Bilingual RST Parsing — DE-MIX Fork

This fork preserves the original repository (see README_original.md) and adds a reproducible path to train on a mixed German RS3 corpus (PCC, APA-RST, PARADISE).

Quick links:

• Original README: README_original.md
• German mixed-corpus guide: README_DEMIX.md

Highlights in this fork:

• DE-MIX corpus wiring in dmrst_parser/data_manager.py (fixed splits under data/de_mix_file_lists/).
• German label normalization germanMixed_labels in dmrst_parser/src/corpus/relation_set.py.
• Utility to inventory RS3 relations: utils/collect_rs3_relations.py.
• Segmentation evaluation utility with JSON output: utils/eval_segmentation.py (see README_DEMIX.md).

Final (segmentation reference):

• Checkpoint: saves/de_DE-MIX_default+base_linear_hints_40 (base + linear + sentence hints).
• Token-boundary F1 on a 33-doc slice: Macro 0.776, Micro 0.744 (JSON under results/seg_eval/).

To train on DE-MIX, follow README_DEMIX.md.