A Novel Distributional Approach to Multilingual Conceptual Metaphor Recognition

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1752–1763, Dublin, Ireland, August 23-29 2014. A Novel Distributional Approach to Multilingual Conceptual Metaphor Recognition Michael Mohler and Bryan Rink and David Bracewell and Marc Tomlinson Language Computer Corp. Richardson, Texas, USA {michael,bryan,david,marc}@languagecomputer.com Abstract We present a novel approach to the problem of multilingual conceptual metaphor recognition. Our approach extends recent work in conceptual metaphor discovery by combining a complex methodology for facet-based concept induction with a distributional vector space model of lin- guistic and conceptual metaphor. In the evaluation of our system in English, Spanish, Russian, and Farsi, we experiment with several state-of-the-art vector space models and demonstrate a clear benefit to the fine-grained concept representation that forms the basis of our methodology for conceptual metaphor recognition. 1 Introduction The role of metaphor in language has been defined by Lakoff et al. (1980; 1993) as a cognitive phe- nomenon which operates at the level of mental processes, whereby one concept or domain is viewed systematically in terms of another. For example, the phrase “to cure poverty” is a metaphor which subtly conveys a wide variety of information to the listener. In order to mentally process this phrase, we must first recognize that a metaphor is being used and that “cure” (as a medical term) is being used figura- tively. Then, we assume some relationship between “poverty” and “things that can be medically cured“ which leads to the conceptual mapping “POVERTY as DISEASE.” This conceptual mapping enables the listener to transfer a variety of properties and associations between the two concepts, such as their as- sociation with a feeling of helplessness, the existence of sustained efforts to end them, the potential for them to spread, and their mutual relationship with ill-health and death. Therefore, by identifying the con- ceptual domains associated with this linguistic metaphor, we are able to reason about the target domain (POVERTY) using concepts and terms associated with the source domain (DISEASE). Any natural language processing system capable of processing metaphor in text with human-level competence must, therefore, overcome three problems in sequence: 1. the identification of metaphorical expressions (also known as linguistic metaphors (LMs)) 2. the discovery of a conceptual domain mapping or conceptual metaphor (CM) which consists of (a) the conceptual domain of the metaphor target (e.g., POVERTY); and (b) the conceptual domain of the metaphor source (e.g., DISEASE) 3. the real-world interpretation of the metaphorical text which uses the conceptual metaphor frame- work to transfer knowledge between the source and target domains. While a significant amount of recent work has presented interesting and promising methodologies for multilingual LM identification (Shutova and Sun, 2013; Wilks et al., 2013; Strzalkowski et al., 2013), the work presented in this paper is focused on (2), the problem of multilingual CM recognition, which will be made to serve as the foundation for a more fine-grained interpretation of metaphor. This work is licenced under a Creative Commons Attribution 4.0 International License. Page numbers and proceedings footer …

Content truncated to 500 words