A few examples of games we've already built - there are many more. Drop your own data into one, or we design a brand-new game around your task. Flip any card to see what it collects.
Read the sentence, tap the right diacritic on each letter.
Each round becomesعلمعَلِمَ
Tashkeel
Read the sentence, tap the right diacritic on each letter.
Written Arabic usually drops short vowels - so the same letters can be many different words. Restoring diacritics (tashkeel) is one of the hardest, highest-value problems in Arabic NLP.
OCR Check
See the original scan and the machine reading. Match or wrong?
Arabic OCR - especially handwriting and historical manuscripts - is notoriously unreliable. Human verification turns raw OCR into gold transcription data.
Answer Scale
Weigh a question and its answer: excellent, partial, or wrong.
Rating answer quality is the backbone of RLHF and model evaluation. A balanced crowd of native speakers produces the preference data that aligns models.
Dialect Compass
Read the text, point the compass at its dialect.
Arabic is dozens of distinct spoken dialects, not one language. Labeled dialect data is scarce - and essential for models that understand how people actually talk.
Red Card
You're the referee. Clean, borderline, or a violation?
Safety classifiers need culturally-aware judgments of what counts as toxic or offensive - context that off-the-shelf models routinely get wrong.
Same Meaning?
Two sentences appear. Identical, related, or unrelated?
Paraphrase and semantic-similarity labels train search, retrieval, and deduplication - and good similarity data is hard to come by.
Entity Radar
Lock onto the highlighted term - person, place, org, or date?
Entity tagging powers everything from knowledge graphs to redaction. NER suffers from sparse, inconsistent labels - exactly what a big crowd fixes.
Trivia Runner
Run the road, pick a lane: TRUE or FALSE.
True/false judgments at speed build fact-checking and claim-verification datasets - and keep players in flow while they label.
Fill the Blank
Spell the missing word on a scored keyboard.
Fill-in-the-blank responses generate completion and next-word data, and reveal how real speakers phrase things in context.
Swipe Quiz
Flick toward your answer - four directions, four choices.
Fast multiple-choice rounds collect classification labels across any domain you can phrase as a question.
Bubble Quiz
Pop the right bubble before the timer runs out.
A second MCQ format keeps the same labeling task fresh - variety is what keeps the crowd annotating for hours.
Stack-It
Right answers stack the tower. Wrong ones topple it.
Endless streak play maximizes labels-per-session and naturally surfaces the questions players find hardest.
