Speech, text, RLHF and evaluation data in Turkish, Arabic, Persian, Balkan, Eastern European and Turkic languages, collected and checked by native speakers. Start with a free pilot batch of up to 100 annotated samples or 30 minutes of audio, with a QA report.
Most AI models are trained mainly on English and Western European languages. In Turkish, Arabic, Persian, Balkan, Eastern European and Turkic languages they struggle with:
Native speakers collect, annotate and check every dataset. Projects run under an ISO/IEC 27001 certified information security management system, with in-house project management and QA, and data is delivered in batches so model teams can start on early batches while collection continues.
From speech collection and image labeling to human feedback and custom MT training data: the data your models need to work in regional languages.
Recordings from native speakers, balanced by accent, age, gender and acoustic environment. Collection focuses on dialects and regional languages that off-the-shelf speech corpora leave out.
Native linguists compare and rank LLM outputs, check facts and judge whether an answer suits the culture of the reader. The ratings feed alignment and fine-tuning for regional languages.
Labeled datasets for hate speech, toxicity, misinformation and sentiment. Our annotators understand the slang, code-switching and cultural context that automated systems miss.
Annotated and labeled text for named entity recognition, classification and intent detection in regional languages and dialects, built to your schema or ours.
Parallel corpora, domain-specific datasets and evaluation sets for fine-tuning machine translation engines and large language models on your domain and terminology, with output benchmarked by native linguists.
Images and video frames labeled with bounding boxes, polygons, segmentation masks or classes. Native annotators read the regional signage, packaging and documents that generic teams get wrong.
Free pilot batch
Send us your spec and we will deliver a free pilot batch of up to 100 annotated samples or 30 minutes of collected audio, with a QA report included. You evaluate the output against your own acceptance criteria before committing to production.
Datasheet
What you get, how it is checked and how it is delivered, before you get on a call.
We review your model needs, target languages and data specifications.
Native speakers are drawn from our network of 700+ approved linguists and contributors.
Work runs in batches against the agreed guidelines, with gold tasks and spot checks on each batch.
Final human review and formatting, ready for your pipeline.
Data security
Data leakage is a common concern in enterprise AI projects. Your proprietary data is used only for your project, access is limited to the assigned team and all work runs under NDA and our ISO/IEC 27001 certified information security management system. More in the FAQ on security and compliance.
Client projects. Names withheld under NDA; references available on request.
AI/ML
Content moderation
LLM training
FAQ
Answers for AI and ML teams scoping speech, text or evaluation data in regional languages.
El Turco collects speech data in Turkish, Arabic, Persian, the Balkan languages and the other regional languages it covers, with native speakers recorded across accents, ages, genders and acoustic environments. Arabic projects can record speakers from the Gulf, Levantine, Egyptian and Maghrebi dialect regions and add Modern Standard Arabic where the model needs it. Turkish collection can span regional accents such as Istanbul, Aegean, Black Sea and Eastern speech.
Timelines for El Turco AI data projects are agreed at scoping, because they depend on data type, volume, dialect coverage, contributor profiles and the QA steps required. Data is usually delivered in batches, so model teams can start working with early batches while collection continues. The free pilot batch comes first and sets the baseline for the production schedule.
Yes. El Turco annotators can work inside a client's own platform or annotation tool, or deliver into a custom schema agreed at the start. Guidelines are co-developed with the client and versioned per batch, so label definitions stay stable as the project grows. When the client has no platform of its own, data is delivered through secure transfer or API.
Yes. El Turco linguists evaluate LLM responses and machine translation output in their native languages, scoring accuracy, fluency, cultural appropriateness and terminology against an agreed rubric. Evaluation sets can be built from the client domain content and rerun after each model update. Teams then benchmark progress in Turkish, Arabic, Persian and the other languages El Turco covers.
El Turco contributors and annotators are native speakers drawn from a network of 700+ approved linguists and contributors across Türkiye, the Middle East, the Balkans, Eastern Europe, the Caucasus and Central Asia. Each contributor is matched to the language, dialect and profile the project needs. Annotators are calibrated on the client guidelines before production and tracked against gold tasks throughout the work.
El Turco contributors give consent before any recording or annotation starts, and every linguist and annotator signs an NDA. Speech and text data is handled in a GDPR-aligned way under the ISO/IEC 27001 certified information security management system. Usage terms for each dataset are set out in the project agreement, so the AI team knows how the delivered data can be used for model training.
Share your languages, data type and volume. We reply with a spec, a timeline and a quote. Pricing is per-word, hourly or project-based, quoted after scoping.
A project manager reads your brief, usually the same working day.