Focused effort on sight translation

As noted by former studies such as Agrifoglio (2004) and Chang (2008), unnatural expressions in the target language and terrible quality of outputs are not necessarily unusual, if not highly frequent, when performing sight translation.

However, for a really long time, most of the relevant studies have been based on product analysis or quality rating. As a consequence, we know “what” the end results are, but little is revealed about “how” sight translation is done, “where” interpreters encounter problems, or even how they “read” in the process. Do interpreters read texts the way similar to reading for comprehension? Or do they somehow develop a unique approach specific to sight translation? How do they juggle between reading and reformulation? And how does training and experience affect their eye behavior?

In order to answer these questions, scholars start to apply research tools and methodology typically used in other fields (mainly cognitive psychology) to study sight translation. For example, two translation experts and two amateur

German-English bilinguals were recruited to participate in McDonald and Carpenter’s (1981) study in order to find out how idioms are parsed during sight translation. A total of 12 idioms were selected and each put in a context of 85 words. The context consisted of a priming cue suggesting whether the idiom should be interpreted in the literal or idiomatic sense, and the priming cue might or might not be consistent with a disambiguating sentence following the idiom, which was the Area of Interest (AOI) in this study. The paragraphs with the literal meaning of the AOI intended were used as the baseline, and the data of fixations and eye movement patterns were collected to indicate how people parsed sentences into meaningful units and sight-translated them.

All in all, participants first read the text until a meaningful unit was found, followed by a regression to reread and orally translate the unit at the same time during the second pass. The participants then continued reading and refixated the AOI or any troublesome unit during the third (error recovery) pass if any discrepancies in meaning were found.

This three-stage model was on the whole verified by the eye movement patterns of participants with a high level of predicting accuracy. In addition, the average duration of fixations denoted that the reading behavior in the first pass resembled that of normal silent reading. The average translation rates were 64 and 62 words per minute for the experts and the amateurs, respectively, reflecting around two to three

times of the duration spent on reading. Finally, it seemed that expertise did not affect the translation speed, but the quality of the product. As enlightening as this study may be, there are some obvious deficits that deserve our attention. One is that participants are too few to confidently generalize the study’s findings; another involves the statement by the authors: “Individual differences in expertise are reflected in the quality rather than the duration of the translation process” (McDonald & Carpenter, 1981, p. 246) Since the whole paper dealt only with the process of sight translation instead of product itself, it is too big a claim to assert the difference lies solely in quality.

Macizo and Bajo (2004) designed two experiments to examine whether the comprehension process of reading for repetition is identical to that of reading for translation, and direction of translation was also included to identify any possible facilitation effect. As mentioned before, in each experiment, eight professional translators who did not participate in the other experiment were recruited to either read to comprehend or read to translate 96 sentences. The sentence structure consisted of a relative clause and a main clause. In general, the duration of reading for

translation was significantly longer than that of reading for repetition. In terms of translation from L2 into L1, this effect was only observable in the ending area of the relative clause. In addition, when reading L1 texts, participants tended to utilize pragmatic cues to help solve the ambiguities of assigning roles to each noun in the stimuli. Hence, comprehension was facilitated when the pragmatic cue was attached to the verb in the main clause. On the other hand, when reading L2 text, the pragmatic

cue did not seem to be at work. But most importantly, comprehension was not negatively affected in any condition.

Macizo and Bajo (2004) spoke in favor of the horizontal view of sight translation, suggesting that, due to the prolonged reading time manifested in reading for

translation, there must have been some processes, such as reformulation, working at the same time other than comprehending the message. One thing that needs to be kept in mind is that this study presented sentences word by word on the computer screen, which is a far cry from the authentic situation when a reader or translator has the freedom to go over the sentence as many times as desired without fearing that words will disappear and in turn has to try harder than usual to memorize everything. Further, the division of AOIs seems unequal. Hence, using each AOI as a basic unit for

comparisons runs the risk of comparing data on different bases. Lastly, the declaration of a horizontal view is too bold a claim to make, since the definition for a

horizontal/vertical process is so vague that it applies equally to a single word and a whole sentence. Owing to the fact that only one global index (reading time) was used, we have no way of confirming what really happened during the process of sight translation.

A few other studies were concerned with the cognitive load one might bear during sight translation. Alves, Pagano, and da Silva (2011) invited six translation students and six professional translators on board for an experiment investigating whether reading for different purposes impose different levels of cognitive load, reflected in the indices of reading time, the number of fixations, and fixation duration.

The experiment was divided into three modes, including reading for 1)

comprehension, 2) giving summary, and 3) sight translation. All would appear in two conditions, the first of which consisted of three texts with divergent rhetorical

structures on the same piece of news (193 words on average), and the second was comprised of three texts on different topics with the same rhetorical structure (105 words on average). The results of condition one showed that the total reading time for summary was the shortest, followed by reading for comprehension, and sight

translation was the most time-consuming task. In terms of the number of fixations, three modes were roughly the same, while that of summary was significantly fewer than that of sight translation. As for average fixation duration, only sight translation showed significantly longer duration than the summary mode. Switching to condition two, the reading time of the task for comprehension was the shortest, followed by summary and then sight translation. When it comes to the number of fixations, only reading for summary showed significantly more fixations than that of sight translation;

conversely, the average fixation duration of reading for summary was significantly shorter than that of sight translation.

Generally, the two conditions manifest the same trend. The results indicate that sight translation may carry a heavier cognitive load, reflected in significantly longer fixation duration on average and longer task time in total. Nevertheless, the texts chosen varied widely in length across the two conditions, not to mention the genre and the content, and it is risky to directly compare the figures between conditions. In addition, even though the rhetorical structures of three texts in condition one were

manipulated, learning effects might still come into play because they were all about the same news; sticking each text with a certain task throughout the whole study and simply altering the task sequence also threatened the validity and reliability of the data. Last but not least, with the already few number of participants, deleting data from two of them might indeed largely twisted the trend emerging from the results.

Shreve et al. (2010) aimed to verify if sight translation is more sensitive to visual interference than written translation through manipulation of syntactic properties.

Whether reading for sight translation takes up more cognitive resources than reading for comprehension was also examined. Two Spanish texts (A and B, both around 167 words) were used for sight translation, each consisting of a syntactically complex paragraph and a syntactically non-complex paragraph. Participants’ eye movement data were compared with those from a bilingual reading task. In general, the authors found that there were more and longer fixations and more regressions during sight translation than in bilingual reading. Unfortunately, no inferential analysis was performed, so we don’t know if the two types of tasks are fundamentally different, except that the number of regressions in sight translation was almost four times more than that in bilingual reading.

Contrasting reading measures of sight translation from syntactically complex and those from non-complex paragraphs, the authors concluded that text A showed

expected syntactic effects. That is, in text A, the complex paragraph led to greater disruption than the non-complex paragraph. However, statistical analysis proved otherwise. None of the eye movement figures showed statistical difference between

paragraphs with different syntactic manipulation, as did pupil size. Furthermore, for text B, in which the first paragraph contained syntactically complex sentences, eye movement data showed that processing of the easier paragraph was even more effortful. Moving onto the written translation task, the procedure was identical, and the task followed sight translation for every participant. This time, all eye movement data again showed no significant difference between conditions.

The results provided by this study suggest that, firstly, reading for sight translation is not the same as bilingual reading; secondly, the amount of cognitive resource used does not vary as a function of syntactic complexity, and cognitive load does not differ between sight translation and written translation. However, a few factors may have led the authors to this precarious interpretation. To begin with, comprehension level was not tested, and it makes little sense to analyze one’s data if s/he can’t even understand the content. Secondly, the text used in another bilingual reading task was of 104 words, largely different from the texts of 167 words for sight translation. Adjusting the data based on the number of words as the authors did carries the assumption that all the data may only increase or decrease proportionately. Further, although the authors contrasted the first paragraph of text A (non-complex) and that of text B (complex) and found significant difference, between-text difficulty was not controlled, so that reliable gap might have come from the disparity between texts, not syntactic complexity. Needless to say, the fixed sequence of text A preceding text B may unfaithfully tilt the results. Above all, syntactic complexity was not operationally defined, leaving all interpretation on a shaky basis.

Trying to understand sight translation more from another aspect, Chmiel and Mazur (2013) aimed at examining the influence of training on cognitive load manifested in sight translation by interpreting trainees. Eighteen participants were recruited, among which ten had received one year of interpreting training and eight for two years. Variables manipulated in the study included sentence structure (simple SVO and complex non-SVO) and frequency, for which a few low-frequency words were selected. It was assumed that participants with more training might endure less cognitive load while performing sight translation, reflected through less fixation counts, shorter fixation duration, and less total time on the whole, and also less in the number and duration of fixations when processing low-frequency words. The results indicated that one year of training might not be sufficient enough to reflect a

significant difference between two groups of participants on all the indicators adopted by this study, including fixation counts, fixation duration, and total time.

Interestingly, in view of the reading pattern, it seems that simple sentences exerted more cognitive load than complex sentences, with more and longer fixations for the former. Since this finding was in stark contrast with previous studies and

intuitive expectation, the authors replaced complexity with readability as the indicator, defined in terms of number of clauses per sentence and word length. This time, more readable sentences (complex non-SVO sentences in this research) showed less counts and length in fixation, suggesting that readability might be a more adequate index for examining cognitive load. The authors speculated that the non-significant findings between two groups of participants might come from the fact that less experienced

participants were indeed more competent, proven by better scores on their final exam.

However, the final exam of a specific course does not necessarily provide adequate evaluation of individuals’ competence, since each exam may be based on different purposes and standards. This again tells us the importance of clear standards for participant selection. Besides, this study focuses itself only on the perception stage, and therefore it is plausible that the real difference might become obvious once we take final product into account.

Still another study looked into an issue that is particularly related to the industry.

Dragsted and Hansen (2009) put sight translation under test to see if it could boost translators’ working speed and maintain acceptable quality at the same time. The authors recruited four interpreters and four translators with at least ten years of experience. The test material amounted to 1289 words, which is probably too long and is likely to lead quickly to fatigue when coupled with an eye tracker, especially for a translation assignment. As indicated by the results, the quality of the

sight-translated output was not unanimously worse than written translation. Also worthy of consideration is the fact that the reading patterns were quite dissimilar between sight translation and written translation. Focusing on sight translation performed by interpreters and translators, one may easily notice the different patterns shown by the two groups; while interpreters generally proceeded with smoother patterns, eye movements of translators were more hesitant and quite often disrupted.

Quality comparison between the two groups showed immense deviations, which even held true among within-group individuals. Yet, a lack of statistical analysis strips us off the ability to conclude how each group did or if any reliable difference exists.

Sight translation as a topic has already been explored with various kinds of research tools, though in total this area still constitutes only the tip of the iceberg in interpreting studies. While most studies have been trying to infer what happens during the process through analyzing oral output, some non-traditional studies have taken a different path. This is worth celebrating because understanding how sight translation is completed is as crucial as what the final product looks like, if not more.

Notwithstanding, a few reminders are in order. To begin with, many of the studies did not control variances inherent in the materials, and sometimes the sequence for

stimuli presentation was not balanced, either, leading to a high possibility of interference from learning effects and biased results. Further, the number of

participants or the amount of data was severely insufficient for most papers to observe certain trends or even make inferences, and criteria for selecting participants (e.g.

language competence) were seldom established, so from time to time within-group variance was even more overwhelming than inter-group difference.

A related issue is that some studies recruited translators to perform sight translation, a task that is also practiced by some translators but almost never under time constraint or without a chance to modify one’s own output. In this way, the observed phenomena might deviate from those of real prospective sight translation practitioners. Thirdly, descriptive statistics was still the norm, and thus we as readers could hardly tell whether the disparities between various tasks show up by chance or indeed emerge as a function of the nature of those tasks. One last observation is that, as mentioned previously, product-oriented studies can only speculate the reasons

behind certain phenomena based on subjective judgment. Recently, more research methods and tools from other disciplines have joined in this line of research on sight translation, and we started to see fruitful results coming from advanced tools such as eye trackers.