3.2 ngu-Epenthesis P-1.

A meaningless morpheme /-ngu-/ is automatically inserted between a prefix P and a following morpheme (prefix or stem) Q under certain conditions. Q must begin with a stop /b dh d d j g/ at the (early) stage when the rule applies. P can be a derivational prefix or compound initial of any phonological shape; it can also be an inflectional prefix but in this event must be terminated by a consonant. The rule may apply more than once within a word.

We exemplify with the verb /=bura-/ 'to sit' as our stem Q. Most inflectional prefixes (marking pronominal subject) end in a vowel and thus do not cause ngu-Epenthesis: /nga=buri-Ø/ 'I sat'. However, a handful of these prefixes end in a consonant, namely /n/ as in 1Sgb /ngan-/ and here the epenthesis does apply: //ngan=bura-y///nga-ngu=buri-Ø/ 'I was going to sit'. (Note that the prefix-final /n/ is deleted by a later rule.)

When a derivational prefix or compounding initial intervenes between the inflectional prefix and the stem, epenthesis always applies to stop-initial roots like /=bura-/ as in /-anyji-ngu=bura-/ 'to sit with (someone)' (Comitative /-anyji-/ .

With noun stems, epenthesis does not apply when the only prefix is an inflectional noun-class prefix: /gulmung/ 'belly' takes prefixed forms /mana-gulmung/ and /ama-gulmung/ without /-ngu-/ However, some nominal constructions involve an intervening derivational noun-class prefix, and these all happen to end in a consonantal archiphoneme which we represent as /G/ (the inflectional noun-class prefixes end in vowels). Whether because of this /G/ or because of the difference between derivational and inflectional prefixes, stems like /gulmung/ require /-ngu-/ in the derivational forms, e.g., //mana-maG-gulmung///mana-ma-ngu-gulmung/ 'belly (of certain objects)'. Compounds in which the second element is a noun are not very numerous, but when they do occur /-ngu-/ is added if the second element begins in a stop: /ngul-ngu-jir/ 'pubic hair' from /-jir/ 'fur, body hair' (see 4.7, 14.5).

The epenthesis rule does not apply in pronominal or demonstrative forms, but this is presumably due to the fact that no phonologically appropriate combinations occur in these morphological classes. We therefore formulate the rule as follows.

(P-1)ngu-Epenthesis
? →-ngu- / P-__-Q
where P is an inflectional (pronominal) prefix ending in a consonant, or a derivational prefix or compounding initial (of any form), and Q is a root or a pre-root derivational/compounding element beginning in a stop.

There are a few exceptions, perhaps reflecting the fact that this is clearly an 'unnatural', highly morphologised rule. Consider, for example, the clan name /nung-garangari/ based on place name /garangari/ and Gentilic prefix /nuN-/ With ngu-Epenthesis we should finally end up with /nu-ngu-garangari/ but here the epenthesis clearly does not apply.

This, and a scattering of similar examples, can be analysed formally as exceptions to P-1, but there is an alternative: they could be considered instances where a stem-initial stop is lenited (e.g., /g//w1/ as derivational/compounding final, so that ngu-Epenthesis cannot apply; a later rule then hardens the continuant back to stop articulation (/w1//g/ / in this case). In fact, the lenition and hardening rules are independently motivated (see P-16 and P-18, also P-19), and there are some verb forms where this is clearly a legitimate excuse for nonapplication of P-1. In other cases, like /nung-garangari/ there is no independent evidence that these particular stems are subject to the lenition rule (which is lexically restricted).

There are also a tiny handful of noun stems which appear to take /- ngu-/ in forms with derivational/compounding initial but which begin in consonants other than stops: /ngumij/ 'ants', derivational noun-class forms like /ma-ngu-ngumi 'scraps, bits and pieces' (prefix //maG//). Such forms are clearly isolated and may involve semantic shifts which make synchronic connections with the simple stems dubious.

Overall the exceptions are few in contrast to the great number of stems for which the rule works neatly. How can we explain the existence of such an unusual rule, in which a morpheme of distinctive phonological shape (not merely a schwa or the like) functions as an epenthetic element?

In a number of cases ngu-Epenthesis permits retention on the surface of important consonantal oppositions threatened with loss due to other rules. Specifically, Hardening P-18 converts stem-initial continuants /w1 w2 y r r lh/ to stops /g b j d d dh/ when preceded by a stop or nasal. Were it not for ngu-Epenthesis, stem-initial oppositions like that seen in //=jura-// 'to push' and //=yura-// 'to transport' would be neutralised on the surface when preceded by a prefix (inflectional or derivational) or compound initial ending in a stop or nasal. In fact, however, //=jura-// will always become /-ngu=jura-/ in such environments (plus some others where homophony is not threatened). This functional explanation accounts for the otherwise strange fact that /-ngu-/ is inserted only before underlying stops, and accounts (in part) for further details of the rule.

Etymologically, /-ngu-/ may be traceable to *-gu-, a morpheme found at the end of some complex pronominal-prefix combinations in near and intermediate protolanguages. Originally a dative marker, then a postconsonantal oblique (accusative/dative) allomorph with a semi-epenthetic character, it occurs in various functions in nearby attested languages (Mara -gu- in pronominal prefixes, Ngandi -gu- and Warndarang -ngu- in similar functions). The alternation of g with ng in these cognates is irregular, but the innovative nasal form may be due to cluster assimilation in such forms as *ngan-gu- (→ *ngan-ngu-).