Mapping Policy Interdependencies in EU Electromobility: An NLP-Assisted Approach to Identify Policy Synergies and Trade-Offs in Textual Data

Climate policy is becoming increasingly complex due to a multitude of policies interacting across governance layers and sectors. Empirical research has shown that policy-making processes can result in policy outputs that create trade-offs with other public policies. For example, a negative interaction exists between vehicle efficiency standards and Zero Emission Vehicle (ZEV) purchase subsidies, because efficiency standards may reduce the total cost of ownership (TCO) of internal combustion engine (ICE) vehicles, making them comparatively more attractive than purchasing and maintaining ZEVs and thereby weakening the signal of ZEV subsidies. However, no systematic data collection mechanisms currently exist that would enable a comprehensive study of such inconsistencies within a given policy domain. As a result, no systematic databases on EU climate policy interdependencies are available that distinguish between synergies and trade-offs. Textual data offer clues on policy interdependencies: legislative documentation, peer-reviewed literature, and media reports contain scattered information on how climate policies interact. Nevertheless, systematic assessments of policy interactions across diverse and voluminous text corpora are lacking. To address this gap, the paper poses the following research question: How can textual analysis be used to systematically identify and classify synergistic and conflicting interactions between EU electromobility policy instruments? The research objective is achieved by applying natural language processing (NLP) techniques on a corpus of EU legislative acts for the automotive sector to retrieve their policy elements. Subsequently, legislative documentation, peer-reviewed literature, and media reports are sought and annotated based on the subject matter, claimed evidence for inter-policy causal effects, and assumed policy elements. The approach illustrates how a multi-mode, qualified network between legislative acts can be built and interpreted. The results aim to uncover whether legal, media, expert, and scientific assessments of policy coherence diverge. The contribution of the paper is primarily methodological, employing a bottom-up approach to develop a reproducible annotation scheme and an NLP model, with further implications for fields such as legal AI and NLP for the detection of policy interdependencies in textual data.