How do we interpret?

Revisiting the old idea of language as code and Seleskovitch’s "théorie du sens" with its concomitant notion of deverbalization. A new understanding of what conference interpreters do.

1. Introduction

A couple of years ago (May 2003), Lingua Franca, the European Parliament interpreters’ bulletin, carried an article («Comment faut-il traduire») by Ann-Marie Widlund-Fantini, expounding Danica Seleskovitch’s «théorie interprétative de la traduction» (or «théorie du sens») [1]. A central tenet of this theory is the notion of deverbalization («déverbalisation»), i.e. the ability of the translator/interpreter to perceive the meaning («sens») of a text/an utterance in its proper context and thus convey its underlying message, as distinct from mere transcoding («transcodage»), or word-for word translation.

It is hard to argue against the value of this notion as a pedagogical tool that has served to nurture thousands of interpreters. Who could think of a better way to focus the mind of young trainees on the essence of what people say rather than the particular words they choose to employ? Not to forget how effective this simple argument has been to press the point that, in conferences, the interpreter must be involved as a full participant rather than just as a mere translating machine. Yet, one might be entitled to ask, without intending any disservice to the memory of Danica Seleskovitch, a great teacher in her own right, whether this view of interpreting can be still considered valid today.

To attempt to answer this question, it is perhaps best to start with the prevailing view of language at the time the «théorie du sens» was first formulated: that of language as pure code. This will lead us to transcoding and its unexpected persistence in human communication; the possible lessons to be learned through the training of novice interpreters; the exploration of a possible mechanism through which deverbalization might take place and finally to a modern linguistic approach to communication on the basis of relevance theory, which could serve to highlight the risk of communication breakdown in specific conference situations.

2. Language as code

An implicit motivation for the «théorie interprétative de la traduction» might have been to shore up the interpreting profession by showing that its practitioners were much more than trained parrots, merely transcoding words from a source into a target text. Cryptography and the view of human language as code were already very much in vogue through the 1940s and 1950s, in the wake of the celebrated cracking of the Enigma code at Bletchey Park by Alan Turing during World War II [2]. This view can only have been reinforced when, a few years later, a young cryptographer, Michael Ventris, successfully took up a challenge that had defeated scores of scholars: deciphering Minoan Linear B, the script, used for the lingua franca of the Eastern Mediterranean between 1500 -1200 BC, which turned out to be none less than Greek [3]. Ventris was tragically killed in a car accident shortly thereafter, but his original handwritten notes are still to be seen among the Herakleion Archeological Museum’s rich collection of Linear B tablets.

This was also the time when Warren Weaver (a distinguished mathematician and founder, together with Claude Shannon, of statistical communication theory) made the following cryptic statement (considered to have signaled the dawn of machine translation) in a 1949 memorandum to the Rockefeller Foundation [4]:

I have a text in front of me which is written in Russian but I am going to pretend that it is really written in English and that it has been coded in some strange symbols. All I need to do is strip off the code in order to retrieve the information contained in the text.”

In fact, for the kind of information that the US military establishment was then most interested in, namely Russian scientific articles, Warren’s statement turned out to be sufficiently accurate. Of course, when machine translation was applied to more general texts its limitations become glaringly apparent, as even the most casual user of translation engines on the Internet can still attest today. Yet, it is remarkable how, on the basis of a purely lexical and syntactical approach, transcoding from one language to another without an ounce of understanding of the texts they are translating, machine translation systems can sometimes produce valid output.

3. Transcoding

Such transcoding is of course anathema to professional interpreters, whether brought up in the fold of the «théorie interprétative de la traduction» or not, even though we might all privately admit having occasionally resorted to such an indignity when under severe pressure.

This does not seem to deter the average layperson from a more relaxed view of translation and interpreting. During summer holidays spent in my home country, I never fail to be amused by the imaginative «English» entries in «bilingual» tourist restaurant menus. What about, say, «chickenspit» as a main course? Well, this is too elementary; you have already guessed that it really means chicken roasted on a spit. Try «reddened veal» instead? Again you might have correctly deduced that it is just veal in red sauce. But what about «eyelid soup»? (That’s right, «eyelid soup»!) Now, the corresponding entry in Greek is «soupa kritharaki», which would normally translate to something like «barleycorn soup» in English. So where do the eyelids come from? It turns out that «kritharaki» has a second, colloquial meaning in Greek; this is the name given to small warts that can grow on your eyelid. Hence, «soupa kritharaki» equals «eyelid soup», QED!

Such idiosyncratic language skills are guaranteed to provide light relief, but this is not the point. What is much more remarkable is the touching but naïve belief they illustrate, namely that a dictionary is all it takes to translate from one language into another. We know better than to expect linguistic perfection - rather than just good food - from tavern owners; yet even sophisticated foreign language speakers are never entirely immune from native language interference. A mercifully long time ago, during a mission of the European Parliament transport committee, the committee members, staff and interpreters were spirited to a bridge construction site in a minivan and issued with hard hats which would have to be returned at the close of the visit there. Accordingly the committee chair, a Greek native speaker but also an inveterate user of both English and French in committee meetings, invited his colleagues «à laisser vos crânes dans le minibus». This fairly radical exhortation send a chill down the spine of all, save those lucky souls who knew that in Greek the word for a hard hat («kranos»), is tantalizingly similar to that for a skull («kranio») which translates to «crâne» in French.

Most of the favorite stories among interpreters hinge on such slips of the tongue. But the tongue can also play darker games, as in the case of aphasic bilinguals (bilingual is used here in the sense of being able to speak two OR more languages), people whose speech faculties have been impaired as a consequence of brain damage. Some aphasic bilinguals exhibit an odd behavior known as spontaneous translation: when asked a question in one of their languages, for example, “What time is it?”, they will respond not by answering the question but by translating the question into their other language(s), e.g. “Quelle heure est-il?”, just like the interpreter in the stock joke, mechanically translating his wake-up call. Another bizarre behavior of some aphasic bilinguals is known as paradoxical translation: during their recovery, they often have a strong language, in which they can communicate and a weak language, which they have great difficulty using. The paradox is that while they can hardly carry on a conversation in their weak language, they can easily translate from their strong language into their weak language [5].

4. From trainee to professional interpreter

But if transcoding is the prevailing, «default» mode of translation for the average layperson and if, as the aphasic bilingual cases suggest, it is practically hard-wired in the brain of everybody who can speak at least a second language, where does that leave deverbalizing? Rather than being a normal, mainstream human activity as the «théorie du sens» would have it, deverbalizing does not seem to extend much beyond the relatively small population of trained translators and interpreters. All of a sudden, their craft appears in the light of a freak activity (or perhaps it would be more accurate to say: an activity reserved to freaks, a statement many interpreters would fully agree with). So how does one acquire this craft in the first place?

The answer might be: «with great difficulty». For a variety of reasons, over the past 20 years, the Greek booth at the European Parliament (EP) has acquired a lot of experience, both in occasionally training interpreters from scratch but also, especially, in «inducting» novice interpreters, with scant if any previous experience of simultaneous interpreting, to the EP environment. Training the latter consists in providing an introduction to parliamentary procedures and thematic areas through classroom exercises in both consecutive and simultaneous mode. It also includes extensive dummy booth practice (mostly in committee meetings) under the constant supervision of a trainer sitting next to them. Such dummy booth practice has proved indispensable in smoothing the transition from the relatively relaxed and forgiving atmosphere of the classroom into the real world of arcane parliamentary discourse. The opportunity to watch professional interpreters in other booths (including one’s own trainers) work, struggle and occasionally get bloodied, is not an inconsequential bonus to the trainee.

One might have expected trainee «induction» to follow a smooth, linear course of continuous improvement in simultaneous technique. It rarely does. Trainees typically go through three stages.

First a transcoding, “automatic pilot” stage during which trainees tend to translate single words in isolation, using their more common meanings, irrespective of context, often rushing to interpret the first couple of words in a sentence without waiting for a full semantic unit to be completed. Or they might «telescope» two such semantic units into one. When they are confronted with the lack of coherence in their output, their most frequent excuse is “but this is just what the speaker said.”

Later rather than sooner, the limitations of the automatic pilot technique become evident and the trainee realises that no further progress can be made without a change of tack. This second stage is typically signaled by a phase of withdrawal: the trainee may continue to translate isolated, innocuous snippets from the input text while refusing to deal with its more ambiguous passages, requiring her to commit herself to a single, coherent line of argument. Tension and friction ensue often escalating to a personal confrontation with one’s trainers. Some trainees simply decide to throw in the towel.

Those that don’t eventually advance to the third stage and to what trainers often call the «déclic». The trainee has at last accepted the conference interpreter’s predicament: the message to be transmitted is not some impersonal abstraction but what she herself has understood, right or wrong, no more, no less. Irrespective of poor sound conditions, faulty grammar, stilted syntax, foreign accents, etc. the show must go on.

At this stage, the capacity to deverbalize has finally been acquired by the trainee, not as a product of methodical, purely intellectual exercises but rather as the outcome of a series of struggles and convulsions involving the whole personality. Before the trainee can rein in the urge to transcode, making it instead contingent upon the understanding of the wider context of each utterance, her brain has perhaps undergone partial recabling.

5. Deverbalizing: The Kintsch & van Dijk model

According to Seleskovitch and Lederer [6], the interpreter transforms the input speech into «chunks» of pure meaning, «forgetting» the original words used. These deverbalized chunks or «units of meaning» are subsequently combined with both immediate and global context (world knowledge) before the output phase. But the «théorie interprétative de la traduction» does not propose a concrete mechanism to explain how this actually takes place.

A possible framework for the meaning formation process could be sought within the Kintsch & van Dijk model for discourse comprehension and production [7, 8]. This model postulates that the surface structure of a text is temporarily stored in memory as a set of micro-propositions, out of which deverbalized macro -propositions are constructed over a characteristic time scale of about 400 msec, on the basis of three “macrorules”: deletion, generalization and construction:

  • Deletion consists in throwing away less relevant information as in: «A black stallion galloped by» transformed into «A horse passed by».
  • Generalization proceeds by replacing specific statements with more general ones, e.g.: «Some demonstrators threw bricks at the police » and «Others lobbed Molotov cocktails into nearby shops» with «The demonstration turned into a riot».
  • Construction works by producing new statements which contain an inference not explicit in the original utterance: «Beckham executed the free-kick; the Bernabeu exploded in applause» can lead to: «Beckham scored for Real Madrid».

In the Kintsch & van Dijk model, combining the input text or utterance with local and global context is not a sequential, algorithmic process. Meaning construction instead proceeds by first going through a “creative anarchy” stage, in which the input text activates a number of immediate memory associations in the reader/listener, followed by a second one where these are severely culled, and only those satisfying a number of consistency constraints survive [9]. Discourse understanding is then much less the result of a deterministic process than that of the sudden “falling together” of the pieces of a puzzle.

6. The Cooperative Principle and Relevance Theory

But how are such immediate memory associations activated? How is the listener to know in which context an utterance should be understood? Most models of discourse comprehension implicitly rely upon a “code” model for human communication, where each utterance contains the totality of the speaker’s intended message, encoded in language, to be decoded by the audience using an identical version of this linguistic code.

An alternative approach to human communication is proposed by the inferential model [10] introduced by the philosopher Paul Grice, according to which a communicator provides evidence of her intention to convey a certain meaning, which is then inferred by the audience on the basis of the evidence provided. Speakers and hearers follow a cooperative principle, whereby speakers shape their utterances in such a way as to be understood by hearers, by giving just enough but not too much information.

The Cooperative Principle has been further refined within the framework of Relevance Theory (RT) [10]. According to RT, human communication is a two-step, “ostensive-inferential” process. In the first step, the speaker “ostents” his intent to bring to the fore a layer of information or set of assumptions. According to the Principle of Relevance, every such act of ostension carries the presumption of its own relevance. The listener then completes the process of understanding the speaker’s utterance by an inference of its meaning consistent with the context pointed out by the speaker as being the most relevant one.

The way in which a speaker will provide evidence of his intent, both verbally and non-verbally (facial expression, pose, gestures) is, of course, specific to his own language community. Familiarity with the patterns of language use in each such community, as well as proper access to the verbal and non-verbal context provided by the speaker is a must if the interpreter is to successfully arrive at the right inferences. Moreover, in the RT framework, it becomes the interpreter’s responsibility to maintain the communication chain by re-ostenting the speaker’s intentions in the manner appropriate to the listener’s language community, rather than just providing a textual translation of the original. [11]

7. A fragile communication chain

If the relevance principle is indeed valid, then it is hard to think of a better communication setting in which it should be applied than parliamentary meetings, with their ritual and highly stylized turns of speech and the liberal use of innuendo. In such meetings, the communication chain is particularly fragile, and can be disrupted in a number of situations.

A first such situation concerns speakers not using their native language, or at least a language in which they are sufficiently fluent, but rather one of the most widely used, “vehicular”, European Union languages such as German, French and especially English, as is often the case following the recent EU enlargement. This choice may be involuntary, as a result of only having a limited number of passive languages to choose from in a meeting, or even voluntary, in the (misguided) belief that speaking a widely used language enhances communication. These speakers will invariably be unable to fully exploit the relevance mechanism in the vehicular language they are using so as to be able to put their point across in an effective manner. Even worse, their statements might be the product of direct transcoding from their native language and therefore completely opaque to the interpreter, unless the latter happens to understand the latter language as well.

A second example of communication breakdown arises in a remote interpreting setting, where, unlike the meeting participants, interpreters have only limited access to non-verbal context in the meeting room. A further, perhaps even more important constraint derives from the inability of interpreters (invisible to their listeners) to properly re-ostent the speaker’s message, in the absence of visual feedback from their listeners to indicate that communication is proceeding smoothly or otherwise. The net effect of remote interpreting where the interpreter no longer shares the same reality with her audience is not much different from that of working in a void with nobody listening: loss of motivation, dissociation from the meeting, ultimately indifference. Here the difference between the seasoned interpreter and the novice, who might actually feel more comfortable in such a setting, can become dangerously blurred.


[1] Ann-Marie Widlund-Fantini, «Comment faut-il traduire»:
[2] Andrew Hodges, «Alan Turing: The Enigma» (Walker & Company, October 2000); Robert Harris, «Enigma» (Arrow, May 1996)
[3] John Chadwick, «The Decipherment of Linear B» (Cambridge University press, September 1990); Simon Singh, «The Code Book» (Fourth Estate, May 2002)
[4] Doug Arnold et al., «Machine Translation: an Introductory Guide» available online:
[5] Franco Fabbro, «The Neurolinguistics of Bilingualism» (Psychology Press, January 1999)
[6] Danica Seleskovitch, «Langage, langues et mémoire» (Lettres modernes, 1975); Marianne Lederer, «La traduction simultanée» (Lettres modernes, 1981)
[7] Jennifer Mackintosh, «The Kintsch and Van Dijk model of discourse comprehension and production applied to the interpreting process» available online:
[8] Helge Niska, «Text Linguistic Models for the Study of Simultaneous Interpreting» available online:
[9] Walter Kintsch and Teun van Dijk, «Toward a model of text comprehension and production” (Psychological Review 85 p363 - 94, 1978); Walter Kintsch, “The Role of Knowledge in Discourse Comprehension: A Construction – Integration Model” (Psychological Review 95, v2 p163 - 182, 1988)
[10] Deirdre Wilson and Dan Sperber, «Relevance Theory» (in G. Ward and L. Horn (eds.) Handbook of Pragmatics, Blackwell 2003), available online:
[11] Robin Setton, «Meaning Assembly in Simultaneous Interpretation» (in Pöchhacker and Shlessinger (eds.) The Interpreter Studies Reader, Routledge 2002)

Panayotis (Takis) Mouzourakis is a staff interpreter in the Greek booth for the European Parliament.

Recommended citation format:
Panayotis MOUZOURAKIS. "How do we interpret?". March 21, 2005. Accessed July 5, 2020. <>.