Skip to content
Published on

What Interpreters Actually Train: What Can Be Confirmed in Public Curricula and Techniques

🇰🇷KO🇺🇸EN🇯🇵JA🇨🇳ZH
Split View필사 모드
Share
Authors

Introduction

Writing about interpreters mostly comes in two kinds. One is admiration, the other is secrets of the trade. Admiring pieces have nothing to verify, and pieces selling secrets usually lose their sources the moment you try to verify them. This series attempts a third kind. It tries to work out what can actually be confirmed in documents about interpreter training, and to mark where those documents stop speaking.

Let me be honest about one thing first. Information in this field sits at very different levels. The curriculum of an interpreting graduate school is published on the school website, so it is easy to check. Explanations of why a given form of training works, by contrast, are usually the recollections of a working interpreter, or a study with a sample somewhere between ten and forty people. Give the two equal weight and the whole piece becomes a lie.

So this post attaches an evidence class to its claims, sentence by sentence. It will be mildly irritating to read, but that is the only reason this series exists.

1. The evidence classes this series uses

The same labels appear across all four posts.

ClassMeaningAttitude toward reliability
Institutional documentA document that a school, an association or similar body has published under its own nameEasy to fact-check, but it does not prove effectiveness
Meta-analysisAn effect size computed by pooling several studiesThe strongest evidence available in this field, but it inherits the limits of the original studies
Single studyOne experiment or one corpus analysisTake the direction only, do not believe the magnitude
Theory or modelA frame that tries to explain a phenomenonNot a finding in itself
Practitioner accountThe experience of a working interpreter or teacherTreated as a hypothesis only
FolkloreWidely repeated, but the source cannot be tracedSomething to be tested against, not evidence

One more thing in advance. Some of the literature cited here sits behind a paywall, so I read only the abstract. Where that is the case, I say so in the body. Writing about a paper you have not read as though you had read it is the one thing this series must never do.

2. Consecutive and simultaneous are not the same job at two difficulty levels

The International Association of Conference Interpreters (AIIC) draws the distinction on its own site as follows. In simultaneous interpreting the interpreter renders the speaker into another language at almost the same moment, and this requires a soundproof booth, a microphone and headphones. In consecutive interpreting the interpreter waits until the speaker finishes a stretch and then renders it, so speaker and interpreter take turns. Whispered interpreting (chuchotage) is simultaneous interpreting without a booth, and AIIC recommends avoiding it where possible because the ambient noise and the interpreter voice interfere with each other. Evidence class: institutional document.

What matters in this distinction is not the equipment. The two modes use memory in fundamentally different ways. Consecutive listens to a whole chunk, builds the structure, and then reconstructs and speaks. In simultaneous, listening and speaking overlap. The first is a problem of analysis and reconstruction; the second is a problem of resource allocation.

The teaching order supports this. The MA in Conference Interpreting at the Faculty of Translation and Interpreting (FTI) of the University of Geneva runs three semesters, and according to the published curriculum page, consecutive interpreting begins in the first semester while simultaneous interpreting, including simultaneous with text, begins in the second. Placing simultaneous later reads most naturally not as simultaneous being an advanced version of consecutive, but as there being something that has to be secured in consecutive first. That interpretation, though, is my own inference; the school page does not explain it that way.

3. The lag: how ear-voice span is handled

The concept mentioned most often in simultaneous interpreting is the lag. In English it is called ear-voice span, in French décalage. It is the interval, measured in seconds, between hearing something and saying it in the other language.

Here is one set of numbers. A 2025 study in PLoS ONE measured the lag in simultaneous interpreting with text among 15 professional interpreters and 30 trainees working between Chinese and English. At a slow delivery rate the professional average was about 5.31 seconds and the trainee average about 8.33 seconds. At a medium rate the figures were about 5.24 and 6.80 seconds respectively, and at a fast rate about 3.50 and 3.91 seconds. The authors summarise a long lag as generally signalling a heavy processing load, while adding that under conditions with visual material present, lengthening the lag may actually help accuracy.

Evidence class: single study. And the limits of this study were written down by the authors themselves. There are twice as many trainees as professionals, there is only one language pair, Chinese and English, and the condition of reading along with a text is a special one. So these numbers must not be transplanted into general targets for work between Korean and English.

What is really to be had here is not a target in seconds but a concept. A simultaneous interpreter does not acquire the lag by accident; the interpreter regulates it. Sit too close behind the speaker and there is no room to take in the next piece of information; stretch it too far and what has accumulated in memory leaks away. What is trained is not the delay itself but the regulation of the delay.

4. Anticipation: often discussed, thinly evidenced

The story that interpreters know in advance what is coming next appears in almost every introduction to interpreting. The experimental evidence is more cautious than you would expect.

A 2023 study in PLoS ONE compared listening in noise with preparing to interpret consecutively. It ran two experiments with Dutch-English bilingual speakers, with 36 participants in the noise condition (32 in the analysis) and 36 in the clear condition (35 in the analysis). Eye-tracking showed that participants predicted the meaning of the upcoming word even in noise, but did not reliably predict its form. And contrary to the authors' hypothesis, there was no significant difference in the pattern of prediction between the task of simply listening and the task of preparing to interpret consecutively.

The limits the authors state are carried over here as they wrote them. The stimuli were short, simple sentences, the participants had had almost no training in consecutive interpreting, there was a single noise condition, and a single language pair. In other words, this study did not disconfirm the claim that trained interpreters predict better; it was not in a position to test that claim.

Put it this way. That people make meaning-level predictions when listening to language is a relatively well-established phenomenon; that interpreters do this especially well, or get better at it through training, is still at the level of a single study. When introductions to interpreting describe prediction flatly as a characteristic of interpreters, the literature behind that flat assertion did not exist within what I was able to check.

5. Chunking and salami: the technique of cutting sentences

There is a technique the profession calls the salami technique. Instead of processing a long sentence whole, you cut it into units of meaning and move forward by joining short sentences with conjunctions.

Sophie Llewellyn Smith, an AIIC interpreter and trainer, describes the technique on her site as breaking a long sentence or a long idea into small chunks in the target language, and says the key is to use open syntax. That means syntax which leaves plenty of options for what you say next. Concretely, she recommends joining with coordinating conjunctions such as and, but, so rather than subordinating conjunctions such as despite or although. The exercises she proposes are to mark units of meaning by drawing slashes through a text, and to reconstruct each unit aloud as an independent utterance. Evidence class: practitioner account.

On the corpus side there is a related study. A 2023 study in Frontiers in Psychology compared transcripts of United Nations Security Council meetings in three groups: non-interpreted English speech, speech interpreted from Chinese into English, and speech interpreted from Russian into English, each around 110,000 words. The use of formulaic multi-word patterns was higher in the interpreted speech, at 49.4 percent going from Chinese into English against 18.5 percent in non-interpreted English.

This study must not be used as evidence for the effectiveness of chunking practice. It is a corpus analysis, not an experiment with participants, and the authors themselves wrote that the proportion of variance their model explained was very low. What the study shows is the observation that under heavy load interpreters lean more on pre-formed patterns, not the conclusion that memorising those patterns in advance improves interpreting.

6. Note-taking: Rozan's seven principles are a real document

Note-taking for consecutive interpreting is one of the rare areas in this field where the primary source is unambiguous.

The book by Jean-François Rozan was published in 1956 by Georg in Geneva, and appeared again in 2005 in English and Polish translation from Tertium in Kraków, Poland. I confirmed these bibliographic details on the note-taking page of the Interpreter Training Resources site.

The seven principles appear in English translation, reproduced in full, in the proceedings of the 2011 WASCLA Language Access Summit. In order they are: noting the idea and not the word, the rules of abbreviation, links, negation, emphasis, verticality, and shift. Before introducing the seven principles, the original text notes that guidelines are such that the system still works if you do not follow them, only less efficiently. The evidence class is close to institutional document. That means the primary source exists and its contents can be checked directly, not that there is experimental evidence that this method beats others.

To take just the core: verticality and shift are the skeleton of the system. Instead of writing left to right you write top to bottom, and you indent a lower line to line up with the position where an element of the line above would repeat. Done this way, logical relations are revealed by the layout on the page itself, which reduces the need to write out connectives separately.

There is no reason to transplant this wholesale to an ordinary learner. That said, the first principle, noting the idea and not the word, holds exactly as stated whether you are writing meeting minutes or sitting in a lecture, and it costs nothing.

7. Deverbalization: a theory, not a verified finding

The concept cited most often in interpreting theory is deverbalization. It is the central concept of the Interpretive Theory of Translation, established by Danica Seleskovitch while she led the École Supérieure d'Interprètes et de Traducteurs (ESIT) in Paris in the 1970s, which views the process of translation and interpreting in three stages: comprehension, deverbalization, and re-expression. Interpréter pour traduire (1984, revised 2014), written with Marianne Lederer, is cited as the representative work. I confirmed this bibliographic information in the Wikipedia entry covering the theory, and since I could not read the primary sources directly, I note that this is a secondary source.

Deverbalization is the concept that at the moment comprehension occurs, the sound or written form disappears from memory and only the meaning remains. It is something we experience in daily life. You remember the content of a story you heard yesterday but not the exact words the person used.

There is something that must be distinguished here. This is not a finding established by experiment; it is a theory. Evidence class: theory or model. The reason the concept is so widely used in interpreter education is not that it has been proven true, but that it supplies a vocabulary for telling a student not to hold on to the words. Being pedagogically useful and being empirically true are different matters.

8. Shadowing: there is no consensus even inside interpreter education

In Korea, shadowing is introduced almost as a cure-all. Inside the interpreter education that produced the technique, it is a long-standing point of contention.

Christopher Guichot de Fortis, a senior interpreter at NATO headquarters who has taught at several interpreting graduate schools, defines it in his guide to shadowing as listening to live or recorded speech and repeating it word for word, sometimes including the intonation. And he recommends practising with the lag varied between 0.5 and 5 seconds. Up to here it is a useful practitioner account.

But the same text also contains this sentence: a claim that shadowing involves about 80 percent of the neuro-linguistic operations engaged in simultaneous interpreting. I could not find a source for this figure. Unless it can be established which study counted what, and how, to arrive at 80 percent, this has to be read as an impressive metaphor and not as data. Evidence class: folklore.

In the same text, the author himself acknowledges that shadowing is a controversial technique and is frequently misunderstood or undervalued. That is to say, the status of the technique inside interpreter education is not settled. The gist of the criticism is generally that shadowing is a matter of repeating sounds in the same language, so it leaves out analysis, which is the core of interpreting.

What about the experimental side? An open-access paper in the International Journal of Health Sciences in 2022 examined the effect of shadowing with a pre-test post-test design on 36 translation students at the University of Duhok, and reported a significant effect on simultaneous interpreting performance. Evidence class: single study, with a sample of 36 and a pre-post comparison rather than a controlled design. One study of this size does not settle a long-running debate.

9. What the schools actually require

Easier to verify than the training techniques are the thresholds for admission and graduation.

The document AIIC publishes as best practices for conference interpreting training programmes states several things explicitly. Applicants must pass an aptitude test before admission; the curriculum must include theory courses and courses covering professional practice and ethics; both consecutive and simultaneous must be covered; the teaching staff must be practising conference interpreters with a language combination recognised by AIIC and must have had training in interpreter pedagogy; and in the final assessment, candidates must be evaluated on whether their consecutive and simultaneous ability in all working language combinations meets the requirements for entry into the profession.

The Geneva FTI side points the same way. According to the public page, applicants must hold a degree and pass the faculty's oral entrance examination, through which they must demonstrate a high level of language and sufficient knowledge of current affairs. Beyond interpreting technique, the curriculum includes parliamentary procedure, conference terminology, knowledge of international organisations, interpreting theory and research, and professional ethics and standards. Students do autonomous group practice under regular supervision by teaching assistants, and also attend classes given by interpreters from EU institutions.

This passage tells us one of the most important facts in the whole series. The population of interpreters is a population that was already selected before training began. The root of the problem of interpreting research that a later post takes up lies right here.

What has not been verified

Here is what this post could not confirm.

First, none of the techniques introduced above is backed by a comparative experiment showing that it is better than another technique. The fact that something is in a curriculum shows the consensus of a professional community; it is not evidence of effect.

Second, there is no general standard for the ideal length of the lag. The numbers I cited are values for a particular language pair, particular conditions, and a particular sample.

Third, I could not find a source for the claim that shadowing involves 80 percent of the neuro-linguistic operations. I classify it as a widely cited but unconfirmed figure.

Fourth, deverbalization is a theory, and on this occasion I did not find a study that directly measured whether linguistic form actually disappears from memory during interpreting.

Fifth, I could not find comparative research showing that Rozan's note-taking system raises consecutive interpreting performance above other ways of taking notes. I confirmed the primary source and the content of the principles, but not the effect.

Sixth, the commonly cited multiplier for how much consecutive interpreting lengthens a meeting was not on the AIIC pages I read. All that can be said is that, by the definition of speaker and interpreter taking turns, it takes longer.

Closing

To sum up in one sentence what this post confirmed: interpreters are not trained to become people who are good at languages, they are trained to allocate limited processing resources in a particular way. Regulating the lag, cutting sentences, holding on to structure instead of words are all about that allocation.

The next post looks at the research on whether this training actually changes cognitive ability. To give away the conclusion, the literature contradicts itself. And that it contradicts itself is the most honest conclusion available.

If you want to try the shadowing and lag regulation covered here for yourself, you can practise repeating sentence by sentence at /tools/shadowing, and if you want to build short conversational situations there is /tools/conversation-lab.

References

Every link was opened and checked directly on 16 August 2026.

Discuss on Twitter • View on GitHub