292 BELL SYSTEM TECHNICAL JOURNAL 



and profanity, 40 per cent; letters and numbers, 25 per cent; and the 

 sound "er," 15 per cent. 



The words which were obtained by the process of sampling con- 

 versations for specific parts of speech are not, of course, identical with 

 those which would have been obtained had the entire conversation 

 been recorded. The representativeness of the most frequent words, 

 which largely determine the relative frequency of the speech sounds, 

 was investigated by a later test in which a different observer recorded 

 the verbs from 250 conversations. These results will be discussed 

 later, but it may be pointed out here that the word list obtained by the 

 two observers corresponded so closely that it is felt that the samples of 

 parts of speech were recorded with sufficient accuracy and were 

 sufficiently large to justify taking the words obtained as a good repre- 

 sentation of the main body of telephone conversation. 



The kinds of conversations encountered are shown in Table I. The 

 great preponderance of business calls is reflected, as will be shown later, 



TABLE I 

 Types of Telephone Calls on which Observations Were Made 



a. Material 



Business Calls 89.0 per cent 



All other Calls 11.0 per cent 



b. Speakers 



Two Men 86.5 per cent 



Two Women 10.4 per cent 



Man and Woman 3.1 per cent 



in the vocabulary. If a smaller percentage of the calls had been busi- 

 ness in nature and if a larger percentage had been between women the 

 vocabulary would probably have been different. Whether any marked 

 change would have been found is open to some doubt when it is re- 

 called that business may cover a wide range of topics and that in the 

 1,900 conversations from which samples were taken there may have 

 been as many as 3,800 different speakers. Evidence will be given, 

 however, which indicates that the relative frequency of the speech 

 sounds would have been changed very little. 



Words 



The number of conversations on which observations were made was 

 regulated to some extent by the ratio of the number of total words to 

 the number of different words recorded in each class. In the early 

 stages of observing many of the total words recorded were different, 

 making this ratio low, but as the observations continued fewer and 

 fewer new words were encountered. In Figure 1 curves are given 



