Making the Arguer Visible: Reconstruction Practices and the Representation of Speaker Intent

Argument reconstruction is typically treated as a technical procedure aimed at identifying underlying argumentative content. In this article, however, we show that reconstruction actively shapes how the speaker is represented in annotated data and thereby affects how argumentative action is modelled. Our analysis is based on an experimental manual annotation of a purpose-built dataset, combined with a critical review of dominant reconstruction practices employed in existing Argument Mining resources, in particular within the framework of Inference Anchoring Theory. We identify recurrent reconstruction operations and analyse their consequences for the representation of argumentation, especially with regard to the visibility of the speaker as an agent of argumentative action. Our analysis of annotation practices further reveals systematic interpretative patterns in which argumentative meaning is extracted from forms that are ostensibly non-argumentative, such as autobiographical narratives. We show that reconstruction often involves a pragmatic transformation of an utterance’s function rather than merely the completion of locutionary content.