非表示:
キーワード:
-
要旨:
Objects in the real world do not occur in a random manner. For example tea spoons
tend to be near tea cups and there is usually a remote control nearby a TV. A cognitive
system, which is adapted to the environment, would take such co-occurrences
into account and use top-down driven expectancies for faster recognition. Although
many object recognition theories assume a serial bottom-up processing and disregard
top-down feedback from later stages to earlier ones, neuropsychological and imaging
studies suggest recurrent feedback during object identication (e.g., Humphreys, Riddoch
and Price, 1997, Philosophical Transactions of the Royal Society of London B,
1275-1282). We present further evidence for top-down in
uences in object recognition
using a priming approach. In Experiment 1 participants named objects (e.g.,
TV) in two dierent views (canonical vs. non-canonical), which were preceded by a
contextually consistent (e.g., remote control) or inconsistent (e.g., hammer) priming
stimulus. We found clear eects of prime consistency and target viewpoint as well as
a signicant interaction. Presenting a consistent prime prior to a target reduced the
viewpoint dependency signicantly. These results were replicated in Experiment 2 using
a contextual association task. Participants had to judge whether the second of two
sequentially presented objects tends to co-occur with the rst one. Consistent priming
stimuli reduced the viewpoint-dependency of the reaction times to the target objects.
Both experiments provided converging evidence for the view, that the human cognitive
system uses knowledge of co-occurrence information for faster recognition.
