Skweak

When it fits

Use Skweak when manual annotation is costly or task-specific labeled data is scarce, and you can define labeling signals and work with spaCy documents.

Who it is for

• NLP developers preparing task-specific training data
• Teams working with resource-poor languages or specialized text domains
• spaCy users who want to combine multiple weak labeling sources

Documented uses

• Create labeling functions based on heuristics, gazetteers, machine-learning models or crowd annotations.
• Apply multiple labeling functions to unannotated text represented as spaCy Doc objects.
• Aggregate conflicting annotations while estimating labeling functions’ relative accuracy.
• Prepare labeled corpora for training a model in a chosen framework.