«« Computational Discourse Analysis for Interpretation » Sane M. Yagi Meta : journal des traducteurs / Meta: Translators' Journal, vol. 44, n° ...»
« Computational Discourse Analysis for Interpretation »
Sane M. Yagi
Meta : journal des traducteurs / Meta: Translators' Journal, vol. 44, n° 2, 1999, p. 268-279.
Pour citer cet article, utiliser l'information suivante :
Note : les règles d'écriture des références bibliographiques peuvent varier selon les différents domaines du savoir.
Ce document est protégé par la loi sur le droit d'auteur. L'utilisation des services d'Érudit (y compris la reproduction) est assujettie à sa politique d'utilisation que vous pouvez consulter à l'URI https://apropos.erudit.org/fr/usagers/politique-dutilisation/ Érudit est un consortium interuniversitaire sans but lucratif composé de l'Université de Montréal, l'Université Laval et l'Université du Québec à Montréal. Il a pour mission la promotion et la valorisation de la recherche. Érudit offre des services d'édition numérique de documents scientifiques depuis 1998.
Pour communiquer avec les responsables d'Érudit : firstname.lastname@example.org Document téléchargé le 25 November 2016 09:09 Computational Discourse Analysis for Interpretation sane m. yagi SQU, Oman RÉSUMÉ Une méthode informatisée qui permet l’examen attentif du discours dans la langue source et dans la langue cible, milliseconde par milliseconde, a été élaborée afin de permettre une analyse de l’interprétation. En utilisant un programme spécialement déve- loppé à cet effet, l’analyse du discours de l’interprétation peut maintenant être automa- tisée et rendue quantitative. Les discours en langue source et en langue cible peuvent être tracés graphiquement et parallèlement sur l’axe des temps, facilitant ainsi la compa- raison entre le discours et son interprétation simultanée ; on peut comparer la durée de chaque éclat et de chaque pause, le début et la compensation, le degré de simultanéité entre l’orateur et l’interprète et bien d’autres paramètres importants au théoricien de l’interprétariat.
ABSTRACTIn order to perform discourse analysis on interpretation, a computational method that facilitates the scrutiny of SL and TL discourses on a millisecond by millisecond basis was developed. Using an in-house purpose-built program, interpretation discourse analysis can now be automated and made quantitative. Both SL and TL discourses are graphically plotted on a clock-time axis, thereby facilitating a comparison of simultaneously-delivered speech in terms of the duration of each burst and pause. As well, onsets and offsets can be compared to their equivalents in the other language, as can the degree of simultaneity of speech between speaker and interpreter, and various other matters of concern to the interpretation theoretician.
Goldman-Eisler (1968) concluded from 20 years of research that speech is an articulate and finely graded external projection of cognitive processes organized and integrated in time. Speech is the end product which reflects the workings of the mind.
When speech is spontaneous, as is the case in simultaneous interpretation (SI), the relationship between the spoken word and thinking becomes more evident. Since SI consists of concurrent and semi-concurrent cognitive activities (listening, decoding, encoding, and speaking), the time factor is of crucial importance. Therefore, analyzing interpreters’ time-management patterns can reveal a wealth of information about the characteristics of their performance, as well as about the cognitive processes associated with their speech.
This paper will describe a new method of discourse analysis which uses the computer to study source (SL) and target (TL) language speech on the basis of the time structures of their acoustic signals. The software that was developed for this purpose will also be reviewed. Research conclusions derived by this method will be cited to demonstrate its utility for the translation specialist and the discourse analysis linguist.
Before describing this digital method, however, it is necessary to explain where it should be placed among the various methods of analysis used in SI research.
SI METHODS OF ANALYSISResearch into simultaneous interpretation has used three methodologies: translation discourse analysis, cognitive experimental procedures, and psychophysiological techniques. Each of the three methodologies has a different focus, the first concerned with output speech, the second with cognitive processes producing speech, and the third with speech segments requiring intense cognitive activity.
1. Translation Discourse Analysis Generalizations can be made about SI by analyzing interpreters’ output and comparing it with SL discourse. Interpretation teachers and theorists alike can use discourse analysis to assess an interpreter’s performance and to gain insight into the cognitive processes involved in interpretation.
There are two types of translation discourse analysis: quantitative and qualitative. In the first type, researchers reduce the data under consideration to numbers that are manipulated in several ways, using standard statistical methods. In qualitative discourse analysis, generalizations about an individual’s performance and about the essence of SI are made on the basis of introspection, native speaker intuition, and subjective assessment of the degree of convergence or divergence between SL and TL pieces of discourse.
Different kinds of information can be obtained from these two types of discourse analysis. The quantitative method can yield information such as duration of delay, degree of simultaneity, speech burst and pause length, articulation and speech rates, etc. It can also offer the opportunity to study the influence of SL discourse temporal characteristics on TL translation. The qualitative method, on the other hand, is used to infer the possible causes of errors, and to study meaning loss, types of addition, omission, and substitution in a translation. Barik (1969) pioneered the quantitative type of translation discourse analysis but at the same time applied the qualitative method to his data.
2. Cognitive Experimental Procedures Because simultaneous interpretation involves complex mental processes, several experimental tasks have been devised to study some of these individual cognitive processes (e.g. recall, recognition, split-span, shadowing, judgment, etc.). Recall tasks study the interpreter’s ability to access and retrieve information from long-term memory, while recognition tasks study the ability to identify correct information.
Split-span studies investigate the ability to divide attention between two tasks, and shadowing the ability to listen and speak at the same time. Judgment tasks study information encoding specificity, while SI itself is used to investigate subjects’ ability to do concurrent tasks: listening and speaking, recognizing and recalling, and decoding and encoding. Many SI researchers have used these techniques (Triesman 1965;
Goldman-Eisler 1972; Gerver 1971, 1976; Lambert 1983).
270 Meta, XLIV, 2, 1999 Split-Span Tasks This task type is used to inquire into selective attention, which interpreters do a lot of.
It requires subjects to listen to one of two auditory signals that are presented dichotically while ignoring the other. The experimenter checks on such things as the subjects’ comprehension of the attended signal, their recognition of the nature of the non-attended signal, and the interference of one signal with the other.
The relevance of this task to SI is evident in its facilitation of the study of allocating attention to more than one cognitive task. Since interpreters have to divide their attention between decoding the SL discourse and encoding it into the TL, their skill in split-span tasks is essential to good SI performance.
Recall Tasks Recall involves searching and accessing memory to retrieve specific information.
Without it, interpreters would not be able to identify the TL equivalents of the material they listen to. Recall tasks consist of an acquisition period and a test period.
During acquisition time, subjects are given a list of items to remember, then are asked during test time to reproduce these items in response to a clue given by the examiner.
Subjects are tested for their ability at “free recall” (recalling the items in any order), “partial recall” (recalling only some of the items), and “ordered recall” (recalling the items in the order presented).
Recall tasks are useful for studying information organization in memory, retrieval sequence, taxonomic semantic relationships, memory capacity, memory trace decay, etc. They can enlighten researchers as to how subjects store information in memory and for how long, and how they are able to access it. Recall errors are also quite valuable; confusion errors that mix up the serial order of two list items, offer a synonym in place of a list item, or paraphrase items on the list are all informative of how subjects represent information in their memory. Intrusion errors that result from guessing are indicative of the background knowledge subjects use when they fail to recall precisely.
Recognition Tasks Subjects are presented with a list of items during an acquisition period, then are given a list during a test period that includes target items presented previously, as well as foils which function as distracters. The subjects’ task is to identify targets by remembering their association to the acquisition context. Their response can be in a yes-no format, in multiple-choice, or by rank-ordering all probes according to how likely they are to be targets. Subjects often have to indicate confidence in their response using a three-point scale.
Recognition tasks are mainly useful for two purposes: studying memory interference and measuring reaction times. The first makes it possible to investigate the efficiency of the interpreter’s retrieval strategies, while the second is concerned with how long it takes the interpreter to access information. There is no doubt that SI lag is directly influenced by interpreter reaction times, among other things.
computational discourse analysis 271 Judgment Tasks Judgment tasks are often employed to study information-encoding strategies. They are similar to recognition tasks, but generally rely on conclusions that subjects formulate rather than merely remember. After the completion of an acquisition period, the examiner might ask about which of two items was presented more recently or more frequently, which was printed in upper or lower case, which was in the native or foreign language, etc. Such questions can be informative about the degree of specificity in memory encoding.
Shadowing Tasks Shadowing is repeating a stimulus discourse verbatim as it is being delivered. It is the task most akin to simultaneous interpretation because it shares several cognitive processes with it. Both involve listening and speaking concurrently, but the latter requires extensive decoding and language transfer.
Shadowing is used for various purposes, most important of which is reaction time. The fact that subjects have to listen to a stimulus discourse before they are able to output speech means that they maintain a reasonable delay that allows them to keep track of the source. This delay is similar to, though significantly shorter than, the delay kept by simultaneous interpreters. It is for this delay, in particular, that SI researchers use shadowing, primarily as a control task against which SI is measured.
3. Psychophysiological Techniques Interpreter’s perception of SL discourse segments produces arousal, an intense degree of alertness and concentrated attention. It is usually manifested in a low amplitude mental activity, some cardiovascular changes, and overt behavioural responses such as tilting the head towards the sound source. Arousal can be studied by recording the electrical activity of the brain in electroencephalograms (EEG), or by measuring cardiovascular changes using heart rate and blood pressure indices.
The size of the pupil of the eye is also thought to be an indicator of arousal and consequently of mental activity. It is fairly well-demonstrated in the literature that pupillary dilation is an index of mental effort. Some researchers, like Hess (1965), discovered that pupils dilate during the performance of mental arithmetics. Beatty and Wagoner (1978) obtained the greatest pupillary dilation in a hierarchically structured letter-matching task that required higher levels of mental processing.
Matthews, et al. (1991) also found that error rate and pupil dilation amplitude rose with increased task difficulty. Just and Carpenter (1993) explored the intensity of mental processing during sentence comprehension by measuring pupillary dilation during reading. They contrasted the cognitive processing of simpler vs. more complex sentences, and found that the complex ones (object-relative center-embedded and filler-gap sentences) produced a larger change in pupil diameter. The method used for measuring pupillary dilation and employing it as an index of mental effort is called pupillometry.
Some psychophysiological methods have recently been employed to infer the mental effort associated with simultaneous interpreting. Klonowicz (1990) used cardiac activity to compare the mental workloads associated with SI and shadowing. She 272 Meta, XLIV, 2, 1999 found out that both tasks elicited a cardiac mobilization, but cardiac activity soon stabilized in the course of shadowing performance, and confirmed the popular notion that SI relies on vigilant attention. She also looked for psychophysiological correlates underlying SI, and concluded that the cardiac “mobilization effect” correlates positively with reactivity temperament, trait-anxiety, and trait-curiosity.