Nunggubuyu is characterised by a series of alternations in which each of the six stop phonemes is paired with another consonant (lateral, rhotic, or semivowel):
In most instances, the stop is derived from the other consonant by Hardening P-18, a rule which applies to the underlying consonants shown above (in the right column) when they are preceded by a stop or nasal (see top of this page for exx.). However, there is also a more restricted morphophonemic rule in which the stop appears to be underlying and the other consonant derived (Lenition P-16). (Other hardening/lenition rules are P-17 and -19.)
In this publication we use the term continuant (or paired continuant) for this specific set of sonorants related by these rules to the various stop phonemes. Note that this set of six continuants does not form a natural phonetic class, since three liquids are included but two others (/1 1/) never alternate with stops.
The other problem, of course, is that /w/ is related both to /b/ and to /g/. This forces us to recognise two distinct entities which we label /w2// (related to /b/) and //w1// (related to /g/), though this distinction can only be made in morpheme-initial position since the alternations occur there (there are also some cases of morpheme-initial /w/ which cannot be subclassified since they do not happen to occur in a Hardening environment).
It might seem that we could evade several difficulties by assuming that instead of hardening continuants to stops, the phonological rules lenite underlying stops to continuants. This would save us the trouble of distinguishing //w1// from //w2//, since we would have just underlying //g// and //b//, whose lenited outputs would happen to converge as /w/; this would also save us from recognising a non-natural phonological class of paired continuants, since we would have an underlying natural class of six stops. However, this analysis is synchronically quite impossible, except for the lexically restricted set of alternations dealt with by Lenition P-16. The vast majority of stop/continuant alternations dealt with here by Hardening P-18 must have an underlying continuant rather than stop.
To begin with, in such alternations the continuant has a broader distribution than the stop: in the morphemes in question the initial continuant is found in word-initial position, and medially after a vowel, liquid, or semivowel, while the paired stop is found only medially after stop or nasal. Secondly, a hardening rule (continuant stop) makes sense for such items, since a sequence of surface stop or nasal plus continuant is impossible in Nunggubuyu (see discussion of possible surface consonant clusters below, this chapter). There is no such canonical-pattern motivation for an inverted lenition rule for these morphemes (stop continuant), since a stop can perfectly well occur on the surface in word-initial position or medially after vowel or any consonant.
Finally, handling these productive alternations with a hardening rather than a lenition rule allows us to have a perfectly natural set of morpheme-initial consonants in underlying forms. Some morphemes always begin in a stop /b dh d d j g/ which never becomes a continuant; we set these up as having underlying initial stop. Other morphemes show stop/continuant alternations (the stop only medially after stop or nasal); we set these up as having underlying initial continuant. There is no third set of morphemes with invariant initial continuant which resists hardening to stop after stop or nasal (though there are some morphemes which do not happen to occur in such an environment and thus happen to show initial continuant in all attested forms). If we replace Hardening P-18 with a productive lenition rule, we will then have to prohibit /w 1hr r y/ as morpheme-initial consonants (though /1 1/ would be permitted), and would have to recognise one stop series /b2 dh2 d2 d2 j2 g2/ which undergoes lenition vs. a distinct stop series /b1 dh1 d1 d1 j1 g1/ subject to lenition. The analysis would result in a quite horrendous proliferation of indexing, along with an unnecessary and awkward set of underlying constraints.
A few additional remarks about //w1// and //w2// are in order. Aside from being related to different stops, they also differ in that //w1// as stem-initial segment shows a distinctly greater tendency to be deleted than does //w2// in such environments as /a-__a/ (see w1-Deletion P-9). There is a suggestion, then, that /w1/ and /w2/ might be articulatorily distinct as surface consonants, with the consequent possibility of distinguishing them even in morpheme-medial (or -final) position. In environments where w1-Deletion P-9 is applicable, instead of complete deletion one sometimes hears (and/or sees) a half-hearted lip-rounding, giving a faintly audible approximation to surface /w/. However, speakers differ in their treatment of //w1// in these positions (some retaining it, others deleting it entirely, others half-pronouncing it, or fluctuating among these). There does not seem to be any distinction possible in such environments as /u- __a/. In general, attempts to transcribe /w1/ vs. /w2/ morpheme-internally have been unsuccessful, and my feeling is that the fairly systematic distinction between //w1// and //w2// in deletability (by P-9) is limited to morpheme-initial position, where it is reinforced by the important phonological opposition between the two in connection with Hardening P-18.
In some lexical items, initial /w/ fluctuates between /g/ and /b/ as its hardened counterpart, hence dictionary entries give both /w1/ and /w2/ as variants. In the case of /=w1urya-/ 'to pour (liquid) on', the hardened form is always /=gurya-/ in compounds, but can be either /=gurya-/ or /=burya-/ in uncompounded form. The same pattern is found with /=w1uldha-/ 'to sever' and with /=w1irgira-/ 'to be skinny' (i.e., variants with /w2/ occur occasionally in uncompounded forms only). On the other hand, with /-w2ayama-/ 'to proceed, keep moving' and /w2ulhuwulhur/ 'first, previous' the hardened form in compounds is with /b/ in all attested exx.; again, the uncompounded form shows /b/ or /g/. It looks as though the older form is best preserved in compounds, while the uncompounded forms show some tendency to introduce fluctuation. Other stems for which some variation is attested are: /=wajala-/ 'to pull out sedge roots' , /=w1alhalha-/ 'to shake spear with woomera', /=waradar-wu-/ 'to eat leftovers in morning', /=warwargu-/ 'to carry along', /=wayajara-/ 'to lay out (bed)', /=wuyuguyuma-/ 'to carry in arms', /wundul/ 'calm', /=wuriyi-/' 'to lift up', and /=w1udha-/ 'to be up (on roof)', where we show /w1/ or /w2/ in the transcription only if there is a compound attested showing the presumably older form.
With some noun stems, the distinction between initial //w1//, //w2//, and zero may be difficult to determine in underlying transcriptions. The distinction is very clear in verbs, and if the noun happens to be one which can be used as a compound initial (with following verb) we merely add appropriate verbal inflectional prefixes and observe phonological behaviour. If the noun is an adjectival noun (i.e., one which can be used as a predicate with intransitive inflectional prefixes), the same tests are applicable. With nonadjectival nouns which cannot be used as compound initials we may be in trouble. Some such nouns, denoting body parts or the like, may occur with derivational noun-class prefixes (4.7), like MSg/NA /niG-/; this permits recognition of //w2// if we get initial surface /b/, but initial surface /g/ could be either the hardened form of //w1//, or else surface /g/ inserted before stem-initial vowel by Velar-Insertion P-8 (thus for some nouns our dictionary entry shows initial /(w1)a. . ./ or the like indicating that we cannot determine whether an initial //w1// is present or not. Other nonadjectival nouns do not occur with such derivational noun-class prefixes. If the unprefixed form starts with /wu.../, we try the form with (inflectional) noun-class prefix, which will usually tell us whether the surface /w/ is real or is inserted by a low-level rule (P-5) applying in word-initial position, in which case the prefixed form should undergo V-Truncation P-46, If the //w/ is real, we probably will never know whether it is //w1// or //w2//, so we just leave it as /w/ in the dictionary. If the form without prefix starts with /wa.../ or /wi.../, the chances are it is really //w2// at the beginning, since //w1/ is often (but not rigorously) dropped in such positions by w1-Deletion P-9. If the form starts with /a.../, we could have either underlying vowel-initial stem, or else underlying //w1a...// with the //w1// deleted by P-9, and again it may not be possible to go any farther (we write /a.../ in the dictionary entry if there is no definite indication of presence of semivowel).
We might mention that borrowed /w/ from well-integrated English loans may come in as //w2//. English work as a noun is heard as /warg/, and a verbalised form /=w2ardha-/ 'to work' is now common: cf. /ngam=bardhi:-'/ 'I will work'. Since //w1// in this position is often deleted by w1-Deletion P-9, the clear pronunciation of the /w/ in the English word probably favoured //w2// in the borrowing.