Understanding how political elites relate to each other in political discourse is crucial for understanding the dynamics of compromise, conflict, and polarization. This paper studies elite relations from the perspective of politicians’ actor-targeted rhetoric in parliamentary debates, focusing on which parliamentary actors speakers mention and the polarity of these statements. While current research has explored group-based rhetoric in political communication, studies have been limited by their reliance on manual content analysis and predominantly English-language focus, particularly affecting comparative research across languages and political contexts. We address three key challenges in analyzing actor-centric rhetoric: the lack of automation, limited multilingual capabilities, and high resource requirements. Our solution introduces a new method for automatically identifying political actor mentions and their polarity in debates, reducing the need for manual annotations and computational resources. Our approach offers three main contributions: First, we develop a new multilingual dataset (English, German, and Spanish) with annotations for actor-centric utterances and debate polarization, including coding for different political actor categories. Second, we adapt an open-source LLM (LLaMA) using training data from one language, employing LoRA-adapter-training to minimize computational needs. Third, we validate our cross-lingual approach using labeled datasets in German and Spanish, demonstrating that our task-adapted LLM can enhance training data across languages through language-agnostic task representation. We train a compact student RoBERTa model for lightweight analysis and compare our method against various baselines, including zero- and few-shot prompting. Our approach enables efficient, large-scale comparisons of political actors' engagement across languages and national contexts while maintaining methodological rigor.