Skip to content

Relation extraction appears partially broken / incomplete #101

Description

@BorysByne

Summary

The public documentation presents relation extraction as a supported feature, including examples that pass relation descriptions for improved accuracy.

However, in the current implementation, relation extraction appears incomplete and inconsistent with the documented API.

Observed issue

  • Relation extraction is documented as a first-class capability in the local library.
  • The wrapper does route relation tasks through a dedicated relation extraction path.
  • However, relation descriptions provided in the schema do not appear to be preserved or used by the local relation extraction pipeline, despite being documented as part of the API.
  • As a result, the documented relation API and the implemented wrapper behaviour appear inconsistent.
  • In addition, the current relation extraction logic appears to take spans[0] for relation arguments, which suggests the selected head/tail may not be the best-scoring candidate when multiple spans pass threshold.

Why this matters

This makes relation extraction in the local wrapper difficult to trust for downstream use, especially for knowledge graph construction or structured information extraction workflows where relation definitions and argument selection quality matter.

Additional note

It looks likely that some relation-handling code was lost or regressed in a recent commit, although we were not able to reliably trace which commit introduced the issue.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions