This wiki provides a detailed guide for translators on handling LQAs effectively, either as a reviewer or in the role of the translator carrying out arbitration.

• LQA: Language/Linguistic Quality Assurance/Assessment (also known as Scorecarding)
• RCA: Root Cause Analysis
• CAP: Corrective Action Plan
• Arbitration: after the scorecard has been completed, the translator gets a chance to appeal any errors.
• LISA QA: a QA model originating from the Localization Industry Standards Association. Many LQA forms we see at Sandberg follow this model.

An LQA, or Linguistic Quality Assurance, is an assessment of the linguistic accuracy and suitability of translated content.

During LQA, errors in the translation are spotted, categorised and assigned a severity level, resulting in an overall translation score. This score determines if the translation is graded as a PASS or FAIL.

LQAs should not be confused with layout checks, QA of our own work (e.g. running QA checks within a CAT tool), or linguistic sign-off tasks (e.g. checking our translations in context).

The purpose of an LQA is:

• To ensure the client’s quality expectations are met;
• To identify and correct possible quality issues;
• To monitor the performance of the translator or translation team and provide feedback.

Many large LSPs or clients use LQAs as an important step in their quality monitoring. They are usually carried out by in-country reviewers who are experts in the end client’s expectations.

They give us the opportunity to:

• Learn the client’s preferred style or terms for future jobs;
• Dispute errors/changes we disagree with;
• Show the client we’ve done our research;
• Revisit our workflows/processes;
• Identify anything more we need from the client.

It can be tempting to see LQAs as just ‘another bit of admin’ or extra feedback, but it is very important to understand the purpose of LQAs within our industry. It is a service the client pays for because it is seen as an objective measure of quality that they rely on to know that they are getting the translation quality they paid for. LQA scores (and even the scorecards/RCAs themselves) are shared with clients, so we need to ensure the best possible outcome.

Some LQAs are scheduled, some are random. You may be notified in advance that an LQA will be performed on your translation, or you may not.

Retrospective LQAs may also be performed several months after the initial translation has been delivered (e.g. once per quarter on larger accounts).

Live LQAs are sometimes run on large projects. This involves submitting a translated and revised sample while the translation is still in progress. This sample should be representative of our final quality.

LQAs should normally be carried out by senior, experienced linguists. The role of the LQA reviewer is not just to flag errors but also to educate the translators. An LQA reviewer should have a similar project-specific knowledge to the translator and act as a true partner to the translation team since the final quality of the translation depends on cooperation between the two parties. A reviewer should have a high level of attention to detail and be able to follow instructions exactly. Most LQA forms have lengthy instructions and definitions tabs, so you should set aside time to read these thoroughly each time you perform LQA for a different client.

Most LQAs are done offline, in an Excel file sometimes referred to as a scorecard or LQA form. Sometimes we may be asked to complete LQAs online via a client portal. LQA reviewers are not usually asked to edit the translated files, but only to copy-paste errors they spot into an LQA form, categorise the errors and their severity level, suggest corrections, and provide the client with a short overall evaluation. The result is calculated automatically in the form. The correct assignment of the Error Category and Severity is crucial to generating an appropriate LQA result. The LQA form is usually the only deliverable from the LQA reviewer.

An LQA reviewer should:

• Be careful of repeated errors! Clients usually have special guidelines on how to deal with those. Sometimes they may ask for each error to be marked once but other times an error should be marked as ‘global’ or ‘repeated’.
• Always fill out the LQA form in English. Comments should be clear so that a non-native speaker can understand what the issue is. Remember that the client will see this LQA. Also be mindful that your role is to help and educate the translation team and ensure the delivered translation meets the quality expectations. This should be done in a way that is constructive, not critical or rude. Comments should be polite, professional and factual at all times.
• Pay special attention to the match rates when filling out the scorecard. You should only log errors for translated content that is in scope for the LQA.

Arbitration is a crucial step in the LQA process. Ideally, we want to avoid agreeing to any of the errors in the LQA form. Most clients will keep a record of our quality rating over a period of months/years and several LQA fails could result in a poor quality rating. The implications of this could be losing the client’s trust, losing accounts, facing financial penalties, or missing out on future work.

• Appeal as many errors as possible. There is a tendency to simply agree with the reviewer, especially if you are pressed for time. However, even if the reviewer’s translation is an improvement, we really want to avoid agreeing that the original translation contained linguistic errors if it didn’t.
• When appealing, ask for the score to be updated by suggesting a recategorisation of the error type (e.g. from error to preferential) or the severity level (e.g. from major to minor) if you can argue that this is warranted.
• Flag comments where the reviewer has said ‘xx’ sounds better than ‘yy’. Without empirical evidence, this should not be included in the LQA form as an error. At most, it should be marked as a preferential change and not contribute to the overall score.
• If the language in an LQA form is unnecessarily harsh or combative, do not replicate this in your own comments. Keep your language polite, professional and factual, and remember that the client will likely see your comments and form an impression based on them.
• Even if the score is a PASS, we still want to appeal as many errors as possible and there is still plenty that can be learned from any true errors the reviewer has flagged, so spend time looking through the comments.
• Do not always assume the LQA reviewer is right. It can be tempting to assume they know the account better than you and you should accept all their changes but, in reality, the reviewer may have just as much knowledge as you. Trust your experience, and back it up with evidence.
• Always provide evidence when arbitrating. Show the client that we did research during translation. Examples of evidence are:
  • Official websites
  • Examples of use online
  • TB/TM results
  • Client style guides
  • Wherever your research came from
• Double-check the LQA instructions provided to the reviewer in the scorecard:
  • If the reviewer was asked not to log repeated examples of the same error but they have recorded them all, then removing them could help to improve the overall result. You can say “Please remove, as this is a repeated error”.
  • If the reviewer was instructed to only review certain match types but they included out-of-scope segments, then removing these could help to improve the overall result.

The aim of the LQA is to answer the following questions:

• Does the translation accurately reflect the meaning of the source text (i.e. free of mistranslations, additions/omissions and data errors)?
• Is the text adapted to the target culture/standards?
• Is the text formally correct from the language point of view?
• Does the text comply with project glossaries and other reference materials?
• Does the text keep the requested format?

Errors are imperfections of the target language text for which the answer to any of the above listed questions is 'no'.

Only disputable errors (errors that can be verified against a grammar/usage book, approved TMs and glossaries, style guides and translation guidelines for the relevant languages) are classified as errors and included in the error count. Changing the writing style according to personal preferences does not play a role in an LQA.

If you feel any error a reviewer has logged is preferential and not a real error, you should disagree with it and request that the error be re-categorised as preferential. If you simply accept/agree with it, then the severity level will not be changed and the overall score will not improve.