Abstract
Object and word recognition are both processes that transform visual input into meaning. When we read words, the frequency of their occurrence strongly modulates recognition performance. With more object-labels available in large real-world image datasets, one can now estimate the frequency of occurrence of objects in scene image databases ("object frequency"). Here, we apply this new "object frequency" measure to investigate frequency effects in word and object recognition. Besides, we compare these measures with established word frequency measures for predicting behavior in a natural vs. man-made categorization task (Experiment 1) and a same-different priming task (Experiment 2) involving words and images of object concepts. In Experiment 1, only a word-based frequency measure (SUBTLEX word frequency based on movie subtitles) resulted in a facilitatory effect (i.e., faster responses to frequent stimuli) on both words and object recognitions. In Experiment 2, we replicated the facilitatory SUBTLEX-frequency effect, for both stimulus types, when the task included cross-modal priming (i.e., word-to-object or object-to-words) but notduring repetition priming. Likewise, only in the cross-modal priming condition we found an object frequency effect (ADE20K object frequency computed from image datasets). Interestingly, the effect showed faster responses for both stimulus types when objects occur less frequently in image databases, which is opposite to the SUBTLEX-frequency effect. This latter finding might imply that recognition behavior can be more effective when objects are more distinct or diagnostic of a scene. In sum, both object and word recognition are faster when their concepts are often used in our language, but also, the distinctiveness of an object's appearance in the real world is essential to efficient access of semantic representations.