NATURE OF LANGUAGE JONES 501 



the number and kind of errors made. The per cent of the total 

 sounds spoken which are correctly received is called the articulation 

 of the system. For these tests simple syllables are used constructed 

 in a systematic manner from the 36 fundamental speech sounds and 

 arranged in lists of 50 syllables each. A carefully worked out 

 technique is observed in the testing. 



Articulation tests have been made upon the high quality experi- 

 mental system without distortion, but with controls adjusted to 

 give various intensities of output from the threshold of audibility 

 to values considerably above the level of ordinary conversation. 

 The results obtained are shown in Figure 6. The abscissas of the 

 curve represent loudness and are expressed as the ratios by which 

 the speech energy has been decreased from the initial intensity at 

 exit from the mouth. When the volume is reduced to about one 

 ten-billionth of the initial speech intensity, the articulation becomes 

 zero. This point corresponds to the value at which speech becomes 

 inaudible. At about one-thousandth of the initial speech intensity, 

 the articulation becomes a maximum. With louder speech than this 

 perception is less accurate, j^robably due to overloading of the ear 

 mechanism and subjective distortion. These results were obtained in 

 a perfectly quiet room. When the observer is submerged in an 

 atmosphere of noise the speech must be louder in order to get the 

 best hearing conditions. 



The articulation data have been further analyzed in such a way 

 as to show the errors made at various intensities for each of the 

 fundamental sounds. The results for some typical sounds are 

 shown in Figure 7. 



It is observed that in general diphthongs and vowels are morfe 

 easily heard than consonants, and that of the latter the stop 

 consonants are heard with fewer mistakes than are the fricative 

 ones. If all the sounds are listed in order of average articulation 

 the top quarter will contain no consonants and the lower half 

 no vowels. When speech becomes weak, the errors of the con- 

 sonants increase greatly, their articulation values falling off at 

 higher intensities than is the case with the vowels. 



There are some exceptions to these general statements. At 

 moderate volume the short vowel "e" is near the bottom of the 

 list, but at very weak volume 22 sounds are harder to perceive. 

 " L," " r," and " ng " are all more readily heard than " e " at moderate 

 volume, but when very weak they fall below it. "L," which ranks 

 with the diphthong "i," as one of the easiest sounds at moderate 

 volume, is mistaken about two times out of three when very weak. 

 The diphthongs " I," " ou," and the long vowels " 6," "6," " a," 

 all 'have average articulations better than 95 and even when very 

 20397—25 33 



