Toward Transparent ESG Reporting: Analyzing Promise and Constraint Claims

ACL ARR 2025 May Submission4046 Authors

Abstract: In the era of increasing regulatory scrutiny and stakeholder expectations, understanding how companies frame their sustainability commitments and limitations is essential for assessing corporate accountability. This study explores promise and constraint claims made by 1,898 companies worldwide from 2020 to 2024. Promises are forward-looking claims that tend to lack specific, measurable actions or mechanisms for accountability, while constraints - are sentences that mention some impediments, restrictions, or obstacles at a company, society, or governmental level that may restrict the company from fulfilling its promises. Our study provides a rigorous and well-defined definition of promise and constraint claims based on the sustainability reports, the terms that have not been explored before, and offers a comprehensive dataset of 5,386 annotated sentences from 2,221 reports. The research presents a lightweight alternative to resource-intensive models by employing ClimateBERT and fine-tuning it as a ClimateBERT-Promise-Constraint model on the collected data. The analysis identifies the distribution of constraint claims across four primary sectors: natural resources, manufacturing, retail, and information. This work contributes a comprehensive dataset and modeling framework, supporting future research on corporate accountability and transparency within environmental, social, and governance (ESG) disclosures, aligned with emerging regulations such as the EU Corporate Sustainability Reporting Directive (CSRD).

Paper Type: Long

Research Area: Resources and Evaluation

Research Area Keywords: corpus creation, language resources, NLP datasets, evaluation methodologies, evaluation, metrics, reproducibility, statistical testing for evaluation

Contribution Types: Data resources, Data analysis

Languages Studied: English

Submission Number: 4046