"My name is... my name is...": A Linguistic Map for Voice Agents — Midam Kim, ServiceNow

Midam Kim spells her first name for a voice agent, letter by letter. The bot reads it back with an N on the end. She corrects it. The bot replies thank you for your correction, then says her name wrong anyway, because its pronunciation rules are English and her name is not. Moments later it asks for an account number she has to go looking for, cuts her off while she reads the unfamiliar string aloud, announces it cannot find her record, and asks her to repeat the whole thing without trying anything different. She asks for a human. Kim is an ML engineer at ServiceNow and a longtime researcher of speech in the wild, and she uses that call to argue that voice AI failures are not a scattered list of bugs but a structured one, and that linguistics already has the structure. Her map is a grid. Each party has a listening channel and a speaking channel, and each channel operates at four levels: sounds, words, interaction, and mental model. Recognition failures sit in one cell, pronunciation in another, turn taking in a third, intent tracking in a fourth. Every familiar problem lands somewhere, which is the point. The cells are interdependent rather than separable, so fixing sounds without words, or words without timing, does not hold. Her sharpest observation concerns what survives a call. In chat the history sits on screen as text. In speech the sounds vanish as they are spoken, and the only thing that accumulates is the user's mental model, which is what you are really designing for. Speaker info: - https://www.linkedin.com/in/midamkim/ Timestamps: 0:00 - A call that goes wrong, step by step 3:39 - Conversation as a joint activity 5:25 - Replaying the failure through that lens 7:14 - Eight cells: two channels, four levels 8:56 - Why the layers cannot be fixed separately 9:49 - What vanishes, and what accumulates 10:42 - What to actually work on 12:28 - Diagnosing your own agent 14:13 - Speakers adapt, and languages change