Form-based form (3)

id Language Grammatical gender Nominal expletive elements Syntactic functions Morphosyntactic complexity influences word order Predominantly head-marking Morphophonologically conditioned stem alternation Morphologically and/or morphophonologically conditioned affix alternation Phonologically conditioned stem alternation Phonologically conditioned affix alternation
1 Bantawa Bantawa does not display grammatical gender. There is nominal classification (Doornenbal 2009: 66ff.), but this involves a defective, semantically conditioned classification that is purely marked by means of suffixes on specific nouns, while it has no repercussions for agreement. There are no nominal expletives in Bantawa, as proven by example (10).

Doornenbal (2009: 469)
(10) wa ta
water comes
‘It is raining.’
In transitive clauses in Bantawa, an Actor receives ergative case and an Undergoer is marked for absolutive case. Undergoers are almost always cross-referenced on the predicate. Actors can be marked on the predicate as well, as in (11), but this is infrequent.

Doornenbal (2009: 122)
(11) naŋsi-ʔa ɨ-catt-a-ŋ
hailstone-ERG 3.A-hit-PST-1SG
‘The hailstone hit me.’

In intransitive clauses, independent Actor arguments, if expressed, get absolutive case marking (Doornenbal 2009: 213) and, as example (12) shows, so do Undergoers. Furthermore, examples (12) and (13) show that it is optional to cross-reference an Actor or an Undergoer on the predicate in intransitives.

Doornenbal (2009: 213)
(12) samba ker-a
bamboo.ABS break-PST
‘The bamboo snapped.’

Doornenbal (2009: 122)
(13) mɨkmɨkmɨkwa khaw-a
profusely cry-PST
‘He cried profusely.’

In sum, semantic role is expressed in transitive clauses by means of case-marking, and optionally by means of predicate markers. In intransitive clauses, semantic role is neutralised, since both Actors and Undergoers receive absolutive case. Predicate marking is also not informative on semantic roles of arguments. Therefore, a syntactic function Subject must be postulated.
Furthermore, there is an antipassive construction in Bantawa (Doornenbal 2009: 225ff.), so that there is a second reason to assume a syntactic function Subject. An example is given in (14).

Doornenbal (2009: 226)
(14) a. nam-ʔa mɨ-hɨt-yaŋ
sun-ERG 3PL-scorch-PROG
‘The sun is scorching us.’
b. kho-ci-ʔa kha mɨ-hɨt
3-PL-ERG ANTIP 3PL-scorch
‘They burn.’
Word order is quite consistently based on pragmatic considerations (Doornenbal 2009: 13). I have not found evidence for an influence of complexity on morphosyntactic placement. M. Doornenbal (personal communication, September 18, 2013) also suspects that it does not play a role. Bantawa is an agglutinative language, as it has extensive affixation (Doornenbal 2009: 12), but few clitics and particles (Doornenbal 2009: 306ff.). Therefore, I consider it a predominantly head-marking language. All Bantawa verb stems have two allomorphs. The first phoneme of the attached pronominal suffix determines allomorph selection: if this is a consonant, the so-called pre-consonantal stem form is used, if the suffix starts with a vowel, the pre-vocalic stem allomorph is selected (Doornenbal 2009: 126ff.). This stem alternation process is phonologically determined and therefore discussed in Section 2.4.8, but the form that the different stem forms take is morphophonologically determined and should therefore be addressed here.
Verbs can be divided over three classes on the basis of the way that the stem allomorphs are formed (cf. Doornenbal 2009: 129ff.). In class 1, the pre-vocalic stem can be derived from the pre-consonantal stem by adding a -t, e.g. hek-ma ‘cut-INF’ vs. hekt-u ‘cut-3SG’. In class 2, the pre-vocalic stem has an added -s, e.g. nu-ma ‘heal-INF’ vs. nus-u ‘heal-3SG’. Class 3 stems do not take an extra consonant; rather, if a consonant-initial suffix is attached to a pre-consonantal stem ends in a consonant, that consonant is either changed or deleted. If a class 3 pre-consonantal stem ends in a vowel, this vowel is retained in the pre-vocalic stem and will coalesce with the suffix-initial vowel. Since it is not possible to predict the precise alternation that occurs on the basis of the phonological shape of the verb stem, this is non-transparent, morphophonologically based alternation (Doornenbal 2009: 131).
In some third person pronominals, the pronominal marker -sa~-so has to be inserted, e.g. o-ci ‘this-PL’, o-sa-ʔa ‘this-PRN-ERG’. According to Doornenbal’s informants, this marker has no meaning, but is obligatory in some contexts since it ‘sounds better’ – apparently, there is a prosodic motivation (Doornenbal 2009: 101). Furthermore, this syllable undergoes vowel assimilation, e.g. o-sa-ʔa ‘this-PRN-ERG’, o-so-ʔo ‘this-PRN-GEN’ (Doornenbal 2009: 54). Another case of morphophonologically determined allomorphy is a Bantawa demonstrative that shows vowel assimilation, e.g. mu-ju ‘that-LOC.low’, mo-ja ‘that-LOC.level’ (Doornenbal 2009: 54). There are quite some phonologically conditioned stem alternation processes in Bantawa. I do not have space to go into all of them, so I will discuss the most important ones only.
Firstly, a syllable in Bantawa must always have an onset. If a vowel-initial suffix attaches to a vowel-final stem, or a vowel-final prefix attaches to a vowel-initial stem, a glottal stop or glide is inserted to fill the onset, e.g. ɨ-uk-n-ɨŋ ‘NEG.NPST-peel-NEG-1SG, I did not peel it’, is pronounced [ɨʔuknɨŋ] (Doornenbal 2009: 29).
Furthermore, as explained, all Bantawa verb stems have two allomorphs: a pre-consonantal and a pre-vocalic form. The choice for one of these depends on the first phoneme of the pronominal suffix selected, e.g. hek-ma ‘cut-INF’ vs. hekt-u ‘cut-3SG’, and nu-ma ‘heal-INF’ vs. nus-u ‘heal-3SG’ (Doornenbal 2009: 127ff.). Hence, the phonological shape of the suffix determines the phonological shape of the stem.
A nasal turns into a nasalised vowel before an /s/ or a /j/, e.g. /wensi/ ‘raspberry’ is pronounced [weĩsi] (Doornenbal 2009: 47).
Geminates arising at morpheme boundaries are realised as long consonants, e.g. *hen ‘stay’, *kham ‘place’, *ma ‘big’ becomes [hɛnkham:a] ‘world’ (Doornenbal 2009: 40). If the first, stem-final consonant is voiceless, it is voiced, resulting in a voiced geminate: bop ‘round’, bojoŋ ‘*termite’, -ma ‘F’ is pronounced [bobbojoŋma] ‘termite’ (Doornenbal 2009: 41).
Phonologically based assimilation processes described above apply to stems and affixes alike.
2 Bininj Gun-Wok There is no grammatical gender in Bininj Gun-Wok. There are nominal classes, but these are semantically motivated, as proven by example (9): a modifying unit is nominalised by a noun class prefix, which contributes its semantic content to the nominalised element (Evans 2003: 181ff.).

Evans (2003: 186)
(9) na-gohbanj al-gohbanj an-gohbanj
I-old II-old III-old
‘old man’ ‘old woman’ ‘old tree’
Bininj Gun-Wok does not have pronominal expletives. Weather predicates, as well as other prototypical expletive constructions such as existentials, are all like example (10): an argument is incorporated in the verb and the verb as a whole is marked for third person.

Evans (2003: 368)
(10) Ø-dung-mirrhmirrhme-ng
3.PST-sun-get_sharp-PST.PFV
‘It’s hot.’

The third person prefix refers to the semantic argument ‘sun’ in (10). This means that there is cross-reference between the incorporated noun and the verbal prefix, which is normal in Bininj Gun-Wok, as explained in Section 3.1.1. Thus, the verbal marker is not a fully expletive element, but refers to an overt element.
The main marking of argument relations in Bininj Gun-Wok takes place on the verb. Each verb has slots for pronominal prefixes, in which two arguments can be expressed, and a slot in which a noun can be incorporated (Evans 2003: 318). A transitive predicate, e.g. (11), takes a portmanteau morpheme as pronominal prefix, that marks Actor and Undergoer and their hierarchical relationship.

Evans (2003: 417)
(11) nga-na-ng boken kornobolo
1>3-see-PST.PFV two wallaby
‘I saw two wallabies.’

An intransitive predicate is marked by a pronominal prefix, whether the S argument is an Actor, as in (12), or an Undergoer, like (13). The arguments in these examples have different semantic roles, but are marked identically – therefore we can speak of a syntactic function Subject.

Evans (2003: 393)
(12) gabarri-lobme
3PL-run.NPST
‘They are running.’

Evans (2003: 446)
(13) njale Ø-bakme-ng?
what 3.PST-break-PST.PFV
‘What was broken?’

Bininj Gun-Wok does not have a passive construction (Evans 2003: 574).
Word order in Bininj Gun-Wok is predominantly pragmatically determined. For instance, new participants occur in preverbal position, while established participants are post-verbal (Evans 2003: 551ff.). I have not found any examples of an influence of complexity on morphosyntactic placement. N. Evans (personal communication, September 20, 2012) adds that complexity in Bininj Gun-Wok is in fact limited to verbal elements; no unit would really qualify as a heavy NP, or even as a complex noun. Therefore, I conclude that complexity does not play a role in morphosyntactic placement. Evans (2003: 258) distinguishes a few clitics in Bininj Gun-Wok, e.g. =wali ‘in turn’. The overwhelming majority of function markers, however, consists of head-marking units: all verbal morphology is affixing. Hence, Bininj Gun-Wok is a predominantly head-marking language. Several morphophonemic processes exist in Bininj Gun-Wok. For example, when a vowel is followed by a consonant, the vowel is usually short. However, vowels in a small set of nominal roots with the form CV- are pronounced long when used before a third person possessor suffix -no or any other root starting with a consonant, e.g. ga-bo:-yo: ‘3-liquid-lie ‘there is water’ (Evans 2003: 74). There is no declension in Bininj Gun-Wok, but there are conjugational classes: Evans (2003: 345ff.) distinguishes 11 verb classes on the basis of distinct TAM allomorphy. Verbs in Bininj Gun-Wok are either monosyllabic or bisyllabic. The second syllable of a bisyllabic stems determines which TAM inflection the verb gets, e.g. stems ending in -me (e.g. karrme ‘have’, karrme-n ‘have-IPFV’) belong to the first conjugation, and verbs ending in -ke and -we belong to the second (e.g. dowe ‘die’, dowemen ‘die-IPFV’; Evans 2003: 359ff.). Since TAM inflection alternations are conditioned by these stem-final syllables, and are not phonologically or semantically predictable, I group this under morphologically conditioned affix alternation.
Furthermore, in some Bininj dialects, the plural object marker -bani is nasalised after a nasal, e.g. abanbaninang ~ abanmaninang (Evans 2003: 111).
Morpheme-initial /d/ becomes /r/ (spelled ‘rr’) after vowel-final monosyllabic prefixes and after open polysyllables with final stress, e.g. gun-dulk ‘IV-tree’, gau-rrulk-di ‘3-tree-stand, there is a tree there’ (Evans 2003: 106, 107). The altered morpheme can be a nominal or verbal root, but also a prefix.
Morpheme-initial /r/ may be dropped after apical consonants, e.g. a-bal-re ‘3PL.PST-only-go.PST.IPFV’ is /abale/ (Evans 2003: 110).
Two successive syllables cannot both contain a glottal stop, so that one of them is deleted or assimilated to a surrounding consonant, e.g. the reduplicated form of buʔme is not buʔ-buʔme but bu-buʔme (Evans 2003: 112).
Stems of the form CVrrVng undergo vowel drop when another stem or a CVC-suffix is attached to the right, e.g. gun-murrung ‘IV-bone’, murrng-wern ‘bone-many’ (Evans 2003: 114ff.).
The phoneme sequence /iyi/ may be pronounced as [i:], e.g. /gabarriyigan/
[gabari:gan] ‘they go for it’, and two short vowels may become one long one when a /g/ in between is lost, e.g. /nagamarrang/ [na:maraŋ] ‘male skin name’ (Evans 2003: 75). In some dialects, /w/ is dropped after a nasal, e.g. kan-wo ‘give it to me’ > /kano/ (Evans 2003: 110).
There is a tendency for consonants to be voiced in onset position, but voiceless in coda position, e.g. kuk ‘body’ can be pronounced [kɷk] and [gɷk] but not [kɷg] (Evans 2003: 79).
3 Chukchi Chukchi has no grammatical gender, but a semantically based nominal classification system only, as highly animate nouns show a different inflectional behaviour from common nouns (Dunn 1999: 64). There are no expletives in Chukchi, as becomes clear from their absence with the weather predicates in (5) and (6). Chukchi has zero-place predication without dummy insertion.

Dunn (1999: 335)
(5) … loŋ-kətəjɣat-ə-lʔ-ə-n …
NEG-wind_blow-ə-NMLZ-ə-ABS
‘There was no wind…’

Dunn (1999: 340)
(6) … etʔəm anə r-ile-rʔu-ɣʔi
probably so FUT-rain-INCH-THEMSUF
‘… probably it will rain.’
There is reason to assume a syntactic function Subject in Chukchi, as the semantic role of the single argument of an intransitive predicate is neutralised. The Actor argument in (7) and the Undergoer argument in (8) are expressed identically, while example (9) shows that in a transitive context the semantic role distinction is expressed by means of case marking.

Dunn (1999: 199)
(7) ŋeekək wetɣak-wʔe
sister.ABS.3SG speak-3SG
‘The sisterA spoke.’

Dunn (1999: 202)
(8) […] n-ə-mk-ə-qin qejuu-t nʔ-ə-jaɣtal-ə-ŋo-nat […]
ADJ-ə-many-ə-3SG calf-ABS.3PL COND-ə-be_preserved-ə-INCH-3PL
‘… many calvesU would be preserved….’

Dunn (1999: 81)
(9) qənwer jara-lɣ-ə-jŋ-ə-n ŋew-ʔəttʔ-ə-qe-e
finally house-SG-ə-AUG-ə-ABS.3SG woman-dog-ə-DIM-ERG
tejk-ə-nin
make-ə-3SG>3SG
‘Finally, the bitchA made a big houseU.’

Additionally, Chukchi exhibits an antipassive construction that neutralises semantic roles (Dunn 1999: 216). Clearly, semantic roles can be neutralised in Chukchi, which therefore has a syntactic function subject.
Constituent order in Chukchi is not fixed (Dunn 1999: 80). Dunn (1999: 344) classifies Chukchi as a pragmatic word order language, that is, pragmatic roles determine constituent ordering. Since constituent order is free to such a high extent – all possible orderings occur (Dunn 1999: 81) – it is difficult to establish whether constituent weight really bears an influence. At least, we can say that there is no proof that there is any influence on word order, and therefore, I will assume there is none. Chukchi is a typical agglutinative language. It possesses a large number of affixes, while Dunn (1999: 76) distinguishes one clitic only, showing that Chukchi is a predominantly head-marking language. Chukchi has, apart from an epenthetic schwa, three underlying vowels, each of which has a default realisation and a non-default realisation that only appears in the presence of a so-called vowel harmony prosody feature. This feature is a covert property of particular morphemes, e.g. the singulative suffix –n+VH, deriving place nouns from action verbs. When such a morpheme combines with a stem in a Phonological Word, that stem’s vowel receives its non-default realisation, e.g. the default stem /təɬe/ ‘path’ becomes /təɬa-n/, whereas this is not the case with absolutive case suffix
–n-VH (Dunn 1999: 48). This process is not phonologically based, since it is not a specific phonological context that triggers the selection of a certain vowel quality, but the specific morpheme. Note that the term ‘vowel harmony’ is a bit odd, since the VH feature is abstract and does not necessarily belong to a vowel – there is no assimilation of the stem’s vowel to another vowel, but to an abstract feature.
Dunn (1999: 51ff.) describes several cases of morphophonologically determined allomorphy. One example is that around two dozen verb stems have an alternating onset, where a consonant is dropped in one of the allomorphs, e.g. wa ~ twa ‘be’. A dozen other verbal stems undergo more extensive deletion, e.g. rətc ~ tc ‘AUX’. The full form is used when the morpheme is word-initial, otherwise the shorter form is used. The deletions are phonologically predictable, but it is unpredictable to which verbs the deletions apply (Dunn 1999: 52).
As explained above, absolutive singular can be marked by means of five different strategies. Dunn (1999: 105) states that the choice for a specific strategy is partly determined by phonology, partly by the morphological origin of the noun stem and partly lexically conditioned. This means that Chukchi has five declension classes: absolutive singular inflection is determined morphologically. I found no other instances of morphophonological affix alternation in Chukchi. Word-final vowels are almost always reduced or deleted (Dunn 1999: 53). Furthermore, as noted above, one strategy to mark a stem for absolutive singulative is by means of a phonologically determined reduction or deletion of the stem-final vowel, e.g. wala-t ‘knife.ABS-PL’, walə ‘knife.ABS.SG’ (Dunn 1999: 106).
Chukchi syllables follow a quite strict (C)V(C) pattern. A coda consonant can be adjacent to a following onset consonant, but it is not possible to have more than one consonant within the onset or within the coda. If this occurs, an epenthetic schwa is added in between (Dunn 1999: 39ff.). For example, when the infinitive suffix -k is attached to a verb stem ending in a consonant, for instance jəlqet ‘sleep’, the resulting sequence jəlqetk is forbidden: addition of a schwa gives the correct output jəlqetək (Dunn 1999: 73).
A phonological alternation rule exists that changes two vowels into one long vowel when an approximant intervenes, e.g.
/ʔoɾacek/ ~/ʔaacek/ ‘youth, lad’ (Dunn 1999: 42). Furthermore, anterior stops undergo nasalisation when adjacent to a nasal (Dunn 1999: 44). The voiceless velar stop becomes an approximant before other consonants and becomes a uvular if adjacent to another uvular. There are many more such relatively common assimilation processes, which I will not describe here for reasons of time and space (cf. Dunn 1999: 44ff. and 52ff.).
Word-final vowels are almost always reduced or deleted (Dunn 1999: 53). Since words may end in a suffix, this is an alternation that may affect both stems and affixes. The other alternation processes described in Section 4.4.8 apply to affixes too.
4 Dutch Dutch nouns are divided over two genders: common and neuter. Even though gender assignment can be shown to be motivated increasingly semantically (cf. Audring 2009a), the basis of the system is still grammatical, as shown in example (6) where a semantically feminine noun has neuter gender agreement.

(6) het meisje, dat …
DEF.SG.N girl(N) REL.SG.N
‘the girl, that…
Dutch makes use of a dummy pronoun with weather predicates, as in example (7).

(7) het regen-t
3SG.N rain-3SG
‘It is raining.’
Dutch shows neutralisation of semantic roles. In transitive clauses, pronominal arguments are marked for nominative and accusative case, as in (8). Such case-marking disappears in intransitive clauses, e.g. (9)

(8) zij zag haar
she.NOM see.PST she.ACC
‘He saw her.’
(9) a. zij reed
she.NOM drive.PST.SG
‘She drove.’
b. zij viel
she.NOM fall.PST.SG
‘She fell.’

Furthermore, Dutch has a passive construction, e.g. (10).

(10) a. zij begroet-te hem
NOM.3SG.F greet-PST.SG ACC.3SG.M
‘She greeted him.’
b. hij werd door haar begroet
NOM.3SG.M was by ACC.3SG.F greet.PST.PTCP
‘He was greeted by her.’
Weight of a constituent can influence morphosyntactic placement in Dutch, as in example (4), repeated here as (11).

(11) heb je die man gezien [die door rood reed]?
have you that man seen that through red drove
‘Did you see that man that drove through the red traffic light?’
Dutch has clitics, e.g. predicate markers, and particles, e.g. articles. However, head-marking affixes are much more frequent, rendering Dutch a predominantly head-marking language. Morphophonological variation occurs in Dutch as the result of the addition of the diminutive suffix -tje (cf. Snow et al. 1980). The stem-final consonant can be deleted or altered under the influence of -tje, e.g. kast ‘cupboard’ /kɑst/ becomes kast-je ‘cupboard-dim’ /kɑʃə/, where the stem-final consonant is deleted, thus obscuring the morpheme boundary. Dutch shows suppletive affix alternation in the plural affix, which alternates between -en and -s depending on morphophonological properties of the stem. Hence, nouns can be divided over two declension classes, based on their plural inflection.
Verbs can be divided over conjugation classes depending on the past tense inflection they take. The majority of verbs is inflected for past tense by means of a suffix, while so called strong verbs do so by means of vowel alternation of the stem (cf. Section 5.3.3). Hence, this affix alternation is morphologically motivated.
An example of morphophonological affix alternation is the diminutive suffix -tje, which can take the forms -etje, -pje, -kje, -je, -ʃe, or
-tje (cf. Snow et al. 1980). The alternations only occur with the diminutive suffix, so that selection of the appropriate allomorph cannot be explained as purely phonologically conditioned.
Dutch exhibits phoneme deletion in degemination, e.g. krabpaal ‘scratching post’ /krɑp/ + / pal/ becomes [krɑpal]. Furthermore, there is assimilation of place and manner of articulation in Dutch. For example, when tand ‘tooth’ /tɑnt/ and pasta ‘paste’ /pɑsta/ form a compound, the result is [tɑmpɑsta], where the stem-final /t/ is deleted and the nasal is labialised. There is assimilation of voicing in Dutch affixes, e.g. the past tense singular suffix on weak verbs is -te after a stem-final voiceless consonant, but -de after a voiced consonant or vowel.
5 Egyptian Arabic Egyptian nouns are either masculine or feminine (Gary & Gamal-Eldin 1981: 75), as reflected in agreement marking on modifiers and predicates (cf. Section 6.1.2). Femininity is marked on the noun by means of a suffix -a, masculinity is not overtly expressed on nouns. Humans and animals are assigned a gender on the basis of their biological sex, but otherwise there are no semantic assignment rules, so that this qualifies as a grammatical gender system. I have not found examples showing semantic agreement, so that I will assume that gender is grammatical. There are no nominal expletives in Egyptian Arabic (Gary & Gamal-Eldin 1981: 25). An example of a weather predicate is given in (7). The predicate ‘to rain’ has a semantic argument, i.e. ‘the world’. This argument can be implicit, in which case the verb will still be inflected for third person feminine, showing that ‘the world’ is actually invoked (M. Hegazy, personal communication, July 22, 2013). Therefore, we can say that there is no construction in Egyptian involving a semantically empty dummy subject.

M. Hegazy (personal communication, July 22, 2013)
(7) (ʔid-dunya) bi-t-matˤtˤar
DEF-world PRS-3SG.F-rain.PFV
‘It is raining.’ (Lit.: “The world rains.”)
In an Egyptian Arabic transitive clause, the person, gender and number of the Actor are marked on the predicate. In intransitive clauses, semantic role is neutralised, since both Actors, e.g. in (8), and Undergoers, e.g. in (9), are cross-referenced on the predicate.

Gary & Gamal-Eldin (1981: 60)
(8) s[ā]mi giri-ø
S. run.PST.PFV-3SG.M.A
‘Sami ran.’
(9) muna maat-it
M. die.PST.PFV-3SG.F.A
‘Mona died.’

There is a passive construction in Egyptian (Gary & Gamal-Eldin 1981: 89), in which it is the expression of an Actor argument in a by-phrase is uncommon, but in some cases possible, as illustrated by (10).

Gary & Gamal-Eldin (1981: 90)
(10) ʔiz-zamaalik ʔit-ʁalab-ø gon-een min il[-]ʔahli
DEF-Z. PASS-beat-3SG.M.A goal-DU from DEF-A.
‘The Zamalek (club) was beaten two goals by the Ahli (club).’
There is heavy shift in Egyptian Arabic, e.g. in example (11). Default word order would place the indirect object (‘to Sami’) after the direct object (‘the books that …’), but since the direct object is so heavy, this ordering is reversed.

Gary & Gamal-Eldin (1981: 54)
(11) ʕali warra (li) sāmi ʔi[k-]kutub ʔik-kitiir ʔilli gab-ø-ha
A. show (to) S. DEF-book.PL DEF-many that bring.PST.PFV-3SG.M-3SG.F
ʔil[-]ʔisbuuʕ ʔilli faat min maʕrad ʔik-kitaab
DEF -week REL pass.PST.PFV-3SG.M from fair DEF-book
‘Ali showed Sami the books that he brought last week from the bookfair.’
Egyptian Arabic has many affixes, i.e. prefixes, suffixes, circumfixes and transfixes (cf. Section 6.2.4), but only few particles and clitics (cf. Gary & Gamal-Eldin 1981: 59ff.,89ff.). Therefore, it qualifies as a predominantly head-marking language. As explained above, Egyptian lexical words consist of three consonantal roots or radicals that are inflected by means of a transfix. However, several verbs have one or more ‘weak radicals’, i.e. glides instead of consonants. The inflection of such verbs is different from that of verbs with three strong radicals, for example, if the second radical is weak, it may appear as a vowel, māt ‘to die’, or as a glide, mawwit ‘die.CAUS, to kill’ (Woidich & Heinen-Nasr 2000: 153). These are phonological changes to the stem, i.e. to the consonantal roots, that only apply to a specific set of verbs, and are therefore morphophonologically based. The feminine suffix is -it in possessive constructions, but -a in all other contexts, e.g. madrasa kibira ‘a big school’, madrasit ʔawlaad ‘boy’s school’ (Gary & Gamal-Eldin 1981: 76).
The final consonant of the passive prefix ʔit- and the determiner prefix ʔil- is assimilated to the initial consonant of the root it attaches to, but only if this initial consonant is a dental, palatal or velar, e.g. ʔid-dars
‘DEF-lesson’ (Gary & Gamal-Eldin 1981: 127).
Plurality is usually marked by means of suffixes, but on some nouns it is marked by means of an exceptional transfix (Gary & Gamal-Eldin 1981: 74), which is phonologically unpredictable, e.g. kitaab ‘book.SG’, kutub ‘book.PL’ (Gary & Gamal-Eldin 1981: 59). Since the vowel qualities are not phonologically predictable, this is counted as morphologically based affix alternation.
Several assimilation and dissimilation processes exist in Egyptian, of which I will name a few here. Firstly, a semi-vowel is introduced to avoid vowel clustering, e.g. li ‘for' +
-i ‘me’ = lijja (Gary & Gamal-Eldin 1981: 128). An epenthetic vowel /i/ is inserted between a word-final consonant and a word-initial consonant, e.g. kunt taajih is pronounced [kuntitaajih] (Gary & Gamal-Eldin 1981: 129).
A stem-final dental consonant is deleted when a pronominal suffix -t is attached, e.g. xad ‘take’ + -t ‘1SG’ becomes xat (Gary & Gamal-Eldin 1981: 129).
Long vowels are shortened when the syllable in which they occur is closed, i.e. when a suffix or part of a circumfix consisting of a consonant attaches that results in a consonant cluster, e.g. gireet ‘I ran’, ma-grit-ʃ ‘I didn’t run’ (Gary & Gamal-Eldin 1981: 128).
Alternations applying to stems apply to affixes alike. For example, the vowel of a suffix can be elided when attached to the vowel of a transfix, e.g. when yi- ‘3SG’ connects to āxud ‘take’, the result is yāxud (Woidich & Heinen-Nasr 2000: 89).
6 Fongbe Fongbe does not have grammatical gender (Lefebvre & Brousseau 2002: 37). In weather predicates, Fongbe does not employ expletives, cf. example (8).

Lefebvre & Brousseau (2002: 245)
(8) jì jà
rain fall
‘It is raining.’

Fongbe does exhibit expletives with other verb constructions, translating for instance as ‘it is good/convenient/possible’, ‘it remains’, ‘it seems’, ‘it is not’ (Lefebvre & Brousseau 2002: 67; E. Aboh, personal communication, July 9, 2014), as demonstrated above in example (6) and additionally in (9). The pronominal element é refers to the subordinate clause here and is therefore referential in a different sense than ‘regular’ pronouns.

Lefebvre & Brousseau (2002: 67)
(9) é nyɔ́ ɖɔ̀ kɔ́kú ní yì
it be_good COMP K. SBJV leave
‘It is good that Koku leaves.

This type of expletive elements is not taken into account in this study (cf. Section 4.4.2).
Fongbe has neutral alignment, i.e. in neutral sentences, no pragmatic, semantic or morphosyntactic roles are marked morphosyntactically. For instance, the Actor in example (10) and the Undergoer in (11) are unmarked for their role, nor is pragmatic or semantic information expressed on the predicate.

Lefebvre & Brousseau (2002: 241)
(10) kɔ̀kú lɔ́n
K. jump
‘Koku jumped.’

Lefebvre & Brousseau (2002: 242)
(11) kɔ̀kú kú
K. die
‘Koku died.’

In (12), the Actor and Undergoer arguments of the transitive predicate are unmarked as well. Hence, we cannot say that there is neutralisation of semantic role in intransitive clauses, but rather that no role is expressed at all.

Lefebvre & Brousseau (2002: 247)
(12) kɔ̀kú xò àsíbá
K. hit A.
Koku hits Asiba.

Lefebvre & Brousseau (2002: 274) state that Fongbe has a verbal passive construction, in which the grammatical object is promoted to subject position. However, it is not possible to express an Actor argument in a by-phrase, since Fongbe lacks a preposition ‘by’. Therefore, there cannot be a full neutralisation of semantic roles.
In sum, semantic roles of constituents appear to follow from word order only, as Actors precede Undergoers by default. If word order is different, this is for a pragmatic reason (cf. Section 7.4.4), for example because an element is focused, in which case a discourse marker is usually obligatory. This shows that in Fongbe, it is possible to distinguish semantic functions on the basis of word order, and pragmatic functions on the basis of word order and discourse markers, but there are no true markers of syntactic functions.
Word order in Fongbe is largely determined by pragmatic considerations (cf. for example Lefebvre & Brousseau 2002: 306ff., 335ff.). E. Aboh (personal communication, October 17, 2013) stresses that it is in fact impossible to separate influence of complexity from influence of pragmatic considerations, because deviations from default word order always have a pragmatic effect, so that it is not possible to determine whether the deviation was triggered by the complexity of the phrase or to achieve this pragmatic effect. However, what we can say is that there is no proof that complexity does have an influence. Since there are languages in which such an influence is clearly present, Fongbe can in comparison with such languages be classified as transparent with respect to this feature. Fongbe is an isolating language, as it hardly shows inflectional morphology (Lefebvre & Brousseau 2002: 183). Function marking is performed by means of morphosyntactically and phonologically independent morphemes. Therefore, Fongbe qualifies as a predominantly phrase-marking language. I have not found instances of morphophonologically based stem alternations in Fongbe. The diminutive suffix –i shows vowel assimilation: àčú-ví ‘rat-DIM’ is pronounced [àčúvú] (Lefebvre & Brousseau 2002: 25). There is tone spreading in Fongbe, meaning that the tone of a unit, whether lexical or grammatical, is influenced by the tone of preceding units (L&B 2002: 22). Lefebvre & Brousseau (idem) say that spreading is limited to the tonal domain, which is a morphosyntactically defined unit. Since I have no reason to believe that morphosyntactic units are not parallel to phonological units, I will see this as a phonological process.
Another assimilation process in Fongbe is affrication, a superficial phenomenon that only occurs in rapid speech (Lefebvre & Brousseau 2002: 25). It involves the affrication of t and d preceding i, sometimes including palatalisation, e.g. tì ‘squeeze’ is pronounced [čì].
Sonorants, as well as /b/ and /ɖ/, are nasalised in Fongbe when they precede nasal vowels, e.g. /bɔ̃̀/ > /mɔ̃̀/ ‘to see’ (L&B 2002: 27). This is an alternation between phones rather than between phonemes – it is a phonetic process.
There is tone spreading in Fongbe that affects affixes as well, e.g. àsá ‘thigh’ + mɛ̀ ‘inside’ > àsá-mɛ̂ (Lefebvre & Brousseau 2002: 22). I am not sure whether this is a tonemic or an allotonic alternation but in any case, it concerns a form-based form process.
7 Georgian There is no grammatical gender in Georgian (Vamling 1989: 16). Georgian does not make use of a pronominal dummy subject in weather predicates, as demonstrated in (9).

Hewitt (1995: 89)
(9) p’arask’ev-sa-c tov-d-a da k’vira-sa-c
friday-DAT-too snow-IPFV-AOR.3 and Sunday-DAT-too
‘It snowed on both Friday and Sunday.’
The marking of argument functions in Georgian is extremely complicated, since the way arguments are expressed depends on the conjugation class of the verb (cf. Section 8.4.7) and on the TAM specifications. Regarding the latter: Georgian verbs are marked for tense, aspect and mood by means of a combination of morphological markers (i.e. specific thematic suffixes and the use of person and number inflection from either a set A or a set B affixes), which can be grouped in three series. These different series require different markings of argument relations. For example in (10), the present tense entails a nominative Actor argument and a dative Undergoer, while in (11), the aorist perfective marking requires an ergative Actor and a nominative Undergoer.

Vamling (1989: 19)
(10) is xat’-av-s surat-s
NOM.3SG paint-THEMSUF-PRS.3 picture-DAT
‘He paints a picture.’
(11) man da-xat’-a surat-i
ERG.3SG PFV-paint-AOR.3 picture-NOM
‘He painted a picture.’

Apart from the inflections described so far, argument relations can additionally be marked by vocalic prefixes. For example, the prefix u- indicates that there are three semantic roles: an Actor, an Undergoer and a Benificiary (Vamling 1989: 18). Hewitt (1995: 170) calls these different argument scenarios ‘versions’.
The semantic roles of the different arguments that play a role in a clause are hence coded in many different ways. Independently expressed arguments get obligatory case-marking, reflecting their semantic function. Secondly, dependent on the TAM series, some arguments are marked on the predicate by means of pronominal affixes. The TAM series also determine whether arguments are marked by set A or set B affixes (cf. Section 8.4.7), and which semantic role they have (e.g. Actor, Undergoer or Beneficiary). Finally, the vocalic prefixes additionally express the amount of arguments and their configuration.
The important question now is whether these markings are fully pragmatically or semantically motivated, or that it is relevant to assume syntactic functions. As it turns out, semantic role can be neutralised, i.e. arguments with different semantic roles can get identical morphosyntactic expression. Consider for example (12), (13) and (14), in which the pastry as Undergoer is marked identically as the Actor Nino. Example (14) proves that the semantic functions are distinguished in transitive clauses, so they can get different case marking.

Hewitt (1995: 550)
(12) k’at’a-m mo-šard-a
cat-ERG PFV-urinate-AOR.3
‘The cat urinated.’

Harris (1981: 43)
(13) namcxvar-i ga-mo-cxv-a
pastry-NOM PFV-VEN-bake-AOR.3
‘The pastry baked.’
(14) namcxvar-i ga-mo-v-a-cxv-e
pastry-NOM PFV-VEN-1SG-PRV-bake-AOR.1
‘I baked pastry.’

Georgian furthermore has a passive construction, justifying the postulation of a syntactic function Subject. As shown in (15), the Undergoer argument is expressed in a passive clause in the exact same way as an Actor in an active clause, thus neutralising the semantic distinction.

Harris (1981: 103)
(15) a. ʒaɣl-i u-k’ben-s bavšv-s
dog-NOM PRV-bite-3SG child-DAT
‘The dog is biting a child.’
b. bavšv-i da-k’ben-il-i-a ʒaɣl-is mier
child-NOM PFV-bite-PFV.PTCP-NOM-COP.3SG dog-GEN by
‘The child has been bitten by a dog.’
Word order in Georgian is to a large extent free (Hewitt 1995: 528, Vamling 1989: 30). I have no evidence for influence of complexity on morphosyntactic placement and will assume it does not exist in Georgian. Georgian has many bound morphemes, that are in large majority specialised for one type of lexemes: either nouns or verbs. Some clitics are found as well, for example postpositional clitics Hewitt (1995: 69ff.). Since the majority of bound morphemes consist of affixes, Georgian should be seen as a predominantly head-marking language. Particular verbal stems undergo alternation under the influence of the addition of a particular thematic suffix (see Section 8.4.7), e.g. the stem -rkv- ‘to name’ becomes -rtv- when the thematic suffix -ev is attached (Hewitt 1995: 27).
A second case of morphophonologically induced irregular stem formation in Georgian is the addition of a vowel under the influence of a certain series of TAM inflections (cf. Section 8.4.7), e.g. šl-i ‘erase-THEMSUF, you erase it’ versus c’a-v-šal-e ‘PFV-1SG-erase-3SG, I erased it’.
In a number of nominal stems, a vowel is dropped when used with specific suffixes, e.g. c’el-i ‘year-NOM’, c’l-is ‘year-GEN’ (Hewitt 1995: 35). Hewitt (ibid.) claims that it is not predictable from phonology which stems undergo such syncopation.
Four classifications play a role in the verbal morphology of Georgian. I will first discuss them and show how they interact, before I go into the question whether they are transparent or not. First of all, three or four verb classes are distinguished on the basis of their inflectional behaviour (cf. for example Vamling 1989: 24). Secondly, there are three so-called ‘series’ for marking tense and aspect (present, aorist and perfect; cf. Vamling 1989: 20 and Van Valin 1990: 240), each containing three ‘screeves’, i.e. particular TAM markings (cf. Hewitt 1995: 122). For example, TAM series I is the pattern of inflection of the present indicative screeve, the imperfect indicative screeve, the future indicative screeve, etc. (cf. Hewitt 1995: 217ff.). Thirdly, two sets of pronominal affixes exist, viz. ‘set A’ or ‘v-set’ and ‘set B’ or ‘m-set’ (Vamling 1989: 22, Hewitt 1995: 128). Finally, Georgian verbs are in some TAM contexts marked by so-called thematic suffixes. Under certain conditions, a verb gets one of the thematic suffixes -av, -ob, -eb, -ev, -am or -i (Vamling 1989: 19). The selection of a particular thematic suffix is not governed by a phonological, morphological or semantic principle – it is determined lexically.
These classifications interact with each other in a complex way. For instance, Vamling’s verb class 1 requires a nominative subject, obligatorily cross-referenced on the verb by means of a v-set affix. It optionally gets a dative object, which has to be cross-referenced on the verb by an affix from the m-series. This is true for all three TAM series. However, a class 3 verb requires a nominative subject and v-set affixation in the present tense TAM-series (called series I), but gets a dative subject and m-set affixation in the perfect TAM-series (called series III). One recognises the enormous complexity of this system, illustrated in examples (16) and (17) with one verb, ‘to paint’ (verb class 3). In (16), the present tense and imperfective aspect require series I TAM inflection: apart from specific pronominal affixes, this triggers the presence of a thematic suffix, which for ‘to paint’ happens to be -av.

Vamling (1989: 19)
(16) is xat’-av-s surat-s
NOM.3SG paint-THEMSUF-PRS.3 picture-DAT
‘He paints a picture.’

In (17), ‘to paint’ is used in a series II TAM inflection context (i.e. the aorist perfective screeve). This triggers different case-marking, different pronominal affixes and requires no thematic suffix.

Vamling (1989: 19)
(17) man da-xat’-a surat-i
ERG.3SG PFV-paint-AOR.3 picture-NOM
‘He painted a picture.’

A verb from class 2 behaves yet differently, as illustrated by example (18). The subject is now in dative case, the object is marked by a nominative marker, and TAM series I requires no thematic suffix in this verb class.

Vamling (1989: 26)
(18) mas me v-u-q’var-var
DAT.3SG NOM.1SG 1SG-PRV-love-1
‘He loves me.’

The three verbal classes are clearly opaque conjugations, since they are differentiated on the basis of their inflectional behaviour. The existence of two sets of pronominal affixes is opaque as well, since there is no consistent semantic distinction expressed by this affix alternation – a semantically identical argument can in some cases be expressed by affixes from the m-set, and in another context by the v-set. This is again a case of morphologically determined affix alternation: suppletive affixes exist for the expression of arguments on predicates in different contexts. As for nominal inflection, Georgian shows no affix alternation.
The division of tense and aspect over three TAM series is based on a semantic principle: series I expresses present and future tenses, series II expresses the aorist and series III codes perfect aspect (cf. Van Valin 1990). Even though the series cannot be associated with specific markers, these semantic categories relate in a one-to-one fashion to morphosyntactic marking – each pattern of case-marking, cross-reference, and the use of a thematic suffix is related to a specific TAM series. Hence, this is transparent, semantically based inflection.
There is one more affix alternation process to be mentioned here: the verbal affix h- ‘3.IOBJ’ assimilates to s- when followed by an alveolar, post-alveolar or palatal fricative (Hewitt 1995: 27). Since this does not occur with all h’s in this context, this cannot be seen as a purely phonologically conditioned process.
Adjectives ending in -a or -o get an epenthetic -v- when suffixed, e.g. brma ‘blind’, a-brma-v-eb ‘you-blind-v-THEMSUF’ (Hewitt 1995: 22).
Geminates are retained in most cases in Georgian, but degemination does sometimes occur, e.g. gaxlavvar, a highly polite version of ‘I am’, is pronounced as /gaxlavar/ (Hewitt 1995: 26).
Phonologically based alternations apply to stems and affixes alike. A case in point is deletion of vowels that are adjacent to another vowel. For example, nouns ending in a consonant get a nominative singular suffix -i, but this -i is elided with nouns ending in a vowel, so that those appear unmarked for nominative singular (Hewitt 1995: 34, 38). This has made some authors believe that two noun classes should be distinguished, one with nouns ending in a consonant, and one with nouns ending in a vowel. However, since the process is fully phonologically predictable, this is not a case of morphological affix alternation but of phonological affix alternation (cf. Vamling 1989: 16 in agreement).
8 Huallaga Quechua Huallaga Quechua does not have grammatical gender (cf. Weber 1989: 35). Huallaga Quechua has no nominal expletives, as demonstrated by example (7).

Weber (1998: 529) in Grández Ávila (2011: 45)
(7) tamya-yka-n fiyupami
rain-IPFV-3SG strongly
‘It is raining strongly.’
The number and person properties of the Actor argument are marked on the predicate by means of suffixes (Weber 1989: 10). Furthermore, there is nominative-accusative case marking to distinguish Actors from Undergoers, cf. (8).

Weber (1989: 15)
(8) hwan tumas-ta maqa-n
J. T.-ACC hit-3
‘John hits Tom.’

However, such markings are neutralised in intransitive clauses, as shown in (9). This requires postulation of a syntactic function Subject.

Weber (1989: 179)
(9) wamra pun̄u-n
child sleep-3
‘The child sleeps.’

Furthermore, there are three constructions that Weber (1989: 233) considers to be passives. At least one of these can have an oblique Actor argument, so that it qualifies as neutralisation of semantic roles; cf. (10) for the crucial example. It should be noted that addition of a by-phrase appears through elicitation only – speakers do not use this spontaneously. The other two passive constructions do not appear to have this option at all.

Weber (1989: 245)
(10) yanapa-sha ka-nqa chay runa qam-pita
help-PTCP be-FUT.3 DEM man you-ABL
‘That man will be helped by you.’
Huallaga Quechua has a default SOV word order, but this is quite free (Weber 1989: 15). If some part of a sentence is relatively complex, it can be realised post-verbally to prevent ‘stack overflow’ (Weber 1989: 282). This is demonstrated in example (11).

Weber (1989: 282)
(11) maria qallari-na-:-ta muna-n
M. begin-NMLZ-1.POSS-ACC want-3
allqu-ta qara-y-ta kani-ma-sha-n-ta
dog-ACC feed-INF-ACC bit-1.OBJ-NMLZ-3.POSS-ACC
‘Mary wants me to begin to feed the dog that bit me.’
Quechua makes extensive use of suffixation (Weber 1989: 9) and is therefore commonly seen as an agglutinative language. However, many of the alleged suffixes are in fact clitics, for example the topic marker =qa in (12).

Weber (1989: 408)
(12) kapas watachaw huk kuti armakun inteeru kwerpun-ta=qa
maybe in_a_year one time bathe.3PL entire body.3.POSS-ACC=TOP
‘Maybe they bathe their entire body once a year.’

Furthermore, Weber (1989: 75) distinguishes a number of suffixes that “may apply to words of any category”, which is in line with Grández Ávila (2011), who argues that case markers should also be seen as clitics, since they are phonologically dependent but morphosyntactically indifferent to the nature of their hosts. I will follow her in her analysis that Quechua is predominantly phrase-marking.
Stem-final high vowels of particular stems are lowered to /a/ in the adjacency of particular suffixes (Weber 1989: 29, 464). E.g. miku- ‘eat’ becomes mika-mu-shun ‘eat-afar-FUT.1PL.INCL, let’s eat over there’. Since this happens with particular affixes only, it is considered a morphophonological phenomenon.
The absolute spatial expressions hana ‘top’, washa ‘level’, and ura ‘below’ lose their final -a when combined with -qa, which relates the location to the hillside. For example: hana-n-chaw ‘top-3PL-rock’, han’qa-POSSX ‘above X on the hillside’ (Weber 1989: 256, 460).
As explained in Section 9.3.1, Quechua makes use of verb markers that express number of subject and object, as well as tense (Weber 1989: 95ff.). As argued, the person and tense markers are separable, but they do undergo some adaptations, partly due to phonological rules. However, person markers also alternate under influence of the optional future tense marker -paq, e.g. Ø-Ø-n-chii ‘3.OBJ-PRS-1.SBJ-PL.INCL’, -Ø-ra-n-chii ‘3OBJ-PST-1.SBJ-PL.INCL’, but -Ø-shun(-paq) ‘3.OBJ-1PL.INCL-FUT’. As said earlier, Weber (1989: 100) analyses this as cumulation, but since there is a separate future tense marker, I rather see this as alternation of person suffixes in the context of this future tense suffix. The person markers themselves do not, in my opinion, express future tense.
As with stems (cf. Section 9.4.6), the final vowel of particular suffixes is lowered to /a/ in the adjacency of particular suffixes (Weber 1989: 29, 464), e.g. the reflective suffix in sha-ku-mu-n ‘come-REFL-afar-3SG’ is pronounced as ka: [shakamuŋ].
To avoid illegal consonant clusters, e.g. clusters of three or more consonants, the epenthetic suffix -ni is inserted before nominal suffixes, e.g. atoq ‘fox’ with -yoq ‘POSS’ is optionally pronounced [atoqniyoq] (Weber 1989: 465). Since -ni does not appear before verbal suffixes, this process is morphophonological rather than phonological.
/sh/ becomes /s/ preceding /ch/ or /y/, e.g. hanash-cha ‘farther_up-LOC’ becomes [hanaschaw] (Weber 1989: 457). High vowels are optionally lowered to mid vowels in the neighbourhood of /q/, e.g. /suqta/ ‘six’ becomes [soqta] (Weber 1989: 458). Syllable-final -n assimilates its place of articulation to a following phoneme, e.g. /qanra/ ‘dirty’ is [qaŋra] (Weber 1989: 458). Long vowels are shortened when the syllable is closed, i.e. when it is followed by a syllable with a coda consonant (Weber 1989: 467). If an /i/ precedes a -y, they fuse into a lengthened /i:/, e.g. chaki-ykan becomes chaki:kan (Weber 1989: 461, 462). Sometimes, the y- is simply dropped, e.g. nawi-yki becomes nawiki (Weber 1989: 462). Several other assimilation processes occur. Suffixes of the form CV may lose their vowel if they are word-final and attached after a short vowel, e.g. mana-mi ‘not-EVID’ can be pronounced as [mánam] (Weber 1989: 459). Several other similar processes occur.
9 Japanese There is no grammatical gender in Japanese (Hinds 1986: 171, 226). Japanese does not have dummy subjects (Hinds 1986: 73), as illustrated by example (4) of a weather verb with a semantic argument.

Hinds (1986: 277)
(4) ame ga fut-ta
rain NOM fall-PST
‘It rained.’
Alignment in Japanese is largely based on word order (Hinds 1986: 188). Furthermore, there are case marking particles that mark pragmatic and semantic functions (Hinds 1986: 191ff.). Example (5) gives an example of a transitive clause.

Iwasaki (2002: 85)
(5) kodomo-tachi ga sensoo no kowasa o kangaeru
child-PL NOM war GEN horrors ACC think
‘Children think about the horrors of war.’

Semantic functions can be neutralised in intransitive clauses, as proven by example (6), in which the Undergoer argument is marked by nominative case.

Hinds (1986: 281)
(6) oyu ga waite-iru
water NOM boil-PROG
‘The water is boiling.’

Furthermore, example (7) shows that semantic role can be neutralised by a passive construction, in which the Undergoer is marked as an Actor, obscuring the semantic difference.

Iwasaki (2002: 126)
(7) sakana ga neko ni tabe-rare-ta
fish NOM cat DAT eat-PASS-PST
‘The fish was eaten by the cat.’
Hinds (1986: 151) mentions that dislocation of elements is possible and results in emphasis on the moved element. This indicates that there may be pragmatic factors influencing word order, but is not informative on the influence of heaviness.
Iwasaki (2013: 13, 237ff.) describes how Japanese constituent order is determined by pragmatic principles. He confirms (S. Iwasaki, personal communication, October 17, 2013) that heavy shift does not occur.
Japanese is an agglutinative language, as it has a rich array of suffixes, especially verbal ones (Iwasaki 2002: 9). It has no clitics (Hinds 1986: 361), but there are quite some phrase-marking particles. Since both head-marking and phrase-marking occur non-marginally, it cannot be determined whether Japanese is transparent or opaque with respect to this feature; I will leave the matter undecided. Verbal stems ending in /m, n, b/ assimilate to attached consonants, e.g. nom ‘drink’ + -ta ‘PST’ becomes /nonda/. The final consonant of verbal stems ending in /w, r/ changes to /t/ when a t-initial suffix is attached, e.g. kaw ‘buy’ + -ta ‘PST’ becomes /katta/. If a verbal stems ends in a velar, this velar is palatalised to /i/ if a t-initial suffix is attached, e.g. kak ‘write’+ -ta ‘PST’ becomes /kaita/ (Hinds 1986: 420). Since these alternations occur in verbal stems only, they are morphophonologically based. Several other morphophonologically motivated stem alternations could be illustrated here, but the examples given suffice to demonstrate the existence of such processes in Japanese. Japanese has a conjugation system, since verbs can be divided over four classes according to a thematic vowel that is inserted with some classes between the stem-final consonant and particular inflectional suffixes (Hinds 1986: 172ff.), as shown in the partial paradigm in (8) (cf. Section 10.4.9 for an explanation of the additional /r/ in taberu and yabereba).

(8) yom- ‘read’ tabe- ‘eat’
yom-a-nai read-a-NEG.IMP tabe-nai eat-NEG.IMP
yom-i-tai read-i-DES tabe-tai eat-DES
yom-u read-INF tabe-ru eat-INF
yom-eba read-COND tabe-reba eat-COND

The classes are not phonologically predictable so that this process is morphologically motivated.
After Chinese numerals, particular counter suffixes undergo phonemic alternations that other suffixes do not undergo. For example, the initial h of counter suffix -hon ‘long, cylindrical objects’ changes to a p after Chinese numerals like ichi ‘one’, which becomes ichi-pon and then, according to a rule described in Section 9.18, ippon (Hinds 1986: 231).
Vowels can be inserted to prevent adjacency of two consonants or to prevent a word from ending in a consonant. This is visible in loanword adaptations, e.g. English ‘street’ assimilated to Japanese sutorito (Hinds 1986: 423).
Word-initial morae with a low pitch may be deleted in casual speech, e.g. atashi ‘1.F’ may be pronounced tashi (Hinds 1986: 422). Several other similar deletion processes occur, mainly in frequent items.
When a stem-final combination of a voiceless consonant and a high vowel is followed by another voiceless consonant, the vowel is dropped and the stem-final consonant adapts to the other consonant, e.g. ichi or iti ‘one’ + ko ‘thing’ becomes ikko (Hinds 1986: 421). Hinds refers to this as a ‘general process of gemination’, so that I assume that it is a phonological rather than a morphophonological process.
Stem-final nasals adapt their place to an attached consonant, e.g. shin ‘new’, shimbun ‘newspaper’ (Hinds 1986: 406 and cf. Iwasaki 2013: 21ff.). High vowels /i/ and /ɯ/ are devoiced in between two voiceless consonants and between a voiceless consonant and a pause, e.g. [kɯ̥tsɯ̥] ‘shoes’ (Iwasaki 2013: 21). Several similar adaptation strategies occur.
Finally, Japanese exhibits downdrift, causing some syllables to have a lower pitch in the context of other low pitched syllables, e.g. yo.ku ‘often’ shows an HL pattern, ka.ku ‘write’ is HL, but the combination yoku kaku ‘write often’ has the pitch contour HL LL’ (Hinds 1986: 416).
Some verbal suffixes have an initial /r/, which is deleted when the suffixes attaches to a consonant-final stem, e.g. tabe-ru ‘eat-NPST’, nom-u ‘drink-NPST’ (Iwasaki 2002: 60).
Suffix-initial consonants become voiced when they attach to a voiced stem-final consonant, e.g. nom ‘drink’ + -ta ‘PST’ becomes nonda (Hinds 1986: 420).
10 Kayardild There is no grammatical gender in Kayardild (cf. Evans 1995: 122). There is no pronominal expletive element in Kayardild, as demonstrated by the weather predicate in (9).

Evans (1995: 326)
(9) balmbi-wu warrngal-warri-r-i-ju, wambaji-wa-thu
tomorrow-MODC.PROP wind-PRIV-FAC-MID-POT clear-INCH-POT
‘Tomorrow it will become calm and clear.’
Evans (1995: 95) argues that syntactic functions Subject and Object are relevant in Kayardild. In transitive clauses, Actor and Undergoer are distinguished by means of accusative case marking on the Undergoer, as in (10). In intransitive clauses, such roles are neutralised: the Actor in (11) and the Undergoer in (12) receive the same nominative case marker.

Evans (1995: 1)
(10) dangka-a raa-ja bijarrba-y wumburu-nguni
man-NOM spear-REAL dugong-MODC.LOC spear-INSTR
‘The man speared the dugong with a spear.’

(11) dangka-a jawi-j
man-NOM go_fast-REAL
‘The man went fast.’

Evans (1995: 136)
(12) mutha-a dangka-a yuuma-th, buka-wa-th
many-NOM person-NOM drown-REAL rotten-INCH-REAL
‘Many people drowned and died.’

Furthermore, Kayardild exhibits a passive construction, in which addition of an oblique Actor is possible, as in (13).

Evans (1995: 350)
(13) ngada ra-yii-ju mun-da balarr-inja maku-nth
NOM.1SG spear-MID-POT buttock-NOM white-OBL woman-OBL
‘I will be injected in the buttocks by the white woman.’
The order of phrasal constituents is relatively free in Kayardild: all possible word orders are attested so that it is impossible to ascribe a basic word order to the language (Evans 1995: 92-93). Pragmatic principles do play a role, as for instance new participants are relatively often placed in the beginning of the sentence (Evans 1995: 93). I have no evidence that the complexity of elements plays any role in this, and will therefore assume that it does not. As said in Section 11.3.1, Kayardild employs suffixation as its most important function marking strategy. There are clitics (Evans 1995: 389) and particles (Evans 1995: 378ff.) as well, but these are scarce in comparison to suffixes. Hence, Kayardild is a predominantly head-marking language. Stem-final apico-alveolar d alternates with lamino-dental th or lamino-palatal j when preceding the nominative suffix -a, e.g. ngirrnguth-inja ‘fly-OBL’, ngirrngud-a ‘fly-NOM’ (Evans 1995: 74). There are morphologically based alternations in Kayardild in case suffixes. Evans (1995: 126) mentions for instance alternation of the initial consonant of the locative suffix, e.g. daman-ki ‘tooth-LOC vs. kuwan-ji ‘firestick-LOC’ (Evans 1995: 126). This distinction is not phonologically predictable.
Furthermore, Kayardild exhibits two conjugational classes: a dental and a palatal class (Evans 1995: 267ff.). As the names suggest, these are largely phonologically based, apart from verbal stems ending in -a – conjugation membership of those verbs is determined by the presence of particular derivational suffixes, and partly lexically.
A velar stop is palatalised before /i/ and likewise, /ng/ is fronted before /i/, e.g. ngimiy ‘night’ is pronounced [ŋjime͡i] (Evans 1995: 54). The labio-velar glide w is sometimes delided word-initially preceding high vowels, e.g. wuranda ‘food’ can be pronounced [‘oɹant] (Evans 1995: 58). This example also illustrates that word-final a can be deleted before planned pauses at the end of a breath group (Evans 1995: 63). Several other cases of allophony are described in Evans (1995: 54-65) – it is not relevant to describe them all here.
Stem-final laminal stops assimilate to ny if preceding an m or ng, e.g. yarbuth + -ngarrba becomes yarbunyngarrba (Evans 1995: 72). A stem-final ng assimilates its place to a following consonant, e.g. kang + juldajulda becomes kanyjuldajulda ‘correct speech’ (Evans 1995: 73). A stem-final /r/ in combination with suffix-initial /r/ fuse into an /l/, e.g. birdin + raja becomes birdilaaja ‘mis-spear’ (Evans 1995: 73). It can be shown historically that the /n/ is lost, while the /r/ alternates with /l/, so that this is both an alternation in the stem and in the affix. Further phonologically conditioned processes exist, but it is not germane to describe all cases.
Case inflection of Kayardild nouns can be grouped in 6 classes (Evans 1995: 124ff.). The declensional class inflection that a noun takes is in most cases predictable on the basis of its phonology. Some noun stems are ambiguous between two classes so that their inflection class is not fully predictable, but still, the selection is based on their phonology and not on their morphology.
A suffix-initial w is nasalised under the influence of a preceding nasal, e.g. kinyin + -warri becomes [kinyinmarri] ‘non-existent’ (Evans 1995: 72).
A stem-final /r/ in combination with suffix-initial /r/ fuse into an /l/, e.g. birdin + raja becomes birdilaaja ‘mis-spear’ (Evans 1995: 73). It can be shown historically that the /n/ is lost, while the /r/ alternates with /l/.