order to get a better picture of sight translation, we need a tool with high temporal resolution to reflect on-line changes of cognitive operations and in the meantime allowing oral rendition during the experiment. Sight translation relies on reading to gain information, and therefore an eye tracker seems to be the best fit for the time being, since some other prominent research methods are not compatible with the requirements. For example, speaking can easily confound the data collected with ERPs, while PET is rather expensive and low in temporal resolution, though accurate in pinpointing locations (see Mitchell, 2004 for further detail).
The advantages provided by an eye tracker are numerous. First of all, it directly monitors the eye. Based on Just and Carpenter’s (1980) immediacy assumption and eye-mind hypothesis, when the eye fixates on a word, processing automatically begins,
and the aforementioned morphological, phonological, lexical information and even word meanings, grammatical features, contextual constraints will start to come into play until all processing required for comprehension is completed. In addition, the mind (attention) will direct the movement of the eye, and hence we can be quite assured that what the reader is reading is practically what he or she is (at least partly) dealing with. Although Inhoff and Radach (1998) warned the possibilities of
pre-processing of the word ensuing the fixated target word, and Radach and Kennedy (2004) also illustrated the workings of the spillover effect (the cognitive burden of the previous region carried over to the current target word), we can still be confident that monitoring the time spent on the fixated area offers useful information about the cognitive workings of the brain. For one, it’s hard to imagine any reader would choose to fixate on certain places with deliberate intention to suppress all related processing. For another, even with the pre-processing and spillover effect, a large part of the time spent still reflects the processing of the current target, which stays in the foveal vision that fits best for information extraction (see also Rayner, 2009).
With an eye tracker, we are able to document moment-to-moment changes of the eyes and see on-line variations of cognitive load/effort or processing strategies, and even identify difficulties with data about when and where eyes move. Where interpreters’ eyes are fixating, for how long, and how they move while producing output simultaneously are interesting enough, but equally fascinating is what they try to do when they are not speaking. With more detailed information on the whole process, we may be one big step closer to the nature of sight translation.
To understand reading, scholars in this field have designed different indices to map behaviors of the eye to different stages (first-pass vs. non-first-pass) and tried to decipher the meaning behind each phenomenon. Among all kinds of indices,
duration-related indicators are of major concern, while others, such as direction of movements and word skipping, serve as supplementary information, providing a more complete picture when taken into consideration all together. Below we introduce the main indices used in reading and sight translation research.
First-pass fixation measures mainly include first fixation duration and gaze duration. These are the fixations made after an eye first enters a specified region of interest and before leaving it, and, whether alone or aggregated together, are
considered as relatively immediate/early indices of cognitive processing when larger chunks are used as the unit for analyses (see for example Hyönä, Lorch & Rinck, 2003). Put separately, first fixation duration indicates “the duration of the first fixation on a target word during first pass sentence reading, irrespective of whether the target word receives one or several fixations” (Inhoff & Radach, 1998). When there is only one first-pass fixation on a target word, first fixation duration equates to gaze duration.
As the first contact with new information, first fixation duration represents adequately the pre-lexical or lexical processing of a certain word. Some scholars doubt if first fixation duration really reflects the sole influence of the currently fixated word. Their suspicion does have some merit, since we have seen empirical evidence for preview and spillover effects. Taking it into consideration, a word may possibly be
processed before even being fixated, and, moreover, we may see lagging effects from earlier text in disguise of first fixation duration of the current word (Rayner, Juhasz,
& Pollatsek, 2005; Tsai, Yen, & Wang, 2005).
Gaze duration is defined as “the sum of all fixations on the word prior to moving to another word” (Rayner & Liversedge, 2004). Similar to first fixation duration, gaze duration is sensitive to some pre-lexical, lexical, and post-lexical features of words and texts. Looking from a broader perspective, gaze duration still manifests relatively early stages of processing; but when taking words as AOIs, one may find some factors taking its toll on gaze duration but not on first fixation duration (“later” effects) and vice versa (“early” effects that appear immediately but do not last long). Gaze duration shows a linearity pattern of variation as a function of word length, since some re-fixations are required for longer words compared to short words. At this point, it’s worth mentioning one closely related index: re-fixation rate. Re-fixation rate means “the ratio of the frequency of multiple target fixations over the frequency of multiple and single target fixations during first pass reading” (Inhoff & Radach, 1998; see also Tsai et al., 2005). Re-fixation rate can easily manifest impacts of word frequency, word difficulty, word length, and initial landing location.
In translation and interpreting studies, gaze duration may be a great index
reflecting the efforts needed to process words (McDonald & Carpenter, 1981) or even to start reformulation (cf. Chen, 2013). However, the question is, the indicator alone cannot tell us what the interpreter is really doing. For example, in Chen’s (2013) experiments, early fixation time of experts was found to roughly equate to that of
novices, contrary to general assumption that experts process information faster.
Nonetheless, the quality was found to significantly supersede that of novices as a group. Chen inferred that the experts might have performed a greater deal of work within similar amount of time.
Mean fixation duration is probably one last important first-pass index. It is seen as the mean duration of all fixations that appear during first-pass reading. As
informative as other figures, mean fixation duration is susceptible to distortion. For example, fixations on a more difficult word may be more than those on an easy word, but with shorter duration for each fixation. In this way, the mean of multiple fixations has a chance to be less than single fixation duration on an easy word.
Early indices are affected by orthographic, phonological, morphological
properties, word frequencies, personal familiarity, and even contextual constraints, to name just a few (further detail see Rayner, 2009; Rayner et al., 2005; Tsai, Lee, Tzeng, Hung, & Yen, 2004). It’s wide consensus that first-pass measures mostly reflect the cognitive engineering of word identification and lexical processing.
All the measures that emerge after the eye first leaves a target region will be labeled as second-pass reading. A few fundamental indices used for later stages of processing include rereading time, go-past time, and total viewing time, plus an important and related probability index: regression rate. Rereading time, according to Hyönä et al.
(2003), is computed by adding all the subsequent fixations other than first-pass reading on a target word, representing later-stage processing after the word is identified.
Go-past time (in some studies also named regression path duration) is defined as
“the time from first entering a region until moving past that region forward in the text”
(Rayner & Liversedge, 2004). Some scholars use this index to capture the effects of some phenomena which are generally not easy to be identified through standard measures, including syntactic ambiguity, inconsistency, or the amount of processing load accumulated from earlier to the current region (see also Vonk & Cozijn, 2003).
Go-past time may be especially useful in studies of sight translation, since some structural disparity and the different habit of placing main or subordinate message in the front between two languages may force the interpreter to go back when s/he has to restructure the whole sentence or digest again the heavier-than-expected information that need to be knitted closely with the current content.
Regression rate, which stands for the percentage of readers moving their eyes back to previous regions, is tightly bound with go-past time and conveys similar messages. An issue with go-past time is that readers don’t necessarily go back for clues to solve problems at hand. They sometimes go further down, trying to gather more information to see if they can construct a more coherent structure of the content (see also Rayner & Liversedge, 2011).
Lastly, total viewing duration (or total viewing time) sums up “all fixations in a given region” (Rayner & Liversedge, 2004). Clearly as the name suggests, total viewing time represents the assumed total cognitive effort expended on a target word or region. In reading, larger amount of total viewing time either indicates more difficulty encountered or more important information embedded. Since sight
translation involves reading as a necessary part of the task, the same index has the same potential of identifying the culprit that prevents interpreters from completing their task with flying colors. Further, when combined with other indicators, we might even find whether interpreters and readers, with different goals in mind, mostly care about or are boggled down by similar constituents or not.
In order to outline the cognitive efforts in different stages during sight translation, two first-pass indices (first fixation duration and gaze duration) and three
non-first-pass indices (go-past time, rereading time, and total viewing time) were selected for this study. In this way, we may be able to see word retrieval in early stages, meaning integration and information searching (or problem solving) in later stages, and the total time spent on words that reflect efforts to achieve appropriate mental representation.