S. nasals vanished at word boundaries before a single consonant; [-{mnŋ}·{fθxsmnŋl}-] > [-ø·{fθxsmnŋl}-]
One interesting feature of Sindarin is that its nasal mutations are distinct from those of Welsh. In particular, the nasal mutations of p, t, c are ph, th, ch rather than voiceless nasals “mh, nh, ngh” as in Welsh. This is especially peculiar given that Sindarin mostly underwent the same medial phonetic developments as Welsh, whereby (for example) nt > nth > nnh (long voiceless nasal) > nn.
To explain this, David Salo proposed a special Sindarin development occurring only at morpheme boundaries, where nasals were lost before other (single) consonants (GS/§4.187). This would mean that the historical development of the definite plural of têw “letter” would be in·tîw > in·thîw > i·thîw, with the nasal loss explaining why the th- remained rather than further developing into a voiceless nasal. This nasal loss occurs for the plural definite article in with most plural forms, whether or not they undergo nasal mutation:
The full form generally appears before vowels and consonant clusters that resist mutation:
The lack of nasal loss before clusters can be explained by the fact that the nasal preceded two consonants rather than one; such consonant clusters resist nasal mutation for similar reasons.
The mechanics of the nasal loss is mostly straightforward. Before other nasals, the n simply assimilated to the following nasal and shortened. Such shortening occurred at the end of words, so it makes sense it would occur at the beginning of word as well, for example: in·m > im·m > i·m. Before s (and probably also l), the nasal assimilated to following sound and again shortened as was the cases with ss (and ll) at the end of words, for example: in·s > is·s > i·s.
Voiceless stops and spirants require a more specialized explanation. Here perhaps the result was [in·t] > [in·θ] > [iθ·θ] > [i·θ], with a backwards assimilation of the nasal to a voiceless spirant rather than the forward assimilation of the spirant to a (voiceless) nasal as occurred medially: [-nt-] > [-nθ-] > [-n̥n̥-] > [-nn-]. Thus it may be that at word boundaries, the nasal was backwards assimilated to a voiceless spirant, and then the long voiceless spirant shortened. This is essentially what David Salo suggested (GS/§4.112, §4.184).
Assuming this is all true, we should be able to find evidence of similar changes in compounds. Some of the best examples are found among the kings of the realms of Arnor and later Dúnedain chieftains, whose names mostly seem to be compounds with an initial elements of ara(n) “king”. For example the 1st chieftain of the Dúnedain was named Aranarth, probably aran + arth “realm” also seen in Arthedain “*Realm of the Edain”. Other names include:
Only the last etymology is confirmed, but all three seem to support nasal loss before voiceless spirants at morpheme boundaries rather than (for example): aran+pant > aramphant > **arammant. Another similar example is N. ifant [*iphant] “aged” = în “year” + pant “full” (Ety/GENG-WĀ, YEN), a word that reappears untranslated in Sindarin (WJ/192).
N. nasals vanished at word boundaries before a single consonant; [-n·{fθxsmnŋl}-] > [-ø·{fθxsmnŋl}-]