Elamur Research⌕
☰
home / papers / 2026 / Documenting Corporate Harm: A Semantic Action Trajectories Approach to the Opioid Industry Document Archive Shared Task

Documenting Corporate Harm: A Semantic Action Trajectories Approach to the Opioid Industry Document Archive Shared Task

Paper · 2026 · Ben Miller

Summary

The paper addresses how to systematically analyze the Opioid Industry Document Archive (OIDA), a litigation archive that is huge and heterogeneous. Its goal is to trace how different actors' possible actions change over time. The author randomly sampled 10,000 OCR documents (about 238 million tokens) and ran semantic role labeling with the Emory Language and Information Toolkit (ELIT). ELIT is a deliberately non-neural, interpretable pipeline, and it was used to pull out subject–predicate pairs. Each pair was linked to years mentioned in the text and placed in one of six periods between 1980 and 2019. Subjects were sorted into a 17-group actor ontology, first by rules and then by a constrained language model for ambiguous cases. Each actor group's predicates received LIWC category scores, which were then condensed into a low-dimensional discourse space by choosing features with a random forest and applying PCA. Mapping trajectories through this space shows that named firms and their opioid products follow almost the same path, so companies and drugs appear to play interchangeable roles. Over time, discourse moves away from products and patients toward organizational coordination and regulators. The approach gives a transparent way to study institutional behavior in large legal archives.

Key points

  • Extraction output: 246,073 SRL predicate instances, 184,831 subject–predicate pairs, 52,305 normalized subjects and 4,548 unique predicates. After filtering, 86,414 subject–predicate–year triples remain, concentrated in 2010–2014 (49,488).
  • LLM adjudication: 267 candidate subjects were classified using a 0.9 confidence cutoff. This recovered names such as Kadian and Eluxadoline and added 10,086 triples.
  • Feature pipeline: LIWC categories feed random forest selection and PCA. Within-period z-scores reduce the effect of temporal imbalance between periods.
  • Main finding: the corporate_entities and commercial_products groups have nearly identical semantic trajectories.
  • Email check and limitations:
    • Email-style markers rise from 0% (1980–1994) to 18.8% (2010–2014). This suggests the shift is partly, but not entirely, a change in communication medium.
    • Pre-2000 data are sparse.
    • OCR noise remains a limitation.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of CanterburyChristchurch · New Zealand5 works · 274 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksGeneral Social Sciences · 227 worksPolitical Science and International Relations · 182 works

Cite

Cite as Ben Miller. 2026. Documenting Corporate Harm: A Semantic Action Trajectories Approach to the Opioid Industry Document Archive Shared Task. Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.nlpcss-1.10

@inproceedings{miller2026documenting,
  title = {{Documenting Corporate Harm: A Semantic Action Trajectories Approach to the Opioid Industry Document Archive Shared Task}},
  author = {Ben Miller},
  booktitle = {Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science},
  year = {2026},
  doi = {10.18653/v1/2026.nlpcss-1.10},
  url = {https://doi.org/10.18653/v1/2026.nlpcss-1.10}
}

References

11 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. G. Caleb Alexander, Lisa A. Mix, Sayeed Choudhury, Rachel Taketa, Cecilia Tomori, Mehdi Mooghali, Andrew Fan, Sarah Mars, Daniel Ciccarone, Michael Patton, Dorie E. Apollonio, Laura Schmidt, Michael Steinman, Jeremy Greene, Pamela Ling, Andrew K. Seymour, and Stanton Glantz. 2022. The opioid industr… find
  2. Daniel Gildea and Daniel Jurafsky. 2002. Automatic labeling of semantic roles. Computational Linguistics, 28(3):245–288.
  3. Han He, Liyan Xu, and Jinho D Choi. 2021. Elit: Emory language and information toolkit. arXiv preprint arXiv:2109.03903. — [arXiv:2109.03903](https://arxiv.org/abs/2109.03903) find
  4. Xinxin Li, Huiyao Chen, Chengjun Liu, Jing Li, Meishan Zhang, Jun Yu, and Min Zhang. 2025. Llms can also do well! breaking barriers in semantic role labeling via large language models. In Findings of the Association for Computational Linguistics: ACL 2025, pages 23162–23180. find
  5. Weeda Mehran, Ben Miller, and Stephen Herron. 2025. Nothing in common? analysis of moral, psychological, and social factors in the identity construction of far-right and violent jihadi extremists. Studies in Conflict & Terrorism, pages 1–24. find
  6. Ben Miller, Jennifer Olive, Shakthidhar Gopavaram, Yanjun Zhao, Ayush Shrestha, and Cynthia Berger. 2015. A method for cross-document narrative alignment of a two-hundred-sixty-million word corpus. In 2015 IEEE International Conference on Big Data (Big Data), pages 1673–1677. IEEE. find
  7. James W Pennebaker, Ryan L Boyd, Kayla Jordan, and Kate Blackburn. 2015. The development and psychometric properties of liwc2015. \#\# Acknowledgements I thank the organizers of the NLP+CSS OIDA Shared Task; the United States Bankruptcy Court for the Southern District of New York, whose orders in the… find
  8. James Pustejovsky, José M Castano, Robert Ingria, Roser Sauri, Robert J Gaizauskas, Andrea Setzer, Graham Katz, and Dragomir R Radev. 2003. Timeml: Robust specification of event and temporal expressions in text. New directions in question answering, 3:28–34. find
  9. Yla R Tausczik and James W Pennebaker. 2010. The psychological meaning of words: Liwc and computerized text analysis methods. Journal of language and social psychology, 29(1):24–54. find
  10. Nalini Vadivelu, Alice M Kai, Vijay Kodumudi, Julie Sramcik, and Alan D Kaye. 2018. The opioid crisis: a comprehensive overview. Current pain and headache reports, 22(3):16. find
  11. George Kingsley Zipf. 1949. Human Behavior and the Principle of Least Effort. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/miller-2026-documenting-corporate-harm-semantic-action/