We’re excited to introduce our updated text editor, which supports more complex NLP use cases and enables more breakthroughs than ever. With this new update, you can create issues and comments to boost collaboration in your labeling pipeline and help models gain insights on the relationship between entities.
Annotation relationships enable complex text use cases
ML teams can now train models on more complex text projects with NER annotation relationships. This new feature enables labelers to create and define relationships between entity annotations in unstructured text, consolidating labeling workflows and reducing the total number of language models needed.
With annotation relationships, ML teams can train models that rely on coreference resolution and dependency parsing use cases like information extraction, document summarization, machine learning, question answering, and more. By labeling relationships between entities in text, models can be trained to understand connections between words through named entity recognition (NER) and parts of speech (POS) tagging.
Labelers can create annotation relationships by dragging and dropping or via hotkeys.
Issues & comments streamline collaboration and quality assurance
Feedback and review are critical to maintaining quality data for every data type, so we’ve expanded support for issues & comments to the text editor. Labelers, reviewers, and stakeholders can now seamlessly start conversations and resolve roadblocks to improve text training data quality and production.
Issues & Comments eliminates the need for disconnected workarounds when asking questions or sharing feedback, such as using free text annotations, shared Slack channels, or spreadsheets. All conversation around your labeling project can now be found directly within the labeling interface, and quickly added to or resolved.
A sneak peek at future improvements
We’re always looking for ways to advance your workflow and make labeling as easy as possible. Here are a few other features we will release to improve your experience:
- Ability to label and read right-to-left languages (e.g. Hebrew and Arabic)
- Annotate multiple instances of an entity without needing to re-select the tool
- Improvements to speed up labeling entities in text
We also support model-assisted labeling in the text editor, letting you use your own model to pre-label text data to improve speed and accuracy of labeling. If you use Hugging Face 🤗 open source models for NLP applications, you can use the model-assisted labeling workflow to jumpstart your model building process.
The updated text editor makes it easy for teams to collaborate and use Labelbox for more complex NLP use cases. To learn more about how to use issues & comments and annotation relationships, read our documentation.