 編 號 ： NSC 100-2410-H-003-001- 執 行 期 間 ： 100 年 01 月 01 日至 100 年 12 月 31 日 執 行 單 位 ： 國立臺灣師範大學科學教育研究所 計 畫 主 持 人 ： 顏妙璇 計畫參與人員： 碩士班研究生-兼任助理人員：王珮靜 碩士班研究生-兼任助理人員：許立懿 報 告 附 件 ： 出席國際會議研究心得報告及發表論文 公 開 資 訊 ： 本計畫可公開查詢中 華 民 國 101 年 03 月 31 日
中 文 摘 要 ： 中文書寫系統中詞與詞之間並無空格等視覺線索隔開，這使 得中文讀者如何在一連串的字當中進行詞彙辨識和斷詞成為 一個重要的議題。在本研究計畫中探討每個字放在四個可能 的詞內位置（單字詞、多字詞詞首/中/尾）的機率，可否做 為中文讀者在閱讀句子時進行詞彙辨識和斷詞的統計線索。 在兩個實驗中，請大專學生閱讀理解句子並記錄眼動。每個 句子中有一雙字目標詞 C12，其後的詞 C34 亦為雙字詞。在 材料中操弄目標詞的詞尾 C2 和後一詞的詞首 C3 做為詞尾/首 的機率高低。在實驗一中採用邊界典範（Rayner，1975）發 現 C2 的詞內位置機率可在週邊預視時取得，當目標詞的詞尾 字 C2 較常放置於詞尾時，預視的幫助較大。然而 C2 的詞內 位置機率對目標詞凝視時間的影響較不顯著。在實驗二中則 是探討因位於邊界的 C2 和 C3 的詞內位置機率高低造成斷詞 的困難度，如何影響目標詞 C12 的辨識。結果發現交互作 用，當位於邊界的其中一個字係放在不常放的位置時(如 C2 較常放在詞首，但卻是目標詞的詞尾)，才能觀察到另一字的 詞內位置機率幫助的效果。兩個實驗結果顯示，在閱讀中文 時，詞內位置機率對詞彙處理是有影響的，但此幫助的效果 僅在上述的情況下才顯現。顯示了中文閱讀中詞彙辨識歷程 的複雜性，需要更多研究來釐清。 中文關鍵詞： 中文閱讀；詞彙辨識；斷詞線索；眼球運動
英 文 摘 要 ： Chinese sentences are written character by character without visual cues, such as inter-word spaces, for word boundaries. This raises the issue of how Chinese readers recognize words, most of which contain one or two characters, embedded in a series of characters. The goal of this project was to investigate whether Chinese readers are sensitive to statistical cues such as within-word character position probability (CPP) when recognizing and segmenting words during sentence reading. The probabilities of each character being used as a single-character word, the beginning, middle, or end character of multi-character words were calculated from the Academia Sinica balanced corpus. In two experiments, college students were instructed to read sentences for comprehension at their normal pace while their eye movements were recorded. A pair of two-character words (C12 and C34) was embedded in each sentence in which the CPPs of the ending character C2 and the beginning C3 were
manipulated. In Experiment 1, it was found that CPP could be extracted from the parafovea by using the boundary paradigm (Rayner, 1975). However, the CPP effect on target gaze durations was moderate. In Experiment 2, the effects of the CPPs of characters (C2 and C3) surrounding the word boundary between C12 and C34 on the foveal processing of C12 was examined. A complementary interaction was observed. The
facilitation effect of the CPP of one character was found only when the other character was located at its less probable position (e.g., the ending
character of the present word was frequently used as word ending). Taken together, character position probability had an influence on word processing during Chinese reading. The results revealed the complicated nature of word recognition during Chinese reading that requires further investigation.
英文關鍵詞： Chinese reading, word recognition, word segmentation cues, eye movements
行政院國家科學委員會補助專題研究計畫
成果報告
□
期中進度報告
詞內位置機率對中文句子閱讀的斷詞與詞彙辨識歷程的影響
計畫類別：
個別型計畫
□
整合型計畫
計畫編號：NSC 100 － 2410 － H － 003 － 001
－
執行期間：100 年 01 月 01 日至 100 年 12 月 31 日
執行機構及系所：國立臺灣師範大學 科學教育研究所
計畫主持人：顏妙璇
共同主持人：
計畫參與人員：許立懿、王珮靜
成果報告類型(依經費核定清單規定繳交)：
精簡報告
□
完整報告
本計畫除繳交成果報告外，另須繳交以下出國心得報告：
□
赴國外出差或研習心得報告
□
赴大陸地區出差或研習心得報告
出席國際學術會議心得報告
□
國際合作研究計畫國外研究報告
處理方式：
除列管計畫及下列情形者外，得立即公開查詢
□
涉及專利或其他智慧財產權，
□
一年
□
二年後可公開查詢
中 華 民 國 101 年 03 月 30 日
I
Abstract
Chinese sentences are written character by character without visual cues, such as inter-word spaces, for word boundaries. This raises the issue of how Chinese readers recognize words, most of which contain one or two characters, embedded in a series of characters. The goal of this project was to investigate whether Chinese readers are sensitive to statistical cues such as within-word character position probability (CPP) when recognizing and segmenting words during sentence reading. The probabilities of each character being used as a single-character word, the beginning, middle, or end character of multi-character words were calculated from the Academia Sinica balanced corpus. In two experiments, college students were instructed to read sentences for comprehension at their normal pace while their eye movements were recorded. A pair of two-character words (C12 and C34) was embedded in each sentence in which the CPPs of the ending character C2 and the beginning C3 were manipulated. In Experiment 1, it was found that CPP could be extracted from the parafovea by using the boundary paradigm (Rayner, 1975). However, the CPP effect on target gaze durations was moderate. In Experiment 2, the effects of the CPPs of characters (C2 and C3) surrounding the word boundary between C12 and C34 on the foveal processing of C12 was examined. A complementary interaction was observed. The facilitation effect of the CPP of one character was found only when the other character was located at its less probable position (e.g., the ending character of the present word was frequently used as word ending). Taken together, character position probability had an influence on word processing during Chinese reading. The results revealed the complicated nature of word recognition during Chinese reading that requires further investigation.
II
摘要
中文書寫系統中詞與詞之間並無空格等視覺線索隔開，這使得中文讀者如何在一連串的字當中進 行詞彙辨識和斷詞成為一個重要的議題。在本研究計畫中探討每個字放在四個可能的詞內位置（單字 詞、多字詞詞首/中/尾）的機率，可否做為中文讀者在閱讀句子時進行詞彙辨識和斷詞的統計線索。在 兩個實驗中，請大專學生閱讀理解句子並記錄眼動。每個句子中有一雙字目標詞 C12，其後的詞 C34 亦為雙字詞。在材料中操弄目標詞的詞尾C2 和後一詞的詞首 C3 做為詞尾/首的機率高低。在實驗一中 採用邊界典範（Rayner，1975）發現 C2 的詞內位置機率可在週邊預視時取得，當目標詞的詞尾字 C2 較常放置於詞尾時，預視的幫助較大。然而C2 的詞內位置機率對目標詞凝視時間的影響較不顯著。在 實驗二中則是探討因位於邊界的 C2 和 C3 的詞內位置機率高低造成斷詞的困難度，如何影響目標詞 C12 的辨識。結果發現交互作用，當位於邊界的其中一個字係放在不常放的位置時(如 C2 較常放在詞 首，但卻是目標詞的詞尾)，才能觀察到另一字的詞內位置機率幫助的效果。兩個實驗結果顯示，在閱 讀中文時，詞內位置機率對詞彙處理是有影響的，但此幫助的效果僅在上述的情況下才顯現。顯示了 中文閱讀中詞彙辨識歷程的複雜性，需要更多研究來釐清。 關鍵詞： 中文閱讀；詞彙辨識；斷詞線索；眼球運動1
A. Background and literature review
Chinese is a unique language. Unlike most alphabetic languages, Chinese is written character by character without visual cues for word boundaries. It is intriguing how Chinese readers segment and recognize words from a series of characters.
Eye tracking during reading has been a promising measurement for reading comprehension. Because of limit in visual acuity, readers actively move their eyes to sample information for comprehension. Thus, eye movement is a natural behavior during reading. In addition, eye tracking provides moment-to-moment, on-line measurement. For example, viewing durations on words reflect difficulty (such as frequency, predictability) in word processing.
Besides passive recording, researchers can manipulate the characteristics of the presented stimuli by using the eye-movement contingent display change technique. This technique is useful in illuminating the time course of processing at different levels (e.g., orthographic, semantic, character, or word levels). For example, Rayner (1975) developed the classical boundary paradigm. An invisible boundary is set prior to the target stimulus (a word or a character) embedded in a sentence. At the beginning of each trial, the target stimulus is replaced by a preview stimulus. After the participants move their eyes across the boundary, the preview stimulus immediately changes into the target stimulus. A significant reduction in target viewing durations (preview benefit) in comparison to an uninformative preview condition is assumed to indicate successful parafoveal processing. The characteristics of the preview stimuli, i.e., the features shared by the preview and target stimuli, are usually manipulated. Significant preview benefit indicates that the shared features can be processed parafoveally.
Parafoveal processing is an important and informative operation during sentence reading. When only the fixated word was available and all other words were masked during reading, reading rate significantly reduced to 2/3 of that when all words are presented normally. Parafoveal processing not only facilitates subsequent foveal processing but also provides information for the Where decision. In alphabetic scripts, inter-word spaces clearly delineate word boundaries and hence the location and length of the parafoveal word can be obtained. This information is then used to guide the next saccade. There is a preferred landing position around the left of word center during reading alphabetic scripts, suggesting that words are saccade targets. In addition, when the word length was manipulated (Juhasz, White, Liversedge, & Rayner, 2008), for example, a space was inserted so that a long word was separated into two shorter units, the landing position distribution shifted to the center of the short unit. That is, word length information obtained from the parafovea is used to guide the next saccade. There are, however, no such boundary cues in written Chinese, how word is processed thus becomes an interesting question.
Basic eye movement behavior has been observed during Chinese reading. For example, frequency and predictability effects are found at both character and word levels (Rayner, Li, Juhasz, & Yan, 2005;
Yan, Tian, Bai, & Rayner, 2006; Yang & McConkie, 1999). In addition, parafoveal processing at the character level (i.e., orthographical, phonological, and semantic properties of the target characters; Liu, Inhoff, Ye, &Wu, 2002; Tsai, Lee, Tzeng, Hung, &Yen, 2004; Yan, Richter, Shu, & Kliegl, 2009) has been documented in the literature. Furthermore, masking the parafoveal word (the target) with a pseudoword consisting of real characters lengthened the subsequent target viewing durations (Yang, Wang, Xu, & Rayner, 2009; Yen, Tsai, Tzeng, & Hung, 2008). These results suggest that parafoveal information beyond the character level can be obtained from the parafovea.
2
The next questions are (1) what kinds of cues are helpful for Chinese word segmentation and (2) when these cues are used (in the fovea or parafovea). Yen, Radach, Tzeng, and Tsai (in press) manipulated
character position probability (CPP) of the ending character (C2) of a two-character target word (C12). In half of the trials, C2 was more frequently used as word beginning in all two-character words containing it. Thus, this character is less probable to appear at the ‘ending’ position but it is located at this less probable position in the target word. In the other half, C2 is more frequently used as word ending and this character is located at a highly probable position in the target. We found that target words with less probable C2 were fixated longer and more likely to be refixated before proceeding to the right, suggesting that there were difficulty in processing these words because of low CPP. Thus, CPP may be a useful cue for word segmentation. However, this experiment had another manipulation which is out of scope for the present project. Because of that, whether CPP could be processed parafoveally remained unknown.
B. Research aim and predictions
The goal of the present project was to further examine the role of CPP during reading. Experiment 1 aimed to investigate whether CPP can be obtained from parafoveal characters and influence parafoveal word processing. A target word (C12) was embedded in each sentence. The CPP of the ending character (C2) of the target word was manipulated. By using the boundary paradigm, C2 was visible or masked during preview. When the eyes moved across the boundary set before the target (to the left of C1), C2 was revealed and remained visible until participants finished reading the sentence. A preview benefit on the target was expected. Furthermore, if CPP of characters can be obtained from the parafovea, the preview benefit should be larger for words with highly probable word ending than those with less probable word ending. The working hypothesis is words with constituent characters located at their high probability positions are easy to be processed. Thus, if a highly probable C2 is visible, processing of the target word is facilitated compared to when it is masked. However, if a less probable C2 is visible, processing of the target word is still difficult; thereby, the preview benefit in the low probability condition is relatively small. If, however, CPP is mainly processed during foveal processing, then, there is no difference in preview benefit between conditions.
Experiment 2 aimed to investigate the effects of CPP of the characters surrounding the word boundary. The CPP of C2 in the target was again manipulated. In addition, the CPP of the first character (C3) of the post-target word was also manipulated. For example, in the High-High condition, C2 in the target was frequently used as word ending and C3 in the post-target was frequently used as word beginning. It is highly possible that the boundary should be put between C2 and C3. In the opposite Low-Low condition, C2 was frequently used as word beginning and C3 was frequently used as word ending. The boundary between C2 and C3 was very ambiguous in this condition. The ambiguity in the other two conditions, High-Low and Low-High, should lie between the High-High and Low-Low conditions. From the strictly serial word-by-word processing viewpoint (e.g., the EZ Reader model, Pollatsek, Reichle, & Rayner, 2006), properties of the post-target should not influence target word processing. Thus, only the manipulation of C2 could have an effect on target viewing durations. However, because only a few characters can unambiguously signal word boundaries (e.g., 們, the plural marker, is always located at word ending), I reasoned that processing of the next character to the right of the target word is necessary (Yen, Radach, Tzeng, Hung, & Tsai, 2009). Nevertheless, the effect of manipulating C2
3
may still be larger than the effect of manipulating C3. Thus, when C2 is a highly probable word ending, then the effect of C3 may be small. On the other hand, when C2 is a less probable word ending, then C3 may be helpful. In other words, an interaction between C2 and C3 probabilities can be expected.
C. Methods Material
The character position probabilities of each character being used at each possible within-word positions (namely, single-character word, beginning / middle / ending of multi-character words) were calculated from the Academia Sinica Balanced Corpus (1998). Two-character words (C12) were chosen as the target words. The character position probability of the ending character (C2) of the target was manipulated. In the high-probability condition, the probability of C2 being used as word ending was higher than 50%. In the low-probability condition, the probability of C2 being used as word ending was lower than 50%. For each target word, a two-character word (C34) was chosen as the post-target word. The CPP of the beginning character (C3) of the post-target was also manipulated. In the high-probability condition, the probability of C3 being used as word beginning was higher than 50%. In the low-probability condition, the probability of C3 being used as word beginning was lower than 50%. At the end, totally 144 word pairs were chosen, with 36 word pairs in each of 4 combinations (High-High, High-Low, Low-High, and Low-Low). The word pairs were embedded in the middle part of sentences with 26-28 characters. The target words are mid-low frequency nouns (15.3 occurrences per 1 million words). In a pretest, the contextual predictabilities of the target word C12 and the post-target word C34 were assessed by 20 participants who did not participate in the main experiments. In the first experiment, extremely low frequency characters (lower than 1 per 1 million) were chosen as the preview masks. They matched the target character C2 by number of strokes and were similar in visual structure.
The 144 experimental sentences and 6 filler sentences were assigned to 3 blocks. There were 12 sentences in each condition in each block. In the first experiment, 2 lists were created so that half of the stimuli in each combination were assigned to the visible condition and the other half were assigned to the masked condition.
Participants
Thirty-four and thirty college and graduate students were paid to participate in Experiment 1 and 2, respectively. All of them are native speakers of Chinese, with normal or corrected-to-normal vision. Apparatus
The eye movements were recorded by an Eyelink 1000 eyetracking system, manufactured by SR Research, with a sampling rate of 1000 Hz. The sentences were presented in black on a light gray background on a ViewSonic PT795 monitor at a resolution of 1024 768 with the extent of each character set to 32 32 pixels. The viewing distance was 60 cm, at which each character subtended about 0.95.
Procedure
Participants were instructed to read 150 sentences for comprehension at their normal pace. After setting up the eye tracking system, a five-point calibration (top, down, left, right and center) was conducted, followed by a validation routine that verified accuracy. Then, 8 practice trials were presented to ensure that participants understand the task. During the experiment, each trial started with a fixation
4
point presented at the location of the first character of the sentence. Participants read each sentence at their own pace and pressed a button to indicate that they finish reading and understand the sentence. During 30% of the trials, a comprehension question was presented on the screen following the disappearance of the sentence. Participants were asked to decide whether the sentence in the comprehension test paraphrased the one they just read by pressing buttons. Participants were allowed to take a break after 50 trials. Calibration was conducted every 25 trials, after breaks, and when there was a drift from the fixation point presented prior to the experimental sentence. The experiment lasted about 40-60 minutes.
Additional procedure for Experiment 1. An invisible boundary was set at the first pixel of the space
preceding the target word. At first, the target character C2 was masked by an extremely low frequency character in the masked condition (Figure 1); while the original C2 was presented in the visible condition. After the participants moved their eyes across the boundary, the display change was implemented so that the sentence with the target character C2 was presented. The participants were instructed to ignore any disturbance that might occur while they were reading the experimental sentences. In the interview after the experiment, participated reported seeing some changes only 1% of time on average.
身 為 一 個 特 技 表 演 者 ， 必 須 擁 有 技 捯 高 超 又 膽 大 心 細 等 特 質 。 *
身 為 一 個 特 技 表 演 者 ， 必 須 擁 有 技 術 高 超 又 膽 大 心 細 等 特 質 。 *
Figure 1. Illustration of the boundary paradigm used in Experiment 1. The fixation point is indicated by the symbol ”*”, the invisible boundary is located prior to the target word, and the critical area is underlined. The ending character (C2) of the target ”技術” is replaced by an extremely low frequency character ”捯” during preview in the first experiment in the present proposal. In the example, ”高超” is the post-target word.
D. Results and discussion
Gaze duration (GD) is defined as the sum of durations of all fixations in the region of interest (ROI, e.g., the target word) during the first-pass reading. Two-way ANOVA was performed in both experiments.
Experiment 1
Gaze duration on the target C12. As is shown in Table 1, the main effect of CPP of C2 was not
significant [F(1, 33) = 2.086, MSe = 610.513, p > .15]. GDs on the target were 347.7 and 341.6 ms in the high- and low-probability conditions, respectively. However, there was a significant preview benefit [F(1, 33) = 87.480, MSe = 946.422, p < .001] and interaction [F(1, 33) = 5.461, MSe = 664.915, p < .05]. Post-hoc comparison revealed that the preview benefit was significantly larger in the high-probability condition than that in the low-probability condition, suggesting that CPP of C2 may be processed during preview.
5
Gaze duration in the critical area C1234. When the critical area was treated as the ROI, a significant
main effect of CPP of C2 was observed [F(1, 33) = 9.382, MSe = 1414.452, p < .01]. GD in the critical area was significantly shorter when the CPP of C2 was high (597.1 ms) than when it was low (616.9 ms). There was a significant preview benefit effect [F(1, 33) = 42.801, MSe = 1785.101, p < .001]. Although the preview benefit seemed to be larger in the high-probability condition than that in the low-probability condition, the interaction was not significant [F(1, 33) = 1.306, MSe = 2646.733, p > .26].
Taken together, the CPP of C2 can be obtained from the parafovea so that the preview benefit in the high-probability condition was larger than that in the low-probability condition. However, the facilitation effect of the CPP of C2 revealed itself relatively late. The effect of the CPP of C2 on the target word was not significant, but this effect was evident when the whole critical area C1234 was taken into consideration. When C2 was visible during preview, the effect on the target was small (317.8 vs. 322.1 ms) but this effect was significant in the critical area (568.4 vs. 598.2 ms). When C2 was masked during preview, the effect on the target was negative (377.5 vs. 361.1 ms) but again a positive effect was observed in the critical area (625.8 vs 635.5 ms). When C2 was masked during preview, C2 could be processed only during subsequent fixation. A low-probability C2 made C12 difficult to process, so we speculate that there was a tendency to jump to other word to gather more information. This phenomenon is similar to the inverted optimal viewing position effect (Vitu, McConkie, Kerr, & O’Regan, 2001). In the literature, an optimal viewing position (OVP) was found around the word center (see Brysbaert, & Nazir, 2005, for a review). When word center was fixated, the naming reaction time was shorter than when word exterior was fixated. However, an inverted OVP effect was observed during sentence reading (Vitu, et al., 2001). Nuthmann, Engbert, and Kliegl (2005) proposed that when word exterior, a non-optimal position, was fixated, there was a tendency to move to the other end to gather more information about the word. But when the optimal position (word center) was fixated, readers can obtain sufficient information so that they stay there longer than when the non-optimal position was fixated. In the present experiment, if C2 was not available during preview and was difficult to process (in the low-probability condition), more information can be gathered from other words so that GDs in the low-probability condition were shorter than those in the high-probability condition.
Table 1
The Means and Standard Errors (in parentheses) of Gaze Duration (ms) on the Target (C12) and in the critical area (C1234) for Each Condition in Experiment 1.
Target (C12) Critical Area (C1234)
Character Position Probability
Visible Masked Preview
benefit
Visible Masked Preview
benefit High 317.8 ( 7.79) 377.5 (10.41) -59.7 568.4 (18.36) 625.8 (17.73) -57.5 Low 322.1 ( 8.49) 361.1 ( 9.51) -39.0 598.2 (18.65) 635.5 (16.65) -37.3
Experiment 2
6
MSe = 635.176, p > .91] nor effect of C3 probability [F(1, 29) = 2.698, MSe = 543.808, p > .11] was
significant. The interaction was also not significant [F(1, 29) < 1, MSe = 624.359, p > .88].
However, when the GD was regressed on the continuous values of CPPs of C2 and C3, there was an interesting pattern (Figure 2). The CPP of C2 was beneficial when C3 was a low-probability character (black line) and was inhibitory when C3 was a high-probability character (blue line). The same pattern was observed for the CPP of C3. In other words, CPP was beneficial when there was a conflict in the other character’s usual position and its current position.
Gaze duration on all two-character words. In an additional analysis, all two-character words (C12) in
the sentences were taken into consideration. The CPP of the ending character (C2) being word ending and the CPP of the next character (C3) being word beginning were calculated. Gaze durations on these words were regressed on the CPPs of C2 and C3 by using the lmer program (lme4 package; Bates, Maechler, & Dai, 2008) in the R system (R Development Core Team, 2008). An effect was significant if the associated t value was greater than 2. When both CPPs were centered at 50%, a marginally significant effect of C2 CPP [b = -12.15, SE = 6.51, t = -1.87] and a significant effect of C3 CPP [b = 17.71, SE = 4.84, t = 3.66] were observed. In addition, a significant interaction was observed [b = 52.2, SE = 22.12, t = 2.36]. Simple slopes were then examined by centering CPP to 0% and 100%. When C2 was highly probable, the effect of C3 CPP was inhibitory [b = 43.81, SE = 12.33, t = 3.55] while when C2 was less probable, a trend of facilitation was observed [b = -8.39, SE = 11.80, t = -0.71]. Similarly, when C3 was highly probable, the effect of C2 CPP was slightly inhibitory [b = 13.96, SE = 14.87, t = 0.94] while when C3 was less probable, a significant facilitation effect was observed [b = -38.25, SE = 10.41, t = -3.68]. Thus, a complementary pattern was observed. A facilitation effect of CPP was observed when the other character was put in a less probable position.
Table 2
The Means and Standard Errors (in parentheses) of Gaze Duration (ms) on the Target (C12) for Each Condition in Experiment 2.
Probability of C3
Probability of C2 High Low Mean
High 345.2 (10.78) 338.8 (11.46) 342.0 (10.60)
Low 345.3 (10.97) 337.7 (10.90) 341.5 (10.56)
Mean 345.3 (10.42) 338.3 (10.67)
7
E. Conclusion
To conclude, although there is no visual cue for word boundary in written Chinese, readers can use statistical cues, such as character position probability for word recognition. Although CPP is not a robust cue as salient inter-word spaces, it can indeed facilitate word recognition. For example, a facilitation effect was observed only when the other character around the possible word boundary was put in a less probable position, showing a complementary effect. In addition, CPP can be extracted parafoveally and expedite subsequent foveal processing. Among many possible cues for word boundaries, such as neighborhood size (Tsai, Lee, Lin, Tzeng, & Hung, 2006), character position probability seems to be a potential one that demands further investigation.
References
Academia Sinica Taiwan (1998). Academia Sinica balanced corpus (Version 3) [CD-ROM]. Taipei, Taiwan: Academia Sinica, Chinese Knowledge and Information Processing Group.
Bates, D., Maechler, M., & Dai, B. (2008). lme4: Linear mixed-effects models using S4 classes R package (Version 0.999375-27) [Computer software]. Vienna, Austria: R Foundation for Statistical Computing. Brysbaert, M., & Nazir, T. (2005). Visual constraints in written word recognition: Evidence from the optimal
viewing-position effect. Journal of Research in Reading 28(3), 216-228.
Inhoff, A. W., & Radach, R. (1998). Definition and computation of oculomotor measures in the study of cognitive processes. In G. Underwood (Ed.), Eye Guidance in Reading and Scene Perception (pp. 29-53). Amsterdam: Elsevier.
Juhasz, B. J., White, S. J., Liversedge, S. P., & Rayner, K. (2008). Eye movements and the use of parafoveal word length information in reading. Journal of Experimental Psychology: Hunan Perception and
Performance, 34, 1560-1579.
Liu, W., Inhoff, A. W., Ye, Y., & Wu, C. (2002). Use of parafoveally visible characters during the reading of Chinese sentences. Journal of Experimental Psychology: Human Perception and Performance, 28, 1213-1227.
Nuthmann, A., Engbert, R., & Kliegl, R. (2005). Mislocated fixations during reading and the inverted optimal viewing position effect. Vision Research, 45, 2201-2217.
Rayner, K. (1975). The perceptual span and peripheral cues in reading. Cognitive Psychology, 7, 65-81. Rayner, K., Li, X., Juhasz, B. J., & Yan, G. (2005). The effect of word predictability on the eye movements of
Chinese readers. Psychonomic Bulletin & Review, 12, 1089-1093.
R Development Core Team (2008). R: A language and environment for statistical computing (Version 2.7.2) [Computer software]. Vienna, Austria: R Foundation for Statistical Computing.
Pollatsek, A., Reichle, E. D., & Rayner, K. (2006). Tests of the E-Z Reader model: Exploring the interface between cognition and eye-movement control. Cognitive Psychology, 52, 1-56.
Tsai, J.-L., Lee, C.-Y., Lin, Y.-C., Tzeng, O. J.-L., & Hung, D. L. (2006). Neighborhood size effects of Chinese words in lexical decision and reading. Language and Linguistics, 7, 659-675.
Tsai, J.-L., Lee, C.-Y., Tzeng, O. J.-L., Hung, D. L., & Yen, N.-S. (2004). Use of phonological codes for Chinese characters: Evidence from processing of parafoveal preview when reading sentences. Brain and
8
Vitu, F., McConkie, G. W., Kerr, P., & O Regan, J. K. (2001). Fixation location effects on fixation durations during reading: Aan inverted optimal viewing position effect. Vision Research, 41, 3513–3533.
Yan, G., Tian, H., Bai, X., & Rayner, K. (2006). The effect of word and character frequency on the eye movements of Chinese readers. British Journal of Psychology, 97, 259-268.
Yan, M., Richter, E. M., Shu, H., & Kliegl, R. (2009). Readers of Chinese extract semantic information from parafoveal words. Psychonomic Bulletin & Review, 16, 561–566.
Yang, H.-M., & McConkie, G. W. (1999). Reading Chinese: Some basic eye-movement characteristics. In J. Wang, A. W. Inhoff, & H.-C. Chen (Eds.), Reading Chinese script: A cognitive analysis (pp. 207-222). Mahwah, NJ: Lawrence Erlbaum Associates.
Yang, J., Wang, S., Xu, Y., & Rayner, K. (2009). Do Chinese readers obtain preview benefit from word n + 2? Evidence from eye movements. Journal of Experimental Psychology: Human Perception and
Performance, 35, 1192-1204.
Yen, M.-H., Radach, R., Tzeng, O. J.-L., Hung, D. L., & Tsai, J.-L. (2009). Early parafoveal processing in reading Chinese sentences. Acta Psychologica, 131, 24-33.
Yen, M.-H., Radach, R., Tzeng, O. J.-L., & Tsai, J.-L. (in press). Usage of statistical cues for word boundary in reading Chinese sentences. Reading & Writing.
Yen, M.-H., Tsai, J.-L., Tzeng, O. J.-L., & Hung, D. L. (2008). Eye movements and parafoveal word processing in reading Chinese. Memory and Cognition, 36, 1033-1045.
1
國科會補助專題研究計畫項下出席國際學術會議心得報告
日期：100 年 9 月 15 日一、參加會議經過
於 8/20 下午搭機前往巴黎，再搭法國國鐵至馬賽會場，參加開幕式，隨後進行四
天的會議。此次會議論文是在 8/24 的中午張貼壁報發表，有德國波茨坦大學心理系博
士後研究員數名(做類似的議題)前來討論，並有其他資深學者給予建議。會議期間聽
取閱讀與眼動領域內各著名模式(model)最新進展的發表，並聆聽其他領域(場景辨識
scene perception，物體辨識和分類)的研究發表以拓展視野。在茶敘時段，也和認識
的外國學者討論彼此的研究。並參觀儀器廠商的新型眼動儀，了解目前的趨勢。
二、與會心得
歐洲眼動會議是每兩年舉辦一次的大型會議，係由各國的眼動研究者輪流主辦，
參加此會議可更新眼動研究的最新理論和研究發現，並可與知名研究者互動。這次會
議有趣的現象是，在閱讀與眼動領域的模式場次中，幾乎都是由年輕一代的研究生報
告，以往都是由資深研究者(即模式的創始者)發表，顯示新一代研究者的產生與傳承。
此次大會並邀請到閱讀與眼動領域的大師 George McConkie 教授與會，現今的眼動研
究都是 McConkie 教授與他的學生 Keith Rayner 教授在 70 年代的經典研究奠基下來
的，很可惜因為大會時間掌控失誤，沒能讓 McConkie 教授發言。
三、考察參觀活動(無是項活動者略)
無
計畫編號 NSC 100-2410-H-003-001
計畫名稱 詞內位置機率對中文句子閱讀的斷詞與詞彙辨識歷程的影響
出國人員
姓名
顏妙璇
服務機構及
職稱
國立臺灣師範大學科學教育研究所
助理教授
會議時間
100 年 8 月 21 日至
100 年 8 月 25 日
會議地點
法國馬賽的 University of
Provence, Campus St Charles
會議名稱
(中文)第十六屆歐洲眼動會議
(英文)The 16
thEuropean Conference on Eye Movements
發表論文
題目
(中文)中文句子閱讀時預視時間對預視效果之影響
(英文)
Preview effect modulated by preview duration in reading
Chinese sentences
2
四、建議
請貴會繼續提供研究者和研究生參與國際會議的經費補助，在發表自己的研究和
更新最新發展之外，也提供和國外學者面對面交流的機會。
但建議貴會提醒申請者，前往南歐國家宜多注意安全(於第六項詳述)。
五、攜回資料名稱及內容
會議論文摘要及詳細時程各一本
六、其他
法國治安不良，南法的大城馬賽尤其嚴重，幾乎專挑東方人下手。在開會過程中，
同行的學姊的電腦背包就在「會場」被偷走，法國警察局的幫忙有限(英文溝通不良，
不太願意幫忙)，只有做筆錄以利後續辦理保險理賠之用。隨後在大會結束之後，在前
往火車站搭車到巴黎前(六點半天還亮)，就在火車站前十公尺左右的十字路口，本人
不幸遇到搶劫，有黑人突然衝上來用刀割斷背在腹部前的斜背包搶走逃逸，巷內並有
摩托車同夥接應，本人追入巷內大叫搶劫，但巷內的黑人面無表情一付不關己事，在
警局從七點等到十一點多才有警察做筆錄，一進警局就告知櫃台是搶劫也沒用。在警
局也得知有香港人被搶，也有應該是被偷或被搶的新加坡護照送到警局。故當天被迫
留在馬賽，隔日才能前往巴黎補辦護照。
藉由此次出國的機會，也自費參加了 9/5-9 在法國里昂舉辦的歐洲科學教育研究
研討會，了解在科教領域的現今發展，有助敝人在任職單位的研究發展。
Laboratoire de Psychologie Cognitive
Université de Provence – Aix-‐Marseille 1 CNRS -‐ U.M.R. 6146
Université de Provence -‐ Centre Saint-‐Charles 3 place Victor Hugo – Case D – 13331 Marseille cedex 3 Tél : +33 (0) 413 550 960 – Fax : +33 (0) 413 550 998 Courriel : lpc@univ-‐provence.fr
Marseille, July 24
th2011
Letter of acceptation
To whom it may concerned
We are very pleased to announce you that your abstract is accepted as a poster presentation at the
European Conference on Eye Movement – ECEM 2011, which will take place in Marseille, France,
from 21 to 25 august 2011. TITLE:
Preview effect modulated by preview duration in reading Chinese sentences AUTHOR:
Miao-Hsuan Yen1 & Jie-Li Tsai2
AFFILIATION:
1Graduate Institute of Science Education, National Taiwan Normal University, Taipei, Taiwan ; 2Department of Psychology and Research Center for Mind, Brain, and Learning, National
Chengchi University, Taipei, Taiwan
Stéphanie DESOUS
Organising Committee
ECEM
Miao-Hsuan Yen1and Jie-Li Tsai2,3
1Graduate Institute of Science Education, National Taiwan Normal University, Taipei, Taiwan 2Department of Psychology, National Chengchi University, Taipei, Taiwan
3Research Center for Mind, Brain, and Learning, National Chengchi University, Taipei, Taiwan
Preview Effect Modulated by Preview Duration
in Reading Chinese Sentences
Introduction
During reading, information at various levels (e.g., features, characters, and words) can be obtained from a limited area (i.e., the perceptual span) around the fixation point. The when and where of information extraction are intriguing issues. It is particularly interesting to investigate when word-level information can be acquired during preview in reading Chinese, which is written character by character without visual cues for word boundaries.
Parafoveal preview at the character (Liu, Inhoff, Ye, & Wu, 2002; Tsai, Lee, Tzeng, Hung, & Yen, 2004; Yan, Richter, Shu, & Kliegl, 2009) and word levels (Yang, Wang, Xu, & Rayner, 2009; Yen, Tsai, Tzeng, & Hung, 2008) has been observed in Chinese. Concerning the time course of parafoveal processing, Li, Rayner and Cave (2009) suggested that characters within the span are processed in parallel, with processing efficiency decreasing according to eccentricity. Most characters can be combined with many other characters at different positions to form different words. Without visual word boundaries, parafoveal word processing may emerge later during preview. The present study aimed to investigate this issue by varying the preview duration and preview types.
Method
Participants. Twenty-seven college and graduate
students at National Chengchi University and nearby
community were paid to participate in the experiment. All of them are native speakers of Chinese with normal or corrected-to-normal vision.
Apparatus. The eye movements were recorded by the
EYELINK II eye-tracking system, manufactured by SR Research, with a sampling rate of 500 Hz. The sentences were displayed on a ViewSonic PT795 CRT monitor with a resolution of 800 600. The extent of each character was 24 24 pixels. The viewing distance was 60 cm, at which each character subtended about 1.05.
Design. The boundary paradigm was adopted in a 3 by 3
factorial design. Two two-character words were embedded in the sentences as the pretarget-target word pairs. Target delay onset time (TDOT) during pretarget viewing and preview type were manipulated. Upon fixating the pretarget, the target was replaced by one of the three types of previews (identical word, unrelated word, or pseudoword) for 60, 120, or 180 ms (TDOT). The target was then presented for the remainder of pretarget viewing and sentence reading.
Results
Gaze duration on the target. A significant effect of target
delay onset time (TDOT) was observed, p < .01. GD on targets with 180-ms TDOT was significantly longer than those with 60- and 120-ms TDOT (p < .01 and p < .05, respectively). Although the main effect of preview type was not significant (p > .12), it was modulated by TDOT (p < .01). When the TDOT was 180 ms, the 45-ms difference between the identical and the unrelated conditions was significant (p < .01) and the 34-ms difference between the identical and the pseudoword conditions was marginally significant (p < .09).
Linear mixed-effects analysis. Two sets of predictors and
their interactions were specified in the model. The first predictor included the contrast between identical and non-identical previews (ID-nID) and that between unrelated and pseudoword previews (UN-PS). The second predictor was TDOT. A significant effect of TDOT (t = 4.0) and a significant interaction between TDOT and ID-nID (t = -3.0) were observed. Gaze durations on the targets lengthened as TDOT increased. In addition, the difference between identical and non-identical previews enlarged as TDOT increased and was significant in the 120 and 180ms conditions (ts = 2.0 and -3.7, respectively). Neither the contrast between UN-PS nor the interaction with TDOT was significant (ts < 1.3).
Discussion
The time course of parafoveal processing at character and word levels is illustrated in the present study. A significant difference between identical and non-identical previews was observed after 120 ms; while the difference between unrelated and pseudoword previews was not observed in the present study. The results suggest that word-level information might not have been obtained from the span during the initial 180 ms. On the other hand, character-level information is gradually extracted around 120 ms. In addition, together with the result of the linear mixed-effects analysis, the present study replicated recently published findings that the size of preview benefit is modulated by preview duration. Future
studies are necessary to elucidate when word-level
information can be extracted during the end of previous fixation.
Acknowledgement
This study was supported by the grants from Taiwan’s National Science Council (NSC96-2413-H-004-018-MY3, NSC99-2420-H-004-002-, and NSC100-2410-H-003-001-).
Feedback and comments:
Miao-Hsuan Yen ([email protected])
Before pretarget viewing
高醫師在受訪時，不斷強調僇靪可以預防及解決文明病的產生。
Pretarget viewing (preview stimulus presented during the initial 60, 120, or 180 ms)
Identical word 高醫師在受訪時，不斷強調運動可以預防及解決文明病的產生。 Unrelated word 高醫師在受訪時，不斷強調道路可以預防及解決文明病的產生。 Pseudoword 高醫師在受訪時，不斷強調嫌都可以預防及解決文明病的產生。
Pretarget viewing (target presented after target delay onset time)
高醫師在受訪時，不斷強調運動可以預防及解決文明病的產生。
Leaving the pretarget (target presented)
高醫師在受訪時，不斷強調運動可以預防及解決文明病的產生。 60 120 180 ID 336 335 336 UN 330 345 381 PS 322 344 370 250 300 350 400 (ms) GD on the target p = .044 p = .009 p = .009 p = .082 (n=27)
國科會補助計畫衍生研發成果推廣資料表
日期:2012/01/10國科會補助計畫
計畫名稱: 詞內位置機率對中文句子閱讀的斷詞與詞彙辨識歷程的影響 計畫主持人: 顏妙璇 計畫編號: 100-2410-H-003-001- 學門領域: 實驗及認知心理學無研發成果推廣資料
100 年度專題研究計畫研究成果彙整表
計畫主持人：顏妙璇 計畫編號：100-2410-H-003-001- 計畫名稱：詞內位置機率對中文句子閱讀的斷詞與詞彙辨識歷程的影響 量化 成果項目 實際已達成 數（被接受 或已發表） 預期總達成 數(含實際已 達成數) 本計畫實 際貢獻百 分比 單位 備 註 （ 質 化 說 明：如 數 個 計 畫 共 同 成 果、成 果 列 為 該 期 刊 之 封 面 故 事 ... 等） 期刊論文 0 0 100% 研究報告/技術報告 0 0 100% 研討會論文 0 0 100% 篇 論文著作 專書 0 0 100% 申請中件數 0 0 100% 專利 已獲得件數 0 0 100% 件 件數 0 0 100% 件 技術移轉 權利金 0 0 100% 千元 碩士生 2 2 100% 帶 領 研 究 生 學 習 操 作 眼 動 儀 進 行 實驗 博士生 0 0 100% 博士後研究員 0 0 100% 國內 參與計畫人力 （本國籍） 專任助理 0 0 100% 人次 期刊論文 1 1 100% Yen, M.-H., Radach, R., Tzeng, O. J.-L., & Tsai, J.-L. (in press). Usage of statistical cues for word boundary in reading Chinese sentences. Reading & Writing. 研究報告/技術報告 0 0 100% 國外 論文著作 研討會論文 1 1 100% 篇 Yen, M.-H., & Tsai, J.-L. (2011, Aug.). Preview effect modulated by preview duration in reading Chinese16th European Conference on Eye Movements, Marseille, France. 專書 0 0 100% 章/本 申請中件數 0 0 100% 專利 已獲得件數 0 0 100% 件 件數 0 0 100% 件 技術移轉 權利金 0 0 100% 千元 碩士生 0 0 100% 博士生 0 0 100% 博士後研究員 0 0 100% 參與計畫人力 （外國籍） 專任助理 0 0 100% 人次 其他成果
