VISIBLE PATTERNS OF SOUND—POTTER 203 
reading, has been added to the class. His learning rate for visible 
speech has so far compared favorably with that for lip reading. 
During his training he has been tested regularly on his ability to read 
word patterns in the visible speech form and by reading lips. 
Although adept at lip reading, and the lip-reading tests were carried on 
under exceptionally favorable conditions, his score on reading visible 
speech has stayed well above that for reading lips. Incidentally, his is 
probably the first case in which a person with substantially no hearing 
has been enabled to talk over an ordinary telephone circuit without 
the aid of a human “interpreter.” 
Thus far the results of the experimental training encourage con- 
fidence that these speech patterns may provide a practical form of 
directly translated visible speech, but much more training experience 
is needed for complete confirmation. 
A second way to determine the legibility of word patterns, by means 
of so-called visual discrimination tests, was initially devised to get 
around a difficulty. It had been assumed at the start of the visual 
hearing project that the trainees would, after a period of training, be 
able to say whether one pattern presentation was better than another, 
and how the many variables that appear in the translator design should 
be treated. But this assumption proved to be mistaken. After train- 
ing with one form of pattern there was a tendency to dislike others. 
It soon became obvious that an unbiased evaluation method was neces- 
sary as a guide to development. 
The visual discrimination tests produced to meet this situation de- 
pend upon an assumption that any language, aural or visible, is made 
up of many patterns; and that the relative merit of different languages 
or of different representations of a single language depends upon the 
ease with which patterns that make up equivalent vocabularies may 
be identified. In the visible-speech case, one test method is to select 
words that in certain respect look alike, and arrange them in what are 
called “similarity series.” Examples are, “man, ran, van, tan,” etc. 
The words are spoken in groups of three, such as “van-tan-tan” or 
“van-van-tan.” An observer of the patterns produced by these words 
is simply asked to check whether the middle pattern is more nearly 
like the first or last in a group. It is not required that the observer 
have any knowledge of the meaning of the patterns. Ratings are in 
terms of the percentage of correct pattern identifications, taking into 
account the fact that 50 percent accuracy represents pure guesswork. 
The figure so derived is called the “discrimination index.” Poor 
patterns result in a low DI, or if the translator fails to show certain 
sounds clearly, the DI for word groups containing these sounds will be 
relatively low. Figures of this kind permit a quantitative appraisal 
of the performance of different translators. 
725362—47——15 
