READIXG RATES 



503 



o 

 z 

 o 

 o 



LU 

 10 



a. 



a. 



in 

 a 

 n. 

 o 

 5 



4.5 

 4.0 

 3.5 



3.0 



2.5 

 3.5 



3.0 



2.5 



2.0 



2 4. 6 8 10 12 14 16 18 20 22 24 26 28 



SUCCESSIVE READINGS OF NONSENSE WORD LISTS 



Fig. 4 — Confirmaton- demonstration of the effect of familiarity upon reading 

 rate. 



with word order have been removed and the bits/word depend only on 

 the frequency of occurrence of words in prose, which is known. Thus, 

 Shannon^ gives a figure of 11.82 bits/ word which apphes to scrambled 

 prose, provided the prose has the same word frecjuencies as that from 

 which the statistics were derived. The information rates for words from 

 a 5,000-word dictionary (Experiment 1) for the preferred lists, and for 

 scrambled prose are given in Table II. 



The information rate for scrambled prose is less reliable than the 

 others, because we are not sure that the word frequencies used by 

 Shannon apply to the prose used by us, but we used the tjT^e of material 

 cited by the reference he quotes. It is clear that the information rate 

 for scrambled prose is high as compared with most other lists. 



Table II shows the gain which may be made h\ fitting the task to the 

 human being — in this case, by choosing a suitable word list. We may 

 note that the gain appears greater in the case of reader A than in the 

 case of reader B. This need not be experimental error. One would sup- 

 pose that there are optimal lists for individuals. Indeed, if we compare 

 Figs. 3(a) and 3(b) we see that for reader A the word rate for mono- 

 syllables drops by a factor 0.72 in going from the first thousand to the 

 tenth thousand, while for reader B the drop is onlj^ a factor 0.88. This 



