原文速度對英譯中同步口譯產出之影響

全文

(1)國立台灣師範大學翻譯研究所博士論文 Doctoral Dissertation Graduate Institute of Translation and Interpretation National Taiwan Normal University. 原文速度對英譯中同步口譯產出之影響 The Effects of Input Rate on the Output of Simultaneous Interpreting from English into Chinese. 指導教授：劉敏華博士、林世華博士 Advisors: Dr. Liu, Min-hua, Dr. Lin, Sieh-hwa 研究生：丘羽先 Student: Chiu, Yu-hsien. 中華民國一〇六年七月 July, 2017.

(2) 摘要本研究旨在探討原文速度對英譯中同步口譯產出之語言及時間面向的影響，並檢視原文速度與英語能力之交互作用對口譯產出各面向的影響。本研究之實驗材料為三篇英文演講。每篇演講之原文速度操弄為三個水準：每分鐘 100 字、每分鐘 130 字與每分鐘 160 字。. 實驗參與者為 28 位翻譯研究所口譯組學生。這 28 位參與者依照英語能力檢定程度區分成高英語能力與低英語能力兩組。每位參與者翻譯三篇演講，而三篇演講的速度均不同。參與者之同步口譯產出則從六個語言面向與三個時間面向來分析。語言面向包括詞彙的漏譯、片段的漏譯、詞彙的替代、片段的替代、句構干擾與詞彙多樣性。時間面向包括耳口間距（EVS）、無聲停頓之次數與產出速度。. 實驗結果採用線性混合模式（linear mixed model）之變異數分析，以檢視原文速度及原文速度與英語能力之交互作用對九項口譯產出面向的影響。統計分析結果顯示，原文速度與英語能力之交互作用對口譯產出面向的影響均不顯著，而原文速度對詞彙的漏譯、片段的漏譯、詞彙的替代、句構干擾、詞彙多樣性、EVS、無聲停頓與產出速度等八個面向則有顯著影響。. Bonferroni 事後檢定結果顯示，在每分鐘 100 字的速度下，學生口譯員的詞彙漏譯與詞彙替代比在每分鐘 130 字與 160 字的速度下要多，而片段漏譯則是隨著原文速度上升而增加。在每分鐘 100 字的速度下，學生口譯員受到的原文句構干擾比在每分鐘 160 字的速度下顯著要高。不過，學生口譯員的詞彙多樣性則是在每分鐘 160 字的速度下比另外兩個速度顯著要高。就時間面向而言，當原文速度從每分鐘 100 字上升至每分鐘 130 字，學生口譯員的 EVS 縮短，無聲停頓次數減少，產出速度加快。原文速度從每分鐘 130 字上升至 160 字時，學生口譯員的無聲停頓次數持續減少，但 EVS 並未持續縮短，而產出速度也未顯著增加。以上結果顯示，原文速度過快或過慢，對學生口譯員 i.

(3) 的產出都會造成困難。即便英語能力與原文速度之交互作用均未達顯著，英語能力仍然對學生口譯員的同步口譯表現有重要影響。高英語能力組學生的片段漏譯現象顯著較少，處理中英句構差異時採用顯著較多的順譯策略，而產出速度也比低英語能力組學生顯著要快。. 關鍵詞：原文速度、錯誤分析、句構干擾、詞彙多樣性、耳口間距、無聲停頓、產出速度. ii.

(4) Abstract This research seeks to investigate the effects of input rate as well as the interactive effects of input rate and English proficiency on the linguistic and temporal aspects of the output of simultaneous interpreting (SI) from English into Chinese. In the present study, three source speeches were used and the input rate of these speeches was manipulated to be at three levels – 100 wpm, 130 wpm, and 160 wpm – for the SI experiment. Twenty-eight graduate students of interpreting participated in the experiment. The participants were divided into high and low groups according to their English proficiency levels. Each participant interpreted three source speeches and the input rate of each speech was different. Six linguistic and three temporal parameters were adopted to analyze the participants’ SI output. The linguistic parameters include omissions of words, omissions of segments, substitutions of words, substitutions of segments, syntactic interference, and lexical diversity. The temporal parameters include EVS, the number of unfilled pauses, and the output rate. A linear mixed model analysis of variance (ANOVA) was used to examine the effects of input rate as well as the interactive effects of input rate and English proficiency on the nine aspects of student interpreters’ output. The results of statistical analysis indicated that there was no significant interactive effect between input rate and English proficiency on student interpreters’ output, but there were significant effects of input rate on omissions of words, omissions of segments, substitutions of words, syntactic interference, lexical diversity, EVS, unfilled pauses, and the output rate. The Bonferroni post hoc tests revealed that student interpreters made more omissions and substitutions of words at 100 wpm than at 130 wpm and 160 wpm. However, they omitted more segments as the input rate increased. The extent of syntactic interference from the source language was significantly higher at 100 wpm than at 160 wpm. In addition, student interpreters’ lexical diversity was significantly higher at 160 wpm than at the two iii.

(5) slower rates. As for the influence of input rate on the temporal aspects, when the input rate increased from 100 wpm to 130 wpm, student interpreters shortened their EVS, paused less and spoke faster. As the input rate increased from 130 wpm to 160 wpm, their unfilled pauses decreased further. However, their EVS did not shorten further and the output rate did not increase beyond the rate of 130 wpm. These findings suggest that both slow and fast input rates posed difficulty to student interpreters’ SI output. Although no significant interactive effect was found between input rate and English proficiency, English proficiency still played an important role in the SI performance. Students in the high proficiency group made significantly fewer omissions of segments and adopted more linearity strategies when dealing with syntactic differences between English and Chinese. They also spoke faster than the low proficiency group.. Key words: input rate, error analysis, syntactic interference, lexical diversity, EVS, unfilled pauses, output rate. iv.

(6) 謝辭首先要感謝我的兩位指導教授－劉敏華老師與林世華老師。謝謝敏華老師在口譯研究上一路指引與扶持，也給予我莫大的鼓勵與關懷。謝謝林老師不嫌棄我這個數學幼幼班的學生，讓我有機會踏入量性研究與統計的領域。兩位老師總是耐心容忍我天馬行空、混沌不清的想法，也總是不厭其煩的回答和解決我各種疑難雜症。在這段絞盡腦汁的痛苦過程中，兩位老師一直賦予我滿滿的正能量，讓我得以重拾信心和勇氣。能夠有敏華老師與林老師的指導，是無比幸福的事！. 感謝廖柏森老師、汝明麗老師、張嘉倩老師與葉舒白老師撥空擔任口試委員。謝謝老師們在口試時提出寶貴的建議，幫助我找到思考的盲點，讓這本論文能夠更臻於完善，也謝謝老師們給予的諸多鼓勵。. 在進行實驗與分析語料的過程中，因為有許多人的熱心協助，每一個環節才得以順利進行。在此要感謝所有參與預試和正式實驗的口譯學生，以及每一位從中協助聯絡的翻譯所老師。特別感謝吳宜錚老師、李筱雯老師、我的外甥鄭家豪與好友陳欣怡，沒有他們情義相挺，這本論文是沒有辦法順利完成的。也要謝謝一路陪伴的同學林宜瑄，感謝她在口試時的大力幫忙。. 在這段漫長而艱辛的旅程中，還有許多人給予我滿滿的愛與支持，幫助我度過重重難關，讓我能夠勇敢地走完全程。感謝阿嬤、楊阿姨、蔡老師、嚴阿姨。感謝我的媽媽，她總是無怨無悔、全心付出。衷心感謝所有幫助我的一切。. 謝謝老天爺！. v.

(7) Table of Contents Chapter One Introduction ..................................................................................................... 1 1.1 Research background and motivation .......................................................................... 1 1.2 Research purpose ......................................................................................................... 6 1.3 Research questions ....................................................................................................... 7 1.4 Research hypotheses .................................................................................................... 8 Chapter Two Literature Review .......................................................................................... 11 2.1 Characteristics of simultaneous interpreting .............................................................. 11 2.1.1 Concurrent speaking and listening .................................................................. 11 2.1.2 Ear-voice span (EVS) ..................................................................................... 13 2.1.2.1 Measurement of EVS ........................................................................... 13 2.1.2.2 Factors affecting EVS .......................................................................... 16 2.1.2.3 EVS and quality of interpreting ........................................................... 18 2.1.3 Segmentation................................................................................................... 19 2.2 Processing models of simultaneous interpreting........................................................ 20 2.2.1 Moser’s processing model .............................................................................. 21 2.2.2 Gile’s Effort Models ....................................................................................... 24 2.2.3 Psycholinguistic model of the reformulation process in interpreting ............. 26 2.3 Input rate .................................................................................................................... 28 2.3.1 The study of Gerver (1969)............................................................................. 30 2.3.2 The study of Pio (2003) .................................................................................. 31 2.3.3 The effects of input rate on accuracy .............................................................. 32 2.3.4 The effects of input rate on fluency ................................................................ 38 2.3.5 The effects of input rate on EVS and output rate ............................................ 44 2.3.6 The effects of input rate on translation approaches ........................................ 45 Chapter Three Method ........................................................................................................ 49 3.1 Overview of the study ................................................................................................ 49 3.2 Research design ......................................................................................................... 50 3.3 Participants ................................................................................................................. 51 3.4 Materials .................................................................................................................... 55 3.4.1 Critical sentences ............................................................................................ 55 3.4.2 Recording of the source speeches ................................................................... 58 3.4.3 Pilot study of the source speeches .................................................................. 58 3.4.4 Text analysis of the source speeches ............................................................... 62 3.5 Procedure ................................................................................................................... 64 3.6 Data scoring and analysis........................................................................................... 66 3.6.1 Preparation of output transcripts and scoring sheets....................................... 66 3.6.2 Analysis of omissions and substitutions ......................................................... 67 vi.

(8) 3.6.3 Analysis of syntactic interference ................................................................... 75 3.6.4 Measurement of lexical diversity .................................................................... 78 3.6.5 Measurement of EVS ...................................................................................... 80 3.6.6 Measurement of unfilled pauses ..................................................................... 81 3.6.7 Measurement of output rate ............................................................................ 82 3.6.8 Statistical analysis ........................................................................................... 83 Chapter Four Results........................................................................................................... 84 4.1 Omissions of words and segments (OW & OS) ........................................................ 84 4.2 Substitutions of words and segments (SW & SS) ...................................................... 91 4.3 Syntactic interference (STI) ....................................................................................... 97 4.4 Lexical diversity....................................................................................................... 101 4.5 EVS .......................................................................................................................... 105 4.6 Unfilled pauses......................................................................................................... 109 4.7 Output rate ............................................................................................................... 113 Chapter Five Discussion ................................................................................................... 119 5.1 Overview of major findings ..................................................................................... 119 5.2 The role of English proficiency in the effects of input rate ..................................... 121 5.3 The effects of input rate ........................................................................................... 123 5.4 The effects of English proficiency ........................................................................... 136 5.5 The effects of speech................................................................................................ 139 5.6 General discussion ................................................................................................... 142 5.7 Implications for interpreting studies and interpreter training .................................. 145 5.8 Limitations of the study and suggestions for future research .................................. 147 5.9 Conclusion ............................................................................................................... 150 References .............................................................................................................................. 152 Appendix A ............................................................................................................................ 162 Appendix B ............................................................................................................................ 167 Appendix C ............................................................................................................................ 173 Appendix D ............................................................................................................................ 174 Appendix E ............................................................................................................................ 179 Appendix F............................................................................................................................. 181 Appendix G ............................................................................................................................ 185. vii.

(9) List of Tables Table 2.1 Summary of Previous Studies on EVS .................................................................... 15 Table 2.2 Comparison of Error Categories for Output Accuracy ............................................ 34 Table 2.3 Summary of Studies on the Effects of Input Rate on Fluency ................................. 42 Table 3.1 English Proficiency Levels of the Participants ........................................................ 53 Table 3.2 Practice Time of Interpreting Per Week ................................................................... 53 Table 3.3 Self-assessment Results of Language and Interpreting Skills ................................. 54 Table 3.4 Examples of English Critical Sentences and Chinese Target Sentences .................. 57 Table 3.5 Presentation Order of Input Rates for the Pilot Study ............................................. 59 Table 3.6 Subject Judgment of Text Difficulty in the Pilot Study ........................................... 59 Table 3.7 Descriptive Statistics of Each Dependent Variable in the Pilot Study ..................... 60 Table 3.8 Differences among Three Source Speeches in the Pilot Study ................................ 61 Table 3.9 Text Analysis of the Three Source Speeches ........................................................... 63 Table 3.10 Presentation Order of Three Input Rates ................................................................ 65 Table 3.11 Means and Standard Deviations for Each Error Category ..................................... 72 Table 3.12 Correlation Coefficient of Two Raters on Four Error Categories .......................... 72 Table 4.1 Means and Standard Deviations of Omissions of Words Arranged by Input Rate, English Proficiency, and Speech .............................................................................................. 85 Table 4.2 Means and Standard Deviations of Omissions of Segments Arranged by Input Rate, English Proficiency, and Speech .............................................................................................. 86 Table 4.3 Three-Factor ANOVA Summary Table for Omissions of Words ............................. 87 Table 4.4 Three-Factor ANOVA Summary Table for Omissions of Segments ........................ 88 Table 4.5 Estimated Marginal Means of Omissions of Words and Segments among Three Input Rates ............................................................................................................................... 88 Table 4.6 Pairwise Comparisons of Estimated Marginal Means of Omissions of Words ....... 89 Table 4.7 Pairwise Comparisons of Estimated Marginal Means of Omissions of Segments .. 89 Table 4.8 Means and Standard Deviations of Substitutions of Words Arranged by Input Rate, English Proficiency, and Speech .............................................................................................. 92 Table 4.9 Means and Standard Deviations of Substitutions of Segments Arranged by Input Rate, English Proficiency, and Speech .................................................................................... 93 Table 4.10 Three-Factor ANOVA Summary Table for Substitutions of Words ....................... 94 Table 4.11 Three-Factor ANOVA Summary Table for Substitutions of Segments .................. 95 Table 4.12 Estimated Marginal Means of Substitutions of Words among Three Input Rates . 95 Table 4.13 Pairwise Comparisons of Estimated Marginal Means of Substitutions of Words . 96 Table 4.14 Means and Standard Deviations of Syntactic Interference Arranged by Input Rate, English Proficiency, and Speech .............................................................................................. 98 Table 4.15 Three-Factor ANOVA Summary Table for Syntactic Interference ........................ 99 Table 4.16 Estimated Marginal Means of Syntactic Interference among Three Input Rates .. 99 viii.

(10) Table 4.17 Pairwise Comparisons of Estimated Marginal Means of Syntactic Interference . 100 Table 4.18 Means and Standard Deviations of Lexical Diversity Arranged by Input Rate, English Proficiency, and Speech ............................................................................................ 101 Table 4.19 Three-Factor ANOVA Summary Table for Lexical Diversity.............................. 103 Table 4.20 Estimated Marginal Means of Lexical Diversity among Three Input Rates ........ 103 Table 4.21 Pairwise Comparisons of Estimated Marginal Means of Lexical Diversity ........ 104 Table 4.22 Means and Standard Deviations of EVS Arranged by Input Rate, English Proficiency, and Speech ......................................................................................................... 105 Table 4.23 Three-Factor ANOVA Summary Table for EVS .................................................. 107 Table 4.24 Estimated Marginal Means of EVS among Three Input Rates ............................ 107 Table 4.25 Pairwise Comparisons of Estimated Marginal Means of EVS ............................ 108 Table 4.26 Means and Standard Deviations of Unfilled Pauses Arranged by Input Rate, English Proficiency, and Speech ............................................................................................ 109 Table 4.27 Three-Factor ANOVA Summary Table for Unfilled Pauses ................................ 111 Table 4.28 Estimated Marginal Means of Unfilled Pauses among Three Input Rates .......... 111 Table 4.29 Pairwise Comparisons of Estimated Marginal Means of Unfilled Pauses ........... 112 Table 4.30 Means and Standard Deviations of Output Rate Arranged by Input Rate, English Proficiency, and Speech ......................................................................................................... 113 Table 4.31 Three-Factor ANOVA Summary Table for Output Rate ...................................... 115 Table 4.32 Estimated Marginal Means of Output Rate among Three Input Rates ................ 115 Table 4.33 Pairwise Comparisons of Estimated Marginal Means of Output Rate ................ 116 Table 4.34 Estimated Marginal Mean of Output Rate Arranged by Input Rate and Speech . 117 Table 4.35 Pairwise Comparisons Estimated Marginal Means of Output Rate in Speech A 117 Table 5.1 Summary of Major Findings .................................................................................. 120 Table 5.2 Means and Standard Deviations of Linearity as a Function of Input Rate and English Proficiency ................................................................................................................ 121 Table 5.3 Means and Standard Deviations of the Output Rate as a Function of Input Rate and English Proficiency ................................................................................................................ 122 Table 5.4 Mean Word Length of Main Clauses and Adverbials in Critical Sentences .......... 140. ix.

(11) List of Figures Figure 2.1 Moser’s Processing Model of Simultaneous Interpreting ...................................... 23 Figure 2.2 Two Translation Routes .......................................................................................... 26 Figure 3.1 Research Design with Independent and Dependent Variables ............................... 51 Figure 3.2 Screenshot of Type-Token Ratio on CRIE 3.2 ....................................................... 79 Figure 3.3 Screenshot of Word Segmentation on CRIE 3.2 .................................................... 79 Figure 3.4 Screenshot of EVS Markings on Audacity ............................................................. 81 Figure 3.5 Screenshot of Unfilled Pauses Markings on Audacity ........................................... 82 Figure 5.1 Estimated Marginal Means of OW among Input Rates ........................................ 124 Figure 5.2 Estimated Marginal Means of OS among Input Rates ......................................... 124 Figure 5.3 Estimated Marginal Means of SW among Input Rates ........................................ 126 Figure 5.4 Estimated Marginal Means of SS among Input Rates .......................................... 126 Figure 5.5 Estimated Marginal Means of STI among Input Rates ........................................ 128 Figure 5.6 Estimated Marginal Means of N/A among Input Rates ........................................ 129 Figure 5.7 Estimated Marginal Means of Lexical Diversity among Input Rates .................. 130 Figure 5.8 Estimated Marginal Means of EVS among Input Rates ....................................... 132 Figure 5.9 Estimated Marginal Means of Unfilled Pauses among Input Rates ..................... 133 Figure 5.10 Estimated Marginal Means of Output Rate among Input Rates ......................... 135 Figure 5.11 Interaction of Input Rate by Speech on Output Rate .......................................... 136 Figure 5.12 Estimates of OS as a Function of Input Rate and English Proficiency .............. 137 Figure 5.13 Estimates of STI, NI, LI and N/A between English Proficiency Groups ........... 138 Figure 5.14 Estimates of Output Rate between English Proficiency Groups ........................ 139. x.

(12) Chapter One. Introduction. 1.1 Research background and motivation Simultaneous interpreting (SI) is a highly complex skill that requires years of extensive training and professional practices to master. What sets simultaneous interpreting apart from other language tasks is the demand of juggling various concurrent online operations, including listening to speech segments in the source language, retaining source speech segments in working memory, reformulating stored segments into the target language, producing target speech segments, and monitoring the output. Each one of these online tasks competes for limited cognitive resources (Gile, 2009). Therefore, coordination of online tasks and efficient allocation of attentional resources are essential for interpreters to successfully carry out the task of SI. Given the complexity of multitasking in SI, interpreters often work close to the limit of their mental capacity (Gile, 1999). Factors that will increase the processing requirement of any of the above-mentioned online tasks during SI are likely to cause difficulty in the interpreting process. Input rate, the speed at which messages of source speech are delivered, is one of the prominent factors that affect the processing load and performance of simultaneous interpreters (Pöchhacker, 2004). From a cognitive perspective, input rate determines the speed “at which information has to be processed by interpreters” (Christoffels & De Groot, 2005, p. 463). With higher input rates, interpreters have to process more information at a given unit of time, which increases the processing requirement of comprehension, speech reformulation, and production. When the processing load exceeds interpreters’ total capacity, they might not have enough cognitive resources to attend to comprehension or speech production and interpreting quality might thus suffer (Gile, 2009; Pio, 2003). On the other hand, a slow speech can also be a problem trigger. The study by Shlesinger (2003) indicated that with a slower input rate, traces for unrehearsed items in interpreters’ working memory were likely to decay, which could lead to 1.

(13) more forgetting of the source speech information. In the literature of interpreting studies, it is suggested that the optimal input rate for SI should be between 100-120 words per minute (wpm) (Seleskovitch, 1965, as cited in Gerver, 1976). However, in real-life settings, interpreters often have to cope with speeches delivered at much faster speeds. Barghout, Rosendo and García (2015) observed that the average input rate of twenty speeches given at the 16th session of the Human Rights Council under the United Nations was about 149 wpm, with the highest rate at 188 wpm and the lowest rate at 106 wpm. Since the input rate is set by the speaker, simultaneous interpreters can only strive to deal with whatever rate the source speech is delivered. According to a report conducted by the International Association of Conference Interpreters (AIIC) in 2012, “fast speech” was listed as the number one stressor for interpreters around the world (as cited in Barghout et al., 2015, p. 307). It seems that fast input rates have become the norm rather than the exception in conference interpreting settings. An investigation into how the input rate affect interpreters’ performance can shed light on the cognitive constraints interpreters have to cope with and explain why fast input rates are often considered one of the main stressors for interpreters. One of the earliest studies on the impact of input rate was conducted by Gerver (1969). He investigated the effects of input rate on professional interpreters’ output of SI from French into English by varying the rates from 95 wpm to 164 wpm in a source speech. Input rates were found to have a significant effect on interpreters’ accuracy. When the input rate was over 120 wpm, there was a sharp decline in interpreters’ proportion of correctly translated words and an increase in omissions and substitutions. In addition to interpreters’ linguistic performance, Gerver also examined the effects of input rate on several temporal parameters of the output, such as the interpreters’ ear-voice span (EVS),1 mean utterance time, unfilled pause time, the ratio of pause time to speech time, and the output rate. Results showed that. 1. EVS is the time lag between the moment when a segment is heard in the source speech and the moment it is produced in the target speech. 2.

(14) when the input rate increased from 120 wpm to 164 wpm, interpreters’ EVS and ratio of pause time to speech time sharply increased while their mean output rate remained constant. Gerver explained that at higher input rates, interpreters had less time to decode the input and produce the output. When the input rate became faster, interpreters opted to pause longer and more often in order to process items quickly accumulated in the short-term memory buffer. Instead of racing with the speaker, they maintained a steady pace and lagged further behind the speaker, leading to more omissions in the output. Concerning the influence of input rate on interpreters’ accuracy, subsequent studies consistently found that the faster the input rate was the more omissions interpreters made in their output (Barik, 1975; Galli, 1989; Lee, 1999b, 2002; Pio, 2003), but this did not necessarily apply to substitutions, which refers to words or segments that have been inaccurately translated in the target output (Galli, 1989). Pio (2003) took a step further to investigate how the input rate affected output fluency. She compared professional and student interpreters’ SI performance from German into Italian at the input rate of 108 wpm and 145 wpm. Results from descriptive statistics indicated that both professional and student interpreters made more filled pauses and unfilled pauses of more than 3 seconds when interpreting the fast speech. This corroborated Gerver’s (1969) findings that when facing fast input rates, interpreters tended to make longer pauses in order to buy more time of processing. However, the study by Cecot (2001) found otherwise. She observed that professional interpreters made fewer hesitation pauses when interpreting the faster English source speech into Italian. In terms of interpreters’ EVS under the fast source speech, Pio (2003) indicated that half of her participants, regardless of professional or student interpreters, opted to lengthen the EVS while the other half chose to shorten the EVS. This result was not in line with Gerver’s findings. Lee (2002) also observed that when interpreting from English into Korean, professional interpreters tended to shorten their EVS as the input rate became faster or when 3.

(15) the speech proportion of the speaker was higher so as to quickly catch up with the speaker and avoid being overloaded. Yagi (2000) suggested that the length of EVS can be affected by several factors such as the input rate, word order, and language combination. It is possible that the discrepancy of results among these studies stems from differences in language combinations or translation directions. Since few studies have sought to examine the effects of input rate on the temporal aspects of the output, little is known on how interpreters with linguistically distant languages such as English and Chinese would adjust their EVS and output rates to cope with different input rates. Apart from accuracy and fluency of the output, the input rate has also been found to affect the translation approach adopted by interpreters, namely the meaning-based and form-based strategies. In interpreting studies, the meaning-based strategy or deverbalization is defined as an approach through which the meaning is extracted from the source speech while the surface form is forgotten, whereas the form-based strategy or transcoding involves direct replacement of equivalents at the phonological, lexical-semantic, morphological, and syntactic level between the source and target languages (Christoffels & De Groot, 2005; Dam, 1998; Seleskovitch, 1976). The form-based approach is considered economical but it also entails greater risk of linguistic interference (Gile, 2009). According to Lamberger-Felber & Schneider (2008), linguistic interference refers to “the result of the auditive and/or visual influence of the source language or source text on structures/elements of the target text that results in a deviation from the norms of the target language” (p. 218). It is generally assumed that professional interpreters mainly adopt the meaning-based approach and only resort to the form-based strategy when encountering processing difficulty such as fast input rates or when under stress and fatigue (Dam, 2001; Fabbro, Gran, & Gran, 1991; Isham, 1994). The studies by Dam (2001) and Zhang (2010) observed that the input rate did affect interpreters’ translation approaches. However, the effects were found to be the opposite. In Dam’s (2001) findings, the meaning-based approach was associated with more difficult and faster source 4.

(16) texts, while Zhang (2010) observed that the faster the source speech was, the less often interpreters attempted to deverbalize. Another linguistic aspect of the output that has rarely been explored in interpreting literature is lexical diversity. Lamberger-Felber (2001) found great variability in words used by individual interpreters and among the group of 12 interpreters. Timarová, Čeňková and Meylaerts (2015) observed a moderate positive correlation between interpreting experience in years and type-token ratio of interpreters’ SI output from English into Czech when the variance of age was not statistically controlled. This seems to suggest that more experienced interpreters had richer and more diverse vocabulary. However, when the variance of age was partialled out, no significant correlation was found between years of interpreting experience and type-toke ratio of the SI output. To our knowledge, no study has ever examined the effects of input rate on lexical diversity of the output. Taken together, the input rate determines the information load simultaneous interpreters have to bear during the interpreting process and thus has great impact on interpreters’ performance. It is well-established that faster input rates led to more omissions in interpreters’ output. However, results were inconclusive with regard to the influence of input rate on substitutions, EVS, the number of pauses, interpreters’ choice of translation approach, and lexical diversity. The present study is motivated by the need to revisit this issue and clarify how different aspects of SI output are affected by the input rate. Simultaneous interpreting performance is not only influenced by input characteristics, it can also be affected by interpreters’ internal factors such as prior knowledge (Anderson, 1994; Díaz-Galaz, Padilla, & Bajo, 2015), working memory span (Christoffels, De Groot, & Waldorp, 2003), and language proficiency (Tzou, Eslami, Chen, & Vaid, 2012). The study by Tzou et al. (2012) found that second language proficiency affected graduate students’ SI performance from English (L2) into Chinese (L1). Students with higher L2 proficiency performed better in the SI task. The role of L2 proficiency in the effects of input rate on the 5.

(17) SI output, however, has never been explored in previous interpreting studies.. 1.2 Research purpose Against this backdrop, the aim of this study is to investigate the effects of input rate on student interpreters’ linguistic and temporal aspects of SI output as well as the interactive effects of input rate and student interpreter’s English (L2) proficiency. The reason for choosing student interpreters as the participants for this study is twofold. First, most previous studies have been conducted on professional interpreters. Therefore, the findings of Gerver (1969), for instance, might not be generalizable to student interpreters. Second, student interpreters might be more susceptible to the manipulation of input rates than professional interpreters because students have little or no real-life experience of interpreting. They might not be adept at adopting strategies to cope with various input rates. Although in the study of Liu (2001), no significant performance difference was found at the input rate of 150 wpm and 130 wpm between professional and student interpreters in SI from English into Chinese, the impact of input rate on many other aspects of students’ output was not examined. In the present study, an experiment of SI from English into Chinese was conducted by using three English speeches. Each speech was manipulated to be at three levels –100 wpm, 130 wpm, and 160 wpm – for the experiment. A number of linguistic and temporal parameters were adopted to examine student interpreters’ SI output under the influence of three input rates. The linguistic parameters include omissions of words, omissions of segments, substitutions of words, substitutions of segments, the extent of syntactic interference from the source language, and lexical diversity. The first four parameters concern student interpreters’ output accuracy. The extent of syntactic interference is to clarify whether student interpreters were more susceptible to linguistic interference or more prone to adopt the form-based approach under the influence of various input rates. The last linguistic parameter aims at examining whether the input rate had an effect on how diverse student 6.

(18) interpreters’ words were in the output. In addition to linguistic parameters, three temporal parameters were employed to explore the impact of three input rates, including student interpreters’ EVS, unfilled pauses, and the output rate. These parameters would help illustrate the temporal relationship between input and output and shed some light on student interpreters’ processing constraints.. 1.3 Research questions This study seeks to answer the following research questions: 1. Would the interaction between the input rate and English proficiency level affect the omission of words and segments in student interpreters’ output? 2. Would input rates affect the omission of words and segments in student interpreters’ output? 3. Would the interaction between the input rate and English proficiency level affect the substitution of words and segments in student interpreters’ output? 4. Would input rates affect the substitution of words and segments in student interpreters’ output? 5. Would the interaction between the input rate and English proficiency level affect the extent of syntactic interference from the source language in student interpreters’ output? 6. Would input rates affect the extent of syntactic interference from the source language in student interpreters’ output? 7. Would the interaction between the input rate and English proficiency level affect the lexical diversity of student interpreters’ output? 8. Would input rates affect the lexical diversity of student interpreters’ output? 9. Would the interaction between the input rate and English proficiency level affect student interpreters’ average EVS? 10. Would input rates affect student interpreters’ average EVS? 7.

(19) 11. Would the interaction between the input rate and English proficiency level affect the number of unfilled pauses in student interpreters’ output? 12. Would input rates affect the number of unfilled pauses in student interpreters’ output? 13. Would the interaction between the input rate and English proficiency level affect the output rate of student interpreters? 14. Would input rates affect the output rate of student interpreters?. 1.4 Research hypotheses Based on the proposed research questions, the following hypotheses were formed: Hypothesis 1 English proficiency would affect the effects of input rate on the omission of words in student interpreters’ output. Hypothesis 2 Input rates would affect the omission of words in student interpreters’ output. Hypothesis 3 English proficiency would affect the effects of input rate on the omission of segments in student interpreters’ output. Hypothesis 4 Input rates would affect the omission of segments in student interpreters’ output. According to results from previous studies (Barik, 1975; Galli, 1989; Gerver, 1969, Pio 2003), student interpreters were expected to omit more words and segments as the input rate increased. Hypothesis 5 English proficiency would affect the effects of input rate on the substitution of words in student interpreters’ output. Hypothesis 6 Input rates would affect the substitution of words in student interpreters’ output. Hypothesis 7 English proficiency would affect the effects of input rate on the substitution of segments in student interpreters’ output.. 8.

(20) Hypothesis 8 Input rates would affect the substitution of segments in student interpreters’ output. The study by Galli (1989) did not find big differences between the numbers of substitution made by professional interpreters at two different input rates. However, in Pio’s (2003) study, student interpreters made more substitutions when interpreting the faster speech. In the present study, student interpreters were expected to substitute more words and segments at faster input rates. Hypothesis 9 English proficiency would affect the effects of input rate on the extent of syntactic interference from the source language in student interpreters’ output. Hypothesis 10 Input rates would affect the extent of syntactic interference from the source language in student interpreters’ output. Student interpreters tend to adopt more form-based strategies than professional interpreters and may be more inclined to resort to this approach when they are under greater cognitive load (Fabbro & Gran, 1994; Gile, 2009). It is presumed that with the form-based approach, elements from the source speech would be directly transposed into the target speech. Therefore, we assumed that there would be greater extent of syntactic interference from the source language when student interpreters were facing higher input rates. Hypothesis 11 English proficiency would affect the effects of input rate on the lexical diversity of student interpreters’ output. Hypothesis 12 Input rates would affect the lexical diversity of student interpreters’ output. When students were coping with heavier cognitive load, their ability to retrieve words in the target language might be affected. Besides, students were expected to make more omissions when facing fast speeches. Based on these assumptions, their lexical diversity was likely to decrease when the input rate became faster. Hypothesis 13 English proficiency would affect the effects of input rate on the average EVS of student interpreters. 9.

(21) Hypothesis 14 Input rates would affect the average EVS of student interpreters. According to previous studies, some professional and student interpreters had longer EVS when coping with fast speeches while others tended to shorten their EVS (Lee, 2002; Pio, 2003). Since results were inconclusive, in the present study we expected that students would either lengthen or shorten their EVS at faster input rates. Hypothesis 15 English proficiency would affect the effects of input rate on the number of unfilled pauses in student interpreters’ output. Hypothesis 16 Input rates would affect the number of unfilled pauses in student interpreters’ output. Since faster input rates were going to bring greater cognitive load, student interpreters were expected to make more unfilled pauses as the input rate increased. Hypothesis 17 English proficiency would affect the effects of input rate on the output rate of student interpreters. Hypothesis 18 Input rates would affect the output rate of student interpreters. Speech rate is related to the pauses a speaker makes. If we assumed students were going to make more pauses as the input rate increased, their output rates were likely to decrease.. 10.

(22) Chapter Two. Literature Review. This chapter first presents an overview of the characteristics of simultaneous interpreting (SI), including concurrent listening and speaking, EVS and segmentation. The following section covers three prominent processing models of SI, which serve as a framework for explicating the complex processes of SI, the dynamics of processing capacity as well as the two translation routes. The final part provides a review of previous studies on the effects of input rate on the accuracy, fluency, EVS, output rate and translation strategies.. 2.1 Characteristics of simultaneous interpreting Simultaneous interpreting involves source language comprehension, reformulation, and target language production all at the same time. The simultaneity of listening to one language and speaking in another is one of the most salient characteristics of this mode of interpreting (Christoffels & De Groot, 2005; Lee, 1999a). Though simultaneous interpreters engage in concurrent listening and speaking during the task, their target output is not fully synchronized with the source speech. Interpreters need to comprehend a certain amount of information before they are able to deliver the target output. Therefore, there is usually a time lag of a few seconds between a source message and a target message, which is also known as the EVS. The distance of the time lag depends on how interpreters chunk the flow of source speech into segments based on their linguistic and extralinguistic knowledge (Kirchhoff 1976/2002). These features of SI will be discussed in the following section.. 2.1.1 Concurrent speaking and listening The demand of comprehending a language while producing and monitoring another one is what makes SI particularly complex. During this process, attention has to be efficiently shared or switched between different online operations (Christoffels & De Groot, 2005; 11.

(23) Cowan, 2000). Early researchers proposed that in order to reduce the load of concurrent listening and speaking, simultaneous interpreters would attempt to maximize the use of pauses in a source speech for delivering their output (Barik, 1973; Goldman-Eisler, 1972). Gerver (1976), however, argued that since the majority of the pauses were less than one second, they were too short for interpreters to “cram” their production into (p. 183). Previous studies have shown that simultaneous interpreters engaged in concurrent listening and speaking for more than half of the total utterance time. Gerver (1976) analyzed 14 recordings of six French-English interpreters and found that the average simultaneity proportion was 64%. According to Chernov (1979), the proportion of concurrent listening and speaking for interpreters with English-Russian combination was 70.5% of the total speaking time. Lee (1999a) examined 30 samples of English to Korean recordings by 11 professional interpreters and found that concurrent listening and speaking accounts for 61% of the total syllables uttered by the speaker. Researchers also looked into whether the simultaneity of comprehension and production affected the quality of interpreting performance. Based on an experiment on the effects of noise, Gerver (1975) reported that some interpreters were able to correctly interpreted 85% of the source speech when the simultaneity proportion was over 75% of the total speaking time. Lee (1999a), on the other hand, found that the accuracy of the concurrent listening and speaking portion was lower than that of the listening only part. These findings indicated that professional interpreters were able to successfully carry out the complex processes of SI but they were also under greater cognitive strain during the concurrent listening and speaking period. Christoffels (2006) suggested that the demand of producing the target speech while listening to the source text “resembles articulatory suppression” (p. 208), which prevents. 12.

(24) subvocal rehearsal2 and disrupts retention of the source text. This may explain why some studies found that recall of the source speech after SI was poorer than recall after the listening task (Gerver, 1974; Isham, 1994; Lambert, 1988). Since simultaneous interpreters have to listen and speak concurrently for most of the time, efficient allocation of processing capacity is essential for ensuring the quality of performance. Shreve & Diamond (1997) suggested that by making the language processing more automatic and less effortful through advanced preparation for the conference theme, for instance, interpreters would better cope with the cognitive constraint imposed by the SI task.. 2.1.2 Ear-voice span (EVS) One way to cope with the processing load of SI is through regulation of the EVS (Gile, 1997; Gumul & Łyda, 2007). By maintaining a short EVS, the load to retain source messages in short-term memory is reduced. However, interpreters who opt to follow the speaker closely with short EVS may be prone to misinterpretation, false starts or literal translation due to incomplete comprehension. On the other hand, prolonging the EVS and waiting for more information may lead to better understanding of the source messages and more idiomatic output but the risk of memory loss is higher. Gile (1997) suggested that regulation of the EVS is aimed at “optimizing the balance between short-term memory load and speech production requirements” (p. 207). There are perils for either following too close or lagging too far behind the source speech.. 2.1.2.1 Measurement of EVS The time lag between the moment a source segment is heard and the moment the interpreter delivers its corresponding target segment provides a window into the temporal 2. Subvocal rehearsal is a process that refreshes traces of verbal materials in the phonological loop of working memory. When subvocal rehearsal is prevented by articulatory suppression, maintenance of verbal materials is disrupted, which could then result in memory loss (Baddeley, 2003). 13.

(25) relationship between the source speech and the target output as well as how the interpreter copes with online cognitive load. Some early studies measured EVS in number of words (Gerver, 1969; Treisman, 1965) but most studies chose to measure EVS in time units, which is comparable across studies with different language combinations. Table 2.1 provides an overview of the EVS measurement methods in different studies. It seems that the average EVS for SI is around two to five seconds across different points of measurement, language combinations and participants, except for the studies by Lamberger-Felber (2001) and Chang (2009). The average EVS observed by Lamgerber-Felber (2001) and Chang (2009) was both over five seconds. Lamgerber-Felber (2001) measured EVS on segments with long omission (more than 15 words), which might contribute to the long average time lag. Chang (2009) also found that the EVS durations longer than five seconds were related to omissions and explained that the long EVS may be attributed to the difficulty of finding equivalent terms as the participants were working into their second language. As for the point of measurement for EVS, there is less consistency across studies (see Table 2.1). Researchers mentioned that it is oftentimes difficult to match the source segment with the target segment to determine an EVS when there has been linguistic restructuring or when the interpreter has adopted a summary or generalization strategy (Chang, 2009; Timarová, Dragsted, & Hansen, 2011). Oléron & Nanpon (1965/2002) chose words with literal correspondence as the EVS unit, while other early studies measured EVS with a fixed interval such as every five seconds or every fifth word (Barik, 1973; Gerver, 1969; Treisman, 1965). Many later studies regarded the beginning of a sentence as the start of EVS and the first word of the corresponding target segment as the end of EVS even when there were omissions or other translation errors (Chang, 2009; Chen, 2012; Kim, 2005; Lee, 2002; Timarová et al., 2011). To find out whether different methods of measurement lead to disparity in EVS results, Timarová et al. (2011) analyzed a 5-minute segment of SI from English into Czech by a professional interpreter and compared three methods of EVS 14.

(26) measurement by Treisman (1965), Barik (1973) and Lee (2002). They found no statistically significant difference among the mean and median values of EVS but there was a two-second difference in range values. These findings suggest that the average EVS derived from three different methods of measurement are comparable.. Table 2.1 Summary of Previous Studies on EVS Study. Point of measurement. Participants. Language combination. Average EVS. Treisman (1965). every 5 secs. 6 bilinguals. English-French French-English. Oléron & Nanpon (1965/2002). words with literal correspondence. 3 professional interpreters;. Gerver (1969). every 5th word. 10 professional. German-French French-Spanish English-French French-English French-English. 3.1 secs (5.1 words) 2.6 secs (4.3 words) 1.9 secs 2.7 secs 2.6 secs 5.4 secs 4 to 8.5 words. interpreters. Barik (1973). of the source text every 5 secs. 2 professional interpreters; 2 student interpreters; 2 bilinguals 12 professional interpreters. French-English English-French. 2.69 secs 2.53 secs. English-German. 5.8 secs. 30 professional interpreters. English-Korean. 3 secs. Lambeger-Felber beginning of a (2001) segment where at least one. Lee (2002). interpreter omitted more than 15 words sentence beginnings. 15.

(27) Table 2.1 (continued) Study. Point of measurement. Participants. Language combination. Pio (2003). segments where speaker made short pauses of 1 to 2.5 secs. 5 professional interpreters; 10 student interpreters. German-Italian. Kim (2005). sentence. 2 professional. beginnings (excluding those less than 1 sec) beginning of a primary or secondary message in the source language 1. sentence beginnings. interpreters. Chang (2009). Timarová et al. (2011). 5 student interpreters. 1 professional interpreter. Slow speech prof.: 3.21 secs stud.: 3.16 secs Fast speech prof.: 3.16 secs stud.: 3.10secs Japanese-Korean 1.85 secs Korean-Japanese Chinese-Korean Korean-Chinese Chinese-English. 1.58 secs 2.00 secs 1.78 secs 5.7 secs. English-Czech. 1. 3.6 secs 2. 3.8 secs. 2. verbs 3. figures. Chen (2012). sentence beginnings. Average EVS. 3. 2.6 secs 12 student interpreters (no SI experience) 12 student interpreters. 4.0 secs. English-Chinese. 3.7 secs. Note. secs: seconds; prof.: professional interpreters; stud.: student interpreters. 2.1.2.2 Factors affecting EVS Though the average EVS of SI is generally within two to five seconds across different language pairs, studies have found that there was high variability of EVS between different interpreters or within the performance of an individual interpreter (Lamberger-Felber, 2001; Pio, 2003; Timarová et al., 2011). Interpreters’ EVS can be affected by a number of external 16.

(28) factors, such as the input rate, text type, syntactic complexity and information density (Yagi, 2000). In other words, input variables that affect the performance of interpreting can also influence the length of EVS. Gerver (1969) found that professional interpreters lengthened their EVS as the input rate became faster. Barik (1973) examined the EVS of six participants (two professional interpreters, two student interpreters, and two amateurs) for four text types: spontaneous speech, semi-prepared material, prepare oral speech, and prepared written material. There was no significant difference in the EVS measures for the four text types. Nevertheless, he did observe that professional and student interpreters had the shortest EVS for the spontaneous material (a story for picture description) and the longest EVS for the prepared written speech. The two amateurs, on the other hand, showed the opposite trend. Lamberger-Felber (2001) compared the EVS length of 12 interpreters under the conditions of SI with text but no preparation, SI with text and preparation, as well as SI without text. Results showed that for nine interpreters, the EVS for SI with text (the first two conditions) were longer than that for SI without text. Apart from these external variables, it is also assumed that internal factors such as interpreters’ strategies or processing preference would affect the EVS length (Timarová et al., 2011; Yagi, 2000). For example, when interpreters adopt the strategy of anticipation, the EVS would be shorter or even in negative values. Moser-Mercer (1997) stated that expert interpreters tend to have longer EVS than novices, indicating that they might have more comprehensive and global analysis of the source speech. Timarová et al. (2014), however, found a significant negative correlation between interpreting experience measured in working days and the median EVS, showing that the more experienced interpreters were associated with shorter EVS. Taken together, interpreters’ EVS during SI is determined by the combination of these external and internal factors as well as the cognitive constraint imposed by the moment-to-moment processing. This may explain why high variability of EVS has been 17.

(29) observed in different interpreters as well as in an individual interpreter during an SI task.. 2.1.2.3 EVS and quality of interpreting To ensure the quality of interpreting, interpreters cannot stay too close with or lag too far behind the speaker. They must obtain enough information from the source text in order to deliver a meaningful target segment. At the same time, interpreters do not have the luxury to wait for too long for the short-term memory may be overloaded. Previous studies on the relationship of accuracy and EVS consistently indicated that long EVS was associated with lower accuracy. Barik (1973) found a positive correlation between the EVS of five out of six participants and the total number of omissions. Based on the analysis of 30 recordings and 814 sentences of SI from English to Korean, Lee (2002) observed that the accuracy of sentences with EVS longer than four seconds was lower than those with EVS shorter than two seconds (66.8% vs. 82.6%). In addition, EVS longer than four seconds had an adverse effect on the quality of not only the sentence currently being processed but also the following sentence. Similar findings were reported by Chen (2012). Thirteen graduate students simultaneously interpreted an English source speech into Chinese and the accuracy of their performance was evaluated by using error categories adopted from Barik (1975) as well as a five-point scale of fidelity. Results revealed that there was a significant negative correlation between the EVS length and fidelity scores – the longer the EVS, the lower the fidelity score. When the EVS was longer than four seconds, the average fidelity score declined to less than three points. Besides, EVS ranges were found to be related to error types. For example, 85% of the comprehension and delay omissions occurred in sentences with EVS longer than four to seven seconds. Chen (2012) therefore stated that the optimal EVS range is two to four seconds, corroborating the assumptions in early interpreting literature (Gerver, 1976; Paneth, 1957/2002) and the findings of Lee (2002).. 18.

(30) 2.1.3 Segmentation Maintaining an optimal EVS depends on the interpreter’s ability to segment the source speech stream at appropriate places and into manageable chunks (Barik, 1975; Kirchhoff, 1976/2002). Kirchhoff (1976/2002) suggested that the unit of interpreting is determined not only by the linguistic structures of the source speech but also by the equivalence relation between the source language (SL) and the target language (TL). In other words, interpreters, who assume the role of both a listener and a speaker, do not parse the source speech stream the way a listener does. Kirchhoff (1976/2002) further proposed that the minimum unit of interpreting is the “smallest possible decoding unit in the SL for which a 1:1 relationship can be established with a TL segment” (p. 114). To find out how simultaneous interpreters with different language combinations segment the input speech, Goldman-Eisler (1972) analyzed the output of six interpreters working from and into English, French, and German. There were a number of interesting findings. First, results showed that 90-95% of the minimum EVS unit was a complete NP+VP (noun phrase + verb phrase). Concerning interpreters’ general segmentation tendency, it was found that interpreters did not segment the source speech according to the linguistic units determined by the speaker. Goldman-Eisler (1972) proposed three ways of segmentation: encoding a chunk of speech after it ends (identity), starting to encode before a chunk comes to an end (fission) and storing two or more chunks and then encoding (fusion) (p. 134). Among the responses of six interpreters, the proportion of fission, fusion and identity is 48%, 41%, and 11% respectively. This indicated that interpreters tended to translate before a source segment ended with a pause or to combine two or more chunks before encoding. When the input rate was factored in, results showed that there were more identity responses in slow speech (13.8%) than in fast speech (7.7%). Goldman-Eisler (1972) explained that when the speech was slow, interpreters were able to make more use of the speaker’s pauses for delivering the output in the target language. In addition, Goldman-Eisler (1972) also found that 19.

(31) language-specific differences between the source language and the target language were a factor affecting the segmentation strategy. For example, when translating from German, a verb-final language, interpreters tended to prolonged the EVS and stored larger chunks than translating from English or French. In sum, the fact that interpreters need to listen and speak for most of the time during SI makes this task extremely demanding. The speaker sets the input rate, which consequently determines the information load that interpreters need to process per unit of time. Interpreters can only cope with different cognitive demand by varying the EVS throughout the task in order to strike a balance between memory load and production requirement (Gile, 1997). The optimal EVS was found to be two to four seconds (Chen, 2012). When interpreters lagged behind the source segment for more than four seconds, the quality of the currently processed sentence began to deteriorate and the performance of subsequent sentence tended to suffer as well due to memory overload (Chen, 2012; Lee, 2002). Therefore, proper segmentation of the input stream is essential for the success of interpreting. Most of the time, interpreters would not follow the units delineated by the speaker but tended to segment the input stream according to their own ways (Goldman-Eisler, 1972). The unit of interpreting is not words but involves an NP+VP. The size of a chunk during SI, however, would differ depending on external factors such as input variables and language-specific differences. These unique characteristics are what set SI apart from other language tasks.. 2.2 Processing models of simultaneous interpreting Most interpreting researchers regard SI as a multi-task, time-constraint process involving speech perception, comprehension, memory storage, reformulation, production, and output monitoring (Moser-Mercer, 1997). These processes all compete for limited attentional resources, which makes SI intrinsically difficult (Gile, 1997; Massaro & Shlesinger, 1997). To account for the complexity of the SI processes, interpreting researchers 20.

(32) have developed different models since the 1970s (Pöchhacker, 2004). In the following sections, two prominent processing models of SI by Moser (1978) and Gile (1995) will be reviewed first. Then a partial cognitive model proposed by Christoffels & De Groot (2005) on the translation routes will be discussed.. 2.2.1 Moser’s processing model One of the most influential full-process models of SI was developed by Moser (1978), which was based on the information processing model of speech comprehension by Massaro (1975). The model depicts a flow chart of SI processes from auditory reception to target output (See Figure 2.1). Structural components such as the generated abstract memory (GAM, which stores strings of processed words) are represented by boxes. Functional components that describe an operation at a particular stage are represented by intermediate headings. Diamonds represent a decision point at which a rehearsal loop is provided. In a later article, Moser-Mercer (1997) explained that the processes presented in the central area “occur in working memory with constant access and feedback to long-term memory” (p. 8). The function of short-term memory is represented by GAM in Moser’s model. The limit of GAM is presumed to be 7±2 chunks of information, based on the proposed working memory capacity by Miller (1956). There are several features of this full-process model. First is the provision of rehearsal loops at each decision point. The rehearsal loop is designed to account for concurrent operations of SI (Moser, 1978). For example, as the interpreter is preparing the target output of Segment A, he or she is also simultaneously attending to the comprehension and storage of the incoming Segment B. The second feature is the emphasis on the interaction between bottom-up processes (such as word recognition) and top-down processes (access to conceptual representations and prior knowledge) happening in every stage, signifying by the double-headed arrows in the flow chart. The third feature is the mechanism of prediction. 21.

(33) Moser (1978) explained that when the interpreter is able to anticipate the incoming message, the process of current input will be discarded and the TL elements are directly ready for output. This processing model seems to represent the ideal situation of SI, as Moser (1978) mentioned that this model depicts the SI processes of a skilled professional interpreter rather than a beginner. Although the flow chart focuses on the step-by step processes at different stages of SI, Moser (1978) did explain that there is interaction among the processes at each stage. For instance, when the syntactic and semantic processing of the source message is completed and the preverbal concepts for the source message are ready for target language output, more processing capacity is free up for the continuous processing of incoming source messages. However, if there is difficulty in the syntactic and semantic processing of the current source message, the speed of the target output will be affected and there will also be less capacity for the processing of incoming message (Moser, 1978). This implies that the attentional resources of interpreters are limited and each stage of the SI process affects one another.. 22.

(34) Figure 2.1 Moser’s Processing Model of Simultaneous Interpreting (Moser, 1978, p. 355). 23.

(35) 2.2.2 Gile’s Effort Models While Moser’s (1978) model was aimed to represent the comprehensive information processing of SI, Gile (1995, 2009) sought to provide a conceptual framework focusing on the dynamics and the finite nature of processing capacity in interpreting. Gile (1995, 2009) observed that even professional interpreters made errors during SI when there was no apparent source of difficulty in the input. Besides, when being asked to interpret the same source speech for the second time, professional interpreters made errors that did not appear for the first time (Gile, 1999). This suggests that there is “intrinsic difficulty” in the task of SI (Gile, 1997, p. 197). To account for the inherent cognitive demand of interpreting, Gile (1995, 2009) proposed the Effort Models of interpreting, which consists of three basic “efforts” – the listening and analysis effort (L), the production effort (P), and the memory effort (M). Each “effort” represents a non-automatic, resource-consuming operation in the interpreting process (Gile, 1997). In Gile’s Effort Models, SI is composed of these three efforts and a fourth one called the coordination effort (C), represented by this formula: SI = L + P + M + C. The L effort refers to all comprehension processes from auditory reception, word recognition to meaning construction of a source sentence. The P effort covers all speech production processes from generating a target message to be delivered, speech planning to the final implementation of the speech plan, including self-monitoring and self-correction. The M effort refers to memory operations that maintain necessary information from the interval of source speech comprehension to target speech production. Finally, the C effort represents the cognitive resources spent on coordinating the three basic efforts. Gile (1995, 2009) suggested that in most cases, the three basic efforts are simultaneously active, as previous literature indicated that interpreters engaged in concurrent listening and speaking for more than 60% of the time during SI (e.g., Gerver, 1976). According to Gile (1995, 2009), the total processing requirement is the sum of the 24.

(36) requirement of individual processing effort requirement, which changes on a moment-to-moment basis during SI. In order to successfully carry out the task of SI, not only the sum of total capacity needs to exceed the total processing requirement, but the capacity for individual effort must also be equal to or more than its processing requirement. When the capacity available for processing is less than the total processing requirement, interpreters would reach cognitive saturation and their performance would start to deteriorate. Even when there is no cognitive saturation, poor coordination of three basic efforts would also lead to problems and errors in interpreting performance. For instance, interpreters may devote too much attention to find a target equivalent for a certain idea (the P effort). Consequently, there is not enough capacity left for memorizing what has been delivered at the beginning (the M effort) and for the comprehension of incoming messages (the L effort). Gile (1995, 2009) proposed that there are certain problem triggers that can easily lead to cognitive saturation. The leading source of problems of interpreting is high density of the source speech, which is mostly caused by fast input rates and highly dense information content such as enumerations. When facing a highly dense source speech, interpreters need to process more information per unit of time, which could lead to the overload of both the L and P effort. Language-specific differences between the source language and the target language are another problem trigger. For instance, when the source language and the target language are syntactically different, interpreters may have to maintain a great amount of information in memory for a while and change the word order before being able to deliver the target segment. This process may saturate the memory capacity and lead to a loss of information. With an emphasis on the processing capacity management during the SI process, Gile’s Effort Models are useful in explaining the cognitive constraint imposed by variables such as the input rate and in explaining performance variability during SI.. 25.

(37) 2.2.3 Psycholinguistic model of the reformulation process in interpreting The models by Moser (1978) and Gile (1995, 2009) incorporated the comprehension and production processes in SI but did not elucidate how source language messages are recoded or reformulated into target language messages. It is believed that interpreters have two approaches of reformulation. One is meaning-based; the other is form-based (Dam, 2001). These two approaches are illustrated in the studies of bilingualism and psycholinguistics as the vertical route and the horizontal route, or as conceptually-mediated translation and transcoding (Christoffels & De Groot, 2005; De Groot, 2011) (See Figure 2.2).. Figure 2.2 Two Translation Routes (Christoffels & De Groot, 2005, p. 460). Through the vertical route (the gray arrows), the interpreter extracts meaning of the source speech up to the conceptual level, retains the meaning and then produces the output in the target language based on nonverbal concepts. Source language comprehension and target language production are no different from monolingual processes. In addition, the form of the 26.

(38) source language is lost. The horizontal route (the dark arrows), on the other hand, is based on the direct links between two languages at the phonological, morphological syntactic and lexical-semantic level. This route was supported by findings in bilingual studies, which indicated that the two language systems in bilingual memory are at least partly shared at the lexical, syntactic and conceptual level (Bernolet, Hartsuiker, & Pickering, 2013). Through the horizontal route, linguistics units of the target language could be replacing those of the source language at the same time. This suggests that reformulation occurs in parallel with source speech comprehension or even when the source message is partially comprehended. Several empirical studies have been conducted to investigate interpreters’ use of two translation routes or strategies. Isham (1994) compared the verbatim recall of source sentences after SI with the recall after listening and found that some interpreters retained more verbatim memory while others had a poorer recall. Isham thus proposed that the differences might stem from the two translation strategies. Those adopting the form-based strategy had better retention of verbal materials from the source speech than those opting for the meaning-based strategy. Recent studies on sentence-by-sentence interpreting using the paradigm of self-pace reading have provided empirical evidence for the form-based or the horizontal route by showing co-activation of the target linguistic units during source language comprehension (Dong & Lin, 2013; Ruiz, Paredes, Macizo, & Bajo, 2008). For example, Ruiz et al. (2008) compared the self-pace reading time of source texts under two conditions – reading for repeating and reading for interpreting – and found that only in the reading for interpreting condition, syntactically congruent structures between the source and the target language were read faster than incongruent structures. This shows that when the participants were asked to read for interpreting, there was parallel processing of target linguistic units during source text comprehension. Although these findings suggest that interpreters have access to both translation routes, it is believed that interpreting is mostly a meaning-based process through which interpreters 27.