Conceptual Metaphor Theory Guides GANs for Generating Metaphors and Interpretations

Abstract

Metaphor is a distinctive linguistic phenomenon, pervasive in real-life communication. The generation and comprehension of metaphors are intricately connected, making metaphor interpretation and generation enduringly challenging tasks in the field of natural language processing. Existing methods for crafting metaphorical sentences often grapple with the production of metaphorical expressions that innovatively align with real-world examples. Additionally, generating metaphorical interpretations from unlabeled metaphorical sentences presents the inherent challenge of domain specificity. In this paper, we proposed GMAI (GANs Metaphor And Interpretation), an adversarial generative model guided by Conceptual Metaphor Theory. This model seamlessly integrates modules for both metaphor generation and metaphor interpretation, facilitating the creation of metaphorical sentences and their corresponding interpretations. GMAI has the capability to employ information labels for identifying specific words within sentences and transforming them into metaphorical expressions tailored to particular domains. Furthermore, the proposed model can generate a spectrum of metaphorical expressions by modifying domain labels. The downstream metaphor interpretation module, working with the generation module, can get a wealth of novel text. It is trained on sentences generated across various domains in an unlabeled state, thus enhancing the precision of the metaphor interpretation. The experimental findings underscore improvements in the precision and stability of both generated metaphors and their interpretations when compared to prior models.