Just as short final vowels vanished in Sindarin and Noldorin at the end of words, they also vanished in the middle of words at morpheme boundaries. As suggested by David Salo (GS/§4.115), this vowel loss at morpheme boundaries was later than the loss at the end of words. There are a couple of examples that support this ordering of sound changes:

• ✶Gondō̆-lindē > Gondolind > Gond-lind > S. Gonlin, the “true” Sindarin form of Gondolin (PE17/133).
• ᴹ✶asmalindē̆ > ON. ammalinde > em(m)elin > N. emlin “yellow bird“ (Ety/SMAL).

This sound change must also have taken place after voiceless stops voiced after vowels, since the second element of compounds consistently show the soft mutation of p, t, k to b, d, g:

• ✶Turukondō > S. Turgond (PE17/113).
• ON. andatektha > N. andeith “long-mark” (Ety/TEK).
• ᴹ✶minitunda > N. mindon “tower” (Ety/TUN).

This change occurred based on how the morphemes were perceived at the time of the vowel loss. The example of Elbereth is illustrative:

• ✶elen-barathī > el-mbereth > S. Elbereth (MR/387; PE17/022).

If this were a late compound, we would expect to see soft mutation in the compound êl “star” + bereth “queen”, which would therefore become **Elvereth. Primitively the compound was elen + b-, and the final n became m by assimilation to the following b: elem-b-. By the time the final vowels began to vanish, the primitive final n had vanished in the independent Sindarin word el(e) “star”, so the first part of the compound was reanalyzed as el(e)-mb-. Thus the second e vanished, but the nasal cluster mb inhibited the soft mutation of the b, and the result was Elbereth.

Another example is ✶Galadandil [> Gladandil > Gleðendil] > Gleðennil “Lover of Trees” (PE21/83). Here the initial element of the compound was modified so much from its independent form (galadh) that it was no longer perceived as a distinct morpheme, and the vowel was not lost.

Where the vowel loss did occur, this sound change could cause clusters of three and sometimes even four consonants to appear, something that almost never happened in Primitive Elvish or Quenya. In most (but not all) cases, these clusters were later simplified.

Conceptual Developments: Since not all final vowels vanished in Gnomish, perhaps vowel loss at morpheme boundaries was less prevelent in the 1910s, but there are examples showing it happened at least sometimes, in both Gnomish as well as the Early Noldorin of the 1920s:

• ᴱ✶ambarinþǝ > G. amrint “lark” (PE13/110).
• ᴱ✶amba-ros-sa > ᴱN. amros “rising, soaring, lofty” (PE13/159).

GS/§4.115 @@@ but compare Magladhûr, Nauglavir