Stop mutation results from a preceding (ancient) voiced or voiceless stop t and d, generally from a preceding preposition. This stop is typically lost before consonants with various mutational effects. This is a somewhat speculative mutation, since Tolkien described it but we have no actual examples of this mutation in any attested sentences, except for in one rejected note where Tolkien considered making na-chaered stop mutation rather than soft mutation (PE17/147, where he described it as being from nat- or nad-). This mutation was mentioned but not explained in the Quendi and Eldar essay written in 1959-1960:

As the mutations following the preposition o show, it must prehistorically have ended in -t or -d. Possibly, therefore, it comes from *aud, with d of the same origin as that seen in Q oär (see above). Some have thought that it received the addition -t (at a period when *au had already become ǭ > o) by association with *et “out, out of”. The latter retains its consonant in the form ed before vowels, but loses it before consonants, though es, ef, eth are often found before s, f, th. o, however, is normally o in all positions, though od appears occasionally before vowels, especially before o-. The influence of *et > ed is therefore probably only a late one, and does not account for the mutations (WJ/366-367).

This would indicate that ancient d in o[ð] “from” and ancient t in *et > ed “out” both caused a similar set of mutations in Sindarin. Notes on the A Elbereth Gilthoniel poem in another document from around this period more fully describes the mutation:

o “from”. Cf. Q au-, oa “away” < stem awa. But the ancient Sindarin form must have been au(t), with -t as in e(t) “out” since the beginnings of following words beginning with m, n, b, d, g, s, as here menel, galadh, are not changed; but t, p, c become th, ph, ch (PE17/24).

Indeed in this poem o does not mutate either g or m: o galadhremmin “from tree tangled”, o menel “from heaven” (LotR/238). Tolkien also explored the stop mutation in more detail in one of his etymologies for a “and”:

a “and” (< ad(a) = Q az > ar): in S. this a leaves the initials b, d, g, m, n, s unchanged; but changes p, t, c > f, th, ch; adt > att. In older Sindarin gw became w (for gw is only the S. initial form of basic [ancient] w- : as wath “shadow”, initial form gwath, i-wath “the shadow”); but in later S. and on analogy of genuine original g (as in a galað “and a tree”) gw was left unaltered: a gwath “and a shadow” (PE17/41).

The mutations of p, t, c and lack of g- and m-mutation described in this note is consistent with the notes on A Elbereth Gilthoniel described above. Thus the principle effect of stop mutation is that it turns voiceless stops p, t, c into voiceless spirants ph, th, ch. This is consistent with the phonetic change whereby voiceless stops became voiceless spirants after other consonants. In the case of combinations with an ancient voiced stop like ad·t, the combination would already have unvoiced in Common Eldarin to at·t before continuing to long spirant ath·th, ultimately shortening to simple a·th. Stop mutation was mentioned in the Early Noldorin Grammar of the 1920s, and had exactly this mutational effect.

In the Quendi and Eldar essay Tolkien said that “es, ef, eth are often found before s, f, th” for the preposition ed “out” (WJ/367). This is likely the result of the long voiceless spirant failing to shorten, as in: et·th or et·t > eth·th (normally > e·th). An alternate interpretation is these are relatively late innovations that apply only to original initial f, th but not to the f, th resulting from mutation of p, t. The advantage of this alternate interpretation is that it helps differentiate original f, th from mutated voiceless stops: e thol “out of an island (tol)”, eth thôl “out of a helmet (thôl)”.

Other Initial Consonants: The effect of stop mutation on rarer initial consonants is more difficult to determine. David Salo proposed a system in his book Gateway to Sindarin (GS/77) that has been more or less adopted in Neo-Sindarin. In particular, he suggested (a) ancient nasalized voiced stops would be fully restored, (b) voiceless liquids would be voiced and (c) h would become ch.

Regarding ancient nasalized voiced stops, the stop in the preceding preposition would voice, nasalize, assimilate, then shorten, as in: et·mb > ed·mb > en·mb > em·mb > e·mb. Regarding voiceless sounds hw, lh, rh, the preceding stop would eliminate the s (or kh) that produced the unvoicing, resulting in a simple voiced sound ’w, ’l, ’r: et·sl > ed·sl > ed·l. It seems like that the preposition itself would retain its consonant as it does before vowels, and likewise before regular voiced liquids r, l.

Regarding the mutation of initial h, like gw- < w- mentioned above (PE17/41), it arose only from older ch-, so its “mutation” to ch would result from the inhibition of an initial sound change rather than a replication of a medial sound change. Unlike gw-, however, there are no analogous forms from which the h might be restored. Thus it is likely that the stop mutation of h was ch. It is even possible that, like the stop mutation before f, th, its voiceless spirant was lengthened to better distinguish it from the stop mutation of c: e chû “out of a bow (cû)”, ech chû “out of a dog (hû)”. However, Tolkien did not mention this in his list of s, f, th given above (WJ/367).

Summary: I would suggest the following for stop mutations in Neo-Sindarin for prepositions like ed “out” derived from ancient t:

• Initial voiceless stops (p, t, c) become voiceless spirants (ph, th, ch).
• Ancient nasalized stops ([m]b, [n]d, [n]g) are fully restored to mb, nd, ng.
• Voiceless sounds (hw, lh, rh) are voiced to ’w, ’l, ’r.
• h becomes ch.
• Before most consonants, the d in the preposition is lost: e dû “out of a night (dû)”, e thol “out of an island (tol)”.
• Before revoiced ’w, ’l, ’r as well as original liquids l, r, the d in the preposition survives: ed ’lûg “out of a snake (lhûg)”, ed ram “out of a wall (ram)”.
• Optionally, before original f, th, s as well as ch from h, the d of the preposition assimilates to the initial consonant: eth thôl “out of a helmet (thôl)”, ech chû “out of a dog (hû)”.

This last rule is rather speculative, since it depend on a particular interpretation of Tolkien’s words whereby “es, ef, eth are often found before s, f, th” applies only to original spirants and not mutated spirants, even though this isn’t really supported by the expected phonetic developments. It seems to be an optional rule, and it may be better to ignore it since it isn’t clear how it is applied.

Prepositions like o “from” having ancient final d would still cause stop mutations, but the consonant is almost always lost except in rare cases before vowels. Prepositions ending with other ancient stops (p, b, k, g) do not cause mutation at all.

I consider stop mutation a “somewhat optional mutation” for Neo-Sindarin. There is strong evidence that Tolkien did use it, but it is obscure and applies only in a few situations. It is a mutation that a less experienced Sindarin speaker (especially mortal Men) could easily forget to apply.