How is translation quality defined?
For internationally active companies, translations are not only the key to gaining a foothold in important export markets and reaching their customers, they also fulfil important legal and safety-relevant aspects, especially in the field of technical documentation.
High-quality translations are a basic requirement – because the consequential costs and effects of poor quality are far-reaching and can range from damage to the image to loss of customer confidence, administrative costs, loss of profit and loss of warranty, and even risk to life and limb.
But the question of what constitutes high-quality translation cannot be answered simply or in a general way, because quality requirements are individual. What constitutes good quality for one company is already insufficient for the next.
The reason for this is the different requirements – linguistic, legal or formal – to which companies are exposed depending on the industry and the degree of regulation. It is precisely here, however, that the standards of quality are found, paving the way for clients and language service providers to meet these demands in the long term.
Starting point: know requirements and define quality
High translation quality can only be achieved if all parties involved have the same understanding of what requirements need to be met and can thus work towards a common quality goal. In order to do this, it is necessary to identify the specific requirements of a translation, which can be characterised in different ways:
- sector-specific, e.g. legal requirements, guidelines
- text-related, e.g. certain text-type conventions
- individual according to customer needs, e.g. company-specific vocabulary, target group
It is important to establish a common understanding of these requirements on this basis right from the very start. Close collaboration and continuous coordination between the client and the language service provider is essential from the very first step for effective quality assurance and long-term satisfactory quality. The jointly agreed requirements are the decisive criteria for this.
Translations must therefore have certain predefined characteristics in order to meet individual quality requirements. This can start with linguistic, content and technical correctness and extend to the appropriate language register, desired deviations from the source text, target group-specific localisation and aspects such as format and text lengths. Deviations from these characteristics, for example in the form of errors in meaning, incorrect spelling, stylistic problems or inappropriate terminology, are then clearly recognisable quality defects.
This view focuses on quality at the product level or on translation as such. It is important to bear in mind, however, that this is only one way of looking at quality, because the quality of a translation is just as significantly influenced by the translation process and by the translation producers. Delivery reliability, response times, queries, translator qualifications etc. – the quality of a translation can also be defined, checked, evaluated and optimised with regard to these points.
Such comprehensive consideration is recommended right at the beginning of a cooperation in order to optimally identify the framework conditions and workflows.
In all cases, it is essential that there are concrete agreements between the client and the language service provider which clearly define and describe the respective requirements: style guides, glossaries, reference materials or other specifics ensure that work can be done on a common basis with the same vocabulary and rules that replace subjective value judgements with objective and measurable criteria.
Don’t lose sight of the path: check translations
Quality is decided even before the first keystroke of the translation. The first question for clients and language service providers should therefore always be whether the translators have all the information they need to meet the requirements – and whether they have all the necessary skills to do so. Which assumes, of course, that these requirements also exist in an understandable way and are known to all those involved.
To avoid deviations or to detect them during an audit, it is first necessary to establish basic standards from which deviations can be made, e.g. specifications from style guides or company terminology. Well-functioning query management, clearly formulated order emails and the like are also important here.
A translation can be checked at numerous points. Depending on the requirements and evaluation criteria, different quality assurance processes are useful and necessary. For example, the user-friendliness of a translated user manual can only be improved to a very limited extent through revision and usually requires a user test. These user tests, in turn, cannot make any statements about whether customer-specific formats or legal requirements have been adhered to, as this is not their aim either. To be truly meaningful, such tests must also be carried out by the client.
In any case, it is important that it is clear which requirements apply to which quality objective. Multiple test loops do not automatically mean a higher quality, but on the contrary can even have a negative effect on the overall result due to friction losses from over- or under-correction.
Here, too, the well-known principle applies: as much as necessary, as little as possible. In addition to the revision and the user test, there is also, for example, the technical review, proofreading or galley proofing as well as the second and third revision, each of which has a completely different focus.
Audits are usually carried out by human auditors, but they can now also be automated. Various tools that are either already integrated into common CAT tools check certain predefined criteria and detect potential errors, for example in terminology, punctuation, text length, tags or consistency. Style, accuracy or even meaning, on the other hand, cannot yet be checked automatically. Also, due to an often very high number of false positives , the results must always be critically cross-checked. The significance of automatic tests is therefore fundamentally limited.
Far from subjectivity: evaluate translations
If these checks serve in principle to detect errors, the aim of every evaluation is to assess and contextualise these errors. In practice, certain evaluation models or metrics have become established for this purpose, which aim to objectify the process.
Well-known models include SAE J2450, the LISA QA model and the DQF from TAUS. The basic assumption is very simple: no or few errors in the defined error categories certify good quality. To this end, certain pass/fail thresholds are set – a kind of boundary line – which define a minimum quality level that the translation must meet or exceed.
As a general rule, evaluation processes such as review processes should always be adjusted and fine-tuned to the individual requirements in order to be able to make a statement as to whether the respective translation achieves the quality objective. Here, for example, a higher-level criterion can also be the differentiation according to text type or test objective – keyword user-friendliness – for which different error categories or types become relevant. However, only a few predefined metrics allow such a degree of customisation, but can provide basic building blocks for creating your own assessment scheme. Newer models such as the DQF or MQM already follow such a modular principle.
The error analysis based on metrics and error categorisation is, in principle, carried out at sentence level and in this way divides the translation into units to be evaluated separately. This can mean that a translation rated pass or “good” by metric still does not meet the needs of a particular target group. And that the rating of errors, viewed on the whole translation, is disproportionate because it is relativised by the overall impression. The length of the evaluated text can also pose an issue, for example, if the entire translation cannot be evaluated for economic reasons and there is ultimately only one cross-section is evaluated. Furthermore, most metrics only recognise a negative perspective – the lack – and neglect positive aspects, such as commendable solutions by the translators.
Go further: evaluate and use evaluation data
Evaluation is a fundamental way of determining whether a translation has reached a certain threshold value and thus exceeds or falls short of the set quality level. However, it is always advisable to look beyond this value and to include all the data obtained in the quality consideration to make full use of the time and costs involved in the evaluation process.
Where do which errors occur? Which errors occur most frequently? By whom or when do they occur? Is the evaluation really objective? Do the evaluation criteria make sense? These and similar are questions to which the evaluation values provide answers. This in turn can be used to derive sensible measures for optimising quality processes and raising the overall quality level.
This is possible, for example, by making source texts translation-friendly, providing relevant or clear reference material, creating terminology databases, new instructions, revised style guides, introducing controlled natural language, training translators or adjusting settings in computer-aided translation and quality assurance tools.
A kind of conclusion: once you reach the end, the cycle starts again
A perfect result in high translation quality is by no means a matter of course. However, the way there can be mastered well through good preparation and a forward-looking view of changing conditions, clearly defined goals and structured processes – which nevertheless need to be questioned critically. Again and again. Because every translation has its own requirements, you know that now.
Do you want perfect results of the highest quality for your translations? Then talk to our experts about your specific requirements and goals. This is something we are very happy to help you with.