A new methodology for conceptual metaphor detection and formulation in corpora: A case study on a mental health corpus

Abstract

We describe a new methodology for conceptual metaphor detection and formulation in corpora, developed within the framework of the MOMENT project for analysing mental health metaphors. We critically review state-of-the-art methods for metaphor identification in texts, highlighting their main drawbacks for metaphor analysis in large corpora, mainly practical applicability and analytical subjectivity. Our method aims at mitigating existing drawbacks on the basis of applying the following principles: (i) working hypothesis formulation and verification at the metaphorical expression detection stage; (ii) partial use of standard methods for metaphorical focus identification; (iii) use of external expert knowledge in the form of more extensive use of dictionaries and the additional use of metaphor compendia; and (iv) the implementation of strategies for conceptual metaphor formulation, including domain formulation at two levels of generalization. Satisfactory reliability test results were obtained whenwetestedourmethodforinter-annotatoragreementregardingmetaphor detection and formulation using texts about mental disorders as a test corpus.