Coreference Resolution with POS Tagging

This repository contains the code for the COMP542 – Natural Language Processing final project at Koç University. The project explores whether Part-of-Speech (POS) tagging can improve coreference resolution by integrating learned POS embeddings into transformer-based models (primarily BERT).

The project achieves the following:

• Implements a POS-aware coreference model that combines BERT representations with POS-tag embeddings.
• Evaluates on OntoNotes v5.0 (CoNLL-2012) and GAP datasets using binary coreference classification and span index prediction tasks.
• Compares POS-enhanced models against non-POS baselines to assess impact.

For methodology, datasets, training setup, and results, see the project report.