2.8 Distributional restrictions: medial CC clusters.

We deal here with medial clusters of two consonants, both within and across morpheme boundaries. /CCC/ clusters are dealt with in §2.9, below.

On the following page, Table 2-3 summarises the attested /CC/ clusters (note that the table's explanation carries over into the page after that). The table is somewhat simplified, but still perhaps difficult to understand, and for that reason we will try to provide a reasonably clear prose commentary here.

First, we observe that interdentals /dh nh lh/ never occur as first member of a cluster (more generally, they never occur in syllable-final position). The only possible counterexample would be the homorganic nasal-stop cluster /ndh/, if retranscribed as /nhdh/ (the two do not contrast).

Fifth, the "paired continuants" (see above, §2.5) /lh r r w y/ do not occur as surface forms following a stop or nasal; if such underlying clusters occur at a boundary, the continuant is converted into the counterpart stop by Hardening P-18.

Having made these five general remarks (which we will refer to in what follows), we now go over particular types of cluster (stop-stop, nasal-stop, etc.) one by one, indicating the types of boundaries where the clusters occur, the rules which alter underlying clusters, and rules which create surface clusters. We refer to the most significant phonological rules (Chapter 3), but do not aim at completeness.

stop-stop. Possible underlying initials are /d d j g/; possible finals in underlying forms are all six stops. Since no relevant suffix begins with underlying stop, and no inflectional prefix ends with a stop, the only possible underlying stop-stop combinations are compounds and similar derivatives, in practice almost exclusively with a verb as the second morpheme. It is difficult to produce all relevant underlying clusters since relatively few verbs begin with underlying stop and since /d d/ in particular are uncommon at the end of compound initials or derivational prefixes.

Moreover, compounds and derivatives of this type normally trigger ngu-Epenthesis P-1, which inserts a dummy morpheme /-n^gu-/ between the verb and the preceding compounding/derivational element, so the underlying stop-stop sequence is not realised as such. On the other hand, surface stop-stop clusters are created by Hardening P-18 from underlying clusters with a stop followed by a paired continuant across a boundary. Thus underlying //j-b// at a derivational boundary will become //j-n^gu-b// (and eventual //n^y-n^gu-b//) by P-1 (and P-27), but underlying //j-w₂// at any kind of boundary will become surface /j-b/ by P-18. Thus /jb/ is one of a set of stop-stop clusters which occur both in underlying and surface forms at boundaries, but where the underlying and surface exx. happen to involve a completely distinct set of combinations.

Generalisations discussed earlier in this section explain the absence of /dh/ or /b/ plus other stop, and the surface absence of /g/ plus other stop (because of P-29). Geminate clusters like //dd// and //jj// are reduced by Geminate-Contraction P-31. Other clusters involving /j/ plus coronal stop /dh d d/ are unstable as surface forms because the /j/ is often deleted by Palatal-Deletion P-26.

nasal-stop. The homorganic type /mb ndh nd nd n^yj n^gg/ is very common, especially since (at boundaries) underlying //n^g// or archiphoneme //N// (along with one case of //n//) assimilate to a following stop's point of articulation. Aside from these, /n/ and /n/ occur as first elements with several following stops; /n^y/ is a little more restricted and does not occur within morphemes with following nonhomorganic coronal stop /dh dd/, though if such a cluster arises at a boundary it may survive.

At compound/derivational boundaries, an underlying nasal-stop cluster will be broken up by n^gu-Epenthesis (which applies in this morphological environment before stem-initial stop regardless of the segments to the left of the boundary). Thus, once again, the underlying nasal-stop clusters are altered into something else, while surface nasal-stop clusters at boundaries are created secondarily by Hardening P-18. Thus //-dan=balhu-// → /-dan-n^gu=balhu-/ (P-1) 'to cut up guts' shows a would-be //nb// cluster broken up, while the same initial morpheme as independent noun occurs in such case forms as //ama-da:n-w₂aj//→/ama-da:n-baj/ 'around the guts' (with P-18).

Morpheme-internal exx.: /manba/ 'armband', /manba/ 'tree sp.', /an^ybaj/ 'different', /yimbiri/ 'mangrove jack fish', /wundhumug/ 'water lily sp.', /=ma:ndha-/ 'to make (good)', /wundul/ 'not fighting', /wundan^y/ 'young wallaby', /manjar/ 'leaves and branches', /n^ganji/ 'baler shell', /mun^yjuj/ 'green plum', /wungali/ 'exchange', /angi/ '(man's) Fa-in-law', /=w₂an^ygala-/ '(fish) to move around', and /n^ga:n^ggu/ 'frog'. Many of these clusters also occur on the surface at boundaries when a noun ending in nasal (see §2.7, above) is followed by a nominal suffix beginning in underlying paired continuant; at such boundaries we can also get surface /n^yd/ in /wu-dhan^gun^y-duj/ 'in the wind' (from //-ruj//). It is likely that compounds could be elicited with /n^y/ followed by surface /d/ (from //r//) and /dh/ (from //lh//), for example with a compound initial like /-lhan^gun^y-/ 'wind' or /-lhan^y-/ 'tongue' plus a verb like //=rabala-// 'to emerge' or //=lhan^garma-// 'to reach, catch up to', but no such compound happens to be attested (mainly for semantic reasons, no doubt). Underlying nasal-stop clusters which end up as something else are instances where n^gu-Epenthesis has applied (see above), with a compound initial like /-dan-/ 'guts', /-lhan^gun^y-/ 'wind', /-lan^g-/ 'top of head', or /-lhan-/ 'wound, scar' plus any verb beginning in (nonleniting) underlying stop.

liquid-stop. We have already seen that this is the only regular type of word- or stem-final /CC/ cluster (§2.7, above), and it is not surprising that they also occur medially. Initials are /1 1 r r/ (but not /lh/), and finals can be any stop, except that /d/ and /d/ in such clusters are indistinguishable phonetically (our convention is to write /d/ or /d/ on the basis of analogy to other forms of the same morphemes).

The most striking thing about liquid-stop clusters is that all more-or-less homorganic clusters are avoided except when they occur naturally at a boundary. This contrasts sharply with nasalstop clusters, where the homorganic type is very productive in all positions. Thus /ld ld ld ld/ are conspicuously rare within morphemes, while /rd/ and /rd/ are attested but uncommon in this position. On the other hand, the four relevant liquids are all attested both within morphemes and at boundaries with following /b/, /j/, and /g/; all are found at surface boundaries before /dh/ and only /r/ is unattested before /dh/ within a morpheme (perhaps because of an accidental gap or because of phonetic similarity to /ddh/; if we include recent loans we can cite /=w₂ardha-/ 'to work').

Because of n^gu-Epenthesis, underlying liquid-stop clusters at boundaries are normally broken up. Surface liquid-stop clusters at boundaries are basically due to combinations of morphemes ending in //Lg// (or sometimes //Lj//), where //L// is /1 1 r r/, plus a morpheme beginning in a paired continuant which is hardened to a stop by Hardening P-18 (the //g// or //j// would then be deleted by Stop-Deletion P-29 or Palatal-Deletion P-26). Note that a morpheme ending merely in a liquid will not induce Hardening P-18 in a following morpheme (a stop or nasal is required).

Exx. of morpheme-internal clusters are: /(n^g)albalan^g/ 'quickly', /malban^ggari/ 'flying fox sp.', /arbidi/ 'anyway', /marbidi/ 'ray sp.', /aldha/ 'racing', /=yaldha-/ 'to slip away', /=lhagardha-/ 'to form rows', borrowing /=w₂ardha-/ 'to work' (from English), /jurguldumbul/ ’female bandicoot' (uncommon stem, only ex. of /ld/ or /ld/, probably historically a compound), /=n^garda-/ 'to be pried loose', /warda/ 'yam sp.' (/rd/ distinct from simple /d/ only by greater emphasis on r-coloured onglide to stop), /n^guljun/ 'young sleepy cod', /alja/ 'python sp.', /n^gurji/ 'loose possessions', /warjar/ 'catfish sp.', /wulguy/ 'admittedly', /walgara/ 'herring sp.', /argi/ '(some) other', and /=mirga-/ 'to do away with'. Exx. of surface clusters at boundaries can be obtained by adding nominal suffixes to nouns ending in /lg/, /lg/, /rg/, or /rg/ (§2.7, above). Exx. of underlying liquid-stop clusters which are broken up by n^gu-Epenthesis P-1 are cases of any verb root beginning in a stop preceded by a compound initial like /-jir-/ 'hair, fur', /-n^gal-/ slime', /-ambal-/ 'top of head', or /-w₁alar-/ 'stem', but rather few such liquid-stop combinations can be cited from dictionary entries.

By far the most common underlying stop-nasal cluster is the combination of a noun ending in a stop with following Instrumental case suffix /-miri/, as in /yimbid-miri/ 'by means of cypress (wood)', optionally becoming /yimbin-miri/ by P-22. Other possible stop-nasal combinations might occur if a compound initial or derivational prefix were followed by a root beginning with a nasal (n^gu-Epenthesis applies only before stops and so is inoperative here). Some such compound initials are /-ruluj-/ 'shadow', /-lhud-/ 'power(ful)', and /-ran^gag-/ 'wood, tree', but we can cite only a handful of resulting stop-nasal clusters (often nasalised): //-ruluj=na-// → /-rulun^y=na-/ 'to see shadow', //-lhud=ngawi-'-// → /-lhun=n^gawi-' -/ 'to die after struggle'. When //g// is the initial segment it is deleted without trace by P-29 (or possibly by P-30 if Nasalisation P-22 has been allowed to apply).

stop-liquid. When a paired continuant, including /lh r r/, is preceded by a stop (or nasal), Hardening P-18 always converts the continuant into the paired (homorganic) stop. Thus Locative /-ruj/ becomes /-duj/ after any stop or nasal, and verb roots beginning with /lh r r/ shift this segment to /dh d d/ when preceded by any morpheme ending in stop (or nasal). Exx. are //a-ran^gag-ruj// → //a-ran^gag-duj//(P-18) → a-ran^ga-duj/(P-29) 'in the tree', and //-yan^g=lhurba-// → /-yan=dhurba-/(P-18 and -27) ’to obey'. Thus this type of stop-liquid cluster is reasonably common in underlying forms but does not make it to the surface even at boundaries.

This leaves us with the possible combination of a stop with following /1 1/, the liquids which do not become stops by P-18. Such clusters would probably be phonetically awkward in this language, since /1 1/ are virtually unattested as second elements in clusters; also, we have seen that homorganic liquid-stop clusters like /ld/ tend to be avoided, and it is reasonable to think that clusters like /dl/ with a more marked order of the same elements would be resisted even more so. In any event, we cannot cite a single ex. of stop plus /1 1/ within a morpheme, and it is difficult to concoct an ex. involving a boundary. One that does occur as underlying form is /gl/, but /g/ is not of much interest since it is always deleted before another consonant by Stop-Deletion P-29: //-yinag=lala:da-// → /-yina=lala:da-/ 'to have skin of head removed'. We have one ex. of underlying //dl// where the //d// is omitted from the surface: //-lhud=lalagi-'-// → /-lhu=lalagi-'-/ 'to get up powerfully'. I suspect that there is a variant pronunciation retaining the /d/ on the surface. Before setting up specific phonological rules for such clusters we would need more data involving several clusters of stop plus /1 1/, but it is very difficult to elicit relevant forms, which would have to be exx. of the relatively few verbs starting with /1/ (the situation with /1/ is hopeless since no suitable verb is available) preceded by appropriate compound initials ending in /d d j/. I doubt that a representative range of forms even for /1/ are in actual usage, and I suspect informant hesitation and fluctuation in producing elicited combinations.

stop-semivowel. All underlying combinations of this type are automatically altered by Hardening P-18, with //w₁//→/g/, //w₂//→/b/, and //y//→/j/. Thus all underlying clusters of this kind will show up as something else, usually a stop-stop cluster (or a simple stop if the first stop is then deleted by Geminate- Contraction P-31 or some other rule). There are, of course, no surface stop-semivowel clusters within morphemes, either.

nasal-nasal. A fair number of nasal-nasal clusters occur both within morphemes and at boundaries. However, //n^g// is deleted by P-30 before any other nasal, /nh/ does not occur in any clusters, /n^y/ is deleted in some environments before /n/ by Palatal-Deletion P-26, and geminates are reduced by P-31. It is also difficult to elicit actual exx. of some nasal-nasal clusters at boundaries because only /m/ is readily available as a productive suffix-initial nasal, and because /n n n^y/ occur as initial segment in relatively few verbs which can take compound initials.

Attested morpheme-internal clusters are seen in these exx.: /=anma-/ 'to go gathering', /munma/ 'tree sp.’, /man^ymaji/ 'deserted (place)', /=bunn^yurn^yurwa-/ 'to sizzle' (variant form), /munn^gu/ 'shrub sp.', /mann^gulg/ 'sacred (food)', and /min^yn^gu/ 'night'. For some speakers a stem /lha:n^gma/ 'young python' is recorded, the only ex. in the entire corpus of /n^g/ followed by another nasal; other speakers pronounce this /lha:ma(g)/.

At boundaries, the most common exx. involve Instrumental /-miri/ added to a noun like /raman/ 'white emu down', /marin/ 'basket', or /ma:rn^y// 'snake' (but stems ending with /n^g/ drop this segment by P-30 before /m/). It should be noted that underlying stops also optionally nasalise (by P-22) before a suffix or other morpheme beginning with a nasal, hence //yimbid-miri// → /yimbin-miri/ 'by means of cypress (wood)'; this generates additional exx. of nasal-nasal surface clusters at boundaries. It is difficult to elicit exx. of nasal plus /n^y/, since the latter is rare as initial segment of verb roots, but with /n/ instead of /n^y/ as second element we can cite /-malan^y=na-/ 'to see fin' (the /n^y/ seems stable as first segment in this ex., presumably because of the preceding /a/-vowel; some other cases of underlying //n^yn// are reduced to /n/ by P-26).

nasal-liquid. Liquids which have phonologically related stops, namely /lh r r/, become stops /dh d d/ after another nasal by Hardening P-18. For example, Locative suffix /-ruj/ is always /-duj/ after a noun ending in nasal, so no surface nasal-liquid cluster results. The same is true of compounds involving roots beginning in /lh r r/ when preceded by a nasal, except that the very few verb roots beginning in tap /r/ are unaffected by preceding //n// at the end of a pronominal prefix (the //n// then gets deleted without a trace by P-30). Ex.: //-yan^g=lharma-//→/-yan=dharma-/(P-18 and P-27) 'to follow the voice of'.

The only remaining possibilities for surface nasal-liquid clusters are, therefore, cases of nasal plus /l l/. However, these clusters are also unacceptable, and the nasal is deleted by P-30. Since /1/ does not occur in initial position in a relevant verb, our exx. involve root-initial /1/: //-man=lu:lha-// → /-ma=lu:lha-/ '(group) to wade'; //-maran^g=lalagi-' -// → /-mara=lalagi-' -/ 'to raise one's hand'. Pronominal prefixes ending in //n// likewise drop the nasal without trace before roots beginning with /1/ and (unhardened) /r/, as in //n^gan=lalagi-'-n^y// → /n^ga=lalagi-'-n^y/ 'I will get up'.

There are, consequently, no surface nasal-liquid clusters in the data, though perhaps one of the compounds just listed is occasionally pronounced with the nasal retained.

nasal-semivowel. This cluster-type is common in underlying representations at boundaries, but is always altered by Hardening P-18, which converts the semivowel into a stop in the usual way. Ex.: //a-mu:n-yin^yun^g// → /a-mu:n-jin^yun^g/ 'of the foot'. There are no surface nasal-semivowel clusters. (Laminoalveolar /n^y/, like similar consonants in languages like Spanish, is structurally a unit phoneme, not a cluster.)

liquid-nasal. Combinations of /1 1 r r/ plus following nasal are reasonably plentiful. As usual, interdental /lh/ does not occur as initial segment of a cluster. The four non-interdental liquids are attested before most of the nasals, but except for an odd case of /rn/ we do not find /n/ or /n/ after a liquid within a morpheme (when such clusters arise at a boundary, however, they are retained on the surface).

Within morphemes we cite these exx.: /(w)almaraba/ 'crab sp.', /rabara-wulma/ 'whip snake', /n^garma-j/ 'laughter', /warmurn^g/ 'wild cucumber', /=n^gurni-/ 'to look back’, /biln^ylga/ 'unsteady' (variant form), /wumbuln^yin/ 'fishtrap', /arn^yal/ 'grass sp.', /garn^yirimba/ 'many, lot', /raln^gar(g)/ 'cockatoo sp.', /maln^gar/ 'rays', /=dhabilirn^gi-/ 'to slip', and /n^garn^gar/ 'roly-poly'.

At boundaries, we note that Instrumental /-miri/ can follow nouns ending in /1 1 r r/, Among the other liquid-nasal clusters which can occur at boundaries we may mention that in /-lhal=na-/ to see country', since /1n/ does not occur within morphemes. We should also note that surface liquid-nasal clusters may arise when a morpheme ending in /Lg/ or /Ln^g/ (where /L/ is /1 1 r r/) is followed by a morpheme beginning with a nasal. Thus /-lhal=na-/, just cited (above), could also (potentially) mean 'to see flame' from //-lhaln^g-// 'flame' (instead of //-lhal-// 'country'), since the //n^g// disappears by Nasal-Deletion P-30; another ex. is //lhagabunulg-miri// → /lhagabunul-miri/ 'by means of dirt'.

liquid-semivowel. Liquids /1 1 r r/ (but not /lh/) can occur with following /w/ or /y/, which do not undergo Hardening P-18 in such clusters. It should be noted, though, that at boundaries a //w₁/ (but not //w₂//) is often deleted by w₁-Deletion P-9 in some environments.

Attested clusters within morphemes are illustrated by /milwin/ 'gnat', /=du:lwi-/ 'to sing (totem)', /arwar/ 'on top', /marwa/ 'edible root sp.', /=galyaba-/ 'to sit astride', /walya/ 'male', /marya/ vegetable food', and /=w₁urya-/ 'to pour on'.

At boundaries, we can add Similative /-yi:/ §4.29 to a noun ending in a liquid, e.g., /a-lha:l-yi:/ 'like a country'. Suffixes bginning with /w/, such as /-w₂ugij/ 'still, only', often show (optional) deletion of the semivowel by P-28 or P-9, but Dual /-w₂a:/ retains its semivowel, hence /a-lha:l-wa:/ 'two countries', and so forth. There are also a number of relevant compounds such as /-al=yarba-/ '(floater) to run along', /-ar=yi-/ '(water) to lie flat’, and /-n^gal=wagiwa-/ 'to break off' (also attested once with semivowel dropped).

liquid-liquid. Clusters with /l r r/ plus /lh/ are the most important ones; unattested */llh/ may be an accidental gap. /lr/ is also acceptable. Most other liquid-liquid clusters are difficult to elicit, e.g., because /1/ does not occur as initial segment of verb roots or suffixes. Underlying clusters of /1 1 r r/ plus /r/ occur in cases where Locative case suffix /-ruj/ follows a noun ending in a liquid, but here the //r// is deleted by r-Deletion P-25.

Within morphemes we may cite /=n^gallha-/ 'to be stuck in' (also /-lhalamillha-/ '(moon) to rise', /=w₁allha-/ 'to float', where /llh/ differs from simple /lh/ by duration), /=w₂urlha-/ 'to sink, drown’, /arlhag/ 'eucalypt sp.', and /=lhulrungi-/ 'to be harmless'.

These same clusters also occur at boundaries, though //llh// in this position is sometimes reduced to /lh/, to judge by /-anbu=lhaayi-/ '(branch) to extend out' (with //-anbul-//), if correctly transcribed. To the clusters above we may add /rl/ in /yamar-li:n^gu/ 'along side'.

There is one apparent ex. where //r// is deleted after /1/ at a boundary: /=ran^gara-/ 'to wait for', cpd. /-lhal=an^gara-/ 'to take care of (country)', rdp. /-lhal-an^ga=ran^gara-/. However, this may involve contamination, namely, the influence of /-an^ga-/ 'camp'. Surface /lr/ does occur at boundaries in other cases: /-n^{gul=raya-/ 'to urinate', /lhal-rarndar/ 'arid land'.}

semivowel-stop. /w/ is virtually unattested as first member of a cluster, or indeed in any syllable-final position. There are no cases of /w/ plus stop within a morpheme. The only possible way to get such a cluster at a boundary is by adding a nominal suffix to /garbaw(g)/ 'owl sp.' in those idiolects where the final /g/ is present: //a-garbawg-w₂aj//→/a-garbaw-baj/ 'around the owls', etc., where the initial segment of the suffix undergoes Hardening P-18.

There are a couple of fauna terms with /y/ followed by /b g/ within the stem: /waraybaray/ 'lorikeet' (frozen rdp.), /waygawuma/ duck sp.', and /la:ygulhu/ 'catfish sp.' At boundaries, we might get underlying /y/ plus step with compound initial /-ay-/ 'nest' and verb root beginning with stop, but n^gu-Epenthesis P-1 would apply here, inserting /-n^gu-/. There are no underlying or surface /y/ plus stop clusters at stem-suffix boundaries.

semivowel-nasal. The infrequency of semivowels as initial segments in clusters is also the case before nasals. Within a morpheme we can only cite the infrequent term /gunduyn^guru/ 'young whale'.

At boundaries, we can add Instrumental /-miri/ to a noun ending in /y w/, hence /dhuday-miri/ 'by means of water marker', /garbaw-miri/ 'by means of owl sp.' (the forms with /w/ would all be semantically anomalous). Aside from this, we can imagine the odd compound involving initial /-ay-/ 'nest' and a following verb root beginning with a nasal, but no actual ex. can be cited from the corpus (it is possible that Palatal-Deletion P-26 might apply sporadically before coronal nasals).

semivowel-liquid. No exx. of this type of cluster are attested within a morpheme.

At boundaries, underlying semivowel-liquid clusters could arise when a noun ending in a semivowel has Locative suffix /-ruj/, as in /a-gandhaw-ruj/ 'at the bittern sp.'. However, the combination of /y/ plus /r/ at a boundary is subject to yr-Contraction P-14, which commonly deletes the /y/. Other clusters involving /y/ plus a liquid could arise if compound initial /-ay-/ were to precede a verb root beginning in a liquid /l 1h r/, but again deletion may occur, as in the attested form //ay-lhamungur//→/a-lhamungur/ (by P-26 this time); actually this particular ex. involves an adjectival noun as second element.

semivowel-semivowel. Within morphemes, /yw/ but not */wy/ is recorded: the attested exx. are /w₂ayway/ 'flat', /malaywiwi/ 'egret', /waraywaray/ 'lorikeet', and /=w₂uywulha-/ '(women) to perform dance'.

At boundaries, both /yw/ and /wy/ can occasionally be found. There are a fair number of nouns ending in /y/, and we can add a suffix like Pergressive /-w₂aj/ to produce forms like /a-madugay-waj/ 'among the shark sp.'; note also the compound /-ay=w₂ann^ga-/ 'to step on (nest)'. The only way to get /wy/ is to add a suffix like Similative /-yi:/ to one of the two nouns which end in /w/, as in /a-garbaw-yi:/ 'like an owl sp.'.