What's she doing in the kitchen? Context helps when actions are hard to recognize

Specific spatial environments are often indicative of where certain actions may take place: In kitchens we prepare food, and in bathrooms we engage in personal hygiene, but not vice versa. In action recognition, contextual cues may constrain an observer's expectations toward actions that are more strongly associated with a particular context than others. Such cues should become particularly helpful when the action itself is difficult to recognize. However, to date only easily identifiable actions were investigated, and the effects of context on recognition were rather interfering than facilitatory. To test whether context also facilitates action recognition, we measured recognition performance of hardly identifiable actions that took place in compatible, incompatible, and neutral contextual settings. Action information was degraded by pixelizing the area of the object manipulation while the room in which the action took place remained fully visible. We found significantly higher accuracy for actions that took place in compatible compared to incompatible and neutral settings, indicating facilitation. Additionally, action recognition was slower in incompatible settings than in compatible and neutral settings, indicating interference. Together, our findings demonstrate that contextual information is effectively exploited during action observation, in particular when visual information about the action itself is sparse. Differential effects on speed and accuracy suggest that contexts modulate action recognition at different levels of processing. Our findings emphasize the importance of contextual information in comprehensive, ecologically valid models of action recognition.