Is There Anything More Deceptive than an Obvious Fact? Investigating Implicitness in User-Generated Argumentative Text

Abstract

While various attempts towards unveiling implicitness in argumentation have been made, particularly towards improving automatic detection and reconstruction of implicit components and background knowledge, the task remains overly challenging. In this paper, we present, to the best of our knowledge, the first fine-grained typology of implicitness in argumentation, distinguishing among implicature, ambiguity, and presupposition. Applying this typology, we annotate 78 full-length discussions from the Change My View forum, building the largest publicly available dataset of real-world enthymemes with implicitness types labeled. For comparison, we additionally annotate 112 short argumentative texts from the Microtext corpus to examine how text length and complexity influence the automatic analysis of natural arguments. Leveraging these datasets, we establish strong baselines for two tasks: (i) enthymeme detection and (ii) fine-grained implicitness classification, with both encoder-only and large language models, highlighting the challenge of modeling implicit reasoning in long, unstructured discourse.

Resources

Details