The influence of training on output and reading behavior

Knowing that reading purposes do change reading behavior, we now turn our focus to how training exerts its impacts. Table 31 compiles results entailing all tasks and simplifies the raw data, presenting them in logical relationships instead.

Discussion in this section centers on the influence of training, and hence the attention will be given mostly to interpreting students and bilinguals.

Table 31 Between-group comparisons across tasks⁶ (reference table)

Silent reading Reading aloud Sight translation

FFD PRO=BIL=NEW BIL=PRO=NEW PRO=NEW=BIL

GD PRO<BIL=NEW PRO=BIL=NEW PRO=NEW=BIL

GPT PRO<NEW=BIL PRO=NEW=BIL PRO=NEW=BIL*

RRT PRO<BIL=NEW PRO=BIL=NEW PRO=NEW=BIL*

TVT PRO<NEW

BIL=NEW PRO=BIL

PRO=NEW=BIL NEW=PRO=BIL*

Rereading count PRO=BIL=NEW PRO=NEW=BIL PRO=NEW=BIL No. of fixations PRO=NEW

NEW=BIL

*PRO = Experienced interpreters; NEW = Interpreting students; BIL = Untrained bilinguals

*FFD = First fixation duration; GD = Gaze duration; GPT = Go-past time; RRT = Rereading time; TVT = Total viewing time

Globally, interpreting students and bilinguals had similar number of fixations and total time during silent reading and reading aloud. Furthermore, all the reading indices (FFD, GD, GPT, RRT, and TVT) turned out to be similar, and trainees in fact spent numerically longer time on all five reading indices.

The aforesaid findings can prove that the two groups are identical in their general linguistic competence, and the difference between two groups lies only in their method of tackling texts: Trainees spent longer time locally, which were then compensated with shorter task time and fewer fixations globally, while bilinguals had

6Numerically speaking, the task placed on the left of any logical symbol is always lower than the one on the right. In addition, statistical significance is represented with “>” or “<”, and “=” stands for non-significance.

a reverse pattern. Only the participants that passed the threshold of comprehension test were selected was another fail-safe mechanism to ensure same level of

understanding of the texts. Hence, we can say with some confidence that significant differences in indices for sight translation may come from training. As was expected, global indices such as total number of fixations and total time in sight translation showed that interpreting students were more efficient when tackling the task of sight translation, and the accuracy and fluency ratings of the output were also significantly better than bilinguals, further attesting to the effect of training (see Table 32 below).

What led to a shorter total time and fewer total fixations? Is it possible that a more concise style of rendition was responsible? It turns out that this explanation doesn’t add up because total word counts of participants’ output show that there was no significant difference across groups. What about the efficiency of information extraction? Did reading indices for sight translation exhibit a different pattern from the other two tasks? As a matter of fact, this study found that all reading indices, including FFD, GD, GPT, RRT, and TVT all failed to show significant differences between interpreting students and bilinguals. Therefore, efficiency of meaning extraction seems to be similar between the two groups. What stands out here is the different patterns in sight translation. Trainees consistently had non-significant, shorter local indices and even significantly shorter global indices. It’s obviously the effect of training that gave trainees an advantage, securing an overall win over bilinguals in speed.

Another interesting fact is that, while first-pass indices remained close for

trainees and bilinguals, the gap gradually widened along different stages of reading, at last reaching the pinnacle in TVT. During training, interpreting students time and again practiced the skill of chunking to “translate along the text” so as to shorten the time needed to go back to previous regions. Thus, much shorter later-stage indices may hint at the possible influence of training because trainees learned to try to engage in reformulation as soon as possible. This strategy can also adequately explain why trainees had fewer fixations in total.

Aside from fewer fixations, we may turn to other reading behavior in the process of sight translation for possible contributors to a less total time. Indices specifically targeting sight translation are listed in the following Table 32.

Table 32 Sight translation indices and reading behavior across groups⁷ (reference table)

Sight translation indices & Reading behavior in the process

Quality BIL<NEW<PRO

Quality: accuracy BIL<NEW<PRO

Quality: fluency BIL<NEW=PRO

Word count NEW=PRO=BIL

Reading-ahead span (1st utterance) NEW=PRO<BIL

Initial silence NEW=PRO<BIL

Reading-ahead span (non-1st utterance) PRO<NEW<BIL

Percentage of pause PRO<NEW<BIL

Pause length PRO<NEW<BIL

No. of fixations (by pause) PRO<NEW<BIL

Verbal gap duration PRO<NEW<BIL

*PRO = Experienced interpreters; NEW = Interpreting students; BIL = Untrained bilinguals

It was found that the initial silent period and number of fixations before participants began interpreting in each trial were both significantly less for interpreting students. On top of that, the distribution was incredibly dense for the group, meaning that almost all trainees exhibited the same tendency to avoid keeping the audience waiting. Bilinguals, though might also be aware that there was audience waiting, clearly chose an opposite strategy: Most read through the trial before opening their mouth. This might be that, without the knowledge of chunking, bilinguals had good enough linguistic capability to know that they would be better off getting all the information in advance.

7 Numerically speaking, the task placed on the left of any logical symbol is always lower than the one on the right. In addition, statistical significance is represented with “>” or “<”, and “=” stands for non-significance.

Once participants began sight-translating the text, training again showed its impact. Trainees on average read significantly less than bilinguals during each non-first utterance. This finding is reasonably expected, since the strategy of

“translating along the text” gave trainees the ability to translate with minimal

meaningful units at hand, thus largely reducing the need to read too far ahead before coming back to the same text again for reformulation. As we have speculated, some reformulation and production might have been completed in the first pass of reading because rereading counts in sight translation was similar to the number in silent reading for trainees. Thus, by far our findings seem to corroborate McDonald and Carpenter’s (1981) three-stage model, as did Su (2013), but with some exceptions.

Of all the verbal gaps in each trial, 13.9% were sensed as pauses for interpreting students, significantly less than bilinguals’ 19.44%. In addition, average pause length for trainees was significantly shorter. Cross-referencing the average number of fixations each group had during each pause, we found that trainees differed

significantly from bilinguals in 25% of the pauses because both groups had only one fixation in 75% of the pauses. Taking the average pause length and number of fixations together, trainees generally reacted more quickly and required fewer

fixations to help them continue. Still, it seems that the trainees were still getting used to the skill of chunking, and therefore had to slow down sometimes to combat

problems encountered, as the bilinguals did. This extrapolation can at least be partly substantiated by the max values for the two groups on average pause length and the number of fixations in each pause.

Lastly, working pace of sight translation reflected through the average duration of verbal gaps showed that interpreting students was significantly quicker than bilinguals. Interestingly, most bilinguals read through each trial before they began interpreting, while trainees mostly started with just a few peeks, but along the way trainees could still resume interpreting quicker than untrained bilinguals after they paused, which is another clear evidence of the impact of training.

From the statistics, we saw that the reading pattern and the time saved along the process helped lead to fewer fixations and less task time in total, and fewer fixations here also means fewer reading passes, since average duration of each reading index was statistically similar. What is really intriguing is that bilinguals and trainees might have different mechanisms. For most of the bilinguals, first-pass reading was much more about comprehension, as they tended to read through the text first; on the contrary, the same pass of reading might have entailed both comprehension and reformulation (or at least part of it) for trainees.