AI Based Analysis of Environmental Disclosure Tone and Readability

Abstract

This research introduces a novel computational framework for analyzing corporate
environmental disclosures through the dual lenses of linguistic tone and textual readability, employing artificial intelligence techniques that diverge from traditional sentiment analysis and financial text mining approaches. We propose a hybrid methodology
combining transformer-based language models with psycholinguistic dictionaries and
graph-based coherence metrics to quantify not just what is said about environmental
performance, but how it is communicated. Our approach uniquely integrates three
unconventional dimensions: (1) a multi-scale tone analysis capturing micro-sentiment
fluctuations alongside macro-narrative arcs, (2) a readability assessment that accounts
for domain-specific environmental terminology rather than general linguistic complexity, and (3) a coherence metric evaluating logical flow between environmental claims
and supporting data. We apply this framework to a corpus of 2,500 environmental
disclosures from SP 500 companies between 1995 and 2004, revealing previously undocumented patterns of strategic obfuscation in high-risk industries. Results demonstrate
that companies in environmentally sensitive sectors employ significantly more complex sentence structures when discussing negative environmental impacts compared
to positive achievements (p ¡ 0.01), while maintaining an artificially optimistic tone
through selective positive framing. Furthermore, we identify a ’readability gap’ where
environmental performance metrics are presented with 42