Reading Rates and the Information Rate of a Human Channel

497 498 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1957 * Here tracking means successively pointing to a series of marks. t In preliminary experiments the reader's voice was recorded, and it was found that errors in reading aloud occur very seldom if ever. information, as contrasted with, for example, typing, playing the piano, or tracking.* The work presented here, while undertaken independently, is in general similar to and in agreement with that reported for reading rate experiments by Licklider, Stevens and Hayes, and by Quastler and Wulff. However, we have considered some factors in more detail than these workers, and also, contrary to the former group, we find that, under optimal conditions, reading with tracking has a lower information rate than reading alone. The chief problem investigated was: (1) Taking people as they are, with no additional training, how fast, in bits per second, can they transmit information by reading? (2) What principal factors control this limiting rate? The experimental procedure consisted simply of people reading aloud as rapidly as they could typed lists of words. Each list was composed of a single vertical row of 12 groups of 5 words, giving a total of 60 words per page. In each instance, the words were chosen at random from a given vocabulary of words. If n is the number of words in the vocabulary and if the words are chosen with equal probabilities, and if all words are read correctly, f the amount of information which is conveyed or transmitted through the human being measured in bits is log n bits/word When the vocabulary for a particular experiment has much fewer than 60 words, certain words must necessarily be repeated several times within a list.