Translation quality management can be a rather confusing subject due to the lack of unique and standard terminology. For example, in ordinary quality management (i.e. ISO9000s), quality assurance (QA) is “part of quality management focused on providing confidence that quality requirements will be fulfilled” (ISO 9000:2005, 3.2.11), and focuses on preventing errors and defects. Quite differently, what is generally called QA in the translation industry is in fact a quality control (QC) task, very similar to the one that used to take place in an industrial setting, although with some major differences that we’re going to deal with later on. Therefore, when we talk about translation quality management we need to take into consideration two different strands of quality: process-related and product-related (linguistic quality).

Process Quality

For quality assurance, we can choose to apply one of the many standards available, starting with the international ISO 17100,  in the USA, or the Canadian CAN/CGSB-131.10-2008. Other specific translation-related standards have been recently devised, from ISO 11669 for general guidance on translation projects, to ISO 18587 on post-editing. There are also various national standards applicable to interpreting services.

The first standardization attempts date back to 1996 with the Italian UNI 10574 standard describing the requirements for translation and interpretation services. The Italian standard was followed by the Dutch ATA Taalmerk standard in 1997. Other attempts were made in Germany (DIN 2345), Austria (ÖN D1200), the USA (SAE J2450) and China (GB/T 19682 and GB/T 193636.1). Another effort is the ISO 21999, still ongoing, which is centered around models and metrics for translation quality assurance and assessment.

Linguistic Quality

Moving on to linguistic quality, we find ourselves face to face with two main problems, the first being that QA (or better said, QC) is still based on marking errors and the second that the definitions of error categories remain inaccurate. Concepts like accuracy, fluency and adequacy are rather academic definitions that leave a lot of space for subjective interpretation. What are, for example, the errors that influence each category? To what extent? What errors do we need to identify? How should we weigh them? There isn’t one single method that can help us describe, identify, and evaluate errors. This means, in short, that two different evaluators could produce two different reports.

As mentioned at the beginning of this article, the current linguistic QA approach is very similar to old-fashioned manufacturing QC, where every single step and every single product were checked thoroughly. Most industrial sectors have already abandoned this so-called 100% full-inspection methodology in favor of a sampling inspection and have a clear sampling standard in place. They also started to follow programs like Capability Maturity Model Integration (CMMI) or Six Sigma to guide process improvement across a project and the entire organization.

A holistic approach to translation quality

So, how can we switch to a more holistic approach to translation quality that spans the whole process? Here are few suggestions.

• Integrate quality assurance and quality control. ISO 9000 standards provide language service providers and translation departments enough guidelines for a process-based approach offering a certain level of quality.
• Integrate all controls within the translation process. To this end, an excellent starting point is the TQM approach based on W. Edwards Deming’s philosophy (“Getting it right the first time every time”). Processes must follow a repetitive cycle, just like in the now-popular agile methodology. One of the main points of the agile methodology to keep in mind is that mistakes happen: we might not be able to avoid them, but while working on small iterations, mistakes can be easily identified and eliminated.
• Put agents (automatic functionalities) in place within a translation project. These can help identify and notify errors like discrepancies with the term base or the translation memory, empty segments, target segments containing the source text, and other issues.
• Reduce the number of steps. With more collaboration and working with smaller batches, quality control activities can take place almost at the same time as the translation.
• Give all the people involved in a translation project a clear indication of what the final result should be. For example, you should define the end use of a translation and set a clear quality threshold. A project could have a 98%-75% quality threshold or even a low//high precision level. Translation management platforms can be useful in this respect. The project manager can set the quality threshold and, during the translation process, they can keep an eye, for example, on false positives or actual errors and intervene as necessary.
• Set rules to determine when revision is necessary as well as the necessary level of revision within the quality control steps. If you decide to skip the revision stage partially or entirely, you’ll have to set clear rules to be followed by all persons involved. The revisor will know what to look for and the quality threshold will help them speed up their task.
• A major role for machine translation. Two revisors might produce two different evaluations of a translation. If the MT output is subject to automatic post-editing before the actual revision, the text will have reached a certain level of internal consistency before it reaches the post-editor’s screen.

Key to all these steps is to have good data at your disposal, both linguistic data and process metadata. In particular, the more process metadata you have for each project, the more and better elements of information will be available to your project managers. And in the end, every piece of information will contribute to your data repository, your personal gold mine when it comes to define your company’s business analytics and KPIs.