Exploiting a semantic annotation tool for metaphor analysis

In this paper, we describe the development and exploitation of a corpus-based tool for the identification of metaphorical patterns in large datasets. The analysis of metaphor as a cognitive and cultural, rather than solely linguistic, phenomenon has become central as metaphor researchers working within ‘Cognitive Metaphor Theory’ have drawn attention to the presence of systematic and pervasive conventional metaphorical patterns in ‘ordinary’ language (e.g. I’m at a crossroads in my life). Cognitive Metaphor Theory suggests that these linguistic patterns reflect the existence of conventional conceptual metaphors, namely systematic cross-domain correspondences in conceptual structure (e.g. LIFE IS A JOURNEY). This theoretical approach, described further in section 2, has led to considerable advances in our understanding of metaphor both as a linguistic device and a cognitive model, and to our awareness of its role in many different genres and discourses. Although some recent research has incorporated corpus linguistic techniques into this framework for the analysis of metaphor, to date, such analyses have primarily involved the concordancing of pre-selected search strings (e.g. Deignan 2005). The method described in this paper represents an attempt to extend the limits of this form of analysis. In our approach, we have applied an existing semantic field annotation tool (USAS) developed at Lancaster University to aid metaphor researchers in searching corpora. We are able to filter all possible candidate semantic fields proposed by USAS to assist in finding possible ‘source’ (e.g. JOURNEY) and ‘target’ (e.g. LIFE) domains, and we can then go on to consider the potential metaphoricity of the expressions included under each possible source domain. This method thus enables us to identify open-ended sets of metaphorical expressions, which are not limited to predetermined search strings. In section 3, we present this emerging methodology for the computer-assisted analysis of metaphorical patterns in discourse. The semantic fields automatically annotated by USAS can be seen as roughly corresponding to the domains of metaphor theory. We have used USAS in combination with key word and domain techniques in Wmatrix (Rayson, 2003) to replicate earlier manual analyses, e.g. machine metaphors in Ken Kesey’s One Flew Over the Cuckoo’s Nest (Semino and Swindlehurst, 1996) and war, machine and organism metaphors in business magazines (Koller, 2004a). These studies are described in section 4.