Corporate Social Responsibility (CSR) reports are a key communication tool through which organizations disclose their sustainability efforts and principles. However, the lack of regulation in these reports often leads to the use of boilerplate language, potentially resulting in greenwashing. This thesis examines the ability to classify sustainability disclosures as substantive or symbolic using Natural Language Processing (NLP). By employing Stanza, the research replicates and extends previous work done by Huq and Carling (2024). It expands the scope from greenhouse gas (GHG) emissions to the entire ESG spectrum. The methodology leverages dependency parsing as the core technique involved in the classification algorithm. Findings reveal significant trends in regional, organizational, and sustainability area specific substantive disclosure ratios. Despite the relatively high accuracy of classification algorithm, limitations persist in handling complex sentence structures and variability in linguistic patterns. These insights contribute to the understanding of CSR communication quality, highlighting the need for enhanced reporting frameworks and tools to detect greenwashing.