In both Sindarin and Noldorin, the short vowels i and u were lowered to e and o in the syllable before final a, a phenomenon known as a-affection after the corresponding sound changes in Welsh (WGHC/§68). Tolkien himself used this term for the phenomenon:

Since the a-affection which produces geil from *gilyā only operates with final -ā, the form in compounds is gil-. Gilgalad = “starlight” < *gilicalat- (PE17/152).

This phenomenon was recognized as part of Sindarin’s phonological history as far back as Jim Allen’s Introduction to Elvish published in 1978 (pp. 115-116, in an article written by Chris Gilson and Bill Welden). A more recent and more thorough analysis appears in Bertrand Bellet’s Vowel Affection in Sindarin and Noldorin article (VASN), published in Arda Philology 1 in 2005. My analysis is essentially the same as his, but with some additional examples that have been published in the intervening years.

Examples of this sound change are numerous in both Sindarin and Noldorin:

• ✶lisyā > liχı̯ā > leχı̯ > leich > S. laich “sweet” (PE17/148).
• ✶dankĭnā > daŋχen > danghen > S. dangen “slain” (PE17/133).
• ᴹ✶rimbā > ON. rimba > rhemb > N. rhem “frequent, numerous” (Ety/RIM).
• ✶ugrā > ogra > S. oer “nasty” (PE22/160).
• ✶ambuna > S. amon “hill” (PE17/92).
• ᴹ✶tundā > tond > N. tonn “tall” (Ety/TUN).

This sound change was limited to the syllable immediately before the final a, and did not apply to long [ī] or [ū]:

• ✶spindilā > S. †findel “head of hair, fax” (PE17/17).
• ✶(n)dūrā > S. dûr “dark” (PE17/152).

One of the challenges of analyzing a-affection is that it was frequently obscured by other sound changes. A-affection is fairly obvious in the case of ✶dankĭnā > S. dangen and ✶ambuna > S. amon, less so in ✶lisyā > S. laich and ✶ugrā > S. oer. The final a that produced the affection was itself invariably lost when short final vowels vanished. Within the fictional context of Tolkien’s world, a-affection can only be detected by comparing related words that have different vowels, either within Sindarin itself or across Sindarin and Quenya. Fortunately, in many cases Tolkien made it clear when a-affection occurred by providing the primitive forms of Sindarin or Noldorin words, making the existence of primitive final a explicit.

Another obscuring sound change for a-affection is that short [u] often became [o] more generally in Sindarin and Noldorin. It can be tricky to tell when u > o was a result of a-affection, and when it was simply part of the general trend of u > o. There are, however, some conditions in which short u is usually preserved, such as before before nasals: ✶turunku > S. trunc “great stake” (PE21/80) or ᴹ✶tundu > N. tunn “hill, mound” (Ety/TUN). In such cases, the change of u > o must be the result of a-affection; compare with ᴹ✶tundā > N. tonn “tall” above.

Since final -ā was a common feature of Primitive Elvish adjectives, one net result of a-affection is occasional i/e and u/o variation between Sindarin and Noldorin nouns and adjectives. Examples include:

• N. lhinn (n.) “air, tune” vs. N. lhend (adj.) “tuneful, sweet” (Ety/LIN²).
• N. rhim (n.) “crowd, host” vs. N. rhem (adj.) “frequent, numerous” (Ety/RIM).
• N. rhinn (n.) “circle” vs. N. rhenn (adj.) “circular” (Ety/RIN).
• N. crum (n.) “left hand” vs. N. crom (adj.) “left” (Ety/KURÚM).
• S. rust (n.) “copper” vs. S. ross (adj.) “red-haired, copper-coloured” (VT41/10).
• N. tunn (n.) “hill, mound” vs. N. tonn (adj.) “tall” (Ety/TUN).

A-affection also plays a role in the origin of the common Sindarin and Noldorin adjective suffixes S. -eb < ✶-ikwā (WJ/412) and S. -en < ✶-inā (PE17/131). As pointed out by Bertrand Bellet (VASN), the latter suffix often appears as -in in (ancient?) compounds: S./N. Celebrindal “Silver Foot”, S. Glórindol “Golden Head”, N. Melthinorn “Tree of Gold”. Presumably the a-affection did not take place medially in these compounds, so that the in was preserved. Whether the in would also be restored in later compounds is unclear, but examples like S. Calenhad “Green Space” and N. Methendol “*Last Hill” suggests this was probably not the case.

A-affection (unlike i-affection) does not seem to play a general role in verb inflection. In theory, any a-stem verb with i or (rare) u as its base vowel might undergo a-affection in its uninflected present-tense forms, but there are a several examples that indicate this is not the case: linna “sing” (PE17/27), S. minna “enter” (PE17/41), N. thinna “fade” (Ety/THIN), S. ruthra “rage” (PE17/188). Likely these verbs were reformed by analogy with their inflected forms, such as linnon “I sing”. However, there are a couple of examples where a-affection seems to play a role in the development of verbs that had more complex phonological changes:

• ᴹ√SUK > N. sautha- “to drain” (Ety/SUK): apparently [sukta-] > [suxθa-] > [soxθa-] > [souθa-] > [sauθa-].
• √RIK > S. raitha “to try” (PE17/167): apparently [rikta-] > [rixθa-] > [rexθa-] > [reiθa-] > [raiθa-].

These verb forms are hard to explain without assuming a-affection of the base vowel [u] > [o] and [i] > [e], respectively. The second example is contrasted with inflected forms (also on PE17/167) where no a-affection took place, such as past forms rithant and rithas. The verb form raitha is doubly-unusual in that it also shows [ei] > [ai], something that normally occurs only in final syllables in Sindarin. A nearby deleted form reitha indicates Tolkien may have been uncertain of the exact development of this verb.

There are, however, other examples where no such a-affection takes place in uninflected verb forms:

• √RUK > S. gruitha- “to terrify” (WJ/389, 415): apparently [grukta-] > [gruxθa-] > [gruiθa-].
• ✶nuktā- > S. nuitha- “to stunt” (WJ/413): apparently [nukta-] > [nuxθa-] > [nuiθa-].
• ᴹ√RIK(H) > N. rhitha- “to jerk, twitch, snatch” (Ety/RIK(H)): apparently [rikta-] > [rixθa-] > [rīθa-] > [r̥iθa-].

This last example appears only in its Noldorin-style infinitive form: rhitho. Given the sparsity of examples, it is hard to determine what these variations represent: (a) conceptual vacillations on Tolkien’s part, (b) dialectical differences or (c) they are simply examples where the uninflected verb underwent a-affection but for whatever reason was not reformed to match its inflected forms.

With all the complex interactions between a-affection and other sound changes, determining when this change occurred in Sindarin/Noldorin phonological history is especially important. Based on Noldorin examples like rhem above, I assume the sound change occurred after the Old Sindarin/Old Noldorin period, as does David Salo (GS/§4.95). It must necessarily have occurred before short final vowels vanished, because the a that caused the affection later vanished in all cases. There are a few examples that indicate it also took place before the vocalizations of voiced and voiceless spirants:

• ✶ugrā > ogra > S. oer “nasty” (PE22/160).
• ᴹ✶tupsē > N. taus “thatch” (Ety/TUP).

The latter isn’t an obvious example of a-affection, but since final [e] became [a] after [s] (at least in Noldorin), the apparent development is: [tupsē] > [tupsa] > [tuɸsa] > [toɸsa] > [tousa] > [taus]. This phonetic development is only possible if a-affection took place before the vocalization of the labial spirant [ɸ] > [u]; if the vocalization of [ɸ] occurred first, the ultimate result would have been **tûs.

@@@ Example of change at morpheme boundary: ON Certhan

Conceptual Development: There isn’t any clear evidence of a-affection in Gnomish, but there are examples of a-affection in the Early Noldorin of the 1920s, as pointed out by Roman Rausch in his Historical Phonologies of Ilkorin, Telerin and Noldorin around 1923 (HPITN/§4.2.1). Tolkien actually mentioned a-affection in the Early Noldorin Grammar:

Note where e is from i by a-mutation, before an original plural ending the original i reappeared, so we get an apparent mutation e > i (PE13/122).

Some examples of Early Noldorin a-affection include:

• ᴱ✶aurina- > ᴱN. oren “hot (weather)” (PE13/160).
• ᴱ✶ninda > ᴱN. nen(n) “stream” (PE13/123).
• ᴱ✶kyurna > ᴱN. corn “cheese” (PE13/140).

Roman Rausch also suggested that a-affection may have occurred in Early Noldorin even in cases where the a was not final (HPITN/§4.2.1). In particular, he noted the prefix ur- “without, -less”, which appeared as or- when prefixed to words containing an a: ᴱN. urbeth “wordless” and ᴱN. urhonn “heartless” vs. ᴱN. orfang “beardless” and ᴱN. orsarn “stoneless” (PE13/156). As Tolkien described it:

ur- (uru-) privative prefix, “without, -less”; before a noun with a in first syllable or-; with other ur-; with i- yr-. (or- should occur before ai = aı̯, akt etc.; oer, or before ai = a mutated; to yr before eı̯, i) (PE13/155).

Whether this was a general rule or limited to the prefix ur-/or- is unclear.

GS/§4.95 WGHC/§68 @@@