Measuring ESG Disclosure Quality Using Bilingual Natural Language Processing: A Proposed Methodological Framework for Large-Cap Saudi Tadawul-Listed Companies

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Provides a methodological framework and corpus design for bilingual ESG quality measurement, with an initial classifier benchmark for future research.

🏢実務担当者:Offers a structured approach to evaluate ESG disclosure quality which can be adapted for internal audit or investor relations in bilingual report environments.

🏛政策担当者:Demonstrates the need for quality metrics in ESG disclosure regulation, potentially informing standards for bilingual reporting.

Firms' disclosure of environmental, social, and governance (ESG) information gives invaluable insights to investors and helps stock markets function effectively. Existing research offers useful techniques for measuring the quantity of ESG disclosure, but there has been too little focus on quality indicators. Unresolved problems also exist in dealing with bilingual source material. To address these issues, this paper presents a methodological framework for measuring the quality of ESG disclosures in a bilingual (English-Arabic) setting. It designs a natural language processing (NLP) pipeline capable of analyzing ESG disclosure quality in the annual reports of companies listed on the Saudi stock exchange, the Tadawul. The ESG-C framework processes Arabic- and English-language content separately, since both are used in Saudi corporate reports. It measures disclosure quality using a weighted index comprising verifiability, sector specificity, quantitative clarity, and standards alignment. The framework is used to compile a structured corpus of 250 annual reports from 50 Tadawul-listed companies (21.6% of listed firms) spanning the fiscal years 2019-2023. The sample includes only firms with complete, machine-readable reports and a market value of at least SAR 500 million, leading to a large-cap focus and the exclusion of 8 of the 21 industry groups represented on the Tadawul. The study does not empirically test the entire NLP pipeline, validate the combined index, or present empirical results on ESG disclosure quality, but it reports an initial sentence-level ESG relevance classifier with an F1 score of 0.87 on a 300-sentence hold-out set, offering a transparent measurement tool and a well-documented corpus design for future validation.

• semanticscholar http://www.refaad.com/Files/GJEB/GJEB-16-3-9.pdffirst seen 2026-07-05 05:36:08