Bilingual RLHF, Localization & Writing Constraints Practice Test
Register and tone calibration, idiom and localization accuracy, creative constraint adherence, and grammar auditing for AI tutor and linguistic QA tests.
10 scenarios, graded as you go, with the exact grading principle for every answer. Start free, no sign-up.
Timed mock assessment
40 scenarios in 48 minutes, drawn from every topic. No answers shown until the end, then a full score breakdown, your weakest topics and a review of what you missed.
Prices are in US dollars, plus tax where it applies. Stripe, our payment processor and the seller of record, adds tax at checkout.
All sales are final. If something is really wrong, such as a duplicate charge or a purchase that does not unlock what you bought, write to support@annotationprep.com and we will put it right.
One-time payment. Nothing renews.
You have an assessment ticket for this practice test. Start the mock assessment to use it.
Translate "Break a leg at your audition tonight!" into natural Spanish for a friend.
MODEL A
¡Rómpete una pierna en tu audición esta noche!
MODEL B
'¡Mucha suerte en tu audición de esta noche!' (or '¡Mucha mierda!' in theater slang, if appropriate).
Why does Model A fail?
Literal idiom translation: translating 'break a leg' word-for-word into Spanish sounds like wishing physical injury rather than good luck.
Spanish does not have a word for 'audition'.
It used exclamation marks.
Model A is better because literal translations are always preferred.
Show the answer
A. Literal idiom translation: translating 'break a leg' word-for-word into Spanish sounds like wishing physical injury rather than good luck.
Bilingual RLHF evaluations penalize 'translationese'—translating idioms literally when the target language does not share the idiom.
Principle: Localization QA: Idiomatic Equivalence
A prompt asks for a polite customer-support apology in Japanese (日本語) to an enterprise client. Model A uses casual plain form (だ・である / 〜だよ). Model B uses appropriate business honorifics (敬語 / 丁寧語, e.g., 申し訳ございません).
Which passes?
Model A, because casual Japanese is shorter.
Model B, because business customer support in Japanese requires polite/honorific register (keigo), and casual forms are inappropriate.
Tie.
Neither, because apologies cannot be written in Japanese.
Show the answer
B. Model B, because business customer support in Japanese requires polite/honorific register (keigo), and casual forms are inappropriate.
Register mismatch (informal vs. formal/honorific) is one of the top criteria on bilingual AI tutor and OneForma linguistic assessments.
Principle: Localization QA: Register & Honorifics
PROMPT
Rewrite this paragraph in active voice: "The quarterly report was reviewed by the finance team, and three errors were found by the lead auditor."
Select the best evaluation:
'The quarterly report had been reviewed by the finance team, and three errors were discovered.'
'The finance team reviewed the quarterly report, and the lead auditor found three errors.'
'Three errors were found in the quarterly report by the finance team.'
'The finance team was reviewing the report while errors were being found.'
Show the answer
B. 'The finance team reviewed the quarterly report, and the lead auditor found three errors.'
Option B converts both passive clauses ('was reviewed by...', 'were found by...') into crisp active voice where the subjects perform the actions.
Principle: Writing QA: Active vs. Passive Voice
A user asks to translate UI copy into German: 'Save changes' (button label, max 20 characters).
MODEL A
'Änderungen speichern' (20 chars).
MODEL B
'Bitte klicken Sie hier, um Ihre Änderungen zu speichern' (54 chars).
Why does Model B fail?
German grammar is wrong.
It broke the 20-character UI length constraint and added unnecessary prose to a button label.
Model A used an umlaut.
Both failed.
Show the answer
B. It broke the 20-character UI length constraint and added unnecessary prose to a button label.
UI localization tasks strictly enforce character limits and concise imperative button phrasing.