Making Sense of Korean Sentences: A Comprehensive Evaluation of LLMs through KoSEnd Dataset

TL;DR

This paper evaluates large language models' understanding of Korean sentence endings using the new KoSEnd dataset, revealing how explicit linguistic cues can enhance model performance on complex agglutinative language features.

Contribution

Introduces the KoSEnd dataset for evaluating LLMs on Korean sentence endings and analyzes the impact of explicit linguistic information on model performance.

Findings

Models perform better when informed about missing sentence endings.

Parameter count correlates with prediction consistency.

Explicit linguistic cues improve LLM understanding of Korean syntax.

Abstract

Although LLMs have made significant progress in various languages, there are still concerns about their effectiveness with low-resource agglutinative languages compared to languages such as English. In this study, we focused on Korean, a language known for its complex sentence endings, and evaluated LLMs on this challenging aspect. We introduce the Korean Sentence Endings (KoSEnd) dataset, which includes 3,000 sentences, each annotated for the naturalness of 15 sentence ending forms. These were collected from diverse sources to cover a range of contexts. We evaluated 11 LLMs to assess their understanding of Korean sentence endings, analyzing them based on parameter count and prediction consistency. Notably, we found that informing models about the possibility of missing sentence endings improved performance, highlighting the impact of explicitly considering certain linguistic features.