Context Matters! Identifying Argumentative Relations in Essays

Context is crucial for identifying semantic relations between input sequences embedded in a document. This paper studies how different levels of context, coming from discourse markers or from a paragraph-wide context window, impact the capability of language models to identify inference relations. Empirical results for the task of argumentative relation identification show that the proposed paragraph-aware model provides considerable improvements over a context-agnostic baseline, surpassing the usage of discourse markers alone.