Text annotation services
Named entities, sentiment, intent, topics and relations labeled in text by native speakers of the language, following your guidelines and working in your tool. Every batch comes with agreement scores, so you know the labels are consistent before you train on them.
Priya Shah booked a flight with Air India to Toronto on March 14 for $820.
Lukas Weber hat am 2. Mai seinen Vertrag bei Rheinwerk Energie in Köln gekündigt.
حجزت سارة غرفة في فندق هيلتون في دبي يوم الجمعة.
Example entity, sentiment and intent labels on three sentences.
- Label types
- Entities, sentiment, intent, topics, relations and classification
- Languages
- Labeled by native speakers of each language and variety
- Tools
- Your annotation tool, or Label Studio or doccano set up for you
- Quality
- Agreement scores and a reviewed sample on every batch
What text annotation is
Text annotation is adding labels to text so a model can learn from it: marking the names, places and dates in a sentence, tagging whether a review is positive or negative, or classifying what a customer wants. The labels have to be consistent, which is why guidelines, calibration and agreement checks matter as much as the labeling itself.
Language is where text labels go wrong most often. An idiom, a sarcastic review or a name written in another script needs an annotator who is native in the language. For a full overview of our labeling work, see data annotation.
Text annotation tasks
From single labels on short texts to linked entities in long documents.
Named entity recognition
People, organizations, places, dates, amounts and your own entity types, marked span by span.
Sentiment and emotion
Positive, negative or neutral, plus emotion and aspect-level sentiment for reviews and support tickets.
Intent and slot labeling
What a user wants, and the details in their message, for chatbots and voice assistants.
Classification and topics
Documents, posts and messages sorted into your categories, single or multi-label.
Relations and linking
Links between entities, such as who works where, and entities matched to your knowledge base.
PII and sensitive data
Personal details found and tagged or redacted before text is used for training.
How a text annotation project runs
Guidelines are tested on a pilot before volume starts.
Guidelines
Your label set and rules, with examples and edge cases for each language.
Pilot
A few hundred items labeled by several annotators per language.
Agreement check
We measure agreement, review disagreements with you and update the guidelines.
Agreement report QAProduction
Regular batches, with a reviewer checking a sample of every annotator's work.
Delivery
Labels in JSONL, CSV, CoNLL or your tool's export format.
Related AI data services
Questions about text annotation
What is text annotation?
Text annotation is adding labels to text, such as marking named entities, tagging sentiment or classifying intent, so it can be used to train or test a natural language processing model.
Which languages can you annotate?
Major European and Asian languages, Arabic and its dialects, Indic languages and less common languages on request. Each language is labeled by native speakers, and we confirm availability in the scoping call.
Can you work in our annotation tool?
Yes. Annotators can work in your own tool, or we set up Label Studio or doccano for your project and deliver the export your pipeline reads.
How do you measure annotation quality?
Several annotators label the same pilot items, and we report agreement between them for each label and language. In production, a reviewer checks a sample of each annotator's work every batch.
Start with a pilot batch
Send your label set and a sample of text, or book a call. We label a pilot and show you the agreement scores.
- Phone
- +44 20 3290 0116
- Phone hours
- Weekdays 9:30 to 16:00, weekends 9:30 to 12:00, UK time