Contact Us AR
العربية
Data annotation for companies & researchers

Any data you need -
labeled through play.

Whether you're training a model, running a study, or building a dataset, bring us the task. We turn it into a game, real people play it, and you get clean, cross-validated annotations.

Companies

Any data you need -
collected for you.

Bring us any data need - weather labels, Arabic dialects, dog breeds, content safety, anything. We plug it into a game or build a custom one, real people play, and you get clean, cross-validated annotations.

Built-in validation
The same item is shown to multiple players across different games. Agreement is the quality signal - not a single rushed annotator.
Any task you can judge
Classification, ratings, transcription, entity tagging, preferences, moderation - if a person can judge it, we can collect it.
Real people at scale
A standing crowd of players across topics, languages, and regions - not a one-off panel you have to recruit.
Exports that keep your IDs
CSV or Excel with your own ID columns, plus the agreement and vote counts behind every label. PDF or PowerPoint reports.
Academics

Build the benchmarks
Arabic NLP is missing.

Many tasks our games cover are exactly where labeled data is thinnest. Use Avenzoar to collect gold-standard, diverse datasets - with the inter-annotator agreement and provenance a paper needs.

Hard-to-source tasks
Diacritization, dialect identification, handwriting transcription - exactly where labeled data is thinnest.
Agreement & provenance
Every label carries how many players saw it and how strongly they agreed - the metadata reviewers ask for.
Diverse participants
Recruit native speakers by region, dialect, or background instead of a convenience sample.
Built for reproducibility
Documented task design and exportable datasets you can cite and others can rebuild on.
How it works for clients

From your data need to annotated data - in four steps.

01
Bring your data need

Weather conditions, Arabic dialects, dog breeds, product reviews, road signs, content safety - any topic, any language.

02
We turn it into a game

We drop your data into one of our existing game formats, or design a brand-new game around your task.

03
Real people play it

A standing crowd of players annotates your data while having fun. The same item is judged by several players.

04
You get clean data

Cross-validated annotations, tagged and ready: CSV or Excel with your own IDs and the agreement behind every label.

Clients collect data on anything
WeatherArabic dialectsDog breedsProduct reviewsRoad signsContent safetyHandwritingMedical termsLocal slang…or your topic
CSV or Excel with your own ID columns, plus the agreement and vote counts behind every label - and a PDF or PowerPoint report.
Built for research

Build gold-standard datasets at crowd scale.

A few example tasks we've already built games for - the kind where labeled data is scarce and a large, diverse crowd helps most.

Diacritization
Tashkeel

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 / handwriting validation
OCR Check

Arabic OCR - especially handwriting and historical manuscripts - is notoriously unreliable. Human verification turns raw OCR into gold transcription data.

Dialect identification
Dialect Compass

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.

Content moderation
Red Card

Safety classifiers need culturally-aware judgments of what counts as toxic or offensive - context that off-the-shelf models routinely get wrong.

Semantic similarity
Same Meaning?

Paraphrase and semantic-similarity labels train search, retrieval, and deduplication - and good similarity data is hard to come by.

Named-entity recognition
Entity Radar

Entity tagging powers everything from knowledge graphs to redaction. NER suffers from sparse, inconsistent labels - exactly what a big crowd fixes.

Submit a Request

Tell us what data
you need.

Send us the task and we'll come back with how we'd turn it into a game, expected timelines, and pricing.

Any topic, any language
From weather to dialects to dog breeds.
Cross-validated by default
Multiple players judge every item.
Clean CSV or Excel
Your own IDs kept, with the agreement behind every label.
I'm reaching out as a…
We'll only use your details to reply about your data request.

Frequently asked questions

Which Arabic dialects do you cover?

Gulf, Egyptian, Levantine and Maghrebi Arabic, as well as Modern Standard Arabic (MSA). You can target contributors by region, dialect or background, so your data is labeled by people who speak its dialect.

How do you control data annotation quality?

The same item is judged by several independent players, often across different games. Hidden attention checks give each player a trust score, labels resolve by trust-weighted consensus, and anything the crowd disagrees on goes to an expert. Every label ships with its agreement level.

What do we receive?

CSV or Excel files that keep your own ID columns, with the agreement and vote counts behind every label. You can follow progress live in the client portal, and you also get a PDF or PowerPoint report.

How much does data labeling cost?

It depends on the task. Send us what you need and we will come back with how we would turn it into a game, expected timelines and a quote.

Do you use an existing game or build a new one?

Either. We can put your data into one of our existing game formats, or design a brand-new game around your task.

Is Avenzoar suitable for academic research?

Yes. Avenzoar helps you build gold-standard datasets for tasks where labeled data is scarce, such as diacritization, dialect identification and handwriting transcription. Every label records how many players saw it and how strongly they agreed: the inter-annotator agreement and provenance a paper needs.