bert-apc-detector-en

BERT APC Detector (English)

Model Details

• Base Model: bert-base-uncased
• Task: Aboutness Person Coreference (APC) detection via token classification
• Label Scheme: BILOU (Begin, Inside, Last, Unit, Outside)
• Training Data: semi-manually extracted and annotated from British National Corpus (English)

Intended Use

Detect mentions of adnominal person constructions ("we happy singers", "you clever inventors", "you idiot") in English text.

Training Procedure

• Optimizer: AdamW
• Learning Rate: 2e-5
• Batch Size: 16
• Epochs: 3
• Hardware: GPU

How to Use

from transformers import AutoTokenizer, AutoModelForTokenClassification

tokenizer = AutoTokenizer.from_pretrained("gfkpth/bert-apc-detector-ger")
model = AutoModelForTokenClassification.from_pretrained("gfkpth/bert-apc-detector-ger")

Evaluation Results

Evaluated on held-out test set with 1732 example sentences.

Repository

GitHub

Citation

If you use this model, please cite:

@software{hoehn2026bert-apc-en,
  title={BERT APC Detector English},
  author={Georg F.K. Höhn},
  year={2026},
  url={https://huggingface.co/gfkpth/bert-apc-detector-en}
}