My ignorance of the speculations of the meta -language brigade is almost total, Max. But from a limited study of vocal tract mechanics and early language formation I would observe it starts as the simplest full-throat sound (similar to why a baby's early sounds will often include /a/ in such vowel&consonant clusters as dadadadad, mamamamama), modified by the natural diphthong effect of closing the lips and retracting the tongue at the end of the production.