What Makes a Translation Good?

A central question in translation, and indeed in any exchange of services for consideration, is how we determine whether the end result is ‘good’. This is a particularly thorny question in our industry because, in many cases, the customer is not in a position to assess the quality of the translation themselves. They must instead rely on the translator’s credentials and reputation, or, in the case of an agency, trust that the right person has been selected to do the work. Often, the answer seems to lie in the translation ‘brief’: if the translator has fulfilled the requirements set out in the brief, the translation can be considered successful. But in reality, many translation projects begin without a clear or comprehensive brief. Instead, translators are given a handful of familiar requirements—accurate, faithful, natural—and are left to determine what these mean in the context of the particular text.

Yet other factors such as style, tone, audience, context, purpose and constraints can be just as important, even when they have not been explicitly specified.

A good translation, then, is not simply one that finds the right equivalents, but one that makes and justifies the right decisions for the situation in which it will be used.

The problem of judging quality

Perhaps the first problem is that the apparently straightforward criteria we use to describe a good translation are not as straightforward as they seem. What does it mean for a translation to be ‘accurate’? Accurate in terms of the words? The underlying meaning? The information conveyed? The effect on the reader? What does it mean to be ‘faithful’? Faithful to the wording of the source, to its meaning, to the author’s intentions, or to its function? And who gets to decide whether a text sounds ‘natural’?

These questions point to a more fundamental problem: there is no single, objective measure of translation quality that can be applied in every situation.

This does not mean that translation quality is simply a matter of personal opinion. Some translations are clearly better than others, and there are certainly errors that can be identified and corrected. But assessing a translation involves more than counting errors or comparing individual words with those in the source. A translation can be grammatically flawless and still be inappropriate for its intended audience. Conversely, a translation can depart quite substantially from the wording or structure of the source and yet be the better solution for its particular purpose.

The translator is not the only person bringing assumptions and expectations to the process, either. The reader does so as well. Our understanding of language is shaped by our cultural background, experience and expectations, and these inevitably influence how we interpret a text.

The question, therefore, is not simply whether a translation is ‘good’ in the abstract. It is whether it is good for a particular purpose, audience and context.

Translation as decision-making

These considerations inform the translator’s decision-making throughout the process. Some decisions may be highly conscious; others become almost instinctive with experience. But faced with a particular text, the translator may need to ask:

None of these questions has a universally correct answer. The appropriate choice depends on the particular translation situation and, crucially, on what the translation is intended to achieve.

Take, for example, a commercial contract. Here, a relatively literal approach may be appropriate because the precise relationship between terms and concepts matters, and introducing a seemingly natural equivalent can create a different legal meaning. A translator may therefore resist the temptation to make the text more elegant or idiomatic if doing so risks obscuring a distinction that exists in the source.

Ambiguity presents a similar challenge. If the source text is deliberately ambiguous, the translator should generally preserve that ambiguity rather than resolving it according to their own interpretation. At the same time, if the translator encounters an apparent error, contradiction or other problem in the source, this should be flagged to the client rather than silently ‘corrected’. The translator’s role is not to make unilateral decisions about the meaning of a legal instrument that could have substantive consequences.

A marketing webpage presents a very different set of priorities. Here, the purpose is not simply to convey the informational content of the source text, but to engage a particular audience and, ultimately, persuade them to take some form of action. A more dynamic approach may therefore be appropriate. A metaphor that sounds awkward when translated literally might be replaced with a different expression that creates a similar impression. A slogan may need to be substantially rewritten. Sentences may be shortened, information reorganised, and culturally specific references adapted so that the text feels natural and compelling to the target audience.

The source text has not necessarily changed. What has changed is the translation situation.

From equivalence to function

The tension between these different approaches has occupied translation scholars for decades. One traditional way of thinking about translation quality is through the concept of equivalence: the idea that a translation should correspond in some meaningful way to the source text.

But equivalence itself can take different forms. At one end of the spectrum, a translator might seek close correspondence with the form and structure of the source. At the other, they might prioritise the communicative effect or function of the text in the target language.

The distinction becomes particularly important when a literal rendering and a functionally effective rendering pull in different directions.

Functionalist approaches to translation, most notably Skopos theory associated with Hans Vermeer, placed greater emphasis on the purpose of the target text. The Greek term skopos means ‘purpose’, and the theory challenged the assumption that the source text must always be the primary determinant of translation decisions.

This idea has sometimes been described as the ‘dethroning’ of the source text. The point is not that the source text becomes irrelevant, but that its relationship with the target text needs to be understood in relation to the purpose of the translation.

This provides a useful way of understanding why the same source text can legitimately be translated in different ways for different purposes.

However, this does not mean that the translator has complete freedom to do whatever they like. Christiane Nord’s concept of loyalty adds an important qualification: translators have relationships and responsibilities towards the different parties involved in the translation process. They may have obligations to the source-text author, the client, the target audience and, depending on the context, other stakeholders.

These interests do not always align.

Negotiating competing demands

This is perhaps where the translator’s expertise is most apparent. Translation is rarely a matter of applying a single rule. Instead, translators must negotiate between competing requirements and decide which considerations should take priority in a particular situation.

A client may want their brand voice preserved, while the target audience may require a more culturally adapted text. A source author may deliberately use an unusual expression, while the target readership may find a literal rendering confusing or unnatural. A word limit may prevent every nuance of the source from being retained. A legal or regulatory requirement may demand terminological consistency even where a more varied style would be preferable.

Sometimes the competing demands are not merely practical or linguistic. They may involve cultural, ideological or ethical considerations.

What terminology should be used when different communities prefer different terms? Should culturally specific language be retained or adapted? What should a translator do when a client’s preferred terminology conflicts with established usage in the target language? Should potentially problematic language in the source be reproduced, explained or modified?

There are no universal answers to these questions either. What matters is that the translator recognises the competing considerations, understands the consequences of the available choices, and makes those choices deliberately and professionally.

This is why it can be useful to think of the translator not simply as someone who transfers meaning from one language to another, but as a cultural mediator. The translator is mediating between texts, languages, cultures, institutions, clients and audiences. Their task is to make decisions that allow the target text to function effectively within its new context.

So, what makes a translation good?

We can now return to the original question.

A good translation is not necessarily the most literal translation. Nor is it necessarily the most fluent, the most elegant or the one that most closely resembles the source text. Those qualities may be important in some situations and less important in others.

Instead, a good translation is one whose choices are appropriate for its purpose, audience, context and constraints, and which successfully fulfils the function for which it was commissioned.

This also changes how we think about the translator’s role. Their expertise does not lie solely in knowing two languages or being able to find equivalent words. It lies in understanding what the text is doing, recognising the factors that may affect how it should be translated, weighing up competing possibilities, and making informed decisions.

In this sense, translation quality is inseparable from professional judgement.

The question is therefore perhaps not simply Is this translation accurate? but:

Are these the right decisions for this translation?

That is a harder question to answer, but it is also a much more meaningful one and provides a more useful yardstick by which to measure the quality of the translation.

A final question: can we automate the decisions?

Given the advent of AI, this is an increasingly important question. There are many ways in which AI can already support the translation process: checking spelling and consistency, identifying omissions, researching terminology, applying style guides, suggesting alternative formulations and generating draft translations, to name just a few. AI can also be remarkably good at producing plausible solutions to individual translation problems.

But producing a plausible solution is not necessarily the same as making the right decision. Many of the questions raised above depend on understanding the purpose of the text, the expectations of the client, the needs of the target audience and the wider cultural and communicative context. Should this ambiguity be retained? Is this metaphor important enough to preserve? Does this cultural reference need to be adapted? Is the author’s unusual style deliberate? When two perfectly viable translations are possible, which one better serves the purpose of the text?

These are not simply linguistic problems. They are questions of judgement. And while AI can increasingly help us identify options, compare them and even recommend a solution, the responsibility for deciding which solution is appropriate remains a more complex matter.

Perhaps, then, the most valuable role for AI is not to replace the translator’s decision-making, but to augment it: taking care of routine checks, generating possibilities and surfacing potential problems, while allowing the translator to concentrate on the decisions that require contextual understanding, professional judgement and negotiation between competing priorities.

After all, if a good translation is not simply a collection of correct equivalents but the result of a series of appropriate decisions, then the real question for AI is not Can it translate? It is

Can it understand what this particular translation needs to achieve, and make the right choices accordingly?

That is a much more difficult question. And perhaps it is the one we should be asking.