Pipeline for adding Backchannel= and Coconstruct= annotations to SST speaker-view CoNLL-U files.

• src/sst/
  • source SST CoNLL-U files (train, dev, test, merged)
• lexicon/
  • lexical resources (e.g., lexicon/sl_backchannels.txt)
• scripts/
  • numbered workflow scripts (01 to 07)
  • see scripts/README.md for script-level details
• docs/
  • process documentation
  • docs/BACKCHANNELS_EXTRACTION.md
  • docs/COCONSTRUCTIONS_EXTRACTION.md
• output/sst/
  • extracted candidates and generated annotated corpora
  • final package: output/sst/final_bc_coco/

1. Merge corpus

python3 scripts/01_merge_sst.py

2. Extract backchannel candidates

python3 scripts/02_extract_backchannel_candidates.py

3. Apply backchannels (uses filtered rows from the candidate table)

python3 scripts/03_apply_backchannel_annotations.py

4. Extract coconstruction candidates

python3 scripts/04_extract_coconstruction_candidates.py

5. Manual coconstruction annotation (outside script)

• fill is_coconstruction, coconstruct_deprel, governor_token_id for YES cases

6. Apply coconstructions

python3 scripts/05_apply_coconstruction_annotations.py

7. Split final merged file back to train/dev/test

python3 scripts/06_split_final_corpus.py

8. Run strict diff checks

python3 scripts/07_diffcheck_final_vs_src.py

• Backchannels workflow: docs/BACKCHANNELS_EXTRACTION.md
• Coconstructions workflow + manual annotation: docs/COCONSTRUCTIONS_EXTRACTION.md
• Docs index: docs/README.md

• output/sst/final_bc_coco/conllu/sl_sst-ud-merged.conllu
• output/sst/final_bc_coco/conllu/sl_sst-ud-train.conllu
• output/sst/final_bc_coco/conllu/sl_sst-ud-dev.conllu
• output/sst/final_bc_coco/conllu/sl_sst-ud-test.conllu