From Kikkawa Eishi 吉川英史's "Katarimono" ni tsuite (＜語りもの＞について, "About 'katarimono'"), in Nihon ongaku no nagare (日本音楽の流れ, "The flow of Japanese music"), ed. Yamagawa Naoharu 山川直治), translation mine:

When "talking" (kataru), performers communicate events and circumstances in a way that the listener can understand. Their words, therefore, must be understood clearly. As a result, prolonging or freely raising and lowering the pitch of the sounds in the words tends to be avoided. If a performer does want to prolong a sound, they will prolong one that does not carry much meaning. For example, the last sound in a phrase, or particles or verb endings. [...]

The aim of "singing" (utau), on the other hand, is to express emotions, not events and circumstances. There need not necessarily be a listener, and it does not matter of the lyrics are not understood. As a result, words can be freely prolonged, advanced vocal techniques used to raise and lower their pitch, and rhythm of the words can be refashioned into a musical rhythm. This is why even though mikagura, saibara, and folk oiwake are sung in Japanese, they are completely incomprehensible, and sound like long vowel practice: "Aaaa, iiii, eeee."