Lexical and syntactic causatives in Oromo

La formation des causatives en galla est assuree par affixation d'un morpheme specifique, le nombre des morphemes dependant du nombre de sujets logiques dans la proposition. L'auteur montre les effets distincts en galla d'un meme processus morphologique, au niveau du lexique et de la syntaxe. Etude menee dans le cadre de la grammaire relationnelle

1.In Oromo (formerly known as Galla), a Cushitic language of Ethiopia and Kenya, the formation of causatives is accomplished through the affixation of a bound morpheme -s (with surface variants /c/, /is/, and /iis/) and the addition of an appropriate argument:'1 (1) a. aannan-ni daanf-e.The affixation of an additional -s and an argument to an already causativized form results in a multiple causative.
NOM milk boil-cAU-cAu-AGR 'Terfa made Gamtesa boil the milk.' As was shown in Lloret 1987, verbs can be classified according to whether their initial causative forms require one -s morpheme or two (the vowel /i(i)/ between the two -s's in -si(i)s is added by a rule of epenthesis).
* An earlier version of this paper was presented at the 18th Conference on African Linguistics in Montreal in April, 1987.We would like to thank Addisu Tolesa for his invaluable help and cooperation in the compilation and interpretation of the Oromo examples which were elicited between 1985 and 1988 at Indiana University.We would further like to express our gratitude to the following individuals for their valuable comments and suggestions: Robert Botne, William Bright, Greville Corbett, William Davies, Geoffrey Pullum, and Laurie Tuller.We also acknowledge the constructive criticisms provided by the anonymous reviewers for Language.All errors and omissions are the responsibility of the authors.
NOM milk boil-CAU-AGR 'Terfa boiled the milk.' (1 agentive) Hayward's proposal that the number of -s's matches the number of agentive nominals in a sentence also explains the formation of multiple causatives wherein an -s morpheme is affixed to a causativized form and an additional agentive argument is added.

NOM boy run-CAU-CAU-AGR
'Gamtesa made Terfa make the boy run.' (3 agentives, 3 s's) 1.1.While Hayward's account is appealing at first sight, it soon runs into problems.First, it is not the case that the subjects of all so-called 'agentive' intransitives-'cry', 'swim', 'be angry', etc.-can be characterized as actually manifesting the semantic/thematic role 'agent'.For example, from a semantic/ thematic perspective, the subject of the 'agentive' intransitive Deekam-'be angry', which forms its causative with two -s morphemes, does not appear to be any more of an agent than the subject of the 'non-agentive' intransitive fayy-'be healthy', which takes only one -s.
(6) Deekam-'be angry' -> Deekamsiis-'make angry' fayy-'be healthy' -> fayyis-'make healthy' Furthermore, the subjects of transitive verbs of perception ('hear', 'see', etc.), which in most accounts are understood to have the thematic role 'experiencer' rather than 'agent', also form their causative with two -s morphemes.1.2.Second, one finds that the addition of an -s morpheme to a causativized form does not always correspond to the addition of another agentive nominal.The additional -s in 8b, for example, serves to intensify the nature of the causation rather than to add a further causative.

NOM boy sleep-CAU-INT-AGR
'Terfa made the boy sleep (e.g. by giving him a sleeping pill).' 2.1.The first problem can be handled by looking at the number of -s morphemes in causative constructions in terms of grammatical relations rather than of thematic roles.That is, we are still operating in the spirit of Hayward's approach, but we are suggesting that the correlation is really between -s morphemes and grammatical subjects rather than between -s morphemes and thematic agents.Our proposal is that the verbs which form their causatives via a single -s morpheme are all unaccusatives, i.e. verbs whose single argument is initially a direct object; by contrast, verbs which form causatives with two -s morphemes are either transitive (verbs that select an initial subject and direct object) or unergative (verbs that select an initial subject only).Using the relational notation of 1 for subject, 2 for direct object, and P for predicate, the relational networks of clauses containing each of these three classes of verbs are represented in 9.In 9c, the initial direct object of daanf-'boil' advances to subject to satisfy the universal requirement that every basic clause have a final subject (the Final 1 Law; Perlmutter 1980, Johnson & Postal 1980);4 underlyingly, daanf-'boil' and other verbs of this class are subjectless.While the intransitive sentences 9b and 9c are structurally parallel on the surface, underlyingly they are presumed to be grammatically distinct.Note that in our analysis a verb such as Deekam-'be angry', which takes two -s morphemes in the causative, differs from a verb such as fayy-'be healthy', which takes only one -s, not in any intrinsic semantic agentivity of its surface subject but in its language-specific categorization as an unergative as opposed to an unaccusative verb.5The causative construction under examination here is taken to be a union construction in the sense of Davies & Rosen 1988.In the formation of a causative, the morpheme -s, which is itself a predicate, is introduced into the clause along with an additional subject argument.The subject of the embedded predicate (when present) is uniformly revalued as a direct object.We can attribute the realization of each -s morpheme in a causative to the presence of a subject in the clause.Thus, since the verbs Dug-'drink' and iig-'run' each introduce their own subject nominals (in addition to that added by the causative morpheme itself), their causatives show up with two -s affixes.In the case of the construction formed with the unaccusative predicate daanf-'boil', only the causative predicate -s introduces a subject, and therefore, only one -s affix shows up.
2.2.This analysis does not address the second problem for Hayward's treatment of causatives, namely, the existence of causative constructions which add an -s but do not add an agentive argument.The explanation here lies in an understanding of the CAUSATIVE-INTENSIVE construction, which, with the ex-ception of Lloret 1987, has been neglected in the Oromo literature.Attention to the causative-intensive construction, moreover, provides important evidence that allows one to distinguish more accurately the nature of causative phenomena in Oromo.While -s is usually described (and usually functions) as a causative morpheme tied to the number of underlying grammatical subjects, it can, as was shown above in 8, serve as an intensifier (INT).
NOM boy milk drink-CAU-INT-AGR 'Toltu made (forced) the boy to drink the milk.'There are, however, strict limitations on the use of-s in its intensive function.
While the causatives of all transitive verbs and all unergative intransitives allow the addition of the INT, only SOME unaccusative verbs (verbs that form their causative with a single -s) do so.For example, the addition of -s to the causative gogs-'make dry' can result in a double causative or an intensive, depending on the number of overt arguments.The double causative form c'absiis-'make someone make something break', on the other hand, permits the addition of the INT -s.Thus, the affixation of -s in 15 may, but need not, add an argument.be attached directly to simple stems, as is possible with polysemous causative/ intensive formations in Semitic (cf.Moscati et al. 1964:124), argues against the treatment of INT -S and CAU -s as distinct homophonous morphemes (cf.Hamp's 1985:64 claim to this effect for Sanskrit).An analysis more in keeping with the nature of Oromo and other Cushitic languages is that the causative -s and the intensive -s are essentially the same morpheme, but that they are introduced by distinct processes and have different grammatical consequences.The CAU -s is added directly to any stem, where the number of -s's correlates with the number of subjects.The addition of an -s to a causative verb requires the concomitant addition of a subject argument and produces a multiple causative.
The causative-intensive construction, on the other hand, strengthens the nature of the causation by reduplicating the -s found on a causative verb.6Since the process of reduplication-intensification does not involve adding a new, independent CAU morpheme, it does not entail the addition of a new subject argument.This explains the seemingly puzzling non-fit in intensive-causative constructions between the number of -s's and the number of grammatical subjects represented.Our characterization of the causative-intensive as a reduplicative process finds further support in cross-linguistic comparisons.Like Oromo, a number of Southern Bantu languages-among them Nguni, Sotho, Venda, Tsonga, and Shona-form their causative by inflecting a verb stem with the morpheme -is or -es (Doke 1954).7 Some examples from Sotho are provided in 18 (Doke 1954:140-41).
(18) utioa 'hear' -> utloisa 'cause to hear' lula 'sit' -> lulisa 'set down' boolaea 'kill' -> boolaisa 'cause to kill' In some of these languages, the intensifier is often a reduplication of the cau-6 Adding an -s and reduplicating an -s produce verb stems with the same surface form, as in examples 13 and 15; but this doesn't affect the fundamental difference in the processes involved.While 2 + 2 and 2 x 2 result in the same answer, they nevertheless manifest distinct operations.
7 The observations on the parallels between Oromo and the Southern Bantu languages discussed here are due to Robert Botne.The similarity in the form of the causative morphemes in Cushitic and in Bantu is presumably fortuitous.
(19) Causative Intensive Nguni: -isa -isisa Sotho: -isa -isisa Tsonga: -isa -isisa The fact that this is reduplication and not a case of coincidental homophony between the causative marker and the intensive marker becomes further apparent when one examines other Southern Bantu languages.In Oshiwambo, for example, which has the causative morpheme -ik, we find the intensifier -ikik (Viljoen 1978).In Swazi there are three coexistent forms of the intensifier, all of them reduplications: -isis, -elel and -ulol (Ziervogel 1952).Nor is the phenomenon of reduplication encoding intensity peculiar to African languages.Moravcsik (1978:321-22), in a general discussion of reduplicative constructions, cites reduplicative intensifiers in Turkish, Sundanese, Agta, Telugu, Thai, and Tagalog.
3. Given that the INT is necessarily restricted to causative verbs, since it is formed by reduplicating an existent -s causative morpheme, one might question whether those apparent causatives to which the intensifier cannot be attached (e.g.14b) are truly causatives.Suppose that the causative predicate -s, in addition to freely attaching to nearly any clause, also combines derivationally in the sense of Lieber 1981 with a limited class of verbs.8This predicate, like any other, has a lexically specified valence.It minimally requires that its clause contain at least a subject and a direct object.We might represent this valence as in 20.
(20) -s : [1 2] Thus, if causative -s combines in the lexicon with a verb such as c'ab-'break', the resulting derived verb c'abs-will have the structure in 21.

(21) [v [v c'ab ] s ]
Since it is the head of the derived word, any lexical features on the affix -s will be features of the derived form.Since the only relevant lexical feature carried by -s is its valence, it functions derivationally as a transitivizing affix, deriving a transitive verb c'abs-'break' from the intransitive verb c'ab-'break' .9 8 This lexical alternation between unaccusative and transitive predicates corresponds to what Marantz (1984:179-93) calls 'anticausativization'.In support of his claims, we have noted that this alternation is NOT lexically productive and is restricted to a 'semantically coherent class of verbs'.9 Interestingly, Jamison 1976 has argued that the Sanskrit 'causative' formative was originally a transitivizer and that it only later developed into a causative that could be used with transitive stems.In Hausa, the 'causative' morpheme (termed 'efferential' in Newman 1983)  In 22, the unaccusative verb c'ab-'break' is seen to be categorized for an initial 2 (direct object) only.When we derivationally attach the CAU morpheme -s, its subcategorization requirements (an initial 1 and 2) are imposed on the derived form c'abs-in accordance with Lieber's Feature Percolation Convention (FPC) II.The meaning of the derived form is otherwise determined by the embedded stem c'ab-in accordance with FPC III.10 Thus, the formation of gogs-'make dry' involves the SYNTACTIC affixation of causative -s to a clause, but the formation of c'abs-'break (trans)' involves the LEXICAL affixation of causative -s to a verb stem.Accordingly, 23a is a causative union construction as expressed by its relational network, while 23b is a simple transitive clause containing a single predicate.11( 23 This analysis claims that gogs-can take an intensifier because it is a syntactic causative, while c'abs-cannot take an intensifier because it is an uncausativized verb stem.However, as shown earlier in 15, c'absiis-, a syntactic causative built on the transitivized stem, can take an intensifier.In what follows, we will provide two additional motivations for the proposed lexical/syntactic distinction. 10 The Feature Percolation Conventions referred to here are as follows (Lieber 1981:49-50): Convention II: All features on an affix morpheme including category features percolate to the first branching node dominating that morpheme.
Convention III: If a branching node fails to obtain features by Convention II, features from the next lowest labeled node are automatically percolated up to the unlabeled branching node.
n The situation in Amharic described by Gragg 1970 is very similar to the one described here for Oromo.The big difference is that in Amharic the 'lexical' and the 'syntactic' causatives use different markers, a prefix a-for the former and a prefix as-for the latter.3.1.The first argument comes from the behavior of analytic causatives.In addition to the synthetic causative construction discussed above, Oromo has an analytic causative formed by the verb goD-'make' and an embedded clause introduced by the complementizer akka.This construction contains an independent embedded clause, and the arguments of the complement clause maintain their normal case marking.The subject of goD-and the subject of the embedded verb are both marked nominatively.In 24, the synthetic and analytic causatives are contrasted.
(24) a. terfaa-n gamteessaa leenc'a ajjees-(s)is-e.The verb goD-'make' can itself be inflected to indicate intensive, i.e. goaisiis-'make [+ INT]'.12This inflected form of goD-can, expectedly, take a complement clause to form an analytic intensive-causative.However, in the same way that the intensive morpheme -s was seen to affix only to causativized verb forms, the intensive form of the verb 'make' goccisiis-only allows a causativized complement clause.The result is necessarily an intensive double causative.In 26a, we see that goccisiis-cannot take an uncausativized Our analysis predicts that the apparent causatives that we have characterized as being lexically derived should behave with respect to the selectional requirements of goccisiis-'maket + INT]' as non-causativized verbs.This prediction is borne out in a comparison of the verb daanf-'boil', which forms a causative with -s (i.e.adds -s syntactically), and the verb c'ab-'break', which combines with -s derivationally (i.e.adds -s lexically).Significantly, in 28b c'abs-'break (TRANS)' cannot appear as the complement of goccisiis-; that is, it behaves as a non-causativized verb.If, however, c'absitself is causativized (i.e.adds an -s syntactically), then the clause containing it CAN be a complement of gocdisiis-, albeit a rather complex one.3.2.The second outside motivation for the dichotomous lexical/syntactic analysis comes from the morphology of the causative itself.The processes of causative formation outlined above and described in detail in Lloret 1987 lead to further significant evidence in support of this analysis.We have already noted that: (1) simple transitive verbs are causativized by the addition of two -s morphemes, while (2) if causativized verb forms are further causativized, they receive a single additional -s.Given our analysis of lexically produced 'causatives' as grammatically simple transitive verbs, the prediction stated in verify.There is a morphophonemic rule in Oromo that reduces one of the /s/'s when an affix beginning in /s/ is added to a stem that ends in /s/.Accordingly, adding -si(i)s to a transitive verb stem such as c'uqaas-'shake' results in the form c'uqaasis-'make shake'.Now, if one examines the causativized form of the proposed transitivized verb c'abs-'break (TRANS)', it is impossible to determine whether the resultant form c'absiis-contains the suffix -siis, which has been phonologically reduced (as would be expected of a transitive stem), or whether it contains -iis (as would be the case with syntactically causativized forms): (32) c'abs-+ -siis -c'absiisbreak(TRANs)-CAU c'ab-s-+ -uis -c'absiisbreak-CAU-CAU Fortunately, there do exist forms which allow the claim to be tested.If -s is added to a verb stem ending in /1/, it becomes /c/ in accordance with a general palato-affricate rule (Lloret 1987).The affixation of -s to /l/-final verb roots results in examples such as in 33.
(33) gal-+ -s -* galc-'return home' 'bring home' bul-+ -s -* bulc-'spend the night' 'make spend the night' In our framework, galc-behaves like c'abs-'break (TRANS)' and is taken to be a transitivized verb stem, with -s added in the lexicon.By contrast, bulc-, which behaves like gogs-'dry something', is a causativized form to which -s has been added syntactically.Since the s-reduction rule 31 only applies to the juxtaposition of surface /s/'s, it does not apply in these cases, and thus the prediction made in 30 can be tested (i.e., one can tell whether one or two -s morphemes have been added).Consistent with our expectations, we find that the verb form galc-does indeed behave like a simple transitive verb stem in that it causativizes by the affixation of -siis.
(35) bulc-+ -s -> bulciis-'make spend the night' CAU 'make someone make someone (causativized INTR) spend the night' These observations are corroborated by the fact that bulciis-can also be interpreted as the intensive of bulc-, depending upon the number of appropriate arguments present.
-(s)ase (cf.Jacobsen 1982:197-206).The examples in 39 illustrate some of the morphological patterns exhibited in Japanese transitive/inchoative pairs.(39) Intransitive Transitive yakeru 'burn' yaku 'burn' aku 'open' akeru 'open' agaru 'rise' ageru 'raise' kieru 'go out' kesu 'extinguish' odoroku 'be surprised' odorokasu 'surprise' In Oromo, on the other hand, the transitivizing affix and the causative affix are the same.The lexical affixation of -s was shown to produce a change in the argument structure of the resultant verb stem.Such an effect is in accordance with the Projection Principle (Chomsky 1981:29), which states that argument structure changes can only occur in the process of word formation.Further, these argument structure changes are exactly what would be expected when one combines the causative -s lexically with a verb stem according to Lieber's 1981 Feature Percolation Conventions.This lexical/syntactic distinction was supported by the behavior of the causative-intensifier.We characterized the intensifier as a reduplication of the causative morpheme.Remembering that some verb forms inflected with a single -s morpheme could be intensified, while others could not (cf.13 and 14), we attributed this difference to the effect of adding -s lexically in the first case and syntactically in the second.The reason that c'abs-'(make) break' cannot be intensified is that -s has been added to c'ab-derivationally to form a new verb stem.There is no independent causative morpheme in this construction for a reduplicative rule to act on.Thus, the form c'absiis-cannot mean '(make) break [+INT]'.On the other hand, since gogs-'make dry' is a syntactically produced causative, the reduplicative rule CAN operate on -s alone to produce gogsiis-'make dry [ + INT]'.
A final point should be made with respect to the processes described.We have noted that the causative morpheme -s can combine lexically with a limited class of predicates; that -s can combine syntactically with the output of this derivation; and that the syntactically affixed causative morpheme can then undergo reduplication to form a causative-intensive.It is generally recognized that reduplicative morphology operates in the lexicon (Lieber 1981, Moravcsik 1978).If this is the case, then the reduplicative rule of causative intensification would have to take effect in the lexicon even though it is fed by the output of syntactic rules.The Oromo phenomena discussed in this paper thus provide support for the claim made by Borer 1985 that morphological rules can potentially take effect at any stage in a grammatical derivation.
The embedded m-This content downloaded from 129.79.38.50 on Tue, 26 Apr 2016 14:27:38 UTC All use subject to http://about.jstor.org/termspredicate and the embedded direct object of a transitive predicate lose their relational status and