46 3 Phonologyfollowing obstruent by a short V in the imperative. This is the case with
superheavy verb stems that have basic (i.e. Imprt) shapes beginning CvCV... (v
= short vowel subject to Stem-Initial Syncope, V = any long or short vowel).
An example is '(door shutters) be brought together', Imprt nakabba-t, hence
basic lexical form -nvkvbbu- (+ -t). The Lolmpf appears in T-ka (and T-md)
as -t-lqkabbu-t with phonetic [rjk], showing Stem-Initial Syncope followed by
Nasal Assimilation and Default Accentuation. In K-d we get -t-lnkabbu-t,
showing Stem-Initial Syncope followed by Default Accentuation but no Nasal
Assimilation. The K-d informant occasionally had a faint short vowel, or at
least a separate consonantal release, between the two C's, suggesting that
Stem-Initial Syncope here is very low-level indeed. In A-grm we get
-t-inakabbi-t, where Stem-Initial Syncope fails to apply, Default Accentation
targets the unsyncopated schwa, and of course there is no opportunity for
Nasal Assimilation to apply.
In all dialects, Nasal Assimilation does apply to the onsets of light verb
stems, e.g. PerfP -ambasr- 'be taken to pasture at night' and Imprt ambar,
compare LoImpfP -nabbasr- and VblN a-naebar (dialectally also t-a-nabre and
t-a-naebra).
One could infer from these data that Stem-Initial Syncope in the onset of
superheavy stems like -nvkvbbu- (+ -t) is a recent dialectal development.
Mediopassive prefix -m- on verbs, and Agent prefix -m- deriving nouns
from verbs, are replaced by dissimilatory allomorph -n- when the stem
contains a labial consonant {bfm}. It does not matter whether there is an
intervening η or other coronal. Thus agentive a-n-anam 'one who is fond'
(verb -vnvm- 'be fond'), and mediopassive -ann-abda- 'be dislocated'. The
regular labial prefix -m- is observed in agentive ae-m-djrad 'one who can
disappear' and mediopassive -sem-era- 'be opened' (PerfP). For more details,
examples, and exceptions, see §8.3 (mediopassive) and §8.8 (agentive).
Similarly, the Reciprocal prefix, normally -nvm-, is replaced by -n- (§8.4), and
denominal derivational prefix -mvs- is replaced by -nvs- (§8.10) when the
stem contains a labial consonant.3.2.1.3 /d/ —> t before voiceless obstruent
d is devoiced to t before a voiceless obstruent (stop, fricative, or sibilant, but
not h). Thus -atfises- '(udder) be full', Imprt atfas, etc., but LoImpfP -ddffass-
bringing out the lexical d. Likewise, -atfa- 'be poured', Imprt Eetf, but
LoImpfP -daff- and related noun t-e-dasffe 'bank (of pond)'. Contrast -odha-
'fold (tent)''.
d and t may once have been allophones of a single phoneme (see §3.1.1.4,
§3.1.1.8) but they are now at least partially separate.
In the case of the verb 'laugh', the C sequence is either Vts (in the
unbroken cluster ts) or Vdz (when the two are separated by a V), thus PerfP
-atsa- but LoImpfP -dazz-. Since the cluster ts is entirely voiceless, there is no