Structured learning for context-aware spoken language understanding of robotic commands

Service robots are expected to operate inspecific environments, where the presenceof humans plays a key role. A major fea-ture of such robotics platforms is thus theability to react to spoken commands. Thisrequires the understanding of the user ut-terance with an accuracy able to trigger therobot reaction. Such correct interpretationof linguistic exchanges depends on physi-cal, cognitive and language-dependent as-pects related to the environment. In thiswork, we present the empirical evaluationof an adaptive Spoken Language Under-standing chain for robotic commands, thatexplicitly depends on the operational en-vironment during both the learning andrecognition stages. The effectiveness ofsuch a context-sensitive command inter-pretation is tested against an extension ofan already existing corpus of commands,that introduced explicit perceptual knowl-edge: this enabled deeper measures prov-ing that more accurate disambiguation ca-pabilities can be actually obtained.