Summary of the evidence for Onset Maximization

Consonant sandhi as an assimilatory phenomenon makes the most sense if one interprets it not as a word boundary phenomenon with some unaccountable overlap into morphology, but as a property of syllable codas. Therefore if consonant C fails to undergo before word-internal X the changes that it would suffer if a word boundary (ergo syllable boundary) intervened, then the first explanation must be that C is not in a coda, i.e., that CX is a possible onset cluster. Since this often happens in a binary cluster which the CS theory would obligatorily split up, this is strong evidence for the OM theory. It is reinforced by the fact that complex onsets do occur word-initially, and that all medial clusters are analysable as codas (with appropriate sandhi) followed by onset sequences whose pattern is consistent with the patterns evidenced by studying sandhi effects and the inventory of word-initial sequences (as well as rather late evidence from the metre).

Specifically, the onset template accepts:

1. A segment followed by a segment of greater sonority. This pattern is found widely throughout the languages of the world, and is predicted by the Sonority Sequencing Principle (Sievers 1876:111-113), which says that sonority should increase monotonically the closer one gets to the sonority peak, the vowel. The universal sonority hierarchy used in Vogel (1977:135-136) and Clements (1990:292-298) fills the bill exactly: obstruents are less sonorous than nasals, which are less sonorous than liquids, which are less sonorous than glides.
2. Non-palatal voiceless unaspirated stops before s. The evidence comes mostly from word-initial sequences, Epic metre, and the authority of the Taittirîyapraâtis"âkhya. But sandhi tests cannot confirm this, since sounds like k would remain unchanged before s whether in a preceding syllable or not. And the sequence violates Clements's version of the Sonority Sequencing Principle, although it would be acceptable by the original and more common formulation that says that fricatives are more sonorous than oral stops (Sievers 1876:112, Kiparsky 1979:432; cf. Clements 1990:315-316).
3. Voiceless fricatives (the sibilants s", s., s) before voiceless stops, with the dental s assimilating in place to a following coronal. This also violates the Sonority Sequencing Principle, except for those versions that permit sequences that are equisonorous or nearly so (Hermann 1923:5), but seems well justified not only by word-initial sequences and BHS metre but also by the sandhi of s" and s.. The case of s is harder to prove, since it only disarticulates at the end of phonological phrases. Steriade (1982:312-333) argued that s does not form onset clusters, on the basis of reduplication and the fact that s disappears between stops in the morphology. In her view, the fact that a word-initial s before an oral stop is not copied by reduplication processes means that it is not syllabified into the onset, and hence is invisible to the copying process. But in Gothic, the entire s-stop cluster reduplicated, even though otherwise Gothic too only reduplicated one consonant: stai-staut `pushed' vs. gai-grot `wept' (Kiparsky 1979:435), so syllabic invisibility may not be the proper analysis of sT peculiarities in reduplication.
4. Labial sonorants(7) (m, v), and h(8) before sonorants. From the standpoint of the Sonority Sequencing Principle, this is shocking. The glide v should be among the most sonorous of all sounds, and hence not followable by any other segment in the onset. And certainly nasals should not be followed by nasals, if one subscribes to the version of the theory by which onset consonants must strictly increase in sonority. There does seem good evidence for the clusters, however. Words begin with ny, nv, my, mr, ml, vr, vy, hn, hr, hl, hv, hy. While m disarticulates or assimilates before onsets when in a coda, it is found word-internally before n, y, r, l, and v. Similarly av is found before n., y and perhaps r word-internally, instead of monophthongizing to e. And the morphology shows that h is always converted to a stop when in a coda, so hn., hn, hm, hr, hl, hy, hv must be onset sequences.