Abstract
A favorite question in cognitive psychology about the computation underlying object recognition is whether we identify an object as a whole or by its parts. This deceptively simple question has been surprisingly hard to answer. Here we provide a quick diagnostic test, applicable to any object identification task that can be done in the periphery at various sizes. The trick is to define a “part” (in cognitive psychology) as a “conjunction” of features (in perception). That may not seem like much, since it's hard to imagine what a part could be other than a combination of features, but it's enough. Historically, “crowding” describes the fact that a letter in the periphery is much harder to identify in the presence of neighboring letters. Pelli, Palomares, and Majaj (J. of Vis., accepted) showed that this is because whenever the observer judges a property that requires more than one feature (i.e. detects a conjunction) that judgment incorporates features over a conjunction region whose radius is roughly half the viewing eccentricity (distance from fixation). If we place the whole object inside the conjunction region then any attempt by the observer to extract a part will necessarily include features from the whole conjunction region, which embraces the whole object and thus will include features not just from the desired part, but also from the other parts as well. Thus, if there are multiple parts, the computation underlying the extraction of each part will fail. However, if there is only one part then the single conjunction over the whole object will succeed, unscathed by crowding. To recap: if the object is normally identified by parts, then the parts will crowd each other and identification will fail, but if the object is identified as a whole (having only one part) then identification will succeed. Early results show that identification of an isolated word or face is by parts, but that, in context, a word is read as a whole.
NEI grant R01-EY004432 to Denis Pelli