The effects of input rate on fluency

Input rate not only affects interpreters’ accuracy but also has an impact on how fluently an interpreter delivers his or her output. Scholars pointed out that fluency is a hard-to-defined concept and taxonomies for fluency analysis tend to vary from author to author (Mead, 2000;

Pradas Macías, 2006; Rennert, 2010). Among the few studies that explored the effects of input rate on interpreters’ fluency, different parameters of fluency have been employed, but many studied focused on the phenomenon of pauses, which can be divided into two types:

unfilled pauses (or silent pauses) and filled pauses (such as uhm) (Ahrens, 2005). The

category of unfilled pauses can be further divided into subcategories according to its function (communicative or non-communicative) and position (at syntactic junctures or at random places) (Cecot, 2001). For unfilled pauses, the minimum threshold of pause duration for speech analysis tended to differ across studies. In addition, analysis can be done on the frequency of pauses and/or the duration of pauses. All these discrepancies in fluency analysis can contribute to the differences in findings among studies.

Gerver (1969) observed that as the input rate increased from 112 wpm to 142 wpm, the mean unfilled pause time went up from 0.7 seconds to 1.1 seconds and the pause to speech ratio also went up. He suggested that when facing fast input rates, professional interpreters tended to pause longer. Pio (2003) also found that the number of unfilled pauses longer than three seconds and filled pauses in student interpreters’ output increased when the input rate

39

went up to 140 wpm. However, other studies found otherwise. Cecot (2001) examined the occurrence of unfilled pauses and disfluencies in the source and target texts in relation to the input rate. Eleven professional interpreters were asked to interpret two source speeches from English into Italian at the rate of 263.3 spm and 204 spm, respectively. For the analysis of fluency, a detailed coding taxonomy was employed. There are three subcategories of unfilled pauses:

1. segmentation: grammatical pauses with the communicative function of segmenting concepts;

2. rhetorical pauses: pauses occur at grammatical or non-grammatical junctures with a communicative function;

3. hesitation pauses: non-grammatical pauses that do not have a communicative function.

The minimum threshold of unfilled pauses was set at 250 milliseconds, conforming to the limit proposed by Goldman-Eisler (1968). Disfluencies include filled pauses, repeats, restructuring, false starts as well as vowel and consonant lengthening. Cecot (2001) also administered a questionnaire after the experiment to investigate interpreters’ perception of pauses in the source and the target speeches. By comparing the pauses occurred in the source speeches and target texts, Cecot found that interpreters tended to “follow the speaker’ patterns”

for segmentation and rhetorical pauses in both source speeches (p. 76). As for the effects of input rate, more hesitation pauses and more disfluencies were observed at the slow input rate, contrary to the findings of Gerver (1969) and Pio (2003). Interestingly, according to the questionnaire, 81.8% of the participants reported that they believed they had more hesitation pauses in their own output of the first speech because it was faster. This showed that

interpreters’ subjective assessment of the input rate was correct but their awareness of pauses in their own output was inaccurate.

Another descriptive study on silent pauses and disfluencies in SI was conducted by Tissi (2000). Ten students were asked to interpret two source speeches from German into Italian.

40

The input rates were 267 spm and 208 spm respectively. The analysis scheme was different from that of Cecot (2001). Silent pauses were divided into two kinds: grammatical and/or communicative pauses and non-grammatical pauses. The minimum limit of silent pauses was also set at 250 milliseconds. Filled pauses were categorized into vocalized hesitations as well as vowel and consonant lengthening. Repeats, restructuring, and false starts were placed under the name of “interruptions” (p. 113). The source and target speeches were compared according to the occurrences of silent pauses, filled pauses and interruptions. It was found that there were more silent pauses when student interpreters were interpreting the slower source speech, corroborating the findings of Cecot (2001). As for filled pauses and

interruptions, only the ranges in each parameter were provided. Tissi (2001) explained that since there were great individual variations in the occurrence of these categories, it was meaningless to calculate the means. A general trend was observed that student interpreters tended to produce more filled pauses in the slow speech than in the fast speech.

The study by Cecot (2001), Pio (2003) and Tissi (2000) were descriptive in nature.

Given the differences in participants, language combination and taxonomy of fluency analysis, it is difficult to make comparisons and this probably explains why there was inconsistency in their findings. The influence of input rate on fluency was examined in two other studies that applied inferential statistics. Piccaluga, Nespoulous, Harmegnies, &

Mons-hainaut (2005) proposed that silent pauses provide a window to the cognitive load undertaken by simultaneous interpreters and these pauses may be associated with complex processes in interpreting, such as difficulties in comprehension, in finding a translation

equivalent, or in expressing ideas in the target language. In their experiment, two professional interpreters, one student and one bilingual with no interpreting experience were invited to interpret three speeches from French to Spanish and three from vice versa. The three levels of input rate and noise were manipulated in certain portions of each speech. For the input rate, compression of 80%, 70% and 60% of the total speech time was introduced to increase the

41

speed. The minimum threshold of silent pauses in the analysis was set at 200 milliseconds.

Variance analysis with four independent variables (temporal compression, noise, translation direction, and subject) and two dependent variables (the number of pauses and the duration of pauses) was conducted. No significant effects were found for temporal compression on either the number or the duration of pauses. However, for the duration of pauses, a significant interaction was found between temporal compression and translation direction. When the participants were interpreting from French (L2) into Spanish (L1), silent pauses lengthened with each increase of input rate. However, in the direction of Spanish into French, the duration of silent pauses shortened with each increase of input rate. This suggests that the effects of input rate on silent pauses may be moderated by direction of translation.

Plevoets & Defrancq (2016) conducted a corpus-based analysis on the effects of

information load on filled pauses in SI. Information load of the source speeches was analyzed based on the delivery rate, lexical density, the proportion of numbers and sentence length.

One hundred and seven source speeches in French from the European Parliament and corresponding target speeches of SI in Dutch were collected. The mean delivery rate of the source speeches in the corpus is 157.8 wpm but the range was not provided. Results of regression analysis showed that there was a positive trend between the delivery rate and the number of filled pauses. In other words, interpreters produced more filled pauses when the input rate increased. This result coincides with the findings of Pio (2003), but not with Cecot (2001).

Table 2.3 presents a summary of the studies reviewed in this section. It is obvious that the findings were rather inconsistent across studies. Besides the input rate, one possible factor that could affect the occurrence of silent pauses in interpreting output is the number of pauses in the source speech. In the studies of Cecot (2001) and Tissi (2000), the slow source speech contained a lot more silent pauses than the fast source speech. This is probably why both studies found more silent pauses in the slow speech. In Pio (2003)’s study, the differences in

42

the number of long pauses between the slow and fast speeches (14 vs. 2) were not as big. In addition, the minimum threshold of unfilled pauses duration for fluency analysis may also play a role in the differences observed in these studies. Pio (2003) examined silent pauses longer than three seconds instead of using the common limit of 200-250 milliseconds.

According to Pio (2003), 85.5% of these long pauses in the output were related to omissions.

When the input rate increased, there were more omissions along with long pauses in interpreters’ output. As for filled pauses and other disfluencies, results were unclear due to high individual variation in the descriptive data. More studies are certainly needed to clarify the effects of input rate on different parameters of interpreters’ fluency.

Table 2.3 Summary of Studies on the Effects of Input Rate on Fluency

Study Participants Taxonomy of Fluency Input rate Results Cecot (2001) 11 prof. 1. Unfilled pauses

43

Table 2.3 (continued)

Study Participants Taxonomy of Fluency Input rates Results Pio (2003) 5 prof.

corpus-based filled pauses No range of input rates

44

Table 2.3 (continued)

Study Participants Taxonomy of Fluency Input rates Results Tissi (2000) 3. Interruptions

a. repeats b. restructuring c. false starts

3. Interruptions a. N/A b. N/A c. N/A

Note. ST: source text; prof.: professional interpreters; stud.: student interpreters; diff.: differences; min: minutes;

N/A: not available