Paper

This paper is currently under review. You can view the full PDF submission at here.

abstract

Relevant market definitions are central to merger and antitrust analysis, but extracting them at scale from European Commission decisions is difficult because the documents are numerous and heterogeneous in structure and language. We present Lextract, an open-source Python pipeline that combines document collection and filtering with large-language-model-based identification of market-definition sections and individual definitions. The extracted records are stored as structured JSON and power JurisMercatus, a semantic-search interface for market definitions. We evaluate 4,105 definitions extracted from more than 600 Commission decisions using a structural validity procedure derived from 50 manually verified definitions. Overall, 76.6% satisfy all validity criteria; the main failure mode is the absence of a recognized structural marker. The results show that Lextract can produce a large, reproducible collection of structurally well-formed market definitions while also revealing the need for a larger gold-standard evaluation of end-to-end extraction accuracy.

keywords

legal information extraction, competition law, relevant market definition, large language models, European Commission

acknowledgements

I am grateful to Professor Thibault Schrepel of Stanford Law School for his invaluable advisement and guidance throughout the course of this project. This research received no funding from any government agency, university, company, or non-profit organization.