Verbal derivation in Kambaata (Cushitic), with a focus on the encoding of noncausal/causal alternations

Kambaata, a Cushitic language of Ethiopia, has a rich verbal morphology. This paper first sets out to describe its productive verbal derivation processes, i.e. the forms and functions of the causative, the passive and the middle (autobenefactive) derivation. In the second part, the study starts from Haspelmath’s (1993) list of 31 verbal concepts to determine the preferred morphological strategies to express noncausal/causal alternations. The analysis shows that the causative strategy is predominant in Kambaata, but that the decausative strategy follows closely behind. If the whole documented lexicon including the numerous ideophones are considered, the equipollent strategy is a very important third strategy. Kambaata has only a single labile verb but otherwise distinguishes consistently between transitive and intransitive verbs. Finally, two suppletive verb pairs are found in the list: ‘die/ kill’ and ‘finish (intr./tr.)’.


Introduction
Cushitic languages are known to have productive verbal derivations such as causative, middle, passive and inchoative as well as reduplicative means to express event plurality (Mous 2012: 403).In addition to dedicated sections in grammars, there is already a significant body of literature on verbal derivation interjections. 1 Kambaata is a language with morphological stress, i.e. it applies stress to distinguish between word forms but not between lexemes.
Kambaata nominal roots (or stems) are never used in isolation but combine obligatorily with inflectional morphology.Apart from nouns denoting biologically male and female referents, all members of the nominal word class fall arbitrarily into two genders, masculine and feminine.Nouns are marked for nine cases and categorized into numerous (~ 20) inflectional classes, also called "declensions" (Treis 2008: 102-126).Case is marked both segmentally by a suffix and suprasegmentally by a specific position of the stress.In certain declensions, e.g.M3, accusative and nominative are only distinguished suprasegmentally, -ú vs. ´-u.In Table 1, the paradigms of two masculine and one feminine declensions are illustrated.The nominative case is used exclusively in a subject-marking function, whereas the accusative marks the direct object, the citation form and certain adverbial constituents (see also §3). 2 In a Kambaata clause, the expression of (pro)nominal arguments is optional if they are retrievable from the context.A finite verb alone can constitute a complete sentence.Except for the infinite verbal noun, all verb forms index the person of the subject, its number in non-second persons, and its gender and social status in the third person; in addition, verbs are compulsorily marked for aspect, mood and/or dependency status.While the paradigm of independent personal pronouns distinguishes between nine subject forms (Treis 2008: 330f.), verbs differentiate maximally between seven subject indexes: 1sg, 2sg, 3m, 3f=3pl, 3hon, 1pl, 2pl=2hon.In Table 2, the subject indexing of the imperfective main verb and the imperfective converb paradigm is exemplified with the verbal root fan-'open (tr.)'. 3 In most main verb paradigms, subject indexing is discontinuous, with one subject marker (sbj1) before and one subject marker (sbj2) after the aspect marker (e.g.a(a) for the imperfective).As the subject indexes tend to fuse with the marker for aspect, mood or syntactic dependency, the verbal inflectional complex is not segmented in the examples, i.e. fantáant is split into fan-táant {open-2sg.ipfv}but not further into fan-t-áa-nt {open-2sg-ipfv-2sg}.Whereas subject indexing is obligatory, object marking is only possible on certain verb forms and, where it can occur, it is differential and dependent on animacy and information status.Object suffixes follow the verbal inflectional complex (Figure 1).

Verbal derivation
The verbal inflectional morphology, as described in the preceding section, is very regular and, if the defective existential verb yoo-'exist' and three verbs with minimal stem variation were ignored, 4 one could go as far as to say that Kambaata has no irregular verbs.In the formation of finite (or semi-finite) verbs, one only needs to take into account the morphophonological rules that apply without exception at the boundary between the verbal stem and the suffixed inflectional morphology (Treis 2008: 60-72).
In contrast, in the domain of verbal derivation, it is difficult to come up with all-encompassing rules that would predict the occurrence of assimilatory processes between certain roots and extensions, explain the distribution of derivational allomorphs (especially those of the causative) and predict the compatibility (/grammaticality) of certain root-extension sequences.In addition, a number of derived verbs have unpredictable translations.Strong tendencies are observable, and rules can be formulated, as shown below, but the multiple exceptions and idiosyncrasies must be listed and all verbs checked individually.Therefore, this section can only be seen as a first step towards a complete account of Kambaata verbal derivation.As seen in Figure 1, the verbal stem can be either a simple root -see fan-'open (tr.)' in Table 2 or morphologically complex.Derivational morphemes are found before the inflection.In its inventory of verbal derivational morphemes, Kambaata has four productive (predominantly) word classmaintaining morphemes: the causative, e.g.onxah-'come closer' > onxah-is-(caus1) 'make come closer' and fan-'open (tr.)' > fan-siis-(caus2) 'make open (tr.)', the passive, e.g.fan-'open (tr.)' > fan-am-(pass) 'open (intr.),be opened', and the middle, e.g.fan-'open (tr.)' > fa< '>nn-(<mid>)  'open (tr.) for oneself'.Different combinations of these derivational morphemes are possible, e.g. a middle and a passive morpheme are combined to encode reciprocity, e.g.usur-'tie' > usu<'>rr-am-(<mid>-pass) 'tie each other'.It is not rare to find three (in exceptional cases even four) derivational morphemes after a verb root, e.g.af-'take' > aph-ph-(mid) 'take for oneself ' > aph-ph-am-(midpass) 'take each other ' > aph-ph-an-siis-(mid-pass-caus2) 'cause to take each other; connect (tr.)'. 5 Before the discussion of the causative in §3.1, transitive and intransitive verbs need to be defined.This task is all the more important when one recalls, firstly, that arguments do not need to be (and 4. The verbs orooqq-~ oroo-'go away, go out', maccoocc-~ maccoo-'hear' and reh-~ re-'die' take the second variant in front of t-initial inflectional morphemes (e.g.oroo-tóo'u 3f.pfv 'she went away') and the first variant in front of all other inflectional morphemes (e.g.orooqq-éenno 3hon.ipfv's/he (respected) goes away').5.For reasons of space, the other combinatorial possibilities of the causative, the passive and the middle, the non-productive word class-maintaining and word class-changing morphology are not discussed further here.often are not) overtly expressed in Kambaata and that, secondly, the accusative does not only mark direct objects.The occurrence of an accusative NP in the clause of which a verb is the head can therefore not be a sufficient criterion to qualify this verb as transitive.A transitive verb rather has to be defined as a verb that can govern an accusative-marked direct object.A direct object is distinguished from other accusative NPs in that it can become the nominative subject when the verb is passivized.The definition thus excludes three kinds of accusative-marked constituents.Firstly, it excludes those that express the goal or direction of a movement, e.g.kabá '(to) here'.Secondly, it excludes those that are adverbial adjuncts, e.g.temporal adverbials such as ga'-áata 'tomorrow' and ankar-í '(at) night' (3), manner adverbials such as xumm-á '(in) peace, peacefully' and 'with-respect-to'-adverbials of property verbs and adjectives, as in kof-ú lung-{upper.arm-m.accbecome.weak-}'become weak with respect to/as far as the upper arms are concerned'.Thirdly, it excludes those that are so-called "cognate objects", e.g.shum-áta shum-a'-{urine-f.accurinate-mid-} 'urinate urine' and nadajj-á nadad-{anger-m.accget.angry-} 'get anger-angry'.

Causative
The analysis of the Kambaata causative poses challenges on different levels.The causative can be realized by a short (simple) suffix -s (caus1) or by a long (seemingly repeated) suffix -siis (caus2).In §3.1.1,I argue that the distribution of the short causative (caus1) and the long causative (caus2) as primary causativizers is partly grammatically, partly lexically determined.As §3.1.2demonstrates, causative suffixes can also be stacked, in which case the second causative introduces yet another participant.In §3.1.3on the morphophonology of the causative, I show that the different realizations of the short causative -s (caus1) cannot be explained by morphophonological rules alone but also requires listing lexically determined exceptions.While the meaning of causative-derived verbs is predictable overall, there are a number of semantic idiosyncrasies, as seen in §3.1.4.Finally, not all causative verbs are derived from a verbal base, and the last subsection on the causative ( §3.1.5)presents a list of denominal causatives.
+ Inflectional complex: subject index(es) fused with markers of aspect, mood, dependency status ++ Position of negator depends on aspect All caus1-and caus2-derived verbs can be further causativized, resulting in V-caus1-caus2-or V-caus2-caus2-stems.However, any attempt to add a short to a long causative (*V-caus2-caus1-) leads to an ungrammatical result.Double causatives are discussed in §3.1.2.
The tricky question that needs dealing with is: what determines whether a verb belongs to Group caus1 or Group caus2?I am going to argue that membership in these groups is largely determined by the argument structure (transitivity) of the verbal lexeme.However, given the considerable number of exceptions, group membership of quite a few verbal lexemes has to be assumed to be lexically determined, i.e. the choice of the causativization strategy is to a certain extent arbitrary.Group caus1/2 is briefly addressed towards the end of this section.

Group caus1 (majority): Intransitive verbs
The majority of verbs that take caus1 as their first causativizer are intransitive verbs that cannot govern an accusative-marked direct object and cannot be passivized.Examples of intransitive verbs and their caus1 forms are given in (1).

Group caus2 (majority): Transitive verbs
The majority of verbs that take caus2 as the first causativizer are transitive verbs that govern an accusative-marked direct object and can be passivized.Examples of such verbs are given in (4).
(7) Woshichchíin ga'mmisiissókkoonte'e!wosh-ichch-íin ga'mm-isiis-sókkoonte-'e dogs-sgv-m.icpbite-caus2-2sg.appr-1sg.obj'Don't make/let the dog bite me! / Don't have/let me be bitten by the dog!' (elicited) While the distributional rule that links the choice of caus1 and caus2 to the (in)transitivity of their base verb covers a majority of verbs, there are also many exceptions, which will be discussed below.

Group caus2 (minority): Intransitive verbs
A number of intransitive verbs require caus2, while the suffixation of caus1 yields an ungrammatical word form.The intransitive, caus2-taking verbs seem to be especially represented in the semantic domains of (manner of) motion and staying (8), emotions (9), illness/pain and bodily functions (10), and sound emission and speaking (11).

Group caus1 (minority): Transitive verbs
While most transitive verbs require caus2, as illustrated in ( 4)-( 7), there are also a significant number of exceptions to this rule.Prominent counterexamples come from the semantic fields of ingestion The verbs of this group, e.g.luus-'miss', are passivizable (20) and can govern an accusative-marked direct object ( 21), but they take the short causative (caus1) as their primary causativizer ( 22).
oos-úta luus-is-éen farr-áta roshsh-áta children-f.accmiss-caus1-3hon.pfv.cvbbad-f.acclesson-f.accawwan-síis-u xúm-a-a-ba'a follow-caus1-m.nomgood-m.pred-m.cop2-neg1'It is not good to mislead children and make them follow a bad example.' (elicited) Although the causativization leads conceptually to the addition of another participant, the number of syntactic arguments in the clause does not necessarily increase.The causer becomes the subject of the causative verb but usually only one other participant, either the causee or the patient, is overtly expressed.An overt patient is always encoded in the accusative case.An overt causee can be marked for the dative (e.g. with xuud-is-'show' in ( 23)), the accusative (e.g. with luus-is-'mislead' in ( 22)) or another peripheral case (e.g. with iyy-iis-'cause to carry' in ( 24)), depending on the verb and on semantic and syntactic factors that are still to be explored.mát-o ciil-áan ~ciil-íin zambeel-á iyy-íishsh-eemm 10 one-m.obllittle_child-m.loclittle_child-m.icpbasket-m.acccarry-caus1-1sg.prf'I had a little child carry the basket (home from the market) (lit.I had the basket be carried on/by a child).' (elicited) Among the examples in ( 16)-( 19), we find verbs that belong to Group caus1/2.Verbal lexemes of this group permit either the causativization with caus1 or caus2 -but, as the translations show, with important meaning differences.The two causatives can, for instance, target different senses of polysemous verbs, as in the case of ag-, which combines with caus1 in its reading 'drink' > 'feed (a liquid)' and with caus2 in its reading 'rip off' > 'cause to be ripped off' (25).

Double causatives
Causative-derived verbs can again be causativized.The second causative, which adds yet another causer participant, is necessarily a caus2-morpheme; see waal-'come' > waash-sh-(caus1) 'bring' > waash-sh-isiis-(caus1-caus2) 'make bring, cause to be brought' in (26).The added second causer is expressed as the subject of the double causative verb -see the 1sg subject index on the main verb in (26).As Kambaata speakers generally avoid rendering more than two arguments explicit in a noun phrase, the intermediate causer (i.e. the person who was sent to bring the doctor) is not expressed here -either because the referent is retrievable from the context or considered to be irrelevant.
ciil-í íib-u bat-a'-ó=bii little_child-m.genfever-m.nombecome_much-mid-3m.pfv.rel=reassarb-í doktoor-ch-ú waash-sh-isiishsh-óomm do_fast-1sg.pfv.cvbdoctor-sgv-m.acccome-caus1-caus2-1sg.pfv 'I made (someone) bring the doctor/I had the doctor be brought (by someone) quickly because the little boy's fever was very high.' (elicited) Double causativization is not restricted to intransitive verbs such as waal-'come' in (26) but is also possible with transitive verbs such as ag-'drink'.The event expressed by the doubly causative-derived verb in (27) has four participants, a causer who instigates the action, an intermediate causer who facilitates or transmits the order for the action, a causee who realizes the action and a patient who undergoes the action.In (28), the 2sg-subject is the causer; the intermediate causer ('Teshome') is encoded in a subordinate converb clause (marked by square brackets); the causee ('guests') is the accusative-marked object of the causative verb, while the patient ('drinks') remains unexpressed. 11 .imp 'Tell to Teshome (and) make (him) offer (drinks) to the guests!' (elicited) Examples ( 26) and ( 28) have illustrated the re-causativization of caus1-stems.In the same way, caus2-stems can be the input of another causativization process, as shown in (29), in which two of four participants are expressed overtly, the patient ('thief') and the causer (2sg-subject on the verb).
Though considered fully acceptable by the speaker who provided the examples in this subsection and explained possible usage contexts, double causatives are so far not yet attested in spontaneously produced and recorded language data.Note, however, that double causatives are also reported for related languages, e.g.Hadiyya (Sim 1985: 35f.) and Oromo (Dubinsky et al. 1988).See also Kulikov (1993)  for a typological study of second causatives.

Morphophonology
It has become apparent in the segmented and glossed examples above that the causative suffixes are subject to allomorphy.This section explains these morphophonological processes in more detail.
The realization of the short causative (caus1) -s {-s, -is, -sh, -iis} is determined by a mixture of phonotactic and lexical constraints.The first allomorph -s is the default after roots ending in a single sonorant, e.g.xur-'become dirty' > xur-s-'make dirty'.The second allomorph -is is the default after all other single and double consonants e.g.mut-'sprout' > mut-is-'make sprout' and farr-'become bad' > farr-is-'make bad'.
These general tendencies are overruled by lexically determined and thus unpredictable realizations of the short causative.The third caus1 allomorph -sh is found after many roots ending in l, which totally assimilates to the causative morpheme (30). 12Not all verb roots ending in l form their causative with -sh, as in the regular derivation caal-'become tactile cold' > caal-s-'make tactile cold', but the higher the frequency of a verb ending in l, the more likely its causative realization as -sh.
The realization of the long causative (caus2) -siis {-siis, -isiis} is predominately determined by the phonotactic rules of the language.The first allomorph is found after verb roots ending in a single sonorant, e.g.il-'give birth' > il-siis-'assist in giving birth', which creates a licit sonorant-obstruent cluster.To avoid illicit three-consonant clusters, i-epenthesis applies after verb roots that end in a two-consonant cluster, e.g.ass-'do' > ass-isiis-'make do'.After root-final single obstruents, the picture varies and may also be speaker-dependent; epenthesis tends neither to occur after voiced stops, e.g.dagud-'run' > dagud-siis-'make run', nor after the fricatives -s and -f, e.g.gaf-'boil (tr.)' > gaf-siis-'make boil (tr.)'.Epenthesis is required after the glottal stop, e.g.ze'-'herd (cattle)' > ze'-isiis-'make herd (cattle)'.Evidence for the realization of the long causative after other obstruents is inconclusive.As with the short causative (33), we observe the anticipatory assimilation of root-final consonants to the following causative suffix for a small number of verbs, e.g.tuf-'vomit' > tus-siis-'make vomit'.

Causatives with unpredictable meanings and argument structures
While most deverbal causatives express causation in a wider sense and are translatable as 'make/ cause to verb', 'let verb' and 'help verb', some causatives are semantically idiosyncratic (at least in their most common uses) -see, for instance, the verbs in (36)-(38).
14.The examples in (34) might give the false impression that root-final geminate consonants trigger the use of -iis.However, this is not the case, as numerous verb roots which end in CC but still take -is are attested; see, e.g., maccoocc-is-'make hear, make listen (to)' in (17).See also agud-iis-'pretend' in (41) where the -iis-allomorph attaches to a verb root ending in a single C. 15.We find a similar type of stem variation with the verb maccoocc-~ maccoo-'hear, listen (to)', but for this verb, the short causative is derived from the first variant: maccoocc-is-'make hear, make listen (to)'.16.Palatalization and gemination are, of course, also observed with underived verbs; see luus-'miss' > lúushsh-eemm /ˈluːʃːeːmː/ (prf) 'I have missed' in (21)  In (42), the causativization leads to a remapping of semantic roles onto syntactic functions.The theme object of the base verb becomes the theme subject of the causative verb.The experiencer subject of the base verb is expressed in a dative NP with the causative verb.So causativization here actually turns a transitive into an intransitive verb.

Denominal causatives
It is not possible to determine a verbal root for all causative-derived verbs, either because the root has fallen out of use in the history of the language or because there has never been a verbal base in the 17.This verb pair is also a member of the doubly derived verb pairs discussed in §4.5.18. Speakers consulted rejected the translation of moog-siis-as 'cause to bury (s.th./s.o.)'.The argument structure of moog-siis-'attend a burial service' is still unclear, as it is never used with a direct object in my database.19.Ronny Meyer (2022 p.c.) notes differences between the idiosyncratic causatives of Kambaata and Amharic.In Amharic, the verb 'fall ill, have pain' (ammämä) has two morphologically different causatives: asammämä 'cause to fall ill, cause (s.o.) pain' vs. astammämä 'care for a patient'.Similarly, the verb addärä 'spend the night' forms two causatives: asaddärä 'cause to spend the night' vs. astädaddärä 'govern'.Unlike the a-and as-causatives, the astä-causative is unproductive and restricted to some verbs.first place.For the non-exhaustive list of short causatives given in ( 43), one has to assume that they are directly derived from a nominal base, i.e. that the causative has a word-class changing, verbalizing function (see also the denominal passive and middle verbs in § §3.2-3.3).The meaning of the denominal causatives is not always deducible from the translation of the base.However, many can be paraphrased as either 'turn (tr.) into N' or 'apply/use N'.

Passive
Kambaata marks the passive by -am, e.g.hiir-'untie, solve, translate' > hiir-am-'be untied, solved, translated'.In (45), the base verb is used with an overt nominative subject expressing the agent ('I') and an accusative object expressing the patient ('all riddles').In the passivized counterpart (46), the patient ('the book') becomes the subject.The agent is dropped or, although rarely attested in spontaneously produced, non-elicited data, expressed as an instrumental-comitative-perlative adjunct to the passive verb (47).
antab-éechch-ut sul-íin it-án-tee'u 21 chicken-sgv-f.nompredator_sp-m.icpeat-pass-3f.prf'The hen has been eaten up by a sula (undetermined predator).'(elicited) In ( 48), the passive verb seems to govern a direct object ('your hip').However, on closer inspection, the example illustrates a frequently attested construction in which a (whole) patient is expressed as the subject of a passive verb and its affected body part as an accusative constituent, whose syntactic function is likely to be adverbial. 228) Nuré woggáa foo'ákk biixánteent Nur-é wogg-áa foo'-á-kk biix-án-teent last_year-m.genyear-m.acchip-f.acc-2sg.possbreak-pass-2sg.prf'Last year you broke your hip (lit.you were broken (with respect to) your hip).' (AL2006-01-02_kineeto) In contrast to the causative, few formal and semantic idiosyncrasies occur in the formation of the passive.Kambaata has two monoconsonantal verb stems, y-'say' and sh-'kill', of which the former requires a reduplicated passive, i.e. y-am-am-'be said' (*y-am-), and the latter has a suppletive passive ( §4.7).Furthermore, Kambaata has a small number of intransitive verbs which combine with the passive morpheme, albeit without a change in their argument structure.The derived and underived variants of amu'rr-~ amu'rr-am-'get angry', sarb-~ sarb-am-'be/do in a hurry' and da'll-23 ~ da'llam-'speed up, be/do faster' are, to the best of my knowledge, synonymous.The meaning of the passive forms of mar-'go' > mar-am-'walk' and lall-'appear, become visible' > lall-am-'become famous' are unpredictable, though not semantically unrelated to the underived forms.The passive of maccoocc-'hear' > maccoocc-am-can mean 'be heard' in some contexts; it is, however, predominately used to express 'feel', in which case it takes the stimulus as the subject and the experiencer as an indirect dative-marked object (Treis 2009: 320ff.; see also Hayward 1991: 152/III.10 for parallels in other languages).The passive of azzaz-'order' > azzaz-am-is usually translated as 'obey', and less commonly as 'be ordered'.
For a few formally passive but semantically reciprocal verbs, no underived verbal base but only a nominal base is attested: daayy-á 'eating together from one plate' > daayy-am-'eat together from 20.In the 1sg and 3m perfect(ive) verb forms the stem-final consonant is geminated and, if alveolar, palatalized (cf.fn.10).21.As seen in ( 47) and ( 48), the morpheme -am is realized as -an in front of t-initial inflectional morphemes.22.The syntactic function of the body part in ( 48) is comparable to that of the adverbial 'with respect to'-constituents of property verbs and adjectives ( §2).23.See the use of da'll-in (66).one plate', seel-ú '(local) beer' > seel-am-'drink together from one cup/glass' and er-á 'discussion' > er-am-'discuss (with each other)' (recall the denominal causatives in §3.1.5).

Middle
The middle morpheme has two (predominately) phonologically conditioned allomorphs, -' /Ɂ/ and -aqq.The allomorph -' /Ɂ/ is added to verb stems ending in a single consonant.If the consonant is a sonorant, the glottal stop is infixed to satisfy phonotactic rules (49). 24If the root-final consonant is a non-ejective obstruent, the glottal stop assimilates in place to it and triggers its ejectivization (50).
(53) galat-á (m) 'thanks' > galax-x-'thank' -see (56) gam-íta (f) 'pain' > ga<'>mm-'cause pain, hurt, bite' gulub-íta (f) 'knee' > guluph-ph-'kneel (down)' shoom-a (m) 'little hunger' > shoo<'>mm-'get a little hungry, eat a little' hem-íta (f) 'gossip' > he<'>mm-'gossip' The middle derivation does not reduce the valency of the verb.The most productive interpretation of the middle morpheme is to express that the subject of the clause is the beneficiary of the event expressed by the verb.There are apparently no semantic restrictions on the verbs that can be used with an autobenefactive middle marker.In (54), the middle morpheme expresses autobenefactivity with the verb laa'll-'search and call (for a missing animal)'.

Noncausal/causal alternations
The description of verbal derivation from a semasiological perspective in the first part of this article has set the scene for a discussion of noncausal/causal alternations from an onomasiological perspective in this section.In accordance with other contributions to this special issue and the typological literature, I understand under "noncausal/causal alternation" an alternating verb pair whose members share the same basic meaning but differ in the absence/presence of a 'cause' meaning component (Haspelmath et al. 2014: 590), as, e.g., English rise vs. raise, break (intr.)vs. break (tr.) and laugh vs. make laugh.Starting with a list of verbal concepts, which is introduced in §4.1, this section investigates the coding differences between the noncausal and causal members of the verb pairs in Kambaata.The description of the coding strategies proceeds from the most common ( §4.2: causativization, §4.3: decausativization) to the least common strategies ( §4.4: light verb change strategy, §4.7: suppletion).I also take a detour to investigate coding strategies that are unattested in the list of verbal concepts but found in the Kambaata off-list verbal lexicon, i.e. in the entire documented verbal lexicon ( §4.5: double derivation, §4.6: lability).For an overview of the relevant comparative-typological literature on noncausal/causal alternations that serves as the background to the following description of Kambaata, the reader is referred to the Introduction to this special issue.

Data
Haspelmath's (1993) list of 31 verb pairs served as the starting point for the analysis of noncausal/ causal alterations in Kambaata (see Appendix A).While the list is useful to ensure a certain degree of cross-linguistic comparability, its use posed three types of problem for Kambaata: -For the verb pair 'sink (intr./tr.)', a Kambaata translational equivalent was difficult to determine.
Given these complications, the resulting Kambaata verb list contains 44 verb pairs, which are categorized into morphological coding strategies -causativization, decausativization, equipollence, lability and suppletion -according to Haspelmath's (1993) typology and, building on this work, Creissels (this issue). 26The count for each strategy proceeds as follows: -If one verb pair has one pair of Kambaata translational equivalents, then the applied strategy counts as 1; -If one verb pair has two pairs of Kambaata translational equivalents and if the applied strategies in these pairs are different, each strategy counts as 0.5 (e.g.pair 14a/b: 'be destroyed/destroy'); -If one verb pair has three pairs of Kambaata translational equivalents and if the applied strategies in these pairs are all different, each strategy counts as 0.33 (e.g.pair 9a/b/c: 'gather (intr./tr.)').
Figure 2 presents the distribution of the formal strategies across the 31 verb pairs.In the following sections, the morphological strategies attested in the verb pair list are discussed in detail and contrasted with the strategies that are found in the off-list verbal lexicon.

Causativization
Of the 31 verb pairs, 15.66 (50.5%) have the noncausal (nC) form as the underived base and the causal (C) form overtly causative-derived; as shown in selected pairs in (57).
25. harfá=y-'sink (go down in)', hurgúf=y-'sink' and qúrc=y-'sink (in water, mud, etc.)' (Alemu 2016: 450, 516 and 859).Their meaning could not be verified, as they are unknown to the speakers I worked with.26.There are slight differences in the terminology of coding strategies between Haspelmath (1993) and Creissels (this issue).Haspelmath speaks of "anticausative alternation" and "labile alternation" where Creissels uses "decausativization" and "ambitransitivity".In all but one causativized form in the list of verb pairs, the short causative (caus1) is used, which is expected given the preference of caus1 with intransitive verbs ( §3.2).The only verb pair in the list in which the long causative (caus2) is attested is tar-/tar-siis-'spread (intr./tr.)(e.g.disease)'.The lexeme tar-'spread (intr.)'belongs to the lexically determined group of intransitive verbal lexemes requiring caus2 -siis and is thus comparable to the motion verbs in (8).Examples in my corpus and in Alemu  (2016: 140, 383) show that tar-'spread (intr.)'marks the source of motion by the ablative and the goal of motion either by the locative (58) or dative case or the directional marker =b-á.The intransitive use of the underived form is illustrated in (58), where the understood subject of tar-is the grammatically masculine noun 'disease', which serves as the head noun of the relative clause.The causative form tar-siis-'spread (tr.)' in (59) has 'sick enset plant' as its subject and moos-ú 'disease' as the direct object; the goal of motion ('healthy plants') is expressed in a directional phrase marked by the enclitic =b-á.

Decausativization
Of the 31 verb pairs, 10.91 (35.2%) have the causal form as the base, while the noncausal form is overtly passive-derived; see the passive morpheme -am in the noncausal column in ( 60).

Equipollence I: Light verb change strategy
The equipollent correspondence type splits up into two subtypes, the light verb change strategy discussed here and the double derivation strategy discussed in §4.5.Of the 31 verb pairs, 2.47 (7.8%) are expressed by ideophones.The noncausal/causal pair shares an underived ideophonic root, but each member of the pair makes use of a different light verb.In intransitive clauses, ideophones combine with the light verb y-'say', in transitive clauses with the light verb a'-'do' (64); 27 the light verb then encliticizes to the ideophone.
With the exception of one example, 29 the noncausal members of the doubly derived pairs are always middle-derived -and not passive-derived, as one could have expected from the analysis of the 31 verb pairs, where passivization is by far the dominant decausativization strategy ( §4.3);The causal counterparts of the doubly derived pairs are always causative.The middle derivation is realized as -aqq or -' /Ɂ/, according to the rules laid out in §3.3; see, e.g., odd-aqq-'get dressed, put on', afuu<'>ll-(< afuul-Ɂ-) 'sit (down)' and nubaaph-ph-(< nubaab-Ɂ-) 'become old'.Apart from noncausal members being marked by the productive middle derivation, there are also six noncausal verbs in Appendix C that carry the unproductive (frozen) middle suffix -a', e.g.buqq-a'-'be uprooted' (59). 30he majority of doubly derived verbs mark the causal form by the short causative (caus1), which is realized with different allomorphs (67) ( §3.1.3).
Only five doubly derived pairs take the long causative (caus2), among them two verbs from the domain of emotion (68); compare with the examples in ( 9). ( 68) amu<'>rr-'get angry, annoyed' amur-siis-(caus2) 'make angry, annoy' osa<'>ll-'laugh' osal-siis-(caus2) 'make laugh' Double derivation is also found with denominal verbs.Kambaata has a (semi-)productive 31 derivational process that generates noncausal/causal verb pairs from nouns.The members of these pairs are marked by -eeh and -ees, respectively, and can generally be translated as 'turn (intr.)into N, develop/obtain N, have N applied' for the noncausal counterpart and 'turn (tr.) into N, make develop/obtain N, apply N' for the causal counterpart (69).My database contains 18 denominal verb pairs of this type and more than 30 denominal verbs attested with either the -eeh-or the -ees-derivation (and potentially being members of a doubly derived pair of which the missing counterparts simply represent accidental gaps in my database).

Lability
The list of 31 verb pairs includes no labile verbs, which is unsurprising given the overall rarity of this verb type.To my knowledge, Kambaata has only a single labile verb, namely the temperature verb gid-'1.become non-tactile cold, 2. cause to feel non-tactile cold', whose use is discussed in Treis &  Deginet (2019: 244ff.). 32The noncausal, intransitive use with the geographic stimulus ('Hangacca') as the subject is illustrated in (70).The causal, transitive use in which the experiencer is encoded as the direct object is seen in ( 71).The experiencer object ('her') is expressed by an object suffix on the verb 'become/cause to feel non-tactile cold'; if it were expressed in an independent object (pro)noun phrase it would be marked for the accusative case.The unexpressed masculine subject of the temperature verb in ( 71) is understood to be ír-u 'weather, (atmospheric, political) situation'.

Suppletion
Two pairings in the verb list are suppletive.It is not unexpected that reh-'die' and sh-'kill' form a suppletive noncausal/causal pair, as this is common in the languages of the world (see, e.g., Haspelmath  1993: 106).In Kambaata, 'die' and 'kill' are mutually suppletive, which means that 'die' cannot be morphologically causativized (*reh-is-/*reh-isiis-) and that 'kill' (*sh-am-) 33 cannot be passivized.The only acceptable translations of an English passive phrase such as 'a man who was killed' are given in (72).In every verbal paradigm, Kambaata has a verb form that is used for honorific third person subjects ('s/he (respected)') or for impersonal subjects ('one') (see Table 2); it is this verb form that occurs as the second variant in (72).

Conclusion
This paper has provided a first description of the productive processes of word class-maintaining verbal derivation in Kambaata.The description has centered on the tricky and hitherto unresolved distribution of the short and long causative, and has shown that the choice of caus1 (-s) and caus2 (-siis) is partly grammatically, partly lexically determined.Whereas most intransitive verbs are causativized with caus1 and most transitive verbs with caus2, the numerous exceptions to these general rules make it necessary to record the causative forms of many verbs in the dictionary.In contrast to the passive and the middle derivation, causativization gives rise to a number of derived verbs whose meaning and argument structure cannot easily be predicted -an aspect which should also be registered in future dictionary entries of these verbs.Without known exceptions, all transitive verbs can be passive-derived with -am.Passive verbs take the direct object (but never the indirect object) of the corresponding underived verb as their subject; the agent usually remains unexpressed in the passive construction.Kambaata does not formally distinguish between passives and anticausatives.The middle derivation is marked by the phonologically conditioned allomorphs -' /Ɂ/ and -aqq.Unlike the causatives and the passive, it does not reduce the valency of the verb.In its most productive use, the middle morpheme marks that the subject is the beneficiary of the verbal action.Furthermore, it is found on verbs of grooming and bodily care but is rarely the only reflexivizer in prototypical reflexive situations.
The analysis of the verbal derivation system set the stage for the study of the expression of noncausal/causal alternations in the second part of the paper.If Haspelmath's list of 31 verb pairs is taken as an instrument to determine the preferred morphological patterns to express these alternations in Kambaata, the causativization strategy turns out to be dominant (50.5%), with the decausativization strategy (which is essentially a passivization strategy) following closely behind (35.2%).Of the two common equipollent marking patterns, the light verb change strategy is represented in 7.8% of the pairs, namely when verbal concepts are expressed by morphologically invariant ideophones that require the light verbs y-'say' and a'-'do' to inflect.Off list, another equipollent marking pattern occurs: For about two dozen, frequently used middle-and causative-derived verb pairs, no base verb is attested -it is unknown whether it was lost or never happened to exist.The translational equivalents of the Haspelmath list also contain two suppletive noncausal/causal pairs, namely 'die'/'kill' and 'finish (intr./tr.)'.
While the Kambaata version of the Haspelmath list gives a good first impression of the expression of noncausal/causal alternations, the picture becomes much more varied when the entire documented verbal lexicon of Kambaata is considered.A larger lexical database sheds light on the actual importance of the equipollent strategies.Not only are none of the doubly derived verb pairs captured by the Haspelmath list but the significance of the light verb change strategy is also underrepresented.Kambaata can, however, be shown to express a very significant portion of verbal concepts through lexemes belonging to the word class of ideophones, which apply the light verb change strategy to differentiate between the noncausal and causal member of a verb pair.
Creissels' exploratory study (this issue) is based on a convenience sample of 30 African languages.While lability, or in his words ambitransitivity, ranks first among the noncausal/causal strategies in his study, it is notably absent from the Kambaata list.In this case, the extension of the database from the list of 31 verb pairs to the entire documented verbal lexicon does not substantially alter the assessment, because according to the current state of knowledge, Kambaata has only a single labile verb.In other languages of the Horn of Africa that are represented in the World Atlas of Transitivity Pairs and/or Creissels' exploratory study, lability is equally uncommon: Sidaama (Cushitic) 1; Wolaitta (Omotic) 0; Amharic (Semitic) 1; Afar (Cushitic) 0 (Kawachi 2014; Wakasa 2014a and 2014b; Creissels this issue).Kambaata thus seems to confirm a tendency observed in other Ethiopian languages.Furthermore, it shares with Sidaama (Kawachi 2014) a clear preference for causativization (Sidaama: 46.7%, Kambaata: 50.5%) and decausativization (Sidaama: 28.5%, Kambaata: 35.2%). 36The only major difference between the two closely related languages concerns suppletive verb pairs, of which Sidaama has as many as five in the Haspelmath list.
Concerning the preferred formal strategies to express noncausal/causal alternations, the similarities and differences between Kambaata and the other languages of its genetic group or of its linguistic area remain to be explored.Among many potentially interesting aspects, the distribution of the suppletive noncausal/causal alternation of 'finish' in languages of the Horn of Africa would merit examination on a larger scale.43.Literally: 'find each other' (reciprocal)/'make find each other' (causative of reciprocal); see dag-'find' > daq-q-'find oneself' (mid) > daq-q-am-'find each other' (mid-pass) > daq-q-an-s-'make find each other' (mid-pass-caus1).

Figure 2 -
Figure 2 -Derivational strategies in the Kambaata verb pair list.

Table 1 -
Case paradigm of two masculine nouns and a feminine noun in selected declensions

Table 2 -
Paradigms of the imperfective main verb and the imperfective converb: fan-'open (tr.)' Whereas the causativization in the above examples leads to the addition of a causer, no increase in participants is observed in (39)-(42).Instead, the arguments of the causativized verb are chosen from a different semantic field, as shown for transitive base verbs in (39).The underived intransitive verb in (40) takes íru 'land' as the subject, whereas the causative verb has a human subject.The causativization in (41) enables the verb to take a simulative clause as a complement.

The Kambaata list of 31 verb pairs
This is a list of 31 noncausal/causal verb pairs and their Kambaata translational equivalents, followingHaspelmath (1993).The coding strategies (column 5) are abbreviated as: C = causativization, D = decausativization, E = equipollence, S = suppletion; verbal derivational morphemes are highlighted in bold.The data is written in the official Kambaata orthography (cf.Treis 2008: 73-80; Alemu 2016).The middle verb xaacc-is especially used for people coming together/assembling.39.It is difficult to pick a translational equivalent for English spread, especially because the English verb is so polysemous.Here only verbs with the meaning of 'spread = scatter, disperse' and 'spread = circulate' are given; verbs with the meaning 'spread = straighten out, roll out, unfurl' and 'spread = smear, plaster' are excluded.