From a translation request to an advisory decision
6 min read
Lesson objectives
- Separate a symptom from an assumed cause and the decision at stake, then frame a testable advisory question within the team's authority.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
Imagine a product manager saying, 'The Arabic help centre translation is poor. Retranslate every article before launch.' The manager may be right, but the request mixes an observed symptom, a causal explanation, and an expensive remedy. An adviser helps the sponsor make a better decision; neither agreement nor rejection should precede the evidence. First ask which decision is due now. It might be whether to delay launch, repair one journey, or redesign the localization process. Each decision needs different evidence and carries a different cost of delay. If the launch decision is next week, a comprehensive linguistic review delivered a month later cannot resolve that immediate decision, however valuable the review may be later.
Localization extends beyond moving sentences into Arabic. It adapts content, interface behaviour and data presentation for a defined audience and use context. 'Arabic' alone does not identify the market, user group, device or placement of a string. Nor does an Arabic complaint prove that language is the sole cause. An article title may be accurate while search fails to find it; a support message may contain an old link; a step may disappear in a narrow mobile view. Separate direct observations from inferences. 'Help-page exits increased last week' is an observation if supported by reliable logs. 'Translation caused those exits' is a hypothesis that needs testing.
Turn the request into a four-line decision record. First, name the decision, deadline and owner: 'The product manager decides on Thursday whether to launch or defer the Arabic help centre.' Second, state the user outcome that matters, such as completing account recovery without contacting support. Third, name feasible options: full launch, a limited launch of critical articles, or a defined delay for repairs. Fourth, identify evidence that distinguishes those options. Do not collect everything merely because it is available. Choose evidence that might change the recommendation. If complaints cluster around one recovery article, examine that article and the path to it first. If failures span the library, test a broader process hypothesis such as a publishing defect or obsolete translation memory.
Notice the limits of the evidence. Forty tickets are not a denominator for prevalence among all visitors: they are a sample conditional on contacting support. Ten sentences cannot establish the quality of a whole library. Each observation is used only within the scope it supports. To test rival explanations, ask users to reach and complete the task through mobile search and record where they stall; compare that finding with an independent review of the text shown at the same point. If better indexing helps discovery but users still cannot execute the instructions, prioritize article clarity. If both versions produce the same difficulty, investigate a third cause, perhaps a broken product step. This is a conditional judgement, not a promised result.
Before submitting the memo, check three boundaries. Have you defined the denominator and period for each metric? Have you stated what could disprove your preferred explanation? Have you separated operational advice from authority to make the decision? In the first case, if most reports instead concerned corrupted safety instructions across several articles, the priority would become halting and reviewing those articles, not improving search alone. That concrete counterfactual keeps the recommendation revisable. Finally, do not label this diagnosis a standards-compliance audit. The public abstract of ISO 11669:2024 connects translation-project specifications with needs analysis, risk assessment and workflows; it does not provide a ready-made method for judging the quality of an individual text.
Lesson summary
Frame the decision, separate symptom from assumed cause, then test rival explanations.