“Further Results on the n–tuple Pattern Recognition Method”, 1961-03-01 ():
[Rebuttal to 1960; Uhr commentary; Highleyman’s response]
Evidently, Highleyman & Kamentsky did not understand that the parameter n (in the n-tuple method) should be chosen to best suit the particular data being read, because they used only n = 2 in their computations. Some studies at Sandia Corporation in February, 1960, on these same data (provided by Highleyman) with n = 6, 8, and 12 yielded results considerably different from those given…
The variability of the hand written characters used in these studies is high and not well represented by the 50 alphabets used. For example, the last 10 alphabets are considerably different than the first 40. For this reason, it will be necessary to have a much larger sample (perhaps 1,000 alphabets) before one can decide with any certainty how successfully the n-tuple method will read characters with this much variability.