“Language expert” is a title, not a job. This page covers the listings whose full descriptions are about judging language: how native and how fluent a voice sounds, including a voice generated by an AI model, and whether a written answer does what its guidelines asked. It first sets out what those descriptions say, and then walks through five ways the judging goes wrong — which is the blog’s own practice material, not something the listings publish.
One boundary before you read on. On the date in Sources, a listing titled “Portuguese Language Expert” described audio transcription, in spite of the title. That is a different job with different skills, and it is outside this page. Read the description, not the title, to know which one you are applying to.
Opens micro1. You apply and are vetted by micro1, not by this site.
Two kinds of judging the descriptions set out
1. Evaluating speech. The listings titled “Bilingual Expert” that we read — Dutch, Indonesian, Hebrew and Thai among them — describe listening to short audio clips and judging how native and how fluent the speech sounds. The Dutch and Indonesian descriptions say the speech is generated by AI models; the Hebrew and Thai ones say “the speaker or model output”. The evaluator assesses pronunciation, intonation and authenticity and writes the reasons down: the Hebrew and Thai descriptions ask for a justification for each evaluation, written in English. The Indonesian one adds a task the others lack: validating whether the language in the clip is culturally and contextually appropriate.
These descriptions ask for native-level proficiency in the language and a stated minimum level of English. The Dutch and Indonesian ones ask for an understanding of regional variation. A background in phonetics, linguistics, voice or audio work is listed as preferred in the Hebrew and Thai ones.
The Cantonese evaluator listing is broader than the rest: assessing Cantonese text as well as speech for accuracy, fluency and fit to context, giving detailed feedback on model outputs, and taking annotation, labeling and categorization assignments.
2. Evaluating written English. One listing, for linguistics specialists, is about written content and AI-generated responses in U.S. English: reviewing them for accuracy, clarity, coherence and completeness, comparing them against detailed guidelines, and writing short, specific feedback. Its description asks for objective assessment rather than creativity, and native proficiency in U.S. English is a requirement.
Translation is not among the tasks. None of these descriptions lists translating as part of the work. Translation appears only as a kind of prior experience some listings count in your favor.
Before the failures: whose list this is
The five failures below are this guide’s selection and this guide’s reading of what the descriptions ask for. They are not a list micro1 publishes and not its acceptance criteria. The first four are practice for evaluating speech; the last is practice for evaluating written English. Nothing here practices the annotation and labeling assignments the Cantonese description adds.
Where the practice clips come from. Every exercise below needs audio or text, and a clip being online does not make it yours to reuse. Use one of three sources and no other: recordings you make yourself; recordings of people you know who agreed to be recorded for your private practice; or material whose license states that this kind of use is allowed. For synthetic speech and generated answers, use a tool you are allowed to use, under its terms, with text you wrote. Keep the clips to yourself.
The examples below are illustrative exercises created for this guide, not published micro1 assignments or interview questions. They show the kind of judgment the work needs; they are not a preview of what a project will hand you.
Hearing a regional accent as an error
Practice for: evaluating speech. The Dutch and Indonesian descriptions ask for an understanding of regional variation. The failure this guide has in mind: marking a native speaker of another region as non-native because the accent is not the evaluator’s own.
Practice. Gather short clips of three regional varieties of your language, from the sources named above. For each, list what is regional and what, if anything, a native speaker of that region would hear as wrong.
One score for two questions
Practice for: evaluating speech. The descriptions name nativeness and fluency side by side. For this exercise, keep a separate score and explanation for each named question. That is the exercise’s rule, based on the two things the descriptions ask the evaluator to assess; it does not define the project’s scale or establish a speaker’s background.
Practice. Record or generate two clips using the permitted sources above. Score nativeness and fluency separately, and write one reason for each score that you could not reuse for the other.
Grading the recording instead of the speech
Practice for: evaluating speech. Background noise, a clipped ending, or a sentence you disagree with are not properties of the speech. The descriptions ask for a judgment of pronunciation, intonation and authenticity; a score lowered because the audio is poor answers something else.
Practice. Record a native speaker — yourself will do — in a noisy room. Write down everything that bothered you, then cross out each item that is about the recording or the content and not about the speech. What is left is your evaluation.
“It sounds off” with nothing behind it
Practice for: evaluating speech. The Hebrew and Thai descriptions ask for a written justification for each evaluation. A score with no reason gives a second reader nothing to check.
Practice. Have a text-to-speech tool read a paragraph you wrote in your language. Write three observations, each tied to a moment in the clip and a feature — a vowel, a stress pattern, a pause in the wrong place — and none of them the word “unnatural.”
Rewriting the answer instead of assessing it
Practice for: evaluating written English. In written evaluation the temptation is to improve the text: a better word here, a cleaner sentence there. The task in the description is a comparison against guidelines, and a comment that only records your preference does not say which criterion the answer missed.
Practice. Write a five-line guideline for a short answer (length, register, what it must include). Generate an answer with a chatbot you may use and comment on it using only those five lines; each comment names the line it comes from. Anything you wanted to say that fits no line stays out.
A grid for your own work
This grid is ours, not micro1’s; it gathers the five failures into one sheet. The three columns of a row exclude one another: a given piece of work fits one and only one. Rows are read separately — the first four for a speech evaluation, the last for a written one. A row you could not check is at Fail until you check it. A row at Fail sends you back to the part of the evaluation its cell names, not to the whole evaluation.
| Criterion | Fail | Acceptable | Strong |
|---|---|---|---|
| Variety | Marks a regional form as an error. Re-assess the clip, then correct the score | Marks no regional form as an error, and does not name both the variety and the feature that marks it | Marks no regional form as an error, and names both the variety and the feature that marks it |
| Two dimensions | One score for nativeness and fluency together. Correct: score them separately | Separate scores, and at least one of them has no reason that applies to it alone — no reason, or one shared with the other | Separate scores, each with a reason that applies to it alone |
| Object of judgment | Penalizes the recording or the content. Correct the score | Judges only the speech, without saying whether anything was set aside | Judges only the speech, and says what was set aside, or that nothing was |
| Justification | Names no feature. Correct: add the features behind the score | Names features, and at least one has no moment in the clip | Names features, each with its moment in the clip |
| Written evaluation | Rewrites the answer or any part of it, or makes no comment. Redo the evaluation as a comparison against the guideline | Does not rewrite; comments, and at least one cites no criterion | Does not rewrite; comments, and every comment cites its criterion |
Only Written evaluation sends you to Redo: rewriting the answer is a different task from assessing it, so nothing in the first attempt can be kept.
Presenting yourself at the interview
There is an AI interview — step two of the four micro1 lists, between applying and onboarding — and micro1 states that human recruiters also review profiles and match experts to opportunities.
Suggestions from this guide:
- Say which of the two jobs you are applying for, in the listing’s own words: speech evaluation or written evaluation.
- Name your variety. “Dutch, Flanders”, not only “Dutch”: two of the descriptions ask for an understanding of regional variation.
- State your English level honestly. The speech-evaluation listings set a minimum, and the written feedback is in English.
- Bring something that fits the job, and say what it is: a practice evaluation from the exercises above, presented as practice; or editorial work you have done, described truthfully and only as far as you may share it.
- Check the interview language in the invitation or the listing instead of assuming it.
Pay: what the descriptions leave out
The cards on the opportunities page show an hourly range. The descriptions we read for these listings add nothing to it: no statement of whether pay is by the hour, by clip or by accepted evaluation, and no minimum volume. The basis matters. A clip is heard more than once before a judgment is written and justified, so the work runs longer than the audio does, and pay per clip and pay per hour worked can be far apart. The listings leave open which one applies: the unit of payment is unknown until someone at the project states it.
One listing, the written-English one, does state a time commitment:
Immediate availability and ability to commit at least 20 hours per week while maintaining consistently high-quality outputs.
Hours you must be available are not the same as hours paid. Before accepting any of these, ask and get the answer in writing: what the unit of payment is; if it is per clip or per evaluation, what that works out to for an hour of your time; who accepts the work and what happens to work sent back; and whether there is a minimum volume. Until then, do not assume that all the hours you work are paid hours.
This guide gives no rate, which is an editorial decision of ours. The referral terms are a different thing: they forbid anyone promoting the program from making promises about compensation, placement or acceptance rates.
Does a referral link make a difference to the assessment? We saw nothing saying it does. micro1 runs the vetting; this site has no role in it. micro1’s referral terms forbid anyone promoting the program from promising placement, compensation or acceptance, and a page that promises a referral advantage is a warning sign.
Residency, not language: which countries are excluded
micro1 states it does not hire residents of: Afghanistan, Belarus, China, Cuba, Democratic Republic of the Congo, Hong Kong, Iran, Iraq, Libya, Macao, Myanmar, North Korea, the Russian Federation, Somalia, South Sudan, Sudan, Syria, Venezuela, Ukraine and Yemen. Snapshot from the date in Sources.
A language can be sought while a place where it is widely spoken is excluded — a Cantonese listing exists alongside Hong Kong and Macao being on that list. Read it as a residency rule, not a language rule, and check the current version before applying. The written-English listing adds its own geography: the US and Western Europe.