The Boundaries of Discourse Markers – Drawing Lines through Manual and Automatic Annotation

Abstract

Discourse markers are non-propositional linguistic items that are notoriously difficult to identify as well as to categorize. We can observe several borderline phenomena and overlaps with other formal and functional categories, e.g. inserts, adverbials, contextualization cues, pragmatic force modifiers, etc. By way of addressing such overlaps as well as the disambiguation between DM uses and their source categories, the paper presents a comparison of automated and manual annotation of oral discourse markers (DMs). Firstly, an overview of the criterial features of DMs that are relevant to disambiguation are presented. Secondly, the UCREL Semantic Analysis System (USAS) and its disambiguation methods are briefly discussed. In the third part of the paper, manual and automatic decisions about categorization are compared with a view to addressing the margin of error reported to apply in general semantic annotation as well as the question of what formal-functional properties of the relevant DMs might explain possible differences between manual and automatic annotation.

Keywords

• discourse markers
• automated semantic annotation
• manual annotation
• d-function ratio
• inter-annotator agreement