TIME INTERVALS IN TELEPHONIC CONVERSATION 291 



Fig. 7, wherein talkspurts having a given number of resumptions are 

 sorted according to length, indicates that almost all talkspurts exceed- 

 ing 6 seconds in length contain resumptions. On the average, resump- 

 tions seem to occur about once every three and one third seconds in 

 the longer talkspurts. The aggregate of all the resumption pauses 

 within talkspurts amounts to about 17 per cent of the total talkspurt 

 time. 



It will be recognized that in obtaining the curves of resumption 

 times in Fig. 4 a certain amount of arbitrary judgment must be exer- 

 cised. As stated previously, the temporary absence of deflection of an 

 oscillographic trace showing speech energy was regarded as evidence 

 that the talker was pausing. Some assistance in determining whether 

 or not speech energy was present was given by the trace showing the 

 operation of the corresponding echo suppressor. The comparatively 

 slow change in amplitude at the beginning and ending of syllables 

 renders this determination more difficult as shorter and shorter pauses 

 are considered. However, those shorter than about 50 milliseconds 

 within connected speech must be observed with so much amplification 

 that they tend to be obscured by noise on even a very quiet line. From 

 a knowledge of the characteristics of the volume controls and echo 

 suppressors used it is estimated that Fig. 4 represents pauses greater 

 than 50 milliseconds during which the amplitude is lower than about 

 0.2 per cent of the maximum amplitude. 



Conclusion 



Telephonic conversation has been arbitrarily defined in terms of a 

 few elements w^hose specification serves to completely describe its 

 progress in time. These elements have been measured on a particular 

 telephone circuit and the measurements have been presented in the 

 form of distributions which approximate the probability of their 

 occurrence. 



A preliminary investigation under quite different conditions gave 

 results remarkably close to those found in these more extended tests, 

 suggesting that the conversational elements to be found throughout 

 the aggregate of telephone users may not be materially different from 

 those found in this investigation. 



