UC San Diego

This dissertation examines the relationship between grammar and discourse in Korean morphology, focusing on the polyfunctionality of the -lul and -nun markers. The -lul marker primarily serves grammatical object functions while also indicating discourse focus or contrastive focus based on context. In contrast, the -nun marker primarily indicates discourse topic or contrastive topic status and can denote grammatical subject or object roles in place of dedicated grammatical case markers distinguishing subjects and objects. The interplay between the grammatical and discourse functions of each marker leads to complex usage patterns particularly associated with the notion of exhaustive interpretation.

In Chapter 1, I present the empirical and descriptive grammatical and discourse notions relevant for the analyses in the subsequent four chapters. In Chapters 2-4, I present psycholinguistic experiments that investigate usage patterns linked to the -lul and -nun markers’ exhaustive interpretation. Chapter 2 examines how these markers interact with differences in shared knowledge between interlocutors. The -lul marker tends to convey exhaustivity even without identified alternatives in shared discourse, whereas the -nun marker’s exhaustivity relies more heavily on the shared discourse. Chapter 3 explores the cancelability of exhaustivity, finding the -lul marker’s exhaustivity to be cancelable, while the -nun marker’s exhaustivity is non-cancelable. Chapter 4 confirms the optionality of the -lul marker and the exhaustivity it conveys, comparing it to null markings.

Chapter 5 evaluates large language models in terms of psycholinguistic experiments from Chapters 2-4. The outputs of large language models do not fully capture the detailed usage patterns of the -lul and -nun markers’ exhaustivity, highlighting the gap between human speakers and the models. In conclusion, this dissertation demonstrates the complex interaction between the grammatical and discourse functions of morphology, and suggests that systems lacking targeted training in discourse pragmatics do not reflect this complexity in natural language.