Get Started
Home
Topics
Search
Library
Research questionHow reliably can political-evasion classifiers transfer across languages and political contexts for pragmatically ambiguous responses?Political answers can engage with a question while withholding the requested information, and their evasiveness may depend on subtle pragmatic cues. It remains unclear whether taxonomies and classifiers preserve their distinctions across languages and political contexts when those cues are ambiguous.
AI
Evaluation & Benchmarks
Machine Learning
Natural Language Processing
Latest papersRecent research connected to this question, newest first.PolERo: Studying Political Evasion in RomanianThe evidence comes from PolERo, a dataset of 3,574 human-annotated Romanian question-answer pairs drawn from official transcripts of five presidents, evaluated alongside an English dataset using a two-level response-clarity and fine-grained evasion taxonomy. The study examines TF-IDF baselines, fine-tuned encoders, a sliding-window encoder, zero- and few-shot LLM prompting, bilingual training, and machine-translation augmentation; it finds asymmetric cross-lingual transfer and persistent difficulty with ambivalent evasion categories. Findings are limited to the studied datasets, languages, political contexts, and classification settings.research paper · Sep 2, 2026
Related questions
How should LLM safety be assessed when jailbreak vulnerability varies by language and persuasive phrasing?How susceptible are large language models to conspiratorial responses under demographic conditioning?How can classification models adapt to removed labels without retraining on the original data?How can LLMs predict flexible workers’ responses to hypothetical pension policies without costly field experiments?