Now let’s apply some of what we have learned about the visual stimulus and apply to how we perceive letters and words. An informative experimental design is to present letters in various forms of incompletion or distortion and determine how these influence perception. The results are then used to determine the best set of characteristics or features that are responsible for letter recognition. The next set shows a set of letters highly confused with one another and a set of letters seldom confused with one another. In a classic paradigm, letters are presented for a very short duration and subjects have to state which letter was presented. The results are analyzed in a confusion matrix indicating what letters were presented and what responses were given. The slide shows a subset of data from an experiment with uppercase letters. The proportions along the following diagonal correspond to correct judgement. One quickly sees that some letters are more difficult to recognize, as was the letter D. Also informative are errors given by the non-diagonal cells which indicate which letters are confused with others. For example, the result show that the upper case letter D is often recognize as an upper case B. We know that information from some spatial frequencies are more important than others as illustrated in our celebrity demonstration. We can ask a similar question about letters. The next graph shows an illustration of the stimulus generation process of an experiment we will now discuss. Each original letter in the upper left was first decomposed into five spatial frequency bandwidths of one octave each. Each octave bandwidth was then independently sampled and randomly positioned in Gaussian windows so that sparse information was revealed. The information samples were summed across the five spatial frequency scales to produce an experimental stimulus shown in the bottom row. The top row of numbers gives the spatial frequencies contained in each bandwidth changing high to low frequency. These stimulus properties were somewhat complex but we can learn from the results without having the detail of this intricacies. The results of a letter recognition experiment with these stimuli are shown in the next slide. The original recognition answers from the subjects were analyzed to determine how much each spatial-frequency band contributed to perception. The dependent measure is a measure of the contribution of each band. As can be seen in the results, the intermediate spatial frequencies between two and eight Hertz, or cycles per second, made the largest contribution to recognition. Now the investigators perform a feature analysis, asking what property felt the letters are responsible for their recognition. They defined various features, such as intersections of two lines, slanted lines, horizontal lines, and vertical lines, and terminations where one line stops when it intersects another. The next slide reveals how the letter A was defined in terms of these features. From the top and left to right in the illustration, the features of the letter A are intersection, intersection. Slant line tilted right, horizontal line, slant line tilted left, intersection, and two termination features. Other letters were defined similarly. In the next slide showing the results, the importance each of the features is given. We see that terminations have the most influence on the recognition judgements. Horizontals also have a strong influence and it is interesting that horizontals have about four times the amount of influence as verticals. We have learned about the visual stimulus and its perception to set the stage for how visual perception in infants differs from adults. To inform how early reading could be expected of children, it is important to understand how children's' ability to deal with the visual world varies with development. The next slide shows this ability in terms of how they would perform on the Snellen chart, if in fact they could read. Perhaps in some future society, Snellen charts will be used with preschool children because they will know how to read. Of course, we are interested in the children's ability to discern spatial-frequency. As can be seen in this slide, infants in the first few months of life are well below that of adults. An important difference is that infants are maximally sensitive to low spatial frequencies. This means that the letters they see should be in very large type fonts necessarily made up of lower spatial frequencies. The poorer acuity shown by new and recently born infants should not be surprising because their neural connections are only developing as we can see in the representation of their brain neurons and the connections between them. In fact, it was believed that babies were essentially blind at birth until they were prodded into imitating an experimenter or their caregivers. Here is a video of a newborn imitating a tongue protrusion. In the last four or five decades, a cottage enterprise has evolved to demonstrate that infants are capable of most every perceptual and cognitive behavior that adults can conceptualize. To read print, infants have to be capable of controlling their looking behavior and seeing written language. Infants have very good control of their focus at two months of age and can coordinate their eyes and track objects at three months. They can perceive depth at four months and can recognize objects and faces at five months. Most importantly, infants' visual acuity also improves rapidly from birth onward reaching close to adult acuity by eight months of age. We can see what the infant is able to see at 3, 6, and 9 months of age. The 9 month old has a pretty good look at the teddy bears relative to an adult. The wealth of well-documented research and measurement of vision development shows that infants approaching their first birthday have the capacity to resolve critical characteristic of written input in order to learn to read appropriately structured text. This conclusion is substantiated by the many YouTube videos of proud parents showing off their child's reading skill. [MUSIC]
