Integration of Linguistic Features and Computational Methods in the Semantic Classification of Literary Metaphors

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This abstract explores the potential of jointly applying linguistic features and computational methods to the semantic classification of literary metaphors. The objective of the study is to develop a classification framework that helps identify which concept is being depicted in a metaphorical expression, which attribute is mapped from another conceptual domain, and what meaning this transfer creates within the text. To this end, the distinction between literal and contextual meanings of words, semantic relations in word collocations, the source and target domains of the metaphor, and the broader literary context are adopted as key linguistic features. The abstract proposes a procedure for semantically defining metaphors by first extracting metaphorical units from the text and then characterizing the underlying relationship behind the semantic shift. For example, when an individual's emotional state is expressed through words related to temperature, motion, or weight, the depicted state and the transferred attribute are recorded separately. Computational methods can be employed for tokenizing and segmenting text into words and sentences, determining lemmas and grammatical dependencies, comparing similar contexts, and filtering candidate units for classification. In the final semantic interpretation, considering the literary context and the author's individual style remains essential. In the proposed approach, linguistic features define the classification criteria, while computational techniques facilitate the consistent application of these criteria across large volumes of text. The classification of a metaphor’s semantic type must be justified by its contextual usage, with ambiguous cases reserved for expert review. Being methodological in nature, this work does not report a finalized software application or novel experimental accuracy benchmarks. Determining the effectiveness of the proposed approach will require developing an annotated corpus of literary texts and comparing model outputs against expert human annotations in future research.