NewNLP labeling is live in 200+ countries

NLP Training Data
Labeled by Native Speakers

Get your text labeled by real people in the countries and languages you choose: sentiment, categories, names, relevance and translation checks. Set the task, set the pay and approve only the labels that fit.

No subscription 路 Pay per approved label 路 195,000+ workers online

200+

Countries

13,487

Tasks completed

195,000+

Registered workers

~4 Hours

Avg time to results

How It Works

How NLP Data Labeling Works

Launch a text labeling task yourself and start receiving labels from native speakers, without long negotiations.

Person at a laptop writing NLP labeling task instructions
01 路 Create

Add your texts and label rules

Put a batch of texts into the task, list the labels workers can choose from and explain each one with an example.

  • Batch of texts per submission
  • Your own label list
  • Example for every label
Laptop with a world map on screen for country and language targeting
02 路 Target

Choose countries and languages

Select the countries your contributors should come from and the language of the texts. Native speakers catch meaning, slang and tone that others miss.

  • Any number of countries in one task
  • Native speakers by country
  • You set the total volume
Person reading and marking text labels on a laptop
03 路 Label

Native speakers label your texts

Your task becomes available to workers in the selected countries. Each person labels the batch once, so you can get several independent labels for the same texts.

  • Labels for every text in the batch
  • Independent opinions
  • New person in every submission
Person reviewing a table of text labels on a computer monitor
04 路 Review

Approve only the labels that fit

Check every submission, approve the labels that follow your rules and reject the rest. Compare answers from several people to find the most reliable label.

  • Check every submission
  • Approve or reject each one
  • Clean labels for training
NLP Data Types

Types of NLP Training Data You Can Collect

Set up a labeling task for the exact signal your language model needs. Contributors follow your rules and work in their own language.

Sentiment

Positive, negative or neutral, judged by people

Contributors read comments, messages or reviews and mark the sentiment and emotion. Native speakers understand sarcasm and slang that automatic tools miss.

  • Positive, negative, neutral or mixed
  • Emotions you define
  • Sarcasm and slang understood
Create a labeling task
Person reading messages on a phone to judge sentiment
Classification

Every text sorted into the right category

Workers assign each text to one of your categories, such as topic, request type or urgency. Use the labels to train routing, search and moderation models.

  • Your own category list
  • One or several labels per text
  • Clear examples for each category
Create a labeling task
Sorted folders or sticky notes representing text categories
Named Entities

Names, places and products marked in text

Contributors find and list people, companies, places, dates or products mentioned in each text. Train models to extract key information automatically.

  • Entity types you choose
  • Local names and spellings
  • Exact words copied from the text
Create a labeling task
Highlighted words in a printed text for named entity labeling
Search Relevance

Does this result answer the query?

Workers see a search query and a list of results and rate how well each one matches. Improve search, recommendations and retrieval for real users.

  • Query and results in the task
  • Rating scale you define
  • Local intent by country
Create a labeling task
Search results on a laptop screen for relevance labeling
Translation Checking

Translations checked by native speakers

Contributors compare a source text with its translation and mark errors, missing meaning or unnatural wording. Find weak spots in machine translation early.

  • Accuracy and fluency checks
  • Errors marked and explained
  • Native speakers of the target language
Create a labeling task
Two texts side by side for translation checking
Text Correction

Spelling, grammar and wording fixed by people

Workers correct mistakes in texts you provide and rewrite unclear sentences. Collect before and after pairs for correction and editing models.

  • Original and corrected versions
  • Spelling and grammar fixes
  • Natural local wording
Create a labeling task
Person correcting text with a pen for NLP text correction data
Task Example

Texts Labeled in Batches

Put a batch of texts into one task. Every worker labels the whole batch once, and you collect several independent labels for each text.

Your setup

What you set up

Task settings

Choose Text collection, paste a numbered batch of texts into the instructions and list the labels workers can use.

  • Text collection task type
  • Numbered batch of texts
  • Label list with examples

Your view

What you receive

Submission

Each submission arrives with a label for every text in the batch, ready to review and compare with other workers' answers.

  • Labels for the full batch
  • Worker country on every submission
  • Approve or reject each set
Pricing

Estimate Your NLP Labeling Cost

You set the pay per task. One submission is one labeled batch. Adjust the numbers to see what your project will cost, with the 15% platform fee included.

?
?
min
?
$
FAQ

You set the pay for each task, starting from $0.10, plus a 15% platform fee. One submission is one labeled batch, so bigger batches and harder labels need a higher rate. A campaign starts from $0.30.

Sentiment and emotion, text classification into your categories, named entities, search relevance ratings, translation checks and text corrections with before and after versions.

Paste a numbered batch of texts, for example 20 comments, into the task instructions with your label list. Each worker labels the whole batch in one submission; for more texts, create several tasks with different batches and remove any personal data first.

For simple labels one or two opinions may be enough. For subjective labels like sentiment, order three or more submissions per batch and keep the label most people agree on.

Any language spoken by workers in the countries you select. State the language in your instructions; native speakers catch sarcasm, slang and tone that automatic tools miss.

You review every submission and approve or reject it. Define each label with an example, add control texts with an obvious correct label and compare answers from several workers.

The average time to results is about 4 hours. If you need more labels later, extend the same task; workers who already took part cannot submit again.