Filter and Annotate: Towards Automatic Identification of Genuine Metaphoricity

摘要

Natural Language Processing has largely addressed automatic metaphor detection on grounds of (cognitive) linguistic frameworks, especially the Conceptual Metaphor Theory of Lakoff and Johnson [1], which sees metaphor as ubiquitous. In contrast, in this work we view metaphor as an exceptional phenomenon [2]. This change in perspective affects applicability of machine learning approaches for metaphor detection, usage of corresponding features, as well as availability of datasets. We propose a combination of manual annotation and automatic filtering as an approach to conduct first steps into the direction of genuine metaphor detection.