Eye-tracking method in T&I studies

Though eye tracker has been widely utilized in reading-related research, its employment in interpreting studies is only at the beginning. This is because the input and output of most interpretation modes (SI and CI) are both oral, thereby providing no practical way to employ eye tracking in related studies.

Sight translation, which uses written text as input, has been the only possible way to combine eye movement research with interpretation.

The very first research on eye movement and interpretation was conducted by McDonald and Carpenter (1981). Their experiment examined idiom interpretation, parsing strategy and error recovery in ST (called simultaneous translation in the paper). They proposed that idiom can be interpreted in two ways: under the Literal First Model, the reader retrieves the literal meaning of an idiom first. If the literal meaning does not match the prior context, the reader would subsequently retrieve the idiomatic meaning.

However, if reading under the Direct Access Model, the idiomatic meaning of an idiom should be retrievable by the reader directly without first comprehending the literal meaning. The experimenter therefore designed a set of two sentences that start with identical idiom (“kick the bucket” or “break the ice”) but end with different disambiguation segments, one priming the literal meaning of the idiom, the other priming the idiomatic meaning.

Four German-English bilingual interpreters participated in the experiment. They sight translated 44 paragraphs that contained the above mentioned sentences with either literal or idiomatic meaning. Their eye movements and oral production during the tasks were recorded. Chunking strategies employed in comprehending the idioms was examined through eye movement indices. A chunked unit was defined as the area from the first word

34

to the farthest word on the right that was fixated in one reading pass before the eye regressed. If the idiom is chunked as a whole, as in Mike hit the nail right on the head, it suggests the idiomatic meaning was retrieved. If the idiom is chunked at any syntactic boundaries within it, such as Mike hit the nail / right on the head, the literal meaning was retrieved.

The results of the experiment shed light not only on parsing strategies for idioms, but also the process of interpretation. Each phrases in the paragraph received at least two scans. The first scan (called first-pass reading), presented as a series of forward fixations, signified the initial comprehension of the phrase. The second scan (or second-pass reading) was marked by a regression of the eye to previously fixated words in the phrase, then followed by rereading of the phrase. The interpretation production also began during the second scan. Furthermore, if the interpreter realized the idiom was wrongly interpreted after the second scan, his eye would regress back to the idiom in question. During this third scan of the idiom, eye movements showed that the idiom would be chunked differently from before, suggesting the right meaning was being retrieved.

McDonald and Carpenter’s experiment supported the vertical perspective of interpretation, though it was not the original focus of their study. The first scan took the interpreters a total of 950ms, which is very similar to the speed of normal silent reading. The second scan, however, took a total of 3000ms, which was three times longer than the reading pass. This was due to the additional effort needed to produce the interpreted oral output. As for the error detecting / recovering third pass, the total time needed was 4801ms, even longer than the second pass. The long duration may be attributed to detection and recomputation process needed in error recovery.

35

Huang’s experiment (2011) took up what Carpenter and McDonald left, and used the eye tracker to examine the participant’s interpreting process during sight translation. 18 trainee interpreters were instructed to silently read, read aloud and sight translate three pieces of Chinese text comparable in difficulty and length, and their eye movements were recorded. It was assumed that since silent reading concerns only comprehension, reading aloud concerns comprehension and production, and sight translation concerns all three stages of comprehension, reformulation and production, comparing eye movements during these three tasks would provide insights into the interpreting process.

First-pass reading time and probability, go-past time and probability, total viewing time, rereading time and probability were examined for all three tasks. The results showed that for first-pass reading, there was no significant difference between silent reading and sight translation, indicating that during the initial stage of word recognition and comprehension, reading and reading for translation are not much different. However, for eye movement indices that indicate the later stage of reading, such as total viewing time and rereading time, interpreters spent much more time during sight translation than silent reading, which suggested that ST calls for more effort in the later stage of reading.

Huang’s research results supported the vertical aspect of interpretation.

Interpreters likely engage in normal reading comprehension during first-pass reading. The complicated task of reformulating, which calls for extra effort, seems to take place in later reading passes. This research also exemplified how various eye movement indices can be utilized in interpreting studies to reveal the complex interpreting process that goes on in the interpreter’s mind.

36

Eye tracking has been applied more to the studies of translation, as both input and output of translation are visual. These researches mostly focus on the demand of cognitive load during translation, with eye movement indices used as indicators of cognitive processing effort (Pavlovic & Jensen, 2009;

Sharmin, Špakov, Räihä, & Jakobsen, 2008). The following are a few examples.

Pavlovic and Jensen (2009) investigated how directionality of translation may affect the allocation of cognitive effort. Conventional assumption in the translation circle was that translating into the translator’s L2 is more demanding than into L1. Eight Danish-English participants, four translation students and four professionals, translated two articles comparable in readability, one from English to Danish, one from Danish to English. The subject’s total gaze time, average fixation duration, pupil dilation and total task length during both tasks were examined.

In both direction of translation, participants spent more time fixating on the target screen. Average fixation duration on the target screen was longer, and their pupil also had higher percentage of change in this area. The results indicated that the translator distributed more attention to the process of producing the target text and revising, and required more cognitive effort in doing so. However, the directionality of translation did not seem to cause significant differences in eye movement indices, which challenges past belief that translating into L2 is more demanding than into L1.

Sharmin, Spakov, Raiha and Jacobsen (2008) examined eye movement indices to explore the changes in cognitive processing load under the influences of time constrain and text complexity. Though complexity of the

37

source text has been studied in both T&I researches, time constrain has mostly be examined only in studies of SI and not in translation. Three English texts with different degrees of complexity were used as experiment material.

Eighteen trainee translators were asked to translate the three texts into Finnish in six, five or four minutes. The translators’ average fixation duration, average fixation count and viewing time were examined.

The findings showed that when asked to translate within restricted time, the overall viewing time per minute did not differ between different time restrain conditions. However, the average fixation duration on the source screen decreased significantly when the translator was under more time constrain. This indicates that it is easier for translator to adjust the time and effort needed for reading comprehension than that for typing and monitoring the target text. This may be due to the fact that reading speed is easier to adjust than typing speed.

Overall viewing time per minute and average fixation duration became longer as the text complexity increased, supporting the claim that lexical and syntactic complex source text would require extra cognitive effort in reading comprehension. When the source screen and target screen were separately examined, it was shown that average fixation duration was longer on the target screen for all three texts, regardless of time constrain. This was consistent with Pavlovic and Jensen’s finding (2009) that target text processing in translation requires more cognitive effort than source text processing.