Translation Quality Assurance

At Palex, the opinion of who is responsible for translation quality assurance changed several times over the last few years. At first, the translator was the one to get all the scolding after a client sent back bad reviews. After all, the translator is the one who didn’t choose the right terms for financial market transactions, house plants, or medical instruments.

Later, the responsibility was partially shifted to the project manager, because they are the ones that choose the people to work on the project. As such, if the translation is bad, the PM is responsible for putting a linguist on the job who is unfamiliar with the subject matter. This system seemed logical and existed at Palex for several years.

Project managers would react to customer complaints, but their work couldn’t be called proper quality management. They could even refuse to work with a vendor after receiving bad feedback from the client, though the problem may have been solved by creating a glossary of a dozen words. On the other hand, the PM would spend hours drafting a fifty-page instruction manual, while all they had to do was find a new translator.

It wasn’t that the PMs were incompetent. They just didn’t have time to carefully analyze client requests and feedback — as we all know, the daily life of a PM is filled with translator CAT tool issues, a pile of client instructions to go through, and hot deadlines to meet. This routine didn’t give time to sit down and think about how to avoid mistakes.

This is when I became the head of the Palex translation department in 2014. I saw that I had to come up with a way to control quality of all translations going through the agency, which is when I got an idea to completely remove the responsibility for quality assurance from the PMs and shift it to the people for whom this would become the main daily task.

There are three components to every project: cost, deadline and quality. When I started at Palex, project managers were responsible for all of these. But a PM usually is much more concerned with time and money than with quality. If the translation quote is too expensive, the client will refuse to work with us, and we'll start to lose money. If the deadline isn't met, we'll have to make excuses to the client or pay fines. On the other hand, we don't see the consequences of poor translation quality right away. When a client complains about wrong terminology used, we think that it's enough to just change it in the text and redeliver. But if we don't do anything else, mistakes will be repeated, and the client will get more and more dissatisfied.

We tried to balance these components in the hands of one person for several years, because the company needed a universal translation quality control system. But every manager was busy with their projects and dealt with the consequences of mistakes as they came — they rarely had any time to concentrate on problem prevention. We needed to fix this, and to do so, made the radical decision to put QA under the responsibility of another type of specialists — Linguistic Leads.

Linguistic leads at Palex became highly experienced editors, who could tell the difference between translating marketing material and educational texts, and why it’s more difficult to localize software than translate manuals. We picked the best staff linguists with an analytical set of mind, who could solve non-standard problems. These new specialists needed to not only understand why a translator chose the wrong terminology, but what needed to be done to avoid this issue in the future.

We synthesized a conflict between PMs and linguistic leads. The former strived to complete projects faster and cheaper, while the latter didn’t allow for the quality to fall below a set norm. At the same time, the PMs contained the linguistic leads and didn’t let them get lost in the details, overcomplicating production processes.

Linguistic leads started to battle one of the biggest reasons we had issues with quality — using "universal" translators. Before implementing the new QA process, Palex had one person translating tomograph manuals, printer user license agreements, and diaper advertisements. This was happening because PMs were apprehensive to work with new linguists and gave all the projects to trusted vendors, not paying much attention to their specialty.

Now Palex linguistic leads go through all new projects and figure out what we need to translate. There are two tasks they had in this process: classifying the content and assessing risks. Content classification helps choose the linguists who will definitely do a good job translating this specific text. Risk assessment determines what kind of preparation is needed before the project begins.

For example, if the linguistic lead sees that there is a lot of specialty terminology in the text, they recommend creating a glossary first. Based on such text analysis, linguistic leads recommend a sequence of steps to the PMs and prepare a list of suitable vendors.

An approach with an analysis of all texts works well for separate requests but isn’t very useful for the daily influx of requests. We have four linguistic leads but thousands of projects. The leads can’t physically go through everything we get from the client.

So we got a new goal — we needed to create a system in which project managers would quickly find a recommendation for a job, even if it had to be delivered in an hour, or if the request came in late at night.

A solution was found in creating project cards, which are documents where we keep all the information on requests of specific types: from links to glossaries and style guides, to file-sharing sites passwords. Linguistic leads add the workflow that should be followed and recommended vendor lists to project cards, as well.

Workflow is the sequence of steps that each type of content would need to go through for a quality translation. For example, a project card may say that marketing texts have to be transcreated and copy-edited, while reference materials go through MTPE and proofreading steps.

Linguistic leads create lists of recommended vendors for each workflow. This list mentions which of the vendors should be used as translators and which as editors. Each vendor has a status: main, backup, candidate, or reject. Based on these, the PMs understand whom they should contact first and who won’t do well on the project.

Look at the list of recommended vendors for projects of one of our IT clients. The marketing brochure and user manual will be translated and edited by different translator-editor pairs.

Project cards saved the linguistic leads from the necessity to analyze every client request. Now PMs just find the needed project card, choose the appropriate workflow and open the list of recommended vendors. It takes several seconds to get all of the information required to kick off a project.

Now all standard requests are completed without linguistic leads, who are involved only when there’s either a new client, a new language pair, or a new subject matter in the project.

No matter how good our project cards are, there will still be mistakes made. Clients will find them and send them back with feedback. Before implementing the new process, feedback processing stopped at the project manager — they sent requests to the linguists, who corrected the text, which was then sent back to the client.

To improve translation quality, we needed to understand the reasons why mistakes were happening and find the best solution to avoid them in the future. This responsibility was passed on to linguistic leads.

If a client complains about semantic errors in the translation, we need to change the editor. If there are too many typos or grammar mistakes, the proofreading step should be added to the workflow. If a translator keeps on using the wrong term time and again, there’s a need for a glossary.

Based on this analysis, our linguistic leads update workflows and recommend vendor lists in project cards, as well as instructions and glossaries. Sometimes, we hold Quality Calls to tell our vendors if there are new and dramatic client complaints.

Over the last several years, we’ve changed how we approach translation quality management quite a bit. Back in the day, we blamed the translators for all the mistakes. Today, linguistic leads set up workflows and project managers take care of the financial side of the job and meeting deadlines.

This system works and helped us get rid of many problems. We continue to evolve it, and I already know what needs to be done in the near future.

Training program for linguistic leads. Palex’s first linguistic leads were our staff translators. They were the ones that translated most of the projects and they knew all company processes. Their experience helped them see where we were making mistakes all the time and what we needed to do to improve. But we ran out of staff linguists with an analytical mind and project manager qualities quite fast. So we had to hire people who didn’t know anything about our company. Based on this analysis, our linguistic leads update workflows and recommend vendor lists in project cards, as well as instructions and glossaries. Sometimes, we hold Quality Calls to tell our vendors if there are new and dramatic client complaints.

There are two linguistic leads "from the outside" at Palex. We were able to immerse them in our production processes and we are happy with their work. But their training took a long time, because many members of our staff had to spend hours on telling the newbies how everything works in the company. We can’t afford spending this much time and energy on training new people.

So now we document all the useful information that’s necessary to train linguistic leads. What we are getting in the end is a full-blown training program, making it so that our next linguistic lead won’t bother their mentor for smallest things.

Regular vendor assessment. We assess some of our vendors too frequently, while others — almost never. This happens because Language Quality Assessment (LQA) remains to be the tool of project managers, who use it as part of their requests, choosing to check the quality of a translator’s work whenever they see fit.

An incomplete picture of the quality of work of vendors is the reason for mistakes in workflows and recommended vendor lists. To minimize these mistakes, we plan to regularly assess all our active linguists, shifting the responsibility for LQA to linguistic leads. Now they will be the ones to decide who and when should be assessed.

Technologies. Linguistic leads consider a wide range of parameters when choosing any specific vendor: from education to quality of prior work completed and PM reviews. This information about hundreds of linguists must be constantly updated. Today, we have so much information on our vendors that we use technology solutions to quickly and efficiently work with the massive amount of information. We’ve been gradually moving away from Excel and SharePoint and using our TMS more and more.

At Palex, we’ve divided the "project triangle" and included linguistic leads into our production processes, giving them practical tools that help them manage translation quality:

• Content analysis for new requests and recommendations for PMs;

• Standardized workflows and vendor lists for regular requests;

• Correction of workflows, vendor lists, instructions and glossaries based on client feedback.

With that said, we knew right away that project managers and linguistic leads will have very different goals, which would create a healthy conflict between the two. Our employees had to talk to each other and find balanced solutions. PMs can't sacrifice quality to deliver a translation that's much cheaper, and linguistic leads can't insist to have dozens of assessments, which would make it so the delivery of the job goes beyond the deadline. This is how we found that sweet spot between price and quality that is greatly valued by our clients.