Natural Language Processing
Annotated text and audio for language and speech models, with consistent human judgement applied at volume.
Who it is for
Language-model, speech, search and analytics teams that need consistent human judgement applied to written or spoken data, whether that means labelling intent, transcribing recordings or extracting structured fields from free text.
Supported tasks
We offer the following language and audio tasks, each run against written guidelines and reviewed before delivery:
- Text classification: assigning topics, intents or categories to documents, messages or passages.
- Audio classification: tagging recordings or segments by sound type or content.
- Transcription: turning speech into text to an agreed style guide.
- Utterance collection: gathering spoken or written prompts to a brief, for training and evaluation.
- Entity and information extraction: marking names, values and relationships so free text becomes structured data.
- Sentiment annotation: labelling tone and opinion in text.
Our published COVID sentiment story, using tweets and Facebook posts, is an example of this sentiment annotation work.
Clinical text
If your text is clinical, such as notes, reports or similar documents that need review by medical professionals, see Healthcare and medical annotation, where clinician-reviewed annotation is described in full.
Sample output, FAQ and quote
Before production begins we can share a permission-cleared or clearly synthetic example showing the task, the labels and the delivery format, so you can check it against your own guidelines.
Can you handle more than one language? Tell us the languages in your brief and we will confirm what we can staff.How is quality checked? Work is reviewed against your guidelines before delivery, as described in How We Work.Can text and audio work run alongside other services? Yes. Language tasks can be combined with Other Data Services, such as cleansing or model validation, in one project.