RuCAM: Comparative Argumentative Machine for the Russian Language

Maria Maslova, Stefan Rebrikov, Anton Artsishevski, Sebastian Zaczek, Chris Biemann, Irina Nikishina

Abstract: Comparative question answering is one of the question answering subtasks which requires not only to choose between two (or more) objects, but also to explain the choice and support it with arguments. ChatGPT-like models are able nowadays to generate a coherent answer in a natural language, however, they are not fully reliable as they are not publicly accessible and tend to hallucinate. Another solution is a Comparative Argument Machine (CAM), which however, has been developed for English only. In this paper, we describe the development of RuCAM—comparative argumentative machine for Russian, as well as the challenges of the system adaptation for another language. It is the first open-domain system to argumentatively compare objects in Russian with respect to information extracted from the OSCAR corpus. We also introduce several datasets for the RuCAM subtasks: comparative question classification, object and aspect identification, comparative sentences classification. We provide models for each subtask and compare them with the existing baselines.