Contribution of material properties to object identification

Now let us consider the second possibility raised above, namely, that even if material is not sufficient by itself to identify an object, it can still help to compensate for weak shape cues by contributing to object identification. The study of Klatzky et al. (1993), in which people identified real objects from their contours, is relevant to this issue. As was described previously, people could identify objects with about 75% accuracy while wearing a glove that eliminated most cues to material (although coefficient of friction, for example, might still be available), and while being restricted to feeling the objects with an outstretched finger. In the same study, we compared performance when people wore the heavy glove to performance when the fingertip of the glove was cut off, permitting full access to material information. There was a significant improvement in performance when the material information was added. This was mainly for the single finger condition, which enforced sequential exploration and processing. Thus it appears that material can augment coarse shape information from touch, improving the speed and accuracy with which an object can be recognized.

In another study (Klatzky & Lederman 1995), we pushed to its limits the use of material for identifying objects. Subjects in this experiment were allowed only 200 msec of passive fingertip contact with an object in order to identify it. We were interested not only in the level of identification that could be achieved with such limited exposure, but with whether performance would depend on the attributes of the objects that were to be identified. We looked at two attributes: the size of the object, and the property that had been identified (by Lederman & Klatzky 1990) as most important in identifying it by touch alone, called the most diagnostic property. Accordingly, objects in this study were selected so that they represented two levels of size and two types of most diagnostic property. With respect to size, objects were either small - having at least one axis that could be spanned by the middle fingers - or large - with both the principal axes extending beyond the middle fingers. With respect to diagnostic property, objects were used for which texture was more important to identification than shape, or for which shape was more important than texture. Our prediction was that with such a short duration of exposure, and such a small area of the object being explored (that which touched the fingertips), subjects would be best at identifying objects for which texture was most diagnostic, or objects for which shape was diagnostic but that were small enough to be entirely explored by exposure to the fingertips.

In order to constrain subjects to approximately 200 ms of exploration, we had them guide their hand, with fingers extended, down to the object by following a vertical rod. The object was placed under the descending hand so that if it had a particularly informative region (for example, the pouring lip on a pitcher), that region would be contacted. A force-sensitive board placed under the object sensed initial contact, and a tone was emitted, instructing the subject to lift the arm away, after 100 msec. The subject then attempted to generate the name of the object that had been contacted.

We found that even under such limited exposure, objects could be named at levels above chance. Overall, 15% of objects were named correctly, and as expected, the level of performance depended on the object's size and diagnostic property. People performed worse, identifying only 5% of the objects, when they were large and shape was the most important property for identification. This is predictable, because with a large object, the relevant shape cues would extend beyond the region contacted by the hand. Performance was best (25%) when the object was large and texture was diagnostic. In this case, the large size maximizes the textured region that is so important for identifying the object. It is somewhat surprising that people could name 25% of objects for which texture is a critical cue to identity, with so short an exposure time. This supports the idea that material properties can contribute substantially to the process of object recognition by touch.

Another interesting outcome of this experiment was an analysis of confusion errors, in which we determined the similarity between the object that was presented and the object that people named, when they made an error. An analysis of nearly 600 errors of this type revealed that most confusion errors (75%) matched the original object with respect to material. Again, this indicates the high level of availability of material information to the sense of touch.

0 0

Post a comment