Fuzzy Metaphor Identification in Terminology: Artificial Neural Network for Reasoning in Cognitive Linguistics

Abstract

Abstract Metaphor is a cognitive phenomenon used for mediating the dialogue between experts and non-experts. To date, research on automation of metaphor identification has focused chiefly on finding metaphor-related words and metaphorical patterns, as well as establishing conceptual mappings in texts. A neglected area in the field of computational linguistics is finding metaphorical terms. This work provides insights into using an artificial neural network for reasoning in cognitive linguistics. We created a multilayer neural network and trained it using the resilient backpropagation algorithm to determine the metaphoricity level in terms. Within the course of English for professional communication, the computer security students collected terms from modern computer security discourse. Each term in the sample was estimated for the percentage of metaphoricity by cognitive linguists, who adopted acknowledged methods for metaphor identification through the prism of cognitive approach to terminology on the one hand and automation of data retrieval on the other. The latter determined our choice of input parameters. This paper has demonstrated an 86% accuracy in artificial network performance. The data presented here also highlight the aptitude of the input parameters for potentially automated preprocessing. The postprocessing stage, including evaluating the input parameters' significance and error analysis, is done automatically. The strength of our work lies in the fuzziness of human-like artificial decision making. We have obtained satisfactory results demonstrating that artificial intelligence can contribute to evaluating the cognitive potency of terms. These findings add substantially to our previous research into linguistic mediation.