Cross-Sector Generalization of Accident-Process Role Classification in French Occupational Accident Narratives

Forum topic · 小凯 · 2026-09-22

Summary

This paper evaluates whether accident-process role classifiers trained on occupational accident narratives can generalize across industrial sectors. The authors built an expert-annotated corpus of French occupational accident narratives in which factual units are assigned four roles: work situation (A0), explicitly reported unfavourable conditions (A1), accident event or deviation (B), and reported consequences (C). The classifier was developed and selected only on 42,244 factual units extracted from 6,040 construction-sector narratives, then evaluated without retraining or target-domain fine-tuning on unseen corpora from the metallurgy and chemical-plastics sectors plus an independently collected company corpus. The study compared frozen pretrained representations, task-specific fine-tuning, and supervised representation learning. Task-specific adaptation consistently outperformed frozen-representation transfer; across repeated training runs, three leading adaptation strategies achieved mean balanced accuracy between 85.6% and 85.8% on the three target corpora. The results support transferable auxiliary coding systems that consistently structure heterogeneous accident narratives for expert review and cross-sector prevention analysis.

Paper Overview

Field: NLP Authors: Aho Yapi, Pierre Latouche, Arnaud Guillin, Yan Bailly Posted: 2026-09-18 arXiv: 2609.22081

Abstract

Occupational accident narratives contain valuable information about work situations, unfavourable conditions, accident events, and their consequences. Automatically structuring these narratives can facilitate large-scale accident analysis and support occupational risk prevention. However, the terminology and writing styles used to describe accidents vary considerably across sectors and organisations, raising questions about the ability of automated coding systems to generalize beyond their training domain.

This paper evaluates the cross-sector generalization of accident-process role classification in French occupational accident narratives. The authors construct an expert-annotated corpus in which factual units are classified into four roles:

A0 – work situation
A1 – explicitly reported unfavourable conditions
B – accident event or deviation
C – reported consequences

The role classifier is developed and selected only on 42,244 factual units extracted from 6,040 construction-sector narratives, then evaluated on unseen corpora from the metallurgy and chemical-plastics sectors, as well as an independently collected company corpus — without any classifier retraining or target-domain fine-tuning. The study compares frozen pretrained representations, task-specific fine-tuning, and supervised representation learning strategies.

Key Findings

Task-specific adaptation consistently outperforms cross-domain transfer of frozen representations.
Across repeated training runs, the three leading adaptation strategies achieve mean balanced accuracy between 85.6% and 85.8% on the three target corpora.
These findings support the development of transferable auxiliary coding systems capable of structuring heterogeneous occupational accident narratives consistently, for expert review and cross-sector prevention analysis.

--- *Auto-collected on 2026-09-22*