Skip to main content

Still Needs a Human Help

Still Needs a Human is where your team checks translations before they go to the client. You sign in once, and your workspace, profiles, members and plan cover three modules:

  • What's Wrong, the translation QA check. You give it a translated file. It reads the source and the translation side by side and tells you what is wrong: a missing number, a broken tag, a brand name that got translated, a sentence that does not mean what the source means. It never changes your original file. It gives you a list, and you decide; a fix you accept goes into a corrected copy.
  • LQA, scorecards: a scored human review on your own MQM or DQF template.
  • Eval, the engine bake-off: translation engines compared on your own content.

Most of this guide is about the check. LQA and Eval has the other two, and Account and plans has your workspace, members and plan.

Getting startedSign in, drop a file, read the result. Under ten minutes.What it checksThe rule pass, the AI review, the AI check of rule findings, spelling and the score.ProfilesEverything the tool knows about one client: terms, rules, style, Learn.ReportsRead a report, decide on each finding, apply fixes, sign off, share.Ways to run a checkUpload, email, a Dropbox folder, the Chrome extension, an assistant, the API.API referenceEvery public endpoint with its request, response and errors.LQA and EvalScorecards for a scored review and the engine bake-off.Account and plansWorkspaces, members and roles, the plans, billing and your preferences.

Two minutes on video​

The whole check, from one file to a signed report. It works without sound: watch it on the site.

The Quick Start as a PDF​

The first day on paper: check a file, read the report, check a spreadsheet and set up a client. Download the Quick Start (PDF). Print it, or send it to someone before their first day.

Where to start​

The one idea behind it​

Two layers. The first is a rule engine: tags, numbers, placeholders, links, protected terms, punctuation, spacing, length. It is exact and instant. The second is the AI review: it reads every translated segment for what rules cannot see, meaning, omissions, a register your style guide forbids. It checks the rule findings that need a reading, so a false positive is marked as one. Nothing is removed from a report on the AI's word alone and nothing changes a profile until a person approves it. How the checks work.

Also in the app​

Help at the foot of the rail and under your name opens this site in a new tab, so you keep your place in the app. Take the tour in Search or do (press Cmd+K or Ctrl+K) walks you through the panel again in a minute.