PEC-Home: Interpretation of Progressively Elliptical Commands in Smart Homes
Researchers have introduced PEC-Home, a new simulated dataset designed to improve how large-language models (LLMs) interpret progressively elliptical commands in smart home environments. Current home assistants struggle with these shortened commands, which are common in human dialogue as shared context builds. Experiments with models like GPT-4o show that even with dialogue history tools, accuracy for elliptical commands remains lower than for complete ones, highlighting a significant challenge in practical smart home applications due to referential and intention ambiguity. AI
IMPACT This research could lead to more intuitive and efficient interactions with smart home devices by enabling AI assistants to better understand shortened commands.