288 



BELL SYSTEM TECHNICAL JOURNAL 



oscillograms. In all cases recording began at the start of the call, but 

 in some instances recording was stopped before the termination of the 

 call due to lack of recording paper in the oscillograph. The oscillo- 

 grams, ranging in length from 29.6 to 660.8 seconds, represented ob- 

 servations on calls whose mean duration was 430.5 seconds. The 

 speed of recording was such that the time intervals under observation 

 could readily be measured with a precision of ± 0.005 second. The 

 conversational elements were measured with this precision and listed 

 in their order of occurrence for each call. The records for all calls were 





,3 5 



ojf: 



OQ 



OL-t 



0.1 0.2 0.5 I 2 



TIME IN SECONDS 



Fig. 3 — Lengths of talkspurts. 



60 p' 



5o 



UJ < 



2 



40 



20 ^ 



then consolidated and retabulated in terms of the number of instances 

 of each element whose duration could be included within each of a 

 regular progression of time increments. For all three items of data 

 time cells 0.10 second wide were chosen. The data, when thus 

 cellularized, provided the basis for the construction of histograms from 

 which the time-distribution curves were obtained. These distribution 

 curves and their respective summation curves are given in Figs. 3, 4, 

 and 5. Some of the statistically significant quantities ® are tabulated 

 on the opposite page. The values are time intervals in seconds. 



Since most telephonic speech syllables are shorter than 0.3 second 

 the modal value of 0.25 second for the length of talkspurts makes it 

 clear that monosyllabic replies are by far the most numerous. From 



^ The mode is the value which occurs most frequently, i.e., the peak of the dis- 

 tribution curve. 



The median is that value above and below which equal numbers of observations lie. 

 The mean is the arithmetic average of all the values observed. 



