Linguists

The models are fluent. That is the problem.

Phonologists and phoneticians, syntacticians and morphologists, semanticists and pragmaticists, sociolinguists, typologists and field linguists — use your training in how language actually works to train the latest AI models. This is not a translation role and not a native-speaker role; we are hiring the discipline, not the mother tongue. Join CrowdGen’s network of linguists, and we’ll bring you in when there’s work worth your time.

RemoteFlexible hoursIndependent contractor

Who we’re looking for.

A vetted network of practicing linguists.

✓5+ years of formal work in language science — Research, teaching, description, or applied analysis. Graduate research counts toward the five.
✓Formal training in linguistics — Across some or all of phonetics and phonology, morphology, syntax, semantics, pragmatics, and sociolinguistics.
✓Comfort with languages you don’t speak — Reading glossed data, minimal pairs, and grammar sketches, and reasoning about a system from a description of it.
✓Clear written English — The explanation is the deliverable, not a note attached to a score.
✓Bonus skills — Phonetic depth — IPA transcription, prosody, phonemic analysis; field or descriptive experience with an under-documented language; or typological breadth across families.

To join: A LinkedIn profile and a resume showing your education and experience are required to qualify.

What you’d actually do.

The work any experienced linguist would recognize.

01
Build the cases where a fluent surface hides the wrong reading
Scope ambiguity, presupposition, implicature, negation, and quantifier binding, plus the irregular paradigms the model will quietly regularize away.
02
Say which rule the model broke, not just that it sounds off
Produce the correct form and name the agreement, case, scope, or aspect constraint that was violated, because the explicit description is what generalizes into training signal and “unnatural” does not.
03
Develop comprehensive rubrics that measure expert-level quality
The constraint that governs, the readings genuinely available, and the dialect and register variants that are legitimate rather than errors — so output is scored against a described system rather than a native speaker’s impression.
04
Work on languages you don’t speak
Read a grammar sketch or glossed data and check the model’s morphology, phonemic inventory, and prosody against it, the way a typologist or field linguist would.

Questions, answered.

What is CrowdGen?+

CrowdGen is the contributor platform of Appen, a publicly traded data company listed on the Australian Securities Exchange under the code APX. Founded in 1996, Appen has been present at every major inflection point in AI — from early speech recognition to frontier reasoning models. Today, 80% of the world’s leading LLM builders are Appen customers. Through CrowdGen, you can use your professional expertise to contribute to frontier AI development and research.

What happens after I get accepted into CrowdGen?+

Joining the network is a qualification step — once your application is approved, we make matching projects available to you, and we reach out when one fits your background.

Why do you need my LinkedIn and resume?+

Because we vet everyone in the network. Seeing your real background is how we make sure the work we send you is actually matched to your experience. A LinkedIn profile and a resume showing your education and experience are required to qualify.

Ready when you are.

Once you submit your profile, our team reviews it and follows up within 72 hours if your background matches what we’re looking for.

Qualify Now