Can Korean Language Models Detect Social Registers in Utterances?

초록

This article investigates the capacity of the deep-learning algorithm to learn social register features from textual input, focusing on an interdisciplinary exploration at the intersection of sociolinguistics and computational linguistics. Languages encompass diverse aspects that are intricately linked to social contexts, implying that the ability to generate and comprehend linguistic expressions within specific social environments is an integral facet of communication competence. Such social factors influencing linguistic expressions, or social registers, encompass elements such as age, sex and gender, interlocutor relationship, and more. This study expands upon prior research on social registers by employing data-driven methods and deep learning skills to classify transcribed Korean language into predefined register classes based on features of sex, age, and kinship. Specifically, methodologies of transfer learning and in-context learning are utilized for classification tasks. However, both methods demonstrate poor performance in identifying the specified features, highlighting potential challenges in employing deep learning approaches for language use that is socially influenced.

키워드