|
Do we visually recognize things -- words or faces -- by wholes or by parts? Denis Pelli of New York University and Bart Farell of Syracuse University have answered that question in their forthcoming Letter to Nature. Their article, "The Remarkable Inefficiency of Word Recognition," is accompanied by a "News and Views" piece discussing their work.
Using the example of letters and words, Pelli and Farell prove that we read by detecting simple features. This makes word recognition very inefficient. Even for the five most common three letter words -- the, and, was, for, him -- people cannot read the word unless the features of each letter are identifiable. The features in question are simple, much smaller than a letter.
In hundreds of thousands of trials, the researchers tested readers' ability to recognize letters and words displayed at various contrasts. Comparing human performance with that of the mathematically defined ideal observer, they found that our visual system does not directly recognize complex familiar objects -- such as words -- but in fact relies on the detection of smaller elements -- features -- and only then recognizes the object of which they are the parts.
Just as modern radios suppress static, our eye suppresses the 'static' of countless weak features that would otherwise besiege us. Pelli and Farell show that this hush comes at the cost of reduced efficiency in seeing complex objects like words.
"One of the interesting aspects of these findings," says Prof. Pelli, "is their counterintuitive character. Readers feel that they are reading whole words, but our research shows that vision has a bottleneck, and must independently detect simple features in order to see anything. Everything we see is a pattern of features. Even after we read millions of three-letter words, the's, and's, and but's are still just patterns of features. We never learn to recognize them as single features."
Journal
Nature