Maria Durham
2. Analysis of data 1. Method
Four members of a class in the first half of their intensive six-month course in simultaneous interpreting at the Polytechnic of Central London took part in the ex
periment. Before starting the experiment, the subjects, three females and one male,
were told that they would be asked to interpret a passage of French prose simulta
neously. The passage of 558 words, spoken at a rate of 113 w.p.m., was taken from the November 1983 issue of Le Nouvel Observateur and was of a general subject nature. The testing session lasted for about 10 minutes. The subjects were seated in language laboratory cabins (not booths), facing the speaker, who read the text into a microphone; the text was heard by the students through their headphones.
The students’ output was recorded via a microphone attached to the headset, on the bottom track of their cassettes, while the speaker’s output was simultaneously recorded on the top track.
To reproduce my own teaching situation, neither the speaker nor the subjects were native speakers of French or English, respectively, but their profession and their school required a high proficiency of these languages from them. This, in fact, reflects a realistic working situation at an international conference, where most speakers use a language which is not their own.
The interpreting was then transcribed and all disturbances, irregularities of ut
terance, hesitations, false starts, filled pauses (er..., uh), etc. were included. The words pronounced simultaneously were then roughly superimposed on paper to reproduce what can be heard when the tape recordings of the two languages are listened to simultaneously, e.g.:
LUTHER MIT UNS. LE CINQ CENTIEMEANNIVERSAIRE DE LA NAISSANCE DU GRAND
LUTHER WITH US. THEFIFTH HUNDREDTH
RÉFORMATEUR ALLEMAND EST UTILISE PAR TOUS LES ANNIVERSARY OF THE BIRTH OF THE GREATREF... GERMAN REFORMER
CAMPS, A L’EST COMME A L’OUEST. ONNE SAURA IS BEING USED BY ALL CAMPS TO EAST AND WEST ALIKE.
JAMAIS SI LUTHERAVRAIMENT CLOUÉ SES QUATRE-VINGT-QUINZETHESES CONTRE WE WILL NEVER KNOW WHETHER LUTHER
LES INDULGENCES PAPALES SUR LA PORTE DE L’ÉGUSE DE WITTENBERG, REALLY DID NAIL HIS NINETY-FIVE THESES AGAINST PAPAL
OU SI C’EST L’HAGIOGRAPHIE POPULAIRE QUI LUIAPRÉTÉ INDULGENCES ONTHE DOOR OF THECHURCH IN WITTENBERG, OR
CE GESTE THÉATRAL.
WHETHER ITIS POPULAR HAGIOGRAPHY WHICH HAS ATTRIBUTED THIS THEATRICAL
REELS OU IMAGINAIRES, CES COUPS DE MARTEAU N’ONT POURTANT CESSE DE GESTURE TO HIM. WHETHER REAL OR IMAGINARY,
RÉSONNER DANS LA TÉTÉ DES ALLEMANDS. ONN’EN A THESE BLOWS OF THE HAMMER HOWEVERCONTINUE TO RESOUND IN THE
Maria Durham
JAMAIS AUTANT PARLÉ QU’EN CE MOIS DE NOVEMBRE OÜL’ON CÉLEBRE
HEADS OF GERMANS. THERE HAS NEVER BEEN SO
A GRAND RENFORT DE DISCOURS, DEPOSITIONS ET DE MUCH TALKED ABOUT AS THIS NOVEMBER
CEREMONIES TÉLÉVISÉES LE CINQ CENTIÉME ANNIVERSAIRE DE LA NAISSANCE WHEN WITH SPEECHES, EXHIBITIONS TELEVISED CEREMONIES
DU MOENE REBELLE. EN CET AUTOMNE, OÜLES THE FIFTHHUNDREDTH ANNIVERSARY OF THE REBELLIOUS MONK IS BEING
The transcripts were used to compare the interpreter’s version with the speak
er’s and to analyse their respective temporal progression. No qualitative analysis of the interpretation was performed.
Since the number of letters in the transcript did not reflect the length of time needed to pronounce them, parallel visual records (mingograph tracing) of the in
put speech and the most successful interpreter’s speech (subject 2) were used to investigate the length and nature of the segments which the interpreter needs to monitor before s/he can start encoding. First the top track was traced, then the bottom track, and finally the two together for a few seconds, so as to mark the be
ginning of the interpretation. Subsequently, the two sheets were fitted together.
Interesting patterns became evident: the words sometimes had to be written above the sound waves in a stretched out way, but at other times in a condensed fashion, because the letters could hardly be fitted next to each other. Clearly, in the former case, the interpreter must be waiting for something or thinking. Let us consider an example:
figure 1.
Alternatively, s/he can be hurrying to catch up because s/he is lagging behind or is trying to win time:
figure 2.
In another situation, s/he may be trying to use the pauses as much as possible to avoid simultaneity:
figure 3.
2.2. Analysis
The passage is written in a formal style, which means that it carries a higher in
formation load, so there should be a proportionally greater interpreter-to-speaker speech ratio. This would show the conversion of densely organised speech to a spon
taneous mode of expression. With the exception of subject 3 this is not the case.
All four interpreters used fewer words than the speaker, three of them over 60 words fewer than the original. One reason for this might be the languages used. Krusina (mentioned by Gerver, in Brislin 1976) found that the English interpretation is al
ways the shortest. (We have to be careful though about describing rate in terms of words per minute: in some languages more syllables are uttered per unit of time than in others, even though the number of words may be the same. We are con
cerned here with meaning rather than sound, and syllables as phonological units do not necessarily correspond with morphemic and semantic units.)
Another reason for the interpreting being shorter as regards the number of words is the fact that the more the speaker speaks, the more often the interpreter is likely to make omissions and the greater the amount of material s/he is likely to omit.
This is much more so in the case of inexperienced interpreters such as my subjects.
The way in which the subjects segmented the input is also probably due to the
Maria Durham
syntactic similarity of English and French, whereby little syntatic restructuring is needed and the source language/target language transfer is easier. As the expecta
tion patterns are largely similar, longer chunks of information can be recoded. The danger is though that the interpreter might slip into a word for word translation, as happens with subject 4. In an interpretation from French into English, the sen
tence is often translated in an order reversing that of the original delivery. This so
lution is popular with interpreters, e.g. subject 2, who fit a fast résumé into the speaker’s pause.
The variance between the other subjects’ interpreting as compared with the original reflects more than a knowledge of the two languages: it is evidence of non-verbal thinking.
The amount of material needed before the starting of interpreting varies with the position in the sentence of key-words, such as the verb. Delays which can range from 2 to 10 seconds are due to the relative difficulty of organizing the incoming material. While Goldman-Eisler found that the essential ear-voice-span is the NP + VP, my subjects often started translating immediately after the noun-phrase, knowing that they were not ‘mortgaging their grammatical future’. The most fre
quent ear-voice-span units were adverbial expressions, NP + VP + object (+ ad
verb, etc.) and the bare NP (often the first half of a genitive construction). Subject 2 usually stayed behind the speaker at an ear-voice-span sometimes involving two clauses or even a whole sentence. This gave him greater freedom to reorganize the structure of his sentences.
Subject 3 starts with an ear-voice-span of a certain length. Then, probably be- causes he speaks too much, this increases and accumulates until the amount of in
put surpasses her storing capacity. Accordingly, she suddenly finishes the sentence because she must catch up to bring the distance between the target language and the source language down to a manageable proportion. Although she omits mater
ial, she does not leave her sentences unfinished, which is something that cannot be said about subjects 1 and 4. The latter provides the most parrot-like interpreta
tion, while subjects 2 and 3 often display signs of cognitive processes in their cor
rections.
The total duration of the passage was 5.26 minutes. The general rule of simul
taneous interpreting is that the interpreter’s output cannot be longer or shorter than the input plus or minus a few seconds. Subject 2’s output lasted 5.27 min
utes, and the others were of similar length.
As concerns the total duration of the French original, 4.24 minutes was filled with speech and 58.5 seconds was used for pausing, i.e. the speaker was speaking during 80% of the period. Subject 2 spent 3.82 minutes speaking and 1.45 min
utes pausing. The pause criterion was 1 second. The overlap duration between this interpreting and the original was 2.3 minutes, i.e. less than half of the total dura
tion.
There were 75 speech bursts in the source language text, each burst lasting about 6 seconds on average. Subject 2’s output contained 103 speech bursts, averaging 4 seconds each.
The pauses in the source language lasted 1 second on average and numbered 53, while in subject 2’s output there were 69 pauses, with an average length of 2 seconds. Thus, the interpreter’s pauses were twice as long as the speaker’s and he
spent more time pausing and less time speaking than the speaker, although his speech bursts often contained many more words than the speaker’s or, in contrast, merely one word.
The latter fact proves the point of Gerver (in Brislin, 1976) that, when a speak
er speaks at rates between 96 and 120 w.p.m., little can be crammed into pauses of 1 to 2 seconds, though within longer pauses the interpreter might speed up his ar
ticulation rate. The pause time can be expanded at the cost of the speech time and vice versa. Barik (1973) also found that, in order to maintain a steady output rate, the interpreter lags further and further behind, pauses more and speaks in shorter bursts.
Paneth (1957), in contrast, states that interpreters make far shorter pauses than the speaker and talk 25% faster than him. Although the latter may be true at times, my findings, like Bank’s, seem to contradict those of Paneth. There is one point, though, which must be taken into account. It is misleading to speak of a lag in in
terpreting if the interpreter has stopped because of an inability to translate. These pauses are not associated with speech segmentation and are more characteristic of less experienced interpreters such as my subjects. This is also the reason why there is less simultaneity between input and output. Whereas Barik (1973) found that listening and speaking are concurrent for 70% of the time during interpretation from French into English, my subjects interpreted simultaneously for less than 50% of the total time. Even during this period, the subjects tended to stretch out their words (especially subject 2), thereby linking what should normally have been successive segments of speech. This likewise contributes to the amount of‘speech’
per se. Such dragging speech reveals an uncertainty of expression, and poor inter
preting is even more chopped up or fragmented.
3. Conclusion
The importance of memory in interpreting is emphasized by the relation be
tween the quality of the translation and the time-lag. More errors are made if the interpreter stays too close to the speaker or if s/he stays too far behind. The time- lag does not depend on the speaker’s speed, but on the interpreter’s ability to store linguistic meanings in his/her-short-term memory before they become translatable units of meaning. The interpreter tries to increase the amount of information the human brain is capable of processing in the short-term or working memory by re
encoding it into fewer chunks with a higher information density per chunk. His/her competence in keeping a suitable distance behind the speaker is closely associated with his/her ability to segment the incoming message. Only a limited amount of time is available for message decoding, encoding and reproduction since the average time-lag is only a few seconds. If the text has a high information density and a low degree of redundancy, the interpreter has to resort to strategies to delay output.
Simultaneous interpreting is based not only on the time-lag the interpreter keeps behind the speaker, but also on anticipation. Students seem to rely on operating with the time-lag, while experienced interpreters use anticipation as well. The two usu
ally go hand in hand, and whether one is used more than the other depends on the language combination, the speaker and the interpreter’s gifts and temperament. As regards the meaning of the message (and not necessarily the style of the rendering),
Maria Durham
students generally perform better when translating from a dominant into a weaker language rather than vice versa. Their energies are concentrated on what is being said, which (if delivered in a second language) requires greater attention to achieve comprehension. As a result, this reduces their ability to retain the message or to ren
der it quickly. The difficulty in recoding into the weaker language does not seem to be critical. The reason might be that the decoding sequence is more automatic and the encoding is closer to’transposing’ than to ‘interpreting’ since the less experienced subject translates more literally from the dominant into the weaker language.
The interpreter’s speech is less smooth than natural speech and contains more false starts, hesitations and other flows of delivery. There is also more variation in the duration of speech bursts than for the speaker. The organizational constraints on larger segments of speech do not necessarily refer to the interpreter’s output, since s/he does not always engage in sustained speech. There seems to be no system
atic or predictable relationship between the alternation of speech and silence peri
ods in the interpreter’s delivery. Periods of high fluency and high hesitancy, the repeated pattern of long pauses and short speech bursts alternating with short pauses and long speech, reflect cycles of acts of planning and production of speech, like the cognitive rhythm in rational spontaneous speech. Goldman-Eisler (1972) suggested that hesitant steep slopes are related to planning and fluent shallow slopes to the outcome of the planning.
Simultaneous interpreters make greater use of source language pauses than would be expected if the interpreter’s delivery were independent of the intervals of speaking and pausing in the source language. This might be because source lan
guage pauses occur more often between units of meaning. The interpreter consid
ers such units before starting to translate, and is therefore more likely to start speak
ing during source language pauses. S/he also tries to reduce the period of simulta
neous listening and speaking. Less experienced subjects have difficulty with simul
taneous listening and speaking. They pause more and speak less, and their pauses are longer and atypical. The simultaneity of the function, however, does not seem to hinder the performance of cognitive tasks, though it may limit the efficiency of the performance.
Further studies of time-lags in simultaneous interpreting are needed to help recognize familiar patterns and thereby assist future interpreters in how to plan their time and give them the ease of simultaneous listening and speaking that will win the admiration of non-practitioners.
References
Barik, H.C. 1973. Simultaneous interpretation: temporal and quantitative data. Language and Speech 16, 3: 237-70.
Brislin, R.W. ed. 1976. Translation: Applications and Research New York: Gardner Press.
Cenkova, I. 1989. L’Importance des pauses en interpretation simultanée. Melanges de phoné- tique generate et expérimentale. Publications de l’Institut de Phonétique de Strasbourg.
249-60.
Fodor, J.A., Bever, T.G. 1965. The psychological reality of linguistic segments. Journal of Verbal Learning and Verbal Behavior 4: 414-20.
Gerver, D., Sinaiko, H.W. eds. 1978. Language Interpretation and Communication. New York &
London: Plenum Press.
Gile, D. 1991. The processing capacity issue in conference interpretation. Babel 37, 1:15-27.
Goldman-Eisler, F. 1972. Segmentation of input in simultaneous translation. Journal of Psycholinguistic Research 1:127-140.
Goldman-Eisler, R, Cohen, M. 1974. An experimental study of interference between recep
tive and productive processes relating to simultaneous translation. Language and Speech 17:1-10.
Harkonen, J. 1989. The Effect of the Interpretation Situation on Students’ Performance, with Reference to Pauses/Silences and Repairs. In: Empirical Studies in Translation and Linguistics, ed. by S.Tirkkonen Condit, S.Condit. Joensuu: University of Joensuu, 201-14.
Paneth, E. 1957. An Investigation into Conference Interpreting. Unpublished MA thesis, London University.