[not on handout, see intranet]
[pages 291-3 of Walker, 1987]

Letter Recognition

1) Morgan, M.J., Fitch, M.D., Holman, J.G. and Lea, S.E.G. (1976) Pigeons learn the concept of an ‘A’. Perception, 5, 57-66.

Abstract. Pigeons learned to discriminate between ‘A’s and ‘2’s in eighteen different typefaces. They subsequently showed excellent transfer to twenty-two typefaces that they had not previously seen; one pigeon was tested with handwritten letters and responded correctly to them also. Pigeon’s responses to ‘A’s and ‘2’s with parts removed suggested that their performance was controlled by several features, none of which could be considered necessary and sufficient. A test in which birds were shown other letters of the alphabet supported this conclusion. It appears that the original discrimination was learned as what Ryle calls a ‘polymorphous concept’.


  • The experiment was conducted on wild birds: the apparatus usually contained in a Skinner box was inserted in a fourth floor window, facing outwards, with a platform for birds to stand on.

  • Out of a large number of birds who visited the apparatus from time to time, 5 regular attenders, identified by rings on their legs were selected for detailed study.

  • Letter stimuli were back-projected onto the response key with a Carousel slide projector. For training, there were an equal number of ‘A’s and ‘2’s, in random order.

  • Each slide was presented for a varying amount of time, averaging 30-sec.

  • The ‘A’s were positive, which meant that the first key peck after appropiate interval led to the presentation of food (and the end of the trial).

  • When the ‘2’s were presented, pecking the key had no effect and the trial ended immediately at the end of the variable interval. This is a standard successive or “go/no go” discrimination.


  • 3 birds were originally trained with 18 different typefaces, and

  • then there was a transfer test with 22 new typefaces, each presented only once, so there was no chance of separate learning. There is data from the initial training and first transfer test is given here. The same 3 birds were then tested on partial and rotated ‘A’s and ‘2’s, all in Helvetica medum type. (data).

    Two further birds learned the original discrimination, and were given a transfer test using the remaining 25 letters of the alphabet in Helvetica medium. One of them was also tested with handwritten figures.

Conclusion. As stated in the abstract, the data suggest that the birds learned about several different features of ‘A’s and ‘2’s in parallel. A related theory is that of “recognition by components” (Biederman, 1987), supported by Van Hamme et al (1992) and Kirkpatrick-Steger et al (1998).


2) Blough, D.S. (1985) Discrimination of letters and random dot paterns by pigeons and humans. Journal of Experimental Psychology: Animal Behaviour Processes, 11, 261-80.

Blough trained pigeons to discriminate between pairs of letter stimuli presented as computed- generated fonts on a video monitor. Each letter was paired with all other letters in turn with 4 days of trainingg on each pair. The data is present as a hierarchical tree based on errors, showing which letters were most likely to be confused with other letters. Not surprisingly letters which have features in common, such as M, N and W, or D, O and Q, are more likely to be confused, both by humans and by pigeons.