QUDsim: Quantifying Discourse Similarities in LLM-Generated Text
This addresses the issue of repetitive and structurally uniform text generation in LLMs, which is a concern for users relying on them for creative or diverse writing tasks, and is incremental as it builds on existing linguistic theories to quantify structural similarities.
The authors tackled the problem of LLM-generated text being repetitive and lacking creativity by developing QUDsim, a similarity metric based on discourse structures, which revealed that LLMs reuse discourse structures more than humans and diverge from human structural patterns.
As large language models become increasingly capable at various writing tasks, their weakness at generating unique and creative content becomes a major liability. Although LLMs have the ability to generate text covering diverse topics, there is an overall sense of repetitiveness across texts that we aim to formalize and quantify via a similarity metric. The familiarity between documents arises from the persistence of underlying discourse structures. However, existing similarity metrics dependent on lexical overlap and syntactic patterns largely capture $\textit{content}$ overlap, thus making them unsuitable for detecting $\textit{structural}$ similarities. We introduce an abstraction based on linguistic theories in Questions Under Discussion (QUD) and question semantics to help quantify differences in discourse progression. We then use this framework to build $\textbf{QUDsim}$, a similarity metric that can detect discursive parallels between documents. Using QUDsim, we find that LLMs often reuse discourse structures (more so than humans) across samples, even when content differs. Furthermore, LLMs are not only repetitive and structurally uniform, but are also divergent from human authors in the types of structures they use.