【作者】 何雯婷；

【导师】 柴明熲；

【作者基本信息】 上海外国语大学 ， 翻译学， 2014， 博士

【副题名】一项基于自动化机制理论的翻译过程实证研究

【摘要】 根据人们的常识判断，经验丰富的专家往往能快速、高效地完成任务。在翻译研究领域，有一些学者观察到翻译经验丰富的译者翻译速度更快。但另一方面，也有学者提出存在“翻译不是越做越轻松”的现象，翻译经验丰富的译者不一定速度更快，甚至有时候会比新手更慢。关于译者经验与翻译速度之间关系的研究存在着相互冲突的观察结果，需要进一步进行探讨和厘清。完成一项翻译任务的速度只是一种可观察的外部现象，其变化的主要原因是译者的认知负荷水平。从认知心理学的信息加工理论框架来看，翻译作为一种复杂的涉及语言的认知心理加工活动，在整体上是一种认知负荷水平较高的控制性加工过程，速度相对较慢。但不能忽视的一点是，大量的翻译训练在一定程度上有利于提高认知资源的使用效率，使某些加工环节成为自动性加工，从而降低译者的认知负荷，提高翻译速度。可见，译者的翻译速度与其自动化水平有很大关系，然而鲜有学者关注翻译过程中的自动性加工问题。根据认知心理学的自动化机制理论以及德格鲁特、哥普费利希和普拉苏尔等人关于翻译过程加工方式的论述，本研究推断翻译过程中至少存在两种加工形式，一种加工形式是直接提取长时记忆中的表征或进行直接的模式配对，即前例加工。另一种加工形式则是因应具体情景的各种因素进行重组、推理、转换、评估等算法步骤，即运算加工。根据自动化机制理论，促进译者自动化水平提高的主要机制有两种，一是运算加工自身的运算速度加快；二是运算加工转变成前例加工。按照任务内容的一致性和稳定性可将翻译任务分为结构清晰的翻译任务和结构不清晰的翻译任务。由于结构清晰的翻译任务的内容较为重复、稳定、一致，因此译者在每一次执行此类翻译任务后都会留下记忆痕迹或前例表征。随着训练次数的增加，储存在长时记忆中的前例数量增加，从而构建成一个与具体任务紧密相关的知识库，促使运算加工转变成前例加工，从而降低译者的认知负荷，提高效率。而在结构不清晰的翻译任务中，由于内容较为零散和多样化，译者需要不断对一系列新的、未经学习过的行为和规划进行组装，难以构建成一个稳定、可靠的知识库，运算加工无法大量转变成前例加工，而且运算加工自身的速度也难以有大幅提升。因此，译者即使经验丰富，在完成结构不清晰的翻译任务时认知负荷可能仍然较高，翻译速度无法大幅提高。由此，本文提出以下假设：假设1：翻译任务类型对译者经验与翻译速度之间的关系产生显著的调节效应。假设2：若假设1成立，在结构清晰的翻译任务中，经验丰富的译者速度更快的主要原因是部分运算加工转变成前例加工。为了对上述假设进行验证，本研究进行了一项实证研究，共邀请11名有着不同翻译经验的受试进行两种不同类型的翻译任务。利用回溯性有声思维法、敲键行为记录法、屏幕录制法和问卷等手段对受试的翻译过程进行了全面观察。通过R软件变量统计分析、Inputlog程序数据对比分析以及受试有声思维报告数据分析，对以上假设进行验证。验证结果为：假设1：支持。翻译任务类型对译者经验与翻译速度之间的关系产生显著的调节效应。在结构清晰的翻译任务中，译者经验与翻译速度呈显著的正相关关系。在结构不清晰的翻译任务中，译者经验与翻译速度之间的关系不显著。假设2：支持。在结构清晰的翻译任务中，前例加工比例与翻译时长呈显著的负相关关系以及显著的因果关系。更多还原

【Abstract】 It is generally believed that the ability to finish a specific task in a quick and efficientmanner is a natural result of extensive experience. According to some empirical translationstudies, experienced translators work faster than novice translators. But other researchesobserve that experienced translators do not necessarily perform faster and a―translation-does-not-get-easier‖phenomenon seems to exist. In sum, the results of earlystudies on the correlation between translator experience and translation speed show ahighly incoherent picture, which call for a further exploration.Translation speed is an external indicator of translators‘mental mechanism and thelevel of cognitive load. Viewed from a cognitive information processing perspective, thetranslation process, as a complex mental activity highly dependent on the use of language,is mainly composed of controlled processing which is slower and requires a relatively largeamount of cognitive capacity. But it should be noted that a large amount of practice, tosome extent, is conducive to enhancing the efficiency of cognitive resources allocation.With an adequate amount of practice, some processing might becomes automatic, which isless cognitively demanding and much faster. Thus it can be inferred that translation speedis related to translators‘degree of automaticity, but so far few researches have beendevoted to this topic.Based on the automaticity mechanism theories derived from cognitive science as wellas the translation processing types proposed respectively by Anette de Groot, SusanneG pferich and Friederike Prassl, this dissertation argues that there exist two types oftranslation processing, i.e. instance-based processing which involves direct retrieval ofrepresentation from long-term memory and direct pattern matching, and algorithm-basedprocessing which involves taking algorithm steps according to the specific context,including restructuring, reasoning, transfer and evaluation. There are mainly twocompeting mechanisms to explain the shift from controlled processing to automaticprocessing: algorithm-strengthening theories and instance-based theories. The formerargues that the enhancement of automaticity is attributed to faster algorithm-based processing, while the latter advocates the shift from algorithm-based processing toinstance-based processing.Judged from the consistency and stability of content, translation tasks can be dividedinto two types: well-structured translation tasks and ill-structured translation tasks. Eachexperience with well-structured translation tasks leaves a memory trace or instancerepresentation that can be retrieved when the task repeat itself. The number of instancesstored in long-term memory grows with the number of practice trials, building up atask-relevant knowledge base, and some algorithm-based processing will shift toinstance-based processing. In this way, translators can benefit from extensive practice, andtheir translation processing become less cognitively demanding and more efficient. Inill-structured tasks which contains diversified and fragmented data and information,translators constantly need to assemble new and unlearned sequences of behavior andplanning. Therefore it is hard to build up a reliable knowledge base and make a shift toinstance-based processing. Also, algorithm-based processing profit only little from practice.As a result, experienced translators might be still subject to a large cognitive load and notnecessarily work faster when handling ill-structured tasks.Taking the above arguments into consideration, this dissertation proposes twohypotheses:Hypothesis1: Translation task types have a significant moderating effect on thecorrelation between translator experience and translation speed.Hypothesis2: Given that Hypothesis1is supported, the enhancement of translationspeed of experienced translator when handling well-structured tasks is mainly attributed toa shift from algorithm-based processing to instance-based processing.An empirical process-oriented study is conducted to test these hypotheses.11subjectswith different level of experience are invited to fulfill two types of translation tasks.Multiple research tools including retrospective think-aloud protocol, key-stroke logging,screen recording and questionnaire are used to observe the subjects‘translation process.R-project and Inpulog software are used to analyze the collected data, and subjects‘retrospective think-aloud protocol are also analyzed. The results of hypothesis test are as below:Hypothesis1: Supported. Translation task types have a significant moderating effecton the correlation between translator experience and translation speed. In well-structuredtranslation tasks, there is significant positive correlation between translator experience andtranslation speed. In ill-structured translation tasks, there is no significant correlationbetween translator experience and translation speed.Hypothesis2: Supported. In well-structured translation tasks, the proportion ofinstance-based processing has a significant negative correlation with total translation time.更多还原