필사 모드: What an Ordinary Learner Can Actually Take from Interpreter Training

Introduction

To summarise what three posts have established: the content of interpreter training is public and can be checked, whether that training changes cognitive ability is under dispute, and the list of polyglot habits is largely unverified.

That leaves one question. Out of all this, is there anything an ordinary learner with one hour a day can actually take?

There is. But what can be taken is narrower than you would think, and what cannot be taken is more than you would think. And the evidence for the things that can be taken is generally not strong. So the criterion in this post is not "has the effect been proven" but "is the cost low, and is the loss small if it turns out to be wrong". I state that criterion up front because it is a practical judgement made while admitting the shortage of evidence, not a substitute for evidence.

1. First, sorting what carries over from what does not

One honest footnote to this table. "Carries over" means you can imitate the activity; it does not mean you get the same result an interpreter gets.

2. Shadowing: the evidence is weak and the cost is low

As the first post showed, shadowing is not an agreed technique even inside interpreter education. The gist of the criticism is that repeating sounds in the same language leaves out analysis, the core of interpreting. On the experimental side, all there is is a significant effect reported in a pre-post comparison of 36 students at the University of Duhok, a design with no control group.

Even so, my reason for putting this item on the side that carries over is simple. It costs almost nothing. You need earphones and an audio source, you can start with 10 minutes a day, and if it does not work the only thing you lose is those 10 minutes.

It is better to attach a condition, though. The way Guichot de Fortis recommends is to practise with the lag varied between 0.5 and 5 seconds. Put that regulation in and it becomes practice in allocating attention rather than plain repetition. Following right behind and following three seconds later are different jobs inside your head.

There is also a way to take the criticism on board. Instead of repeating verbatim, follow one sentence behind and note the content in a word or two as you go. Doing it this way partly sidesteps the objection that analysis is missing. There is no evidence that this variant is better, though. It is just a logically plausible adjustment.

3. Deliberate delay: making the lag for yourself

In simultaneous interpreting research the lag was a core measure. The difference between experts and trainees showed up in the length and the stability of the lag.

Here is how an ordinary learner brings the concept over. Do not respond immediately to what you heard; wait a set amount of time, then render it. Listen to English audio and summarise it in Korean three seconds later, or read a Korean sentence and say it in English three seconds later.

Those three seconds do two things. First, they force a break in word-by-word correspondence. Respond immediately and you convert the words you heard as they came; wait three seconds and meaning survives rather than form. The concept of deverbalization from the earlier post is aimed at exactly this point. To stress it again, deverbalization is a theory, not a finding established by experiment. What is proposed here is proposed not because the theory is true, but because the constraint of delay itself changes the conditions of practice.

Second, they load memory. During those three seconds you have to hold on to what you heard. There is no guarantee that this load amounts to training. It is, however, a more interesting condition than repetition with no load.

I recommend varying the delay rather than fixing it. One second is easy, five seconds is hard. The point where it becomes hard is the point near your current limit.

4. Chunking and open syntax: cutting sentences short and joining them up

I think this is the most practical item in the series.

The most common failure when rendering Korean into English is trying to translate a whole sentence at once. If one Korean sentence is long, the English sentence gets long too, relative clauses pile up, and it collapses midway.

Interpreters do the opposite. They cut a long idea into small units and join short sentences with coordinating conjunctions. And, but, so are enough; subordinating conjunctions like despite or although are risky in impromptu speech because they lock in the structure ahead and behind. This explanation follows the guidance of Sophie Llewellyn Smith, an AIIC interpreter.

Practising alone works like this. Take one Korean paragraph and draw a slash at each unit of meaning. One unit is a chunk that could stand as an independent utterance. Then say each unit aloud as a short English sentence. Finally, try joining those sentences with conjunctions.

What the corpus research tells us is the observation that under heavy load interpreters lean more on formulaic patterns. Reading that result as "memorise the patterns and you are set" is over-interpretation. That study was a corpus analysis rather than an experiment with participants, and the authors themselves wrote that the explanatory power of the model was low. What is to be taken here is an observation, not a prescription.

5. Noting the idea, not the word

The first of Rozan's seven principles of note-taking is to note the idea and not the word. The original explains it with the example that if you give the same French text to ten excellent English translators, you get ten good translations but the words actually used are all different across the ten. What matters is not the translation of words but the translation of ideas.

For an ordinary learner this applies directly to sitting in a meeting or a lecture. Try to transcribe the sentences you hear and you always fall behind, and once behind you miss the next sentence. Note instead the gist of each chunk in two or three words as it ends, and you keep pace.

There is no need to bring in verticality and shift. Those are a page design for the special job of consecutive interpreting, and they take time to learn. Taking just the first principle is enough.

Let me be honest about the evidence. I could not find a comparative experiment showing that Rozan's system is better than other ways of taking notes. The reason I recommend this item is that the primary source is unambiguous and the cost of doing it is zero, not that its superiority has been proven.

6. Output under time pressure

Gile's tightrope hypothesis holds that because interpreters work near the saturation point of their processing capacity, they break down at small fluctuations regardless of the difficulty of the source text. What he observed was that the same expert, interpreting the same speech a second time, made new errors in passages they had got right the first time.

What this model tells a learner is not the ceiling of their ability but its range. What you can do with room to spare and what you can do under pressure are different abilities. And the real situation is always the latter.

So there is a point to putting pressure into practice artificially. The method is simple: remove the preparation time. Constraints like speaking without stopping for 60 seconds as soon as you draw a topic, starting a sentence within 3 seconds of hearing a question, or leaving the recorder running and not doing a second take.

The evidence class for this approach is theory and model. This post is not presenting experimental evidence that pressure training is effective. That practising only without pressure tells you nothing about performance under pressure is, however, a matter of logic before it is a matter of evidence.

How to design practice conditions is covered in more detail in How to Improve Your English Without the Situations. If this post is about what to carry over, that one is about the conditions under which to do it.

7. What does not carry over

This is the part most learning-method pieces stay silent about.

First, selection. The AIIC document states that applicants must pass an aptitude test before admission, and Geneva FTI requires an oral entrance examination. The student body of an interpreting graduate school is a group already filtered twice. Using their methods does not amount to having passed that selection.

Second, volume of input and time. The Geneva FTI master's is a three-semester full-time programme, and students do group practice under regular supervision from teaching assistants and get placement opportunities at the international organisations in Geneva. That is a different world from the conditions of an employee finding one hour a day. This is a matter of structure, not of will.

Third, feedback. The AIIC document states that teaching staff must be practising conference interpreters with a language combination recognised by AIIC, and must have had training in interpreter pedagogy. You cannot get feedback at that level in solo practice. Listening back to a recording can substitute for part of it, but errors you do not know about you still will not know about on the second listen.

Fourth, the goal. Interpreting is a profession of rendering someone else's words accurately. Most learners want to say their own things. There is no guarantee that training which demands accurate correspondence helps self-expression directly.

8. What remains when the theory wobbles

The theories covered in this series are in poor shape. The explanatory power of deliberate practice halved in the course of replication, the cognitive advantage of interpreters splits by sub-ability, deverbalization is only a theory, and shadowing is contested.

Yet these theories point in common to one practice: producing under constraint. Rendering with a lag, cutting sentences and joining them, speaking for 60 seconds without stopping, all of it is imposing a constraint and generating speech.

This post has not proven that practising production under constraint is effective. But when several approaches that started from different evidence converge on the same practice, that practice has a chance of outliving the individual theories. This is a judgement, not evidence. Please read it that way.

9. A four-week framework

The table below is a proposal, not a prescription. No item in it promises a result.

Two things to watch. First, changing the item every week is an arrangement to reduce boredom, not a claim that this is the optimal order. Second, if you try to judge for yourself what has changed after four weeks, you will mostly be deceiving yourself. Better to keep the recordings from week 1 and week 4 and compare them.

When this advice does not fit

Here are the cases where the proposals in this post do not fit.

First, when a test score is the goal. What is covered here is training for impromptu output, and a test with a fixed format has its own preparation.

Second, when reading and writing are the main need. Most of the items in this post assume real-time speech.

Third, when basic vocabulary and grammar are still thin. Chunking requires something to cut, and time pressure becomes frustration rather than pressure. In this case increasing input is more natural than applying pressure. The basis for this judgement, though, is my experience and common sense, not citable research.

Fourth, when speaking anxiety is severe enough to affect daily functioning. The meta-analysis seen in the earlier post reported a moderate negative correlation between anxiety and achievement, but whether it is causal and in which direction is not settled. So it cannot be asserted that applying heavy pressure helps. In this case it is safer to break the exposure into very small increments.

Fifth, when you already have a method that is working. The conclusion of this series is not that a particular method is superior, but that most writing claiming superiority is short of evidence. That conclusion applies to this post too.

Closing

One thread runs through all four posts: let us separate what can be checked in documents from what merely gets repeated.

What remains after that work is modest. What an ordinary learner can take from interpreter training is a few practice conditions, the evidence for those conditions is not strong, and they are merely cheap. What cannot be taken is the selection, the time, and the feedback structure, and these are not the kind of thing effort can make up for.

Even so, I think this sorting has a use. Someone who knows what has not been verified can change method when it does not work. Someone who believes it has been verified blames themselves. That difference is why I wrote these four posts.

• Previous post: What People Good at Languages Have in Common, and How Much of It Is Verified
• Back to the start of the series: What Interpreters Actually Train
• Designing practice conditions: How to Improve Your English Without the Situations

Shadowing and delay practice can be done at /tools/shadowing, and impromptu output practice at /tools/conversation-lab. To break the four-week framework into stages and manage it there is /tools/skill-path, and to attach repetition to something like a game there is /tools/language-quest.

References

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

• AIIC, Conference interpreting training programmes: best practices (aptitude test, teaching staff requirements, final assessment): https://aiic.org/company/roster/companyRosterDetails.html?companyId=13451&companyRosterId=120
• University of Geneva FTI, MA in Conference Interpreting (three semesters, entrance examination, supervised group practice): https://www.unige.ch/fti/en/enseignements/ma-interpretation
• Christopher Guichot de Fortis (AIIC), A Guide to Shadowing (lag of 0.5 to 5 seconds): https://www.guichotdefortis.com/blog/guide-to-shadowing
• Hikmat & Hasan (2022), A study on the effectiveness of shadowing on SI performance at the university level, International Journal of Health Sciences 6(S9): https://sciencescholar.us/journal/index.php/ijhs/article/view/13349
• Sophie Llewellyn Smith (AIIC), An exercise to practise salami technique (open syntax, coordinating conjunctions): https://www.theinterpretingcoach.com/an-exercise-to-practise-salami-technique/
• Huang, Li & Guo (2023), Chunking in simultaneous interpreting, Frontiers in Psychology, DOI 10.3389/fpsyg.2023.1252238: https://pmc.ncbi.nlm.nih.gov/articles/PMC10502168/
• WASCLA Language Access Summit VII (2011) proceedings, containing the English translation of Rozan's seven principles: https://apps.wascla.org/library/attachment.204723
• Gile (1999), Testing the Effort Models tightrope hypothesis in simultaneous interpreting, HERMES 12(23), 153-172: https://tidsskrift.dk/her/article/view/25553
• Yang, Li & Lei (2025), Multimodal processing in simultaneous interpreting with text, PLoS ONE, DOI 10.1371/journal.pone.0326527: https://pmc.ncbi.nlm.nih.gov/articles/PMC12225871/
• Botes, Dewaele & Greiff (2020), The Foreign Language Classroom Anxiety Scale and Academic Achievement, The Journal for the Psychology of Language Learning 2(1): https://www.jpll.org/index.php/journal/article/view/botesetal
• Macnamara & Maitra (2019), The role of deliberate practice in expert performance, Royal Society Open Science 6:190327: https://pmc.ncbi.nlm.nih.gov/articles/PMC6731745/