[The essay also mentions critically: the four-eyes principle; the use of translation databases; machine translation; and post-editing. The language pair in the foreground is English/German, but much will apply to other language pairs.]

When done properly, translating involves much interpretation. In the case of most business and many technical texts the translator must seek out the underlying messages that need to be communicated to the end user and transfer only as much of the ballast as can be done without obscuring those core statements.

Most writers of both business and technical texts are not expert at writing: their primary competence will normally be in some other field. Or they will have been taught to write badly, i.e. pretentiously.

Texts presented for translation will typically contain repetitions, superfluous qualifications and inconsistencies. Often they contain grammatical errors, and even where these are absent there may be failings in logic and in the flow from one sentence to the next. If these are all reproduced “faithfully” in the translation, then due to the different emphases governing the internal logic of each language the original faults will be magnified, and the effect will be an awkward end product. It will be difficult to read and understand.

The assumption will be that it is the translator who is to blame. One comparison might be with a portrait photographer who by truthfully choosing the brightest and coldest light brings out every blemish on the skin.

There are differences between languages, for example, between German and English, in the level of generality or in sensitivity about time. The preference in English is to be specific, and this is recommended in classic essays and manuals about good style. Whereas German loves an abstraction, English demands something concrete. English has many tenses whereas German mostly uses only two. Hence in the transition from German to English the translator will normally be forced to make assumptions (and sometimes mistakes) about matters that are vague in the source language.

Another obstacle occurs at the level of sentences rather than words: this is the tendency to cram each sentence full.

It will often be possible to put more into a single sentence in one language than another. That is, in one language overloading will not be immediately apparent whereas in the other language it will jar. German can accommodate endless adverbs and substantives, whereas English strings verbs together willy-nilly.

Sometimes it is possible to split a sentence, i.e. make two, separated in English best by a semicolon. Often, though, it will be possible to discard some of what is in the overloaded sentence since it is expressed adequately in an earlier or later passage. In other cases it may be possible, and indeed logically preferable, to place semantic elements from the overloaded sentence in other sentences nearby.

Irrespective of whether written poorly or elegantly, any three or four lines will contain, between the lines as it were, a multitude of implicit statements and assumptions that might need more than a page to enumerate. Mostly, the overlap between languages, especially related languages with similar cultures, is such that this does not matter, and the context does the rest. A translation is not an exhaustive analysis or a forensic exercise.

These considerations demonstrate what is wrong with the much vaunted “four eyes” principle and the reliance on databases (known in the industry as TMs, or translation memories). (Incidentally, the original standard translation for the “Vier-Augen-Prinzip” was the “dual control” principle, applied famously in a military context.)

The idea of the four-eyes principle is that each translation should be reviewed sentence by sentence by another translator. (One could equally argue that every decision made by one doctor should be reviewed by another. Obviously, there will be circumstances where this precaution is appropriate, and others where it functions as an occasional control, or else as a teaching or learning tool. Normally, though, it would represent a needless expense. Crucially, it would involve a dilution of responsibility such that the first doctor – or engineer, or other professional – is less careful and conscientious than they might be, while the second just ticks the box, such that the end result is less – rather than more – reliability.)

Spotting a mistranslation or misinterpretation in a sequence of complex sentences is generally more time-consuming and involves more concentration (that by its nature can only come in spurts) than may be supposed. Indeed, it may be more costly, in terms of effort, and may provide less reliability than translating afresh.

If once it is supposed that semantic elements may have been transposed from one part of a paragraph to another, whereas others have been deliberately left out, or indeed a minor elaboration added in order to convey culturally specific information that goes without saying in the source text but may not be known to the target readership, it becomes clear that serious review is impracticable and mostly disproportionate. The result is that adherence to the four-eyes principle is more pretence than reality.

It is of course, largely, just another way of pulling the wool over the eyes of those charged with procuring translations, especially MBAs and the like who, devoted to their cult of managerialism, imagine that professional procurement is consistent with substantial ignorance of what is being procured and who hold a blind belief in standards, norms, certifications and process.

Translation agencies (“language service providers” – LSPs) drive down their own procurement costs such that, often, they must hire people who cannot yet translate or who otherwise have no motivation or time to deliver more than a minimum, especially as they remain anonymous. The four-eyes principle is an attempt to correct for the inevitable errors.

The spread of machine translation is transforming the so-called “industry” with the creation of a novel skill, namely post-editing, which is quite distinct from that of either translating or proof-reading. Whereas once some translations could be vilified for not being idiomatic, the complaint now may be too much polish at the expense of fidelity to the source text. An end-text may read reasonably smoothly but an attentive reader, e.g. someone who actually needs to understand fully what is being communicated, may notice that something is not quite right. Core messages get distorted while the ballast takes centre-stage.

A further phenomenon should not be ignored. The overzealous proofreader may seek to justify their function by changing wording arbitrarily – i.e. creating the appearance of work done. But this occurs naturally, too.

Three decades ago the professional journal The Linguist printed a cartoon with the words “The Greatest Drive is not Love or Hate, but one person’s wish to Change Another’s Copy”. The word “change” is crossed out and replaced by “amend,” which is turn is replaced by a synonym, then another, ending with “chopped to pieces” being replaced by “change.”

The dual control principle – Vier-Augen-Prinzip – is not observed when the second pair of eyes fails to request consent for – or even inform – the original translator. But this is what is likely to happen when a one-way process is enforced, for example, by an agency. This process then also excludes any learning process for either party. It is by its very nature authoritarian. So much for “continuous improvement” in what used to be known as a liberal profession!

Given that most translation of business and indeed other texts (manuals are another story) is as much about interpretation as about words and grammar, how does one proceed as a translator?

Translation memory software, properly designed and used, is an enormous aid, of course. Apart from much else which is obvious, it facilitates visual focus on the segment (i.e. sentence or sequence of sentences) being translated. The software should make it possible to easily mark or even change the source text (most does not). For instance, one may wish to highlight expressions one has chosen to edit out in the translation, or mark passages that are unclear and need further reflection or consultation. Most current TM software presents the source text in a column to the left of – rather than above – the translation. This is suboptimal and, for those of us with less than perfect eyesight, disastrous.

TM software is an aid only, and it cannot replace intellectual effort. The difficulty is that one thinks differently in different languages, especially as the level of complexity rises. So each translation involves a reconceptualisation in the target language. Ideally, one tries to identify a core statement within each sentence separately to the statements that add qualifying information. The first practical task is to design the sentence structure such that the core statement can be recognized as such immediately, i.e. without conscious reflection. The next task is to arrange the secondary information around this core statement in a way that is not obtrusive, i.e. such that it reads smoothly and the logical links are clear. A third task is to connect the segment with the preceding and following segments. In the source text there will often be connectors (e.g. “however”, “moreover”) that are out of place: they mimic logic where there is none, or where the logic is different (for instance, the word “however” is used although there is no contrast). It is not the task of a business translator to reproduce bad style.

Apart from the reconceptualisation, there is craftwork involved, as in the choice of vocabulary (not: “terminology”), which should not reproduce slavishly the often limited or deviant wording of the writer.

Each author indeed has their own idiolect, i.e. particular words and phrases that they prefer and therefore soon overuse and often misuse. An important aspect of the job of translating is decoding the source text, e.g. recognizing how the author is (usually unconsciously) concealing their message in jargon. Often, what is needed is radical simplification (which may upset the author or the commissioning party).

In choosing vocabulary one should avoid causing even momentary confusion and bear in mind that the readers may not be native speakers.

For instance, the word “provision” has several meanings, and the word might easily appear twice even in a single sentence or paragraph with different meanings. One might use instead “stipulation” and “supply” while retaining the word “provision” for the accounting device. (A good example in German is the word “Anlage”.)

On the other hand, if a single concept occurs frequently, it is stylistically better to vary the expression used if this is unlikely to cause confusion. For instance, “in comparison with the prior year” might be expressed as “year-on-year”, and there are a number of further alternatives. (Best would be to state on each page once, and once only, something like “unless otherwise indicated, here as elsewhere all comparisons are with the prior year period”.

But the German or accountant or legalistic mentality seems to baulk at this superior solution.

These considerations mean that a set of word-for-word equivalences (“literal” translations) that work well for one text may not do so for another.

A further complication is when the author has been deliberately vague for reasons they may be reluctant to disclose; the reasons may or may not be ethically justifiable. Reproducing ambiguity is awkward at the best of times. Often it may be possible to convey it by adding something in brackets preceded perhaps by the word “or”. This will generally be more elegant than the translator adding a comment in [square] brackets or in a footnote.

The advantage of good style, which is a separate matter to clarity, is that it keeps the reader reading and comprehending. This is also an argument against the slavish use of databases of sentence pairs (TMs). Not only does this widespread approach lead easily to disjointed – patchwork – texts, i.e. without any flow; it encourages homogenous sentence structures, the monotony of which soon tires the reader, like a featureless straight road sending the driver to sleep.

Granted, the attitude in many circles where long legalistic reports are produced is that everything needs to be said so that compliance can be claimed; no-one takes account of, or responsibility for, the bigger picture. This is indicative of the absence of professionalism among the very people who are most keen to profess the word. Lip-service suffices.

Contrast this with resort to a single translator, taking responsibility for a text as a whole, and focussed on the needs of users receiving information from a foreign text: one of the best quality-assurance (QA) measures imaginable.



Professional translation of all but the most standardised texts is by its nature individual. Obviously, where there is need for a second opinion, this is something networked freelancers attend to as a matter of course. But splitting up of the process among different persons is likely to generate difficulties of coordination to the detriment of the end product. (Gourmets do not proceed on the principle that a meal cannot have too many cooks!)

Consistency of style and terminology are most efficiently and effectively achieved by a single individual, or at most two or three when the scope or deadline of the assignment necessitate more than one.

In the case of some of the work farmed out by companies, there is an added consideration: Not infrequently, separate sections of a report have been written by different authors, with corresponding inconsistencies of style and terminology. Hence it is recommended that the translation be performed by a single individual who has a sound knowledge of the subject matter and is therefore able to iron out inconsistencies and infelicities.

What does happen is that some translation agencies arrange to have work revised, not least because the price pressure and time constraints they place on sub-contractors inevitably result in inferior work. But this assumes that they have available a revising editor who is capable not only of producing a final text that reads well and idiomatically, but who understands the original and its subject matter so perfectly that they can check that the meaning in the translation corresponds to that in the original.

Whereas the problem thirty years ago was that there were still many translations that read poorly, today the problem is that there are many translations that read perfectly and even reasonably logically, but fail to reproduce the meaning of the original because they have been revised – airbrushed – by someone without the requisite knowledge of the source language and idiom. That is, they express perfectly something that is substantially at variance with the text of the original.



Native speakers in different languages do not simply use different words and grammatical constructions: they think differently, conceptualising even simple matters in markedly distinct ways.

In the case of business texts (as opposed to literary ones), a distinction needs to be made between the semantic core of what is being expressed and the baggage. Inexperienced translators and those without a proper knowledge of the subject area concerned fail to distinguish these correctly, with the consequence that a different meaning or at least a different emphasis is conveyed. By contrast, experienced translators with a robust knowledge of the subject matter automatically revise the texts, removing confusion and repetitions, and even discreetly adding snippets of background information where this will enhance a proper understanding. They improve the style so that the text reads as smoothly and effortlessly as the material permits.


There is a received wisdom that all translations should be performed by persons working into their native language. Two points on this matter: There is a substantial category of persons who are "near-native speakers" and as such are perfectly capable of performing competently translations out of their mother tongue.

Secondly, in some cases such persons may be better placed to grasp the meaning of a source text and to reproduce it articulately than well-qualified translators whose mother tongue is not that of the source language. This applies in particular to convoluted legal texts.

Notwithstanding the above remarks, gold-plating is seldom required. Considerations of cost and time aside, it is the purpose of the translation or, rather, the needs of its readership which should guide us. It is enough if a translation is slightly better than the source text, and this will mostly be the case if the translator is competent.