Data labeling
Text and NLP annotation in English, Swahili and code-switched text
Named entities, intent, sentiment, topics and relations, labeled by fluent readers who understand context, idiom and tone. Built for chatbots, search, document AI and moderation.
- Pilot phase
- Two or more review layers
- You own every label
Language data needs people who read the language.
Text labels look simple until real data arrives: sarcasm that flips sentiment, product names that look like people, and messages that switch between English, Swahili and Sheng mid-sentence. Rules alone don’t cover it. Fluent, trained readers do.
Our NLP teams work from versioned guidelines with worked examples, and a share of every batch is double-labeled so we can measure agreement and resolve disagreements before they spread.
Used for
- Chatbots and virtual assistants
- Customer-support analytics
- Search and recommendation relevance
- Clinical and legal document AI
- Trust and safety classifiers
Industries: Finance & insurance · Generative AI & LLMs · Healthcare & medical imaging
What our text & NLP annotation covers
Combine any of these in one workflow and one guideline.
Named entity recognition
Entity spans and types, nested entities and custom taxonomies.
Intent & sentiment
Utterance intent, sentiment polarity and emotion for conversational AI.
Text classification
Topic, toxicity, spam, quality and relevance labels.
Relation extraction
Links between entities for knowledge graphs and document AI.
Search relevance
Query–result ratings that improve ranking and recommendations.
Translation review
Quality checks on English–Swahili translations and transcripts.
Formats and tools
Delivery formats
- JSONL spans
- spaCy
- CoNLL
- CSV
- Custom schema
Tools we work in
- Label Studio
- Prodigy
- Labelbox
- doccano
- Your in-house tool
Pricing
Text classification from $0.015 per item and named entity recognition from $0.20 per 100 words, with volume rates from 50,000 units.
How we keep it accurate
Every project runs on the same four-check QA process. See how our QA works.
Double-labeled samples
A share of each batch is labeled twice so agreement is measured, not assumed.
Span-boundary rules
Clear rules for where entities start and end, the most common source of NER noise.
Language-qualified reviewers
Swahili and code-switched data is reviewed by fluent speakers, not translated first.
How a project runs
Scope and guidelines
We review a sample, agree classes, edge cases and acceptance criteria, and sign your NDA.
Pilot phase
Pilot phase helps you judge quality, speed and communication.
Certified team
Annotators pass a gold-set test before they are cleared for production.
Production and reporting
Batches with QA reports, in your format, on an agreed schedule.
Text & NLP annotation questions
Which languages do you annotate?
English and Swahili, including code-switched text and Sheng. Other languages depend on reviewer availability, which we assess before any commitment.
Can you build a taxonomy with us?
Yes. We often help define labels, write examples and test the taxonomy during the pilot before production starts.
Do you handle personal data in text?
Yes, under a data-processing agreement. We can redact or pseudonymise personal data before annotation.
What agreement levels should we expect?
It depends on how subjective the task is. We report agreement per label during the pilot and agree a realistic target with you.
Send a sample. See the labeling quality first.
Share a few files and a line about your goals. A project lead replies within one business day with an approach, a timeline and a price.