What the work involves

You receive batches of short Polish audio clips, some spoken by people and some generated or modified by a model, and you judge nativeness and fluency against a project rubric. That means listening for vowel quality and nasal realisation, consonant clusters and palatalisation, stress placement, sentence-level intonation, and the small giveaways that mark speech as non-native, regionally shifted, or synthetic. Each rating is accompanied by a written justification in English — usually a few precise sentences naming the specific phonetic or prosodic features that drove your score, not a general impression.

The second half of the job is consistency. Reviewers on these projects are scored on whether they agree with calibration items and with each other, and on whether their written notes actually explain the number they gave. Where a clip is genuinely ambiguous — heavy Silesian or Kashubian influence, a heritage speaker, a clip too short to judge — you flag it and ask the coordinator rather than guessing. Feedback loops are frequent, and guidelines get revised mid-project; adapting to a rubric change without drifting back to old habits is part of the role.

What the platform screens for

  • Verifiable Polish nativeness, probed through questions about regional variation, phonology and register that a fluent-but-not-native speaker answers vaguely.
  • English writing quality at B2 or above, since every justification is written in English and read by non-Polish-speaking coordinators.
  • Ability to name features, not just react — "the /ɨ/ is fronted and the phrase-final fall is truncated" versus "it sounds a bit off."
  • Rubric discipline — will you apply a guideline you personally disagree with, and escalate rather than improvise?

Logistics

Fully remote and asynchronous, contractor basis, no fixed schedule. Volume arrives in waves tied to customer demand, so hours are irregular rather than steady; contributors commonly commit to a weekly minimum they can actually hit. You need a quiet listening environment and decent closed-back headphones — laptop speakers are not adequate for judging sibilants or low-level synthesis artefacts. Pay in the $30–65/hr band reflects rates observed on comparable micro1 language projects and is not guaranteed for any given assignment.