What the work involves
You build language analysis items and the reference analyses that go with them. A typical task starts from data — a paradigm you construct, a corpus extract, a set of grammaticality contrasts, a transcription — and asks the model to do something a trained linguist can do reliably: identify the phenomenon, parse it, derive the generalization, or predict the next form. Then you write the answer with the argument attached: which analysis wins, what data rules the alternatives out, and where a competing framework would land differently.
The second half is grading. Model output in linguistics is frequently fluent and wrong in a specific way: it names a plausible-sounding phenomenon, cites terminology correctly, and misparses the data underneath. Flagging that gap — plausible prose over a bad structural analysis, or a correct label attached to the wrong constituent — is the core skill being paid for.
What the screen looks for
- Concrete subfield depth. Expect follow-ups that keep narrowing: name the framework, name the diagnostic, describe what data would falsify your analysis. Vague breadth reads worse than one well-defended specialty.
- Data handling. Whether you can build a minimal pair, a paradigm, or a transcription that isolates one variable rather than four.
- Calibration. Linguistics has live disagreements. Screeners want to see you separate "wrong" from "a defensible analysis in another framework."
- Written English precise enough to document an argument — the reference analysis is read by graders and researchers who are not in your subfield.
Logistics
Fully remote and async; no scheduled calls, no fixed shift. Work arrives in batches, and volume can be scaled up or down week to week — expect real variability rather than a steady queue. Minimum commitment is 10 hours per week. Pay is weekly via Stripe at a rate set from your experience and subfield; the $50–100/hr band is what contributors on this platform have reported, not a guarantee for any given applicant.