katrohrb/embodiment-classifier

Embodiment Classifier (English narrative fiction)

Sentence-level classifier assigning one of 11 embodiment categories to sentences from English-language prose fiction. Fine-tuned from roberta-base.

Labels: no_rel, character_move, haptic_material, character_feltness, facial_expr, sight_human, book_paratext, haptic_human_animal, sight_material, segmentation_error, gesture.

Usage

from transformers import pipeline
clf = pipeline("text-classification", model="katrohrb/embodiment-classifier")
clf("She reached out and touched the rough bark.")
# -> [{'label': 'haptic_material', ...}]

Performance

Held-out test set (n=580): micro-F1 0.805 [0.772, 0.835], macro-F1 0.797. 5-fold CV: 0.808 +/- 0.022. Human inter-annotator agreement ceiling: Cohen's kappa 0.760.

Data and code

Codebook, gold annotations, and reproduction notebooks: https://github.com/katrin-rohrbacher/embodiment-paper

Citation

Rohrbacher, K., Reincke, J. & Mahlberg, M. (2027). Modeling Embodiment in Narrative Fiction. Proceedings of the Conference on Computational Humanities Research (CHR 2027), Manchester. Long paper. Forthcoming (accepted).