The truth is no diaper: Human and AI-generated associations to emotional words
Špela Vintar, Jan Jona Javoršek
TL;DR
This paper investigates how human word associations to emotionally loaded Slovenian cues compare with associations generated by three contemporary LLMs. Using the SWOW-SL human norms and the SloEmoLex emotion lexicon, the authors generate and compare associations for 40 cues with zero-shot prompts across Llama-3.3, GaMS-9B, and Claude 3.7 Sonnet, analyzing creativity, overlap, and sentiment. They find a moderate overlap between human and LLM responses, with Claude showing the strongest alignment to human data, but LLMs tend to amplify emotional load and exhibit less creativity than humans. The study highlights the limitations of current LLMs in mirroring human associative processes, suggesting the need for larger cue sets and finer-grained analyses to better understand machine-like versus human-like creativity in language.
Abstract
Human word associations are a well-known method of gaining insight into the internal mental lexicon, but the responses spontaneously offered by human participants to word cues are not always predictable as they may be influenced by personal experience, emotions or individual cognitive styles. The ability to form associative links between seemingly unrelated concepts can be the driving mechanisms of creativity. We perform a comparison of the associative behaviour of humans compared to large language models. More specifically, we explore associations to emotionally loaded words and try to determine whether large language models generate associations in a similar way to humans. We find that the overlap between humans and LLMs is moderate, but also that the associations of LLMs tend to amplify the underlying emotional load of the stimulus, and that they tend to be more predictable and less creative than human ones.