音声学における数学的一方法:特に日本語の母音に関連して

概要

論文の詳細を見る
On the basis of the invariant-theoretical principle that "every phonemically significant factor extracted from the physical attri-butes of a vowel sound should be a quantity invariant under any transformation which leaves the phonemical value of the vowel invariant", under the assumption that (i) the phonemic features of a vowel are well defined by (F1, F2, F3; I1, I2, I3)(Fi: frequency of the i-th formant, Ii: intensity of the i-th formant), (ii) the vowel phonemes are invariant under two transformations A and B: A. Filtering through filters with uneven frequency characteristics, B. Raising frequencies uniformly (FicFi), it is concluded that "the vowel phonemes are characterized by the set of parameters (a, b): a=10 log(F2/3√F1F2F3), b=10 log(F2/3√F1F2F3)" and that "all other phonemically significant quantities must be functions of a and b." It is also pointed out that the vowel diagram (Fig.1), obtained by plotting the five vowels of Japanese pronounced by eight Japanese subjects as the points on the two-dimensional plane with a's and b's as the abscissae and the ordinates, is identical, in its average character, with the vowel chart (Fig.3) of Prof. S. Hattori and others, and that it is essentially the only vowel chart admissible under the above assumption.
日本言語学会の論文