An ontology-based approach to quantifying metaphor-making potential

Abstract

The Conceptual Metaphor Theory argues that metaphor enacts cognition in a manner of crossing domains (Lakoff and Johnson, 1980), which has prompted various investigations into metaphor, particularly relating to its behavior and function in different environments. The paper aims to extend the research scope on metaphor by proposing an intrinsic perspective on metaphor - based on the notion that words possess a Metaphor Making Potential. Specifically, I aim at aspects associated with a word's meta metaphorical mechanism, a typical cognitive process of metaphor which depends first and foremost on a word's ability to cross domains. The upper level ontology, Suggested Upper Merged Ontology (SUMO), with its well developed hierarchy of domain terms that represent world knowledge, assists in revealing the inherent metaphorical property of a word. The lexical ontology, WordNet, decomposes a word's lexical meanings into senses in terms of synsets", which in turn can be mapped to the conceptual ontology SUMO. In this study, WordNet serves as a parser of a word's semantic senses, and SUMO identifies the domains, in which these senses belong. Experiments with English verbs under the framework of WordNet and SUMO ontology have been carried out. Eight sets of data were first drawn from the 400 million-word Corpus of Contemporary American English (COCA). The first set consists of 1000 most frequent English verbs in the general corpus. Each of the additional five sets consists of 1000 most frequent verbs respectively from such registers as "Academic", "Fiction", "Magazine", "Newspaper", and "Spoken". Two additional registers form the last two sets, i.e., "Science and Technology" under "Academic" and "Scientific Fiction and Fantasy" under "Fiction". In seeking ways to operationalize the measurement of an English verb's MMP, two novel domain-based algorithms have been designed along with another conventional depth-based metric. Results from these three algorithms were compared. A program called "MMPCalculator" was built to incorporate WordNet 3.0 and SUMO Merge 1.47. This program performs three algorithms in parallel to generate MMP values of eight groups of verbs. One hypothesis, which is based on Lakoff's view that metaphor is the result of our constant interaction with our physical and cultural environments", was established to determine whether higher frequency verbs show greater MMP values and if so, to what extent they are correlated with each other. Another hypothesis is based on the human intuitive judgment that texts of different genres bear different amount of metaphorical usage of words due to different stylistic features, which was supported by the statistical analysis of the data drawn from a survey. Given that this judgment is true, the second hypothesis was established to determine whether texts of different genres reveal different amount of MMP values. As a study which is both theory- and application-oriented, this paper also tries to show that an ontology-based approach to metaphor evaluation is de facto more objective and operable than an intuitive-based approach in providing insights into a word's metaphorical properties.