Although frequently named objects are generally more fixated and more salient, the number of fixations on an object shows a larger variation for frequently named objects than for rarely named ones. In addition, if only one observer recalls a particular object, they have a slight tendency for a larger fraction of fixations on that object. Since we ask for “keywords,” we may have biased observers to name scene-diagnostic objects. It is therefore possible that rarely named objects could still be remembered, if they were specifically queried. In this view, less expected objects need more fixations (or more salience) to be named. This is in line with the idea that “surprising” (out of context/less expected) events draw attention, whether they deviate statistically (Itti & Baldi,
2008) or semantically (Friedman,
1979) from expectation. Indeed, “implausible” objects (i.e., objects that conflict with scene gist) tend to be recalled better (Pezdek, Whetstone, Reynolds, Askari, & Dougherty,
1989), although they are recognized worse (Davenport & Potter,
2004) and their effective field-of-view is smaller (DeGraef,
1998). Whether semantically implausible objects are fixated earlier or even “pop-out” (Loftus & Mackworth,
1978) has remained, however, controversial. Recent studies that use more complex scenes than Loftus and Mackworth (
1978) and control the saliency of the critical item typically do not find an early preference to fixate implausible objects. Instead, they find that implausible objects are fixated longer, more likely to be fixated again (Henderson, Weeks, & Hollingworth,
1999), and are fixated earlier than plausible objects only after prolonged viewing (Underwood, Templeman, Lamming, & Foulsham,
2008) or if they appear while saccadic suppression suppresses bottom–up attention capture (Brockmole & Henderson,
2008). Under some experimental conditions, the recall of an item is improved by increased numbers of fixations on the object (Hollingworth & Henderson,
2002), although this effect can be restricted to certain aspects of the item and depend on the methods of querying (Tatler, Gilchrist, et al.,
2005). The effect that different object properties (saliency, object properties, fixation frequency, naming frequency, etc.) have on the ability of an observer to recall an item when queried will be an interesting issue for further investigation. The diversity of findings stresses that the querying for keywords in the present study and the unknown motivation of LabelMe participants in Elazary and Itti (
2008) may yield substantially different results from other tasks, such as change detection or item recall.