Template-based approaches have advanced Event Argument Extraction (EAE) by formulating it as a structured prediction problem. However, existing methods mainly rely on either fixed manually designed templates, which suffer from template under-matching in diverse contexts, or exhaustive enumeration of template permutations, incurring prohibitively high computational cost. Moreover, they often overlook the semantic heterogeneity across events. To address these challenges, we propose ATSR, an Adaptive Template Selection and Role-Aware Routing framework for document-level multi-event EAE, integrating an Adaptive Multi-structure Template Selection (AMSTS) module and a Role-Aware Mixture-of-Experts (RAMoE) module. Specifically, AMSTS leverages a set of pre-defined multi-structure templates for each event type and adaptively selects the most compatible one for each event based on learnable weights conditioned on trigger representations. The RAMoE operates over template-aligned slots and routes each role to specialized experts for event semantic modeling. Extensive experiments on three benchmark datasets demonstrate that ATSR achieves state-of-the-art performance with competitive efficiency, while remaining effective in few-shot settings.

The climate crisis is a salient issue in online discussions, and hypocrisy accusations are a central rhetorical element in these debates. However, for large-scale text analysis, hypocrisy accusation detection is an understudied tool, most often defined as a smaller subtask of fallacious argument detection. In this paper, we define hypocrisy accusation detection as an independent task in NLP, and identify different relevant subtypes of hypocrisy accusations. Our Climate Hypocrisy Accusation Corpus (CHAC) consists of 420 Reddit climate debate comments, expert-annotated into two different types of hypocrisy accusations: personal versus political hypocrisy. We evaluate few-shot in-context learning with 6 shots and 3 instruction-tuned Large Language Models (LLMs) for detecting hypocrisy accusations in this dataset. Results indicate that the GPT-4o and Llama-3 models in particular show promise in detecting hypocrisy accusations (F1 reaching 0.68, while previous work shows F1 of 0.44). However, context matters for a complex semantic concept such as hypocrisy accusations, and we find models struggle especially at identifying political hypocrisy accusations compared to personal moral hypocrisy. Our study contributes new insights in hypocrisy detection and climate change discourse, and is a stepping stone for large-scale analysis of hypocrisy accusation in online climate debates.