A new paper argues that despite decades of research, nobody has actually mapped how misunderstandings form, spread, and slip past unnoticed, and that gap matters more now that conversations increasingly run through AI.
The researchers pulled together findings from nine fields that don't normally cite each other, including linguistics and communication theory, and distilled them into eleven distinct failure modes. Each mode is tied to a specific point in the communication process, not scattered loosely across it, and those points form eight analytical layers. The authors formalize those layers by extending information and communication theory, originally built for transmitting signals, to cover how listeners reconstruct meaning. They back the taxonomy with a source-by-source evidence matrix, a coding manual, and nine analyzed dialogue cases so other researchers can check the work.
Real-time, in-person conversation gives people constant small cues, like a raised eyebrow or a quick request to repeat something, that flag confusion early. AI-mediated channels strip most of that out, and the paper's authors say detection tools aren't being built fast enough to fill the gap. A taxonomy that pins down exactly where and how understanding breaks down gives anyone building conversational AI agents a more precise diagnostic than simply logging that a bot misunderstood a user.
It is, admittedly, a research paper about categorizing problems rather than solving them, so builders should not expect a plug-in fix anytime soon.