Keynote

AI creativity where it's too easy to be creative: using LLMs to detect puns and allusions in heraldry

Richard Sproat

Sakana.ai, Japan

In European heraldry and Japanese family crests (kamon), it is common to find motifs that are allusive to, or puns (cants) on the family name. For example the arms of Cockburn of Clerkington in Sir David Lindsay's Armorial are blazoned argent, three cocks gules, i.e., a silver field with three red roosters. The literature on European and Japanese heraldry discusses the topic at length but rarely presents hard numbers: for a sizable corpus of arms or crests, what proportion involves puns or allusive relations? Analyzing a corpus of hundreds of arms, their verbal descriptions and associated family names is labor intensive, but perhaps AI can help. I report on a Large Language Model multi-agent debate-based approach to this problem. While the method is indeed useful, the LLMs sometimes become too creative, offering connections that few humans would make. This is particularly an issue with allusive arms, where it is all too easy for the systems to find some connection, but with cants they also often find phonetic relations that would seem strange to humans. In this domain, where even humans may be tempted to make fanciful connections, one has to be careful of the even greater creativity of LLMs.

Bio

Richard Sproat is a research scientist at Sakana.ai, Japan, working on artificial intelligence in language processing, agentic systems and image understanding. He received his PhD in Linguistics from MIT in 1985, worked as a researcher at AT&T Bell Laboratories, as a professor at the University of Illinois Urbana-Champaign and the Oregon Health & Science University, as a research scientist at Google New York, then Google Tokyo, before joining Sakana.ai. He has published in a wide variety of areas of linguistics and computational linguistics, including work on experimental phonetics, computational morphology, text-to-speech synthesis, text normalization, and finite-state methods in language processing. He has a strong interest in writing systems and symbol systems more generally, with two of his recent books being in this area: Symbols: An Evolutionary History from the Stone Age to the Future (2023), and Tools of the Scribe: How Writing Systems, Technology, and Human Factors Interact To Affect the Act of Writing (with Brian Roark and Suyoun Yoon, 2025), both published by Springer.