The effects of input rate on accuracy

In previous studies concerning the effects of input rate on SI performance, error analysis is a common way to evaluate the accuracy of interpreters’ output. Barik (1971) was among the first researchers to develop a detailed error system for the analysis of an interpreting product. In a later paper, Barik (1973) presented the results of analyzing the output of six French-English participants (two professional interpreters, two student interpreters, and two amateurs) on four types of speeches (spontaneous, semi-prepared, prepared “oral,” and

33

prepared “written”) with the input rate ranging from 132.4 spm to 252.5 spm. He found that the faster the speech was, the higher the omissions were. However, the effects of input rate on other error categories such as substitutions and additions were not discussed in the study.

Like Gerver (1969) and Pio (2003), Galli (1989) also adopted the error analysis

approach to investigate the influence of input rate on professional interpreters’ performance.

She collected a corpus of ten source speeches and ten target texts of SI in real-life medical conferences (six from English to Italian and four from Italian to English). The effects of input rate, translation direction and text type on interpreting performance were examined. The input rate of the source speeches ranged from 106 wpm to 156 wpm. The target texts were coded according to four major categories of “departures”: omission, substitution, addition, and interpretation (p. 65). Subcategories and definitions of each error were illustrated in Table 2.2.

The number of each category of departures was counted per 1000 words of the source

speeches. Then, a regression analysis was used to examine the relationship between the input rate and translation departures. Positive correlations were found between the input rate and the numbers of all types of omission except for omissions of unnecessary information. A positive trend was also discovered between the input rate and translation mistakes (E”’) as well as summaries or interpretations that could lead to loss of information (I-). These findings suggested that errors that are not of high relevance or do not change the meaning of the source text were not affected by variation of input rates. Also, input rates did not affect the number of additions in interpreters’ output, consistent with Pio’s (2003) results.

A comparison of the error categories employed by Gerver (1969), Galli (1989) and Pio (2003) indicated that Galli (1989) adopted a more detailed coding system (See Table 2.2).

Gerver (1969) counted all incidences of omissions and substitutions without distinguishing the relevance or seriousness of these errors. Pio (2003) argued that some deletions might be necessary in order to enhance the clarity of the interpreting output. Therefore, in her study, only omissions with high information relevance that could alter the meaning of the source

34

speech were taken into consideration. Galli (1989) attempted to distinguish the relevance or seriousness of errors in her taxonomy (e.g., O” and O””; E’ and E”’). However, this coding system might not be applicable to other studies for two reasons. First, some of the error categories are specifically tailored to the text analysis of medical conferences, such as the omission of a term of a compound word (O’) and the omission of a term of a list (O’”).

Second, there are ambiguities in the definition of categories. For example, “inaccurate translation” (E’) was defined as substitutions that “do not” alter the meaning of the source text, which is somewhat confusing. It is also difficult to distinguish substitutions with slight changes in meaning (E”) from the category of translation mistakes (E”’).

Table 2.2 Comparison of Error Categories for Output Accuracy

Gerver (1969) Galli (1989) Pio (2003)

O”: omission of words or phrases which are unnecessary for message transmission O”’: omission of a term of a list O””: omission of parts of the texts which make the message incomplete do not change the meaning of ST E”: substitutions which lead to a slight change in meaning

E”’: translation mistakes that change the meaning of ST

Substitution

35

Table 2.2 (continued)

Gerver (1969) Galli (1989) Pio (2003)

Addition

A’: addition of words which do not provide supplementary information

A”: addition of material for clarification

A”’: addition of material to give closure to a sentence

Addition

Interpretation

I+: summaries that do not lead to a loss of information or correct interpretations for unclear I-: summaries which lead to a

loss of information or wrong interpretations for unclear sentences

2. Time sequence errors

Instead of adopting the error analysis, Lee (1999b) used a seven-point scale to evaluate the accuracy of each interpreted sentence. The evaluation was conducted on a corpus of 25 samples of SI from English into Korean by 14 professional interpreters from TV broadcasts and real-life conferences. Significant negative correlations were found between the speakers’

speech proportion and the average accuracy of 25 interpreting samples. This means the higher the input rate was, the lower the output accuracy was.

The above-mentioned studies investigated the impact of input rate on the overall accuracy of the interpreting output. Some studies chose to examine how input rates affected the rendering of specific structures of elements in the source text. Shlesinger (2003) aimed to probe the effects of input rate on simultaneous interpreters’ ability to retain long

left-branching noun phrases in English when interpreting into Hebrew, a head-initial language

36

that places modifiers after a noun. Sixteen professional interpreters were invited to interpret six English source speeches with 180 strings of long noun phrases (four modifiers before one noun) embedded within. Half of the speeches were delivered at 120 wpm and the other half were presented at 140 wpm. Three weeks later, the same participants were invited to interpret again the six speeches with reverse orders of input rate – the first half of the speeches were read at 140 wpm while the other half were delivered at 120 wpm. Shlesinger (2003)

hypothesized that the slow speech is detrimental to the retention of long strings of modifiers because traces of these items in working memory might decay when the delivery rate is slow.

However, no significant differences were found between the numbers of modifiers retained at 120 wpm and at 140 wpm. In fact, in two-thirds of the cases, only one modifier or no

modifier was kept in interpreters’ output irrespective of input rates. Shlesinger (2003) suggested that this might be the result of interpreters’ strategy to select or keep only one or two modifiers even when the speed was slow and when their cognitive capacity had not reached saturation. In this case, interpreters’ deliberate strategy of omission probably played a bigger role than the influence of input rate.

Two more recent studies by Korpal (2012) and Barghout et al. (2015) also saw omission as a deliberate and pragmatic strategy that interpreters may adopt when facing fast speeches, instead of treating it as an error that interpreters make under cognitive overload. Korpal (2012) explored the effects of input rate on omissions of redundant information in the source speech by comparing the SI performance from English into Polish between 11 professional

interpreters and six students. Two source texts were made into slow (130 wpm) and fast versions (177-180 wpm). Each source speech contained 19 areas of redundant information divided into five categories: repetition of exactly the same words, redundancies, cultural allusions, empty fillers, and the speaker’s subjective assessment. Results of t-test analysis showed that input rate had an effect on the omission of students but not on that of

professionals. Students omitted more redundant information in the fast speech. This finding

37

suggested that professional interpreters are less susceptible to the manipulation of input rates.

However, when the numbers of omission were combined regardless of input rates, there was no significant difference between professional interpreters and students. This null result was probably due to small sample size.

In a similar vein, Barghout et al. (2015) studied the influence of input rate on omissions of redundant compound conjunctions and pairs of synonyms. Ten professional interpreters were invited to interpret three source speeches from English into French. Each speech was divided into three segments of three different speeds: 120 wpm, 160 wpm, and 200 wpm. By comparing the proportions of omission at each input rate, results showed that the omission of synonyms did increase with faster rates. Nevertheless, it was found that only one synonym was rendered in the interpreting output for around 50% of the cases across all speeds. This shows that when dealing with a pair of synonym, professional interpreters tended to keep only one of them whether in slow or fast speeches. Consistent with Shlesinger’s (2003) findings, interpreters were selective in their rendering of redundant elements. For compound conjunctions, the effects of input rate were less clear. More omissions of redundant

compound conjunctions were observed when the input rate increased from 120 wpm to 160 wpm, but from 160 wpm to 200 wpm, the omissions were reduced. Barghout et al. (2015) explained that when interpreters were under extreme cognitive load such as coping with an input rate of 200 wpm, they were probably less capable of distinguishing important messages from redundant ones and could only strive to reproduce whatever the speaker had said.

Therefore, there were fewer omissions of redundant information at the highest speed.

In sum, previous studies have found that fast input rates beyond 120 wpm had a negative impact on simultaneous interpreters’ output accuracy. Both professional interpreters and students made more omissions when the input rate was over 120 wpm. The impact of input rate was greater on students than on professionals. Students made more substitutions in fast speeches, but it was not necessarily so for professionals. It was also observed that input rates

38

did not affect every category of errors in interpreters’ output. The most consistent finding is that fast input rates led to more omissions of words and segments. Input rates seemed to have a greater influence on the rendering of important messages than on redundant ones in the source speech. For redundant elements in the source speech, professional interpreters tended to adopt an overall strategy of selective encoding (Liu, Schallert, & Carroll, 2004), which was less affected by changes of input rate.