2.1 The clausal syntax of Shona

I follow Bliss and Storoshenko (Reference Bliss, Storoshenko and Jones2008) in assuming that the pre-verbal subject position in Shona is actually the reflex of information-structure–driven ${\rm \bar A}$movement to a Topic Phrase.Footnote ³ The language shows a strong discourse-sensitivity in the subject position, to the extent that all consultants reject the simple declarative (4a) below as the answer to a subject wh-question Who cooked the pumpkins? The implementation is similar to existing accounts for Kinande in which the subject position is argued to be an information-structure–sensitive head in an augmented CP-domain rather than in TP (Baker and Collins Reference Baker and Collins2006, Schneider-Zioga Reference Schneider-Zioga2007). Based on this, I adopt (4b) as the basic architecture of the Shona clause:

1. (4)

   1. a. Peter a-ka-bik-a ma-nhanga.

      Peter sub.1-rpst-cook-fv 6-pumpkin

      ‘Peter cooked pumpkins.’

   2. b.

The v-head, which is also the spellout location of the passive voice suffix -w in the original Bliss and Storoshenko analysis, here introduces the external argument in its specifier. In the active voice, v has no overt exponence in their analysis. The external argument checks a topic EPP, and the class one ϕ-features of the proper name Peter are spelled out as agreement on the Topic head.

2.2 Object marking in Shona

The nature of object marking in Bantu languages has been a subject of considerable debate over several decades of research. In brief, three positions have been advanced:

• • Object markers are incorporated pronouns moving from argument positions

• • Object markers spell out agreement between a head at the edge of the vP domain and an internal argument

• • Object markers are clitics associated with a DP in the relevant argument position

The only one of these positions that is incompatible with my proposal is the first; my contention is that there is a covert anaphor in the argument position of a reflexive sentence, not that zvi- itself is an anaphor originating in the argument position which is subsequently incorporated. A potential counter-argument to the incorporation analysis is the fact that object markers in Shona can co-occur with a full DP object:

1. (5) Mufaro a-$\emptyset $-ri-bik-a bota.

   Mufaro sub.1-past-obj.5-cook-fv porridge.5

   ‘Mufaro cooked porridge.’

In Sikuku (to appear), it is asserted that if such doubling is possible, then the object marking is agreement-based, with the full DP in the argument position. However, the analysis for Chicheŵa presented in Bresnan and Mchombo (Reference Bresnan and Mchombo1987), where object markers are taken to be incorporated, contains parallel examples with the object marker and the full DP object, so doubling in itself may not be a solid argument against the incorporation analysis. Bresnan and Mchombo observe that in Chicheŵa, a DP argument (subject or object) which is also marked on the verb may be freely positioned in the sentence, essentially acting as an adjunct. As subject marking is obligatory in the language, two base word orders are always possible:

1. (6)

   1. a. Njûchi zi-ná-lúm-a alenje.

      bees sub-past-bite-indic hunters

      ‘The bees bit the hunters.’

   2. b. Zinálúma alenje njûchi. (Bresnan and Mchombo, Reference Bresnan and Mchombo1987: 744–745)

(6a) represents the canonical SVO word order of the language; VOS order is derived by treating the subject marking as an incorporated pronominal, with the subject DP adjoining to the right of VP. When an object marker corresponding to alenje ‘the hunters’ is added to the inflected verb, all six logically possible permutations of S-V-O are attested.

The evidence for Shona is somewhat different. Among my consultants, VOS order with no object marker is judged inconsistently (some consultants accept the order, others reject it). This suggests that in Shona, at least for some consultants, subjects cannot be taken to be adjuncts. The most relevant word order to consider with respect to object marking then is the OSV order, which is the only permutation of elements that requires an adjunct treatment of the object but not the subject under Bresnan and Mchombo's analysis. Again, judgements vary between consultants. Some reject the order completely, while others accept it with or without an object marker. Interestingly, it is the consultants that accept VOS who also accept OSV without an object marker, providing no evidence for an incorporation analysis. Mugari (Reference Mugari2013), using similar word-order tests, reports all six S-V-O permutations to be possible in Shona without object marking, subject to intonation and genre considerations. He argues that Shona is fundamentally an agreement-based system, though he suggests a change in progress toward a pronominal incorporation system as proposed for Chicheŵa, potentially accounting for the variation among my consultants.

While the original Bresnan and Mchombo proposal is not presented in a Principles-and-Parameters framework, the core idea is compatible with the pronominal argument hypothesis (PAH), which is adaptable to the framework (Jelinek Reference Jelinek1984). To conclude that Shona is not a pronominal-argument language, an argument proposed by Davis and Matthewson (Reference Davis and Matthewson2009) against assuming PAH in St’át'imcets (Salish) can be applied. The key observation is that, contrary to a prediction of the PAH, St’át'imcets has some obligatorily unregistered arguments that are not marked in any way on the verbal stem. The same is true in Shona, where object marking is not obligatory, and for any predicate with two (or more) internal arguments, all but one must be unregistered, as Shona strictly allows only one object marker per verb. Furthermore, in such cases, there is complete optionality in terms of which internal argument is registered:

1. (7)

   1. a. John a-ka-bik-ir-a Bill bota.

      John sub.1-past-cook-appl-fv Bill porridge.5

      ‘John cooked porridge for Bill.’

   2. b. John a-ka-mu-bik-ir-a bota.

      John sub.1-past-obj.1-cook-appl-fv porridge.5

      ‘John cooked porridge for him.’

   3. c. John a-ka-ri-bik-ir-a Bill.

      John sub.1-past-obj.5–cook-appl-fv Bill

      ‘John cooked it for Bill.’

The facts in (7), showing that either internal argument in a transitive extended with an applicative may be replaced with an object marker, demonstrate that there is no unique role or position that must be registered. Based on this, it can be concluded that Shona is also not a pronominal-argument language, and we may move forward with the notion that object markers derive from agreement rather than from the incorporation of a pronominal.

Agreement accounts of Bantu object markers share the following elements: some head at or near the edge of the vP domain probes downward for matching ϕ features, which are spelled out on the probe as a prefix immediately to the left of the verb root. It is important at this point to recall that object marking in Shona has a very clear discourse function: it is used to highlight discourse-old referents. To account for this within a broadly minimalist framework (Chomsky Reference Chomsky1995), agreement can be said to operate as follows: a head X bears unvalued ϕ-features, along with an uninterpretable feature F. X probes its c-command domain to locate a YP bearing a matching F feature. For the time being, it will suffice to understand F as being a discourse-related feature which has been variously defined as definite (Mugari Reference Mugari2013), or as topic (Bax and Diercks Reference Bax and Diercks2012). In checking F, the ϕ-features of YP are copied to X. To account for the optionality of object marking in Shona, the uninterpretable feature may or may not be present on the head X. The object bears interpretable versions of this feature, and does not itself require agreement for any licensing. The presence or absence of the probe can be seen as a manifestation of a discourse function highlighting the fact that the checked internal argument carries this F-feature, analogous to the optionality of a feature that may trigger left-dislocation of a topic in English.Footnote ⁴ The basic mechanics are illustrated in (8):

1. (8)

Where accounts differ is on the precise implementation of this mechanism in terms of identifying the X head. In some analyses, dedicated agreement heads are proposed above vP: Visser (Reference Visser2008) for isiXhosa, Riedel (Reference Riedel2009) for Sambaa. Other accounts use v itself as the spellout position, such as Adams (Reference Adams2010) for Zulu. For Shona, v may be the better choice for the position of the marking, ensuring that the relevant c-command domain for potential object marking will not contain the external argument generated at the specifier of vP. However, one consideration to keep in mind is the fact that object marking is possible in passive clauses in this language:

1. (9) Bota ra-$\emptyset $-mu-bik-ir-w-a.

   porridge. 5 sub.5-past-obj.1-cook-appl-pass-fv

2. ‘The porridge was cooked for him.’

If, as has previously been proposed, the v head is the spellout position of the passive suffix -w, a higher agreement head spelled out as a prefix is motivated. This bears the additional explanatory burden of blocking agreement with the external argument position. Conversely, if object marking is spelled out on v, the passive must occupy a dedicated VoiceP position immediately below v. At present, there is no need to decide between these two, as either is feasible and compatible with my analysis of the reflexive.

A recent proposal which is more amenable to a v-based analysis of object marking is that object markers in the Manyika dialect of Shona are in fact clitics. This analysis, put forward in Bax and Diercks (Reference Bax and Diercks2012), is essentially that the v head enters into an agreement relation with a complex DP containing both the argument phrase and an associated clitic. The clitic consists of nothing but ϕ-features, and is incorporated into v once agreement takes place. While their analysis does not include a feature for the discourse-old status of the object, it is clear from their discussion that Bax and Diercks acknowledge an information-structure sensitivity, describing object marking as incompatible with focus, defined as the complement of topic. An adaptation of their schema is presented in (10):

1. (10)

As shown, this approach could also remain coherent with the treatment of v itself as the spellout location of the passive suffix, as the clitic would left-adjoin to the v head, where the verbal complex will have raised, yielding an object proclitic, and the passive -w as a suffix. This account also requires no additional statement obviating object marking of the external argument.

The key commonality between the two analyses is that an agreement relation holds between the object DP and some higher head. Both the agreement analysis and the clitic analysis are compatible with my proposal that reflexive sentences in Shona contain a covert anaphor refl. Under an agreement scenario, my argument is that agreement with the covert refl anaphor, which carries no ϕ-features, would trigger class eight marking to be spelled out on the X⁰ (remaining agnostic as to the identification of that head). Under the clitic analysis, the feature bundle occupies the D head sister to the refl DP, and undergoes cliticization after agreement. I thus take no stance as to the exact mechanism and location of the spellout of the zvi- morpheme in reflexive contexts, maintaining only that it is the result of some agreement, parasitic on a discourse feature, between a higher probe located at the left edge of vP and the internal argument position containing refl.Footnote ⁵

2.3 Implementing the covert anaphor analysis

With the building blocks of the clause structure and object marking laid out, it now only remains to describe the covert anaphor refl itself in more detail. One noteworthy property of the Shona reflexive is that it is subject-oriented: in all cases, the antecedent of the reflexive must be the subject of the sentence. To express co-reference of two non-subject arguments, a more complex structure must be employed:

1. (11) Mufaro a-ka-zivis-a gudo_i ku-na-gudo_i.

   Mufaro sub.1-rpst-introduce-fv baboon to-assc-baboon

   ‘Mufaro introduced baboon_i to baboon_i.’

As shown in (11), the co-indexed arguments are both overtly expressed, with the indirect object appearing in an oblique phrase. This contrasts with a parallel case in English, where a reflexive theme can be bound by a higher internal argument:

1. (12) I showed John_i himself_i in the mirror.

A version of (11) above with zvi- replacing gudo in the object position only has one possible reading:

1. (13) Mufaro a-ka-zvi-zivis-a ku-na-gudo.

   Mufaro sub.1-rpst-obj.8-introduce-fv to-assc-baboon

   ‘Mufaro introduced himself to (the) baboon.’

Consultants report that it is impossible for zvi- to replace the oblique, making (13) the closest available approximation to the English example in (12). The judgements here are clear: (13) only has the reading where the subject is the antecedent.

Cross-linguistically, subject orientation is a common feature of long-distance anaphors, as shown in these Icelandic and Chinese examples from Cole and Sung (Reference Cole and Sung1994):

1. (14)

   1. a. Jón_i sagði Maríu_j [að ϸú elskaðir sig_i/*j/*k].

      Jon told Maria that you loved.subj   self

      ‘Jon told Maria that you loved him.’ (Cole and Sung Reference Cole and Sung1994: 359)

   2. b. Wangwu_i shuo [Zhangsan_j zengsong gei Lisi_k yipian guanyu

      Wangwu say Zhangsan give to Lisi one about

      ziji_i/j/*k de wenzhang.]

      self DE article

      ‘Wangwu says that Zhangsan gave an article about him/himself to Lisi.’ (Cole and Sung Reference Cole and Sung1994: 360)

In the Icelandic case, the local subject is not a possible antecedent for third-person sig due to a ϕ-feature mismatch; only the matrix clause subject is a possible antecedent, despite there being a third person indirect object in the matrix clause as well. Turning to Chinese, ziji shows subject orientation even in the local clause, taking only the matrix or embedded clause subjects as possible antecedents. Ziji is subject to a blocking effect, under which the anaphor can take any antecedent with the same ϕ-features as its nearest potential antecedent, unless a potential antecedent with different ϕ-features appears in a higher clause. This mismatched antecedent and all higher ones are blocked:

1. (15) Zhangsan_i renwei [wo_j zhidao [Wangwu_k xihuan ziji_*i/*j/k]]

   Zhangsan think I know Wangwu like self

   ‘Zhangsan thinks that I know what Wangwu likes himself.’ (Cole and Sung Reference Cole and Sung1994: 363)

In this case, the intervening first person subject blocks ziji from being bound by the higher clause subject. In a later analysis by Sohng (Reference Sohng2004), which I will adopt below, this is a consequence of ziji having no inherent ϕ-features, which also accounts for subject orientation in his analysis.

In contrast to these other subject-oriented examples, the Shona reflexive anaphor is strictly clause-local:

1. (16) John_i a-$\emptyset $-t-i [(imbwa_j) ya-$\emptyset $-zvi-rum-a

   John sub.1-past-say-fv dog.9 sub.9-past-obj.8-bite-fv

   refl _*i/j].

   refl

   ‘John_i said the dog_j bit itself_*i/j.’

(16) only has the reading where John is making a statement about a dog biting itself; there is no possible way to construe this as a statement of John reporting that the dog bit him. As shown, the embedded clause subject is optional, though in (16) the different subject agreement will easily allow native speakers to recover a subject for the lower clause (a class nine pro) who is not John. Matters are more complicated when the clauses match in subject agreement; (17) is reported as ambiguous:

1. (17) John_i a-$\emptyset $-t-I [ pro _i/j a-$\emptyset $-zvi-rum-a refl _i/j].

   John sub.1-past-say-fv pro.1 sub.1-past-obj.8-bite-fv  refl

   ‘John_i said that he_i bit himself_i.’

   ‘John_i said that he_j bit himself_j.’

   *‘John_i said that he_j bit him_i.’

The apparent long-distance reading can be derived from a situation in which the embedded clause subject is in fact a pro that is coreferential with the matrix subject. In this case, the anaphor remains locally bound. Similarly in the second reported reading, a contra-indexed pro is interpreted as the subject of the embedded clause, once again leading to local reflexivity. The reading where the theme of the embedded clause is co-referential with John, but the subject of that clause is contra-indexed is not available.Footnote ⁶ The Shona reflexive anaphor is therefore not only subject-oriented, but also clause-bound, which is likely one of the reasons it has been taken for a detransitivizing reflexive, a position argued against in section 4.

Cole and Sung's Reference Cole and Sung1994 analysis of long distance ziji in Chinese treated the anaphor as an X⁰ element undergoing LF head movement in search of local binding, stopping at any I⁰ head, where the anaphor would be locally c-commanded by the element in [Spec, IP]. In this way, subject orientation is predicted. However, this analysis cannot capture the subject orientation of ziji within its clause of origin, as ziji would not need to undergo any head movement to be bound locally. As a result, the local subject orientation of ziji is stipulated in the 1994 analysis.

A revision to this theory of long distance anaphora in Sohng (Reference Sohng2004) resolves this issue for ziji, linking it to the fact that ziji can in principle take antecedents with any ϕ-features, within the bounds of the blocking effect. Under this analysis, ziji is generated with no inherent ϕ-feature value, with valuation taking place in local Spec-Head agreement after LF movement. The LF movement is not treated as movement of the anaphor itself, but rather movement of a set of formal features associated with ziji to a T⁰ head bearing a [Refl] feature when selected from the numeration. This [Refl] feature drives the LF feature movement. According to Sohng, the formal features of ziji move covertly from head to head, not pied-piping intermediate heads, until they reach a T head bearing [Refl]. However, the first T⁰ position encountered determines the ϕ-features for ziji, explaining the blocking effect: if [Refl] appears on a higher T⁰ head, forcing the formal features to move beyond their valuation point, then any subject with clashing ϕ-features will crash the derivation. This analysis can be easily adapted to Shona. I propose a parallel movement for the features of refl, which have no inherent ϕ-value and must undergo the same LF movement in order to be valued. Unlike Sohng's account of ziji, I do not propose a separate [Refl] feature motivating this movement; the covert movement in Shona is driven by the lack of ϕ-features, and stops as soon as a ϕ-feature value is encountered. In this way, locality of binding is assured.

Rather than requiring us to stipulate the location of this valuation, the situation in Shona provides a principled reason for isolating the subject as the only potential antecedent. Recall that the Shona verbal stem offers two positions for ϕ-based agreement: subject marking and object marking. Subject marking in active clauses is the result of agreement with the external argument which moves to the subject (topic) position for independent reasons. Object marking is an optional function of a head at the edge of vP, identifying discourse-old material in the vP domain. I assume that these are the only two heads in which ϕ-features may be transmitted; the LF movement thus provides two opportunities for ϕ-features valuation. The object marking head itself remains an active probe though, and assuming that the anaphor is lexically-specified to carry the relevant discourse-old feature (since it must be bound, the anaphor itself can never introduce a new referent) the object-marking head selects the obligatorily bound anaphor as the most discourse-old element in its c-command domain before LF. It is from this agreement relation between the anaphor underlying the reflexive and the object-marking head that the zvi- morpheme emerges, class eight being the marker for agreement with an element that lacks ϕ-features. This lack of ϕ-features on the anaphor is also what motivates the subsequent LF movement, and allows it to be bound by an antecedent of any noun class.

As the anaphor itself is moving up the verbal projection in search of ϕ-features, it finds none at the first possible location, the object marking head, and is forced to move onward to have its features valued by the subject at [Spec, TopP]. From this, subject (topic) orientation emerges. A sample derivation of a simple reflexive (18) is given in (19), with XP used as a placeholder for the location where object agreement appears:

1. (18) John_i a-$\emptyset $-zvi-chek-a refl _i.

   John sub.1-past-obj.8-cut-fv refl

   ‘John cut himself.’

2. (19)

Dotted lines indicate probes for agreement: subject agreement is with the external argument, which is also subject to independently motivated EPP-driven movement. In the object case, agreement is with refl, triggering class eight zvi-. In its search for feature valuation, refl then undergoes Sohng-style LF movement, shown with dashed lines, with two possible valuation points: X or Top. The first option being a) featureless and b) circular (refl attempting to get features from itself), it must carry on to Top. This analysis also predicts the unavailability of matrix subject reflexives in the language, as a covert reflexive in the highest argument position could never get ϕ-feature valuation within its local clause, there being no antecedent with ϕ-features. Other class eight subjects are of course possible, but the reflexive anaphor is independently blocked from this position.Footnote ⁷

In concluding this section, one issue that remains is the possibility of generating ‘silent’ reflexives. If the reflexive anaphor is covert, and object marking is optional, one could imagine a derivation in which the reflexive anaphor is present, but fails to trigger object marking. While I have no strong position on how to block this, (at least) two solutions are possible, aligning with the choice one makes as to the nature of object marking. Recall that I have characterized object marking as the result of a probe, whose activation is optional, searching its c-command domain for an element bearing an information-structural feature of being discourse-old. Suppose that the reflexive anaphor bears the relevant feature, but an uninterpretable version of it. Unlike other DP objects, on which the discourse-old feature is interpretable, the reflexive anaphor would be uniquely required to enter into the agreement relation for the derivation to converge. This would block the possibility of an unmarked reflexive, and would also crash any derivation in which the object-marking head agrees with a different discourse-old co-argument of refl. Under the clitic analysis, it could be assumed that unlike other DP objects on which the associated clitic is optional, the reflexive anaphor is obligatorily generated with the associated clitic which must move to the higher incorporated position. Leaving open the exact details of the implementation, the most likely solution lies in specifying that the reflexive anaphor, unlike other DPs, is lexically deficient and must enter into the agreement relation before LF. With this timing, the agreement will take place before the LF movement which values the anaphor's ϕ-features, leading to the spellout of a class eight object marker.

2.4 Against accidental homophony

The lack of ϕ-features on refl has already been discussed in terms of motivating the LF-movement analysis to account for subject-orientation, but an explanation for the emergence of class eight morphology has not yet been provided.

Noun classes one through ten in Shona instantiate a set of five singular-plural pairings, with the odd numbers being singular, and the next even number the associated plural. The seven/eight pair is reserved for generic inanimate objects, with Fortune (Reference Fortune1980) describing the class as indefinite and Brauner (Reference Brauner1995) using the term miscellanea. More than any other, this class pairing is also frequent in the derivation of deverbal and deadjectival adverbs in the language. For this reason, the zvi- morpheme, along with its phonological alternate zva-, is very frequent in the language as a whole. In a 30,500 word corpus of folk stories (Fortune Reference Fortune1974) digitized and electronically searched, the adverb zvino ‘now’, derived from a class eight marked word ‘these things’ (Fortune Reference Fortune1980), occurs 720 times. The most frequent verbal form in the sample with this agreement is in the introduction of quoted speech:

1. (20) Zvi-ka-nz-i, “I-we u-ri ku-famb-a…”

   sub.8-rpst-tell-fv pron-2^nd.sg sub.2^nd.sg-aux   inf-walk-fv

   “‘You were walking…” was told.’

While one might be tempted by this evidence to argue for a similarity between the Shona reflexive and Romance reflexive clitics which are also used to introduce quoted speech,Footnote ⁸ here I pursue a different analysis. I have already argued that (20) cannot be an instance of refl in the subject position, but a closer examination of class eight morphology in general will explain not only this case, but why class eight emerges with the reflexive. In addition to plural sets of inanimate objects such as chairs, class eight is used for plurals made up of elements of different noun classes (Fortune Reference Fortune1980):

1. (21) Nda-$\emptyset $-zvi-tor-a [sadza no-mu-riwo].

   sub.1^st.sg-past-obj.8-take-fv sadza.5 and-3-relish

   ‘I took them (sadza and relish).’

While there is some observed tendency for class two (plural humans) to be used when only one of the two members of a conjunct is human, class eight is also acceptable, acting as the default for conjoined plurals with mismatched ϕ-features.

Class eight is also used as the agreement marker for subordinate clauses in subject and object positions, despite the absence of any sense of plurality:

1. (22) [Ku-tsva-ir-a mu-mba ma-zuva e-se]

   inf-sweep-appl-fv 18-house.9 6-day 6-every

   zva-ka-kosh-a.

   sub.8-rpst-important-fv

   ‘Sweeping the house every day was important.’

1. (23)

   1. a. John a-ka-kumb-ir-a Bill [kuti a-mu-muts-e].

      John sub.1-rpst-ask-appl-fv Bill comp sub.1-obj.1-wake-fv

      ‘John_i asked Bill_j to wake him_i/k up.’

   2. b. John a-ka-zvi-kumb-ir-a Bill.

      John sub.1-rpst-obj.8-ask-appl-fv Bill

      ‘John asked it of Bill.’

In (22), the matrix predicate is important, emerging with class eight subject agreement. (23) shows a simple ditransitive with a human indirect object, and a CP direct object which can be referred to using a class eight object marker as in (23b), assuming prior discourse has established the request to wake him up as discourse-old.

Contributing to the apparently free uses of class eight subject agreement as in (20) is the fact that subject clauses may be dislocated to the right periphery with no change in the subject agreement, as shown in a paraphrase of (22):

1. (24) Zva-ka-kosh-a [ku-tsva-ir-a mu-mba ma-zuva

   sub.8-rpst-important-fv inf-sweep-appl-fv 18-house.9 6-day

   e-se.]

   6-every

   ‘Sweeping the house every day was important.’

The same rightward shifting responsible for (24) can also be used to account for (20): the quoted speech is the subject of the sentence, reflected through class eight agreement, though the quote itself is right-dislocated.

The common thread is that class eight is used as the agreement marker for elements which contain either a clash of ϕ-features, or for elements such as propositions or quotations which do not inherently carry ϕ-features. It is therefore not surprising that refl, with no ϕ-features, also triggers class eight agreement morphology. Thus, the reflexive anaphor triggers class eight object marking, while the reason it triggers object marking at all derives from a lexical specification differentiating it from other DPs. Despite this one distinction in the lexical specification as a trigger for object marking, the zvi- which emerges in reflexive contexts shares a number of properties with non-reflexive object markers, which are the focus of the next section.

3.1 The Shona vP domain

Shona predicates may be modified by a series of suffixes which denote changes to that predicate's valence. Here, three in particular are crucial: the causative, the applicative, and the passive. As shown below, it is possible to get all three marked on one predicate:

1. (25) Mu-sango ma-ka-don-edz-(er)-w-a Mufaro na Shingi.

   18-forest sub.18-rpst-fall-caus-appl-pass-fv Mufaro by Shingi

   ‘Whilst in the forest, Mufaro was caused to fall by Shingi.’

   lit: ‘In the forest was caused to fall Mufaro by Shingi.’

In (25), the unaccusative predicate don ‘fall’ has taken the causative -edz, optionally takes the applicative -er, and finally is marked with the passive -w.Footnote ⁹ The causative has the semantic effect of adding an external causer to a given event, in this case Shingi, which is subsequently demoted by the passive. In addition to introducing benefactives, malefactives, and goals, the applicative may be used to introduce locative elements as it does here with mu-sango ‘in the forest’. The relative ordering of these suffixes is fixed, leading to the following structure for the lower verbal domain, illustrated for an active transitive with applicative and causative marking:

1. (26)

In this structure, the CausO position corresponds to the causee, while the external causer appears in the AGENT position. As noted above, the passive morpheme is tentatively treated as a variant of the v head, which conforms to the observed morpheme order, as expected according to the Mirror Principle (Baker Reference Baker1985). The structure proposed here is non-controversial for Bantu languages, with similar distance between V⁰ bearing the theme and v ⁰ introducing the external argument proposed for Kinande (Baker and Collins Reference Baker and Collins2006) and Zulu (Halpert Reference Halpert, Choi, Hogue, Punske, Tat, Schertz and Trueman2012). Following Julien (Reference Julien2002), verbal suffixes are derived via overt head-movement of the verb, while the prefixes are derived via a mechanism of affix lowering, again governed by the Mirror Principle. Because nothing here relies on the mechanics of verb formation beyond the vP domain, I remain agnostic as to the position of the final vowel and the final resting place of the verb. Minimally, I operate under the assumption that the hierarchical structure of arguments in Shona is derived from the tree structure in (26), and that the verb will move at least as high as v ⁰, generating the canonical SVO order.

3.2 Properties of object marking

Bantu languages show variation in the number of object markers permitted on a single clause, with Tswana cited as having as many as three (but more in principle imaginable) in Creissels (Reference Creissels2002), while Bemba is limited to two object markers, and then only when both are animate, or when the second is first person singular (Marten and Kula Reference Marten and Kula2012). Shona, like isiZulu (Zeller Reference Zeller2012), only allows one object marker per verb:

1. (27)

   1. a. * John a-ka-ri-mu-bik-ir-a.

      John sub.1-past-obj.5-obj.1-cook-appl-fv

      Intended: ‘John cooked it for her.’

   2. b. * John a-ka-mu-ri-bik-ir-a.

      John sub.1-past-obj.1-obj.5-cook-appl-fv

      Intended: ‘John cooked it for her.’

Furthermore, it is impossible to combine reflexive zvi- with an object marker on a ditransitive predicate:

1. (28)

   1. a. * A-ka-ri-zvi-bik-ir-a refl.

      sub.1-rpst-obj.5-obj.8-cook-appl-fv refl

      Intended: ‘He cooked it for himself.’

   2. b. * A-ka-zvi-ri-bik-ir-a refl.

      sub.1-rpst-obj.8-obj.5-cook-appl-fv refl

      Intended: ‘He cooked it for himself.’

This result suggests that the reflexive zvi- is in competition for the object marker position, reinforcing the claim that reflexive morphology should fall under the rubric of object marking. In lexical ditransitives and applicative constructions, Shona places no restrictions on which internal argument may be replaced by an object marker (Mugari Reference Mugari2013), illustrated with my own data repeated from (7) above:

1. (29)

   1. a. John a-ka-bik-ir-a Bill bota.

      John sub.1-past-cook-appl-fv Bill porridge.5

      ‘John cooked porridge for Bill.’

   2. b. John a-ka-mu-bik-ir-a bota.

      John sub.1-past-obj.1-cook-appl-fv porridge.5

      ‘John cooked porridge for him.’

   3. c. John a-ka-ri-bik-ir-a Bill.

      John sub.1-past-obj.5-cook-appl-fv Bill

      ‘John cooked it for Bill.’

Mugari also reports the same object marking symmetry with causatives, while noting that there are further restrictions on object marking when both internal arguments of a causative transitive are of equal animacy. In this situation, only the causee may be object-marked (Mugari Reference Mugari2013).

Recalling the structure presented in (26), it is not the case that only the highest internal argument can be object-marked. Shona allows object marking of any one internal argument per clause, with the felicity condition that the marked argument must be discourse-old, or definite in Mugari's terminology. By treating this object marking as parasitic on a feature associated with information structure, as described in section 2, relativized minimality can be maintained, and there is no need to enter into considerations of equidistance between the two internal arguments and the locus of object marking. If two DP-internal arguments bear the relevant feature, presumably structure will play a role, unless one of these two is refl, whose uninterpretable discourse-old feature must be checked by object marking for the derivation to converge.

In the same way that object marking is symmetric, reflexivity is also symmetric in the language:

1. (30) Bill_i a-$\emptyset $-zvi-rov-er-a (refl _1i) John (refl _2i).

   Bill sub.1-past-obj.8-hit-appl-fv refl John refl

   ‘Bill_i hit John_j for himself_i.’

   ‘Bill_i hit himself_i for John_j.’

The ambiguity of (30), with the reflexive anaphor potentially replacing either of the internal arguments, is exactly what would be expected if reflexivity parallels object marking. In this case, the reflexive anaphor can be construed as appearing in either the applied object position, indicated by refl ₁ yielding the first reading, or the theme position position (refl ₂), triggering the object marker in exactly the same way as either of those arguments would have done in (7b) and (7c). The same symmetry is observed in a case where an unaccusative has been extended with both the causative and the applicative:

1. (31) John_i a-$\emptyset $-zvi-don-edz-er-a (refl _1i) Bill (refl _2i).

   John sub.1-past-obj.8-fall-caus-appl-fv refl Bill refl

   ‘John_i tripped Bill for himself_i.’

   ‘John_i tripped himself_i for Bill.’

Again, construing refl in the higher applied object position yields the reading where John tripped Bill for his own self-serving reasons. The second position for refl is the causee position, with the meaning that John purposefully tripped himself to benefit Bill, elicited in the context of describing a footrace where John intentionally falls to let Bill win.Footnote ¹⁰

Tests for determining whether a reflexive morpheme is part of the object-marking system are given for Kikamba in Kioko (Reference Kioko2005). While these mainly amount to demonstrating that the reflexive morpheme in that language has the same distribution as other object markers, which has been done above for Shona, there is one further test to advance, relating to the behaviour of object markers in imperatives. When an imperative takes an object marker, the final vowel on the verb is -e, rather than the -a found with a full nominal object. This alternation with respect to object marked imperatives has been independently observed for Shona imperatives (Brauner Reference Brauner1995), and is replicated below in (32) with an added reflexive:

1. (32)

   1. a. Gez-a mwana!

      wash-fv baby.1

      ‘Wash the baby!’

   2. b. Mu-gez-e!

      obj.1-wash-fv

      ‘Wash him/her!’

   3. c. Zvi-gez-e!

      obj.8-wash-fv

      ‘Wash yourself!’

The reason for this alternation is unclear, as the final vowel does not undergo any changes in ordinary declarative sentences depending on the presence or absence of an object marker. For the current argument, it is sufficient to note that the reflexive triggers the same final vowel alternation as does a non-reflexive object marker, giving one more piece of evidence that the zvi- morpheme in reflexive sentences is indeed a member of the set of object markers.

The final piece of evidence lies in the judgements of consultants who reliably report sentences using the zvi- morpheme to be ambiguous between a reflexive reading and a class eight object marker reading. One consultant reports being so conscious of this ambiguity that they make an effort to passivize sentences with a discourse-old class eight object, to avoid the object marker. Rather than treating this ambiguity as coincidence, the analysis proposed here explains this alignment of facts by arguing that the zvi- which appears in a reflexive sentence actually is the class eight object marker and not a dedicated reflexive morpheme. The ambiguity thus lies in whether zvi- agrees with refl or with a class eight pro.

3.3 Reflexivized passives

Despite the overwhelming similarities, there is one environment where the parallel between reflexive uses of zvi- and other object markers breaks down: passive sentences.

Woolford (Reference Woolford and Akinlabi1995) makes the claim that only Bantu languages that allow multiple object marking in the active should allow object marking in a passive ditransitive. However, SiSwati is listed as an exception, with the caveat that in passive ditransitives, only the theme can emerge as an object marker; the applied object must have moved to the subject position. The reverse is not possible. As Shona is a single-object-marker language, one should expect either confirmation of Woolford's generalization, or perhaps another instance of the exceptional behaviour of SiSwati. What emerges is something different yet: Shona allows symmetric passivization and object marking in a transitive extended with the applicative, as described by Mugari, and again shown with my own data:

1. (33)

   1. a. Bota ra-$\emptyset $-mu-bik-ir-w-a.

      porridge.5 sub.5-past-obj.1-cook-appl-pass-fv

      ‘The porridge was cooked for him.’

   2. b. John a-$\emptyset $-ri-bik-ir-w-a.

      John sub.1-past-obj.5-cook-appl-pass-fv

      ‘For John was it cooked.’

In a transitive extended with the applicative, no matter which argument becomes the subject in the passive, the other can be object marked. Not only does object marking disregard the structural positions of the internal arguments, but passivization shows the same symmetry. Given that the Shona reflexive is similarly symmetric, a reflexive passive should also be acceptable:

1. (34) % John_i a-$\emptyset $-zvi-rov-er-w-a refl _i.

   John sub.1-past-obj.8-hit-appl-pass-fv refl

   ‘John was hit because he was asking for it.’

The sentence in (34) is a passivization of (30) above, which was ambiguous. It's worth noting that this sentence is not acceptable for all speakers. Some reject it outright, or at least hesitate to give a categorical judgement, while others (interestingly the younger consultants) find it acceptable, but only under the reported reading. Knowing that the Shona passive is symmetric, allowing either of the theme or applied object to be promoted to the subject position, and further knowing that the reflexive can map a subject to either of those internal argument positions, one might even expect a sentence such as (34) to be not only grammatical, but ambiguous like its active counterpart, depending on which argument moved up to the subject, and which position contains refl. Rather, it emerges as marginal, and acceptable only with one reading where John's status as a beneficiary is foregrounded, not his status as a theme. The key question then becomes one of determining the syntactic structure of the acceptable reading. Those consultants accepting this example agree that it puts in the forefront not the fact that John was hit, rather that it was John's own fault he was hit. Speakers paraphrase this as ‘for his own reasons’, reasons assumed to be known to the discourse participants. This suggests an interpretation where it is the applied object, representing the beneficiary, which is being interpreted as the topical subject.Footnote ¹¹ The reflexive would then be replacing the theme. In effect, this would have the structure schematized in (35a), rather than the one in (35b):

1. (35)

   1. a.

   2. b.

With passivized causatives, judgements are slightly less murky. There is more general acceptance of such sentences as grammatical, even among speakers who would reject or hesitate in the face of (34), but again there is only one reading:

1. (36) % Mufaro_i a-ka-zvi-pis-is-w-a refl _i.

   Mufaro sub.1-rpst-obj.8-burn-caus-pass-fv refl

   ‘Mufaro was caused to burn himself.’

Unlike in the applicative, Mugari claims that when a transitive verb is extended with the causative, and subsequently passivized, object marking the unmoved internal argument forces that argument to be interpreted as the theme. In the case of (36), this would mean that it can only be derived from a structure in which Mufaro is the higher causee, and refl the lower theme. This accords with the reported reading, where Mufaro's status as a causee rather than theme is made more prominent. The schematic for this case is therefore similar to what was seen in (35):

1. (37)

   1. a.

   2. b.

Thus, the reflexive zvi- appears with meanings that correspond to the same general placements of object markers in passive sentences, with the one difference that in an applicative context, a sentence which ought to be ambiguous only has one reading. I propose that this one distinction follows from the bound variable nature of refl.

Looking back at (35) and (37), the a) examples schematize the available readings, while the b) readings are blocked. Schematically, the ungrammatical b) examples have the appearance of strong crossover (SCO) violations. However, SCO is typically found in operator-variable relations, and not in reflexives interacting with case-driven A-movement such as passivization or raising, shown in the English (38) below:

1. (38)

Treating the Shona facts as SCO would first require evidence that the reflexive anaphor is in fact a semantically bound variable, and then evidence that passivization is ${\rm \bar A}$ movement.

That the Shona reflexive can act as a bound variable can be shown with a quantified antecedent:

1. (39) Imbwa y-oga-yoga_i ya-ka-zvi-rum-a refl _i.

   dog.9 9-every-redup sub.9-rpst-obj.8-bite-fv refl

2. ‘Every dog bit itself.’

As in the earlier cases, this binding relation is strictly clause-bounded. That the reflexive can be bound by a quantifier serves as a demonstration that refl can be treated as a semantic variable. The bound-variable status of the Shona reflexive can also be demonstrated through the lack of a possible strict subject comparative reading for the sentence in (40), originally proposed in Dalrymple et al. (Reference Dalrymple, Mchombo and Peters1994) as a test for transitivity:

1. (40) Va-kumana va-no-zvi-veng-a refl _i kupfuura va-sikana.

   2-boy sub.2-hab-obj.8-hate-fv refl surpass 2-girl

   ‘The boys hate themselves more than the girls.’

The ambiguity here is between a sloppy subject comparative reading (the boys’ self-hatred exceeds that of the girls), and the object comparative reading where the boys hate themselves more than they hate the girls.Footnote ¹² A strict reading where the boys’ hatred for themselves exceeds the girls’ hatred of the boys is not available, lending more support to the claim that refl acts as a bound variable.

The second ingredient of the SCO story, that the passive is in fact ${\rm \bar A}$-movement, has been argued for independently by Bliss and Storoshenko (Reference Bliss, Storoshenko and Jones2008). First, they show that the subject position in Shona sentences is sensitive to the information-structural status of the DP moved there: focused elements are not permitted in subject position, hence the analysis of subject agreement being in the head of Top(ic)P in the CP domain. Second, they argue that the symmetry of the Shona passive, discussed above in relation to (33), is suggestive of information structure-driven ${\rm \bar A}$-movement rather than case-driven A-movement, which should only allow the syntactically highest internal argument to move to the subject position in a passive. The findings in (35) and (37) support that position. If refl is indeed a bound variable, the ungrammaticality of the b) examples in those schematic representations strongly suggests that the illustrated movement is an ${\rm \bar A}$-movement triggering a SCO violation. Reflexives thus differ from object markers in this one way, failing to show symmetry in passive contexts, because the anaphor cannot be bound by an element which has moved overtly from a position bound by the anaphor itself.

3.4 The mechanics of symmetry and constraining the reflexive

The previous discussion of SCO relies on some underlying assumptions presented in previous work on the symmetric passive in Shona. Here I briefly sketch an implementation of that analysis, showing how it interacts with an analysis of object marking based on agreement (whether spelled out as an agreement prefix or a clitic), and leading to a summary of the constraints on the placement of refl.

The relevant symmetric examples are given in (41), repeated from (33):

1. (41)

   1. a. Bota ra-$\emptyset $-mu-bik-ir-w-a.

      porridge.5 sub.5-past-obj.1-cook-appl-pass-fv

      ‘The porridge was cooked for him.’

   2. b. John a-$\emptyset $-ri-bik-ir-w-a.

      John sub.1-past-obj.5-cook-appl-pass-fv

      ‘For John was it cooked.’

In a context in which both John and the porridge are known, the sentences in (41) are truth-conditionally identical. However, there is a subtle interpretive difference in the speaker's choice of which established referent is most topical moving forward in the discourse, visible in the choice of internal argument promoted in a passive sentence.

Shona is not alone in having a symmetric passive in which either internal argument of a ditransitive can be promoted to the subject position. Presented with these facts in Kinande, Baker and Collins (2006) propose that relativized minimality is parametrized, and does not apply in that language. Given that there is an interpretive difference between the symmetric pair in (41), here I advance an analysis making use of ${\rm \bar A}$-movement for the passive. This analysis proceeds under the same structural assumptions underlying the Kinande situation described above: two elements in an asymmetric c-command relation (here the applied object c-commanding the theme) are both eligible not only for passivization but also for object marking. Given that the former is driven by the EPP, and the latter by the agree mechanism described above, relativized minimality should come into play, and symmetry would not be expected.Footnote ¹³

Recalling that the subject position in Shona is limited to topics, and that object marking also identifies elements with some topic-like characteristics, one might suspect that both internal arguments in (33) bear the same information-related feature. If this were the case, relativized minimality should dictate that only one of the internal arguments may be object marked, with the other emerging as the subject. The first probe seeking a discourse-old element, in this case the one responsible for object marking, would presumably agree with the highest argument given that the two internal arguments are in an asymmetric c-command relationship. That this does not happen suggests that each of the internal arguments bears a different feature, only one of which is probed by the head responsible for object marking, the other being probed by the Topic head deriving the sentential subject.

Recall, from the discussion of the Shona topic head in section 2.1, that simple declarative sentences may not serve as answers to subject wh-questions. Under the standard assumption that the answer to a wh-question is focused, this position more naturally fits the definition of Topic advanced by Bax and Diercks, as the complement of Focus. However, they claim that this is the information status of the object markers. Such a claim is too strong, as it predicts that an object-marked element cannot appear in a thetic statement answering a more general question of What happened? In elicitation, if the context is carefully constructed such that the relevant object referent is active in the discourse before the question is asked, then object marking is acceptable. Rather than Topic, an appropriate description of the object-marking feature labelled F in section 2.2 may be found by elaborating on the condition for object marking cited by Mugari: definiteness. Descriptively, consultants report that the necessary condition for using an object marker is that the referent has to already be present in the discourse. Not so much a matter of topic versus focus, this is more akin to discourse-givenness. The conditions for the use of an object marker accord quite clearly with the status of Activated on Gundel et al.'s (Reference Gundel, Hedberg and Zacharski1993) Givenness Hierarchy. Activated elements are represented in current short-term memory, with one potential source of this activation arising from the immediate preceding discourse. As this activated status is described by Gundel et al. as a necessary condition for all pronominal forms, the reflexive fits into this category, and it more generally accounts for the quasi-pronominal nature of the object markers. While the Topic head probes for a unique non-focused element which I designate as [+Top], object marking is driven by a feature denoting this activated status, which I will designate as [+Act].

In a situation such as (33), where either feature can be borne by either of the asymmetric internal arguments, the speaker has a choice. A [+Top] feature which determines the eventual subject can be assigned to either internal argument depending on which is most topical in the current discourse. Whichever argument bears this feature becomes the subject in a passive ditransitive clause. If the other internal argument is activated, it bears a [+Act] feature. Here, the optionality lies in whether the probe which underlies object marking is present: while any activated internal argument may be object-marked, it is felicitous to omit object marking for activated internal arguments. The one exception to this is refl, which as discussed earlier bears an uninterpretable version of the feature now called [+Act], which must be checked.

While the two features [+Top] and [+Act] are quite similar, I assume that they cannot co-occur on a single item. Assuming that each clause must have a topic, the speaker may determine the [+Top] assignment at the time each clause's numeration is constructed, choosing which is the most appropriate topical element. Subsequently, additional referents may be designated as [+Act] based on the speaker's reasonable assumptions of the addressee's awareness of the referent, bearing in mind that the Givenness Hierarchy is defined in terms of the addressee's cognitive state rather than the speaker's. Treating refl as obligatorily [+Act] is thus no problem, as by the time the addressee hears the overt manifestation of refl, the antecedent will already have been uttered and therefore activated. This addressee orientation sidesteps any concern of whether the referent is activated in the discourse at the moment refl is merged: the referent will be activated in the discourse by the time the addressee processes refl.

This analysis has an important consequence for deriving active sentences in Shona. Bliss and Storoshenko (Reference Bliss, Storoshenko and Jones2008) stipulate that in active clauses, the external argument at the specifier of vP has a privileged status of always checking the Topic EPP by virtue of proximity, even though the passive ditransitive cases show that proximity alone cannot be the factor driving movement to the subject position. The current argument that the subject position and object marking are the result of distinct features permits a more refined analysis. If the proximity-based account is to be maintained, and the external argument position is uniquely (structurally) privileged, this would at least predict that no element that does not bear [+Top] could appear as the external argument; if this were to happen, there would be a clash of discourse features when the EPP is checked. An even stronger implementation would be to bar any element with a clashing feature (i.e. [+Act] rather than [+Top]) from merging at the specifier of vP, but the result would be the same. If v uniquely selected specifiers bearing [+Top], this could remove one of the major barriers against the higher agreement position account of object marking, as the external argument would never bear the feature for which object agreement is probing.

Having clarified the mechanics of the subject position, the full set of restrictions on the distribution of refl can now be enumerated. In an active clause, refl may not appear in the external argument position, as refl, with its [+Act] feature, cannot satisfy the the local Topic head. In matrix clauses, an external argument refl is additionally blocked because there is no potential antecedent. Whether in a matrix or embedded clause, refl additionally cannot act as a derived subject via passivization, again due to the feature conflict in attempting to check the EPP. These constraints restrict refl to an internal argument position, where object marking will occur. In an active clause, the external argument is the default antecedent, while in a passive clause, the antecedent for refl cannot incur a SCO violation on the way to checking with the Topic head.

Through this combination of factors specific to the reflexive anaphor and the externally motivated mechanisms needed to account for symmetric passives, the only position in which refl may be felicitously used is as an internal argument with a local subject antecedent. It thus emerges that while the mechanics are fundamentally different, the result is a distribution of meanings that, at first glance, match a detransitivizing derivational reflexive. Eliminating that possibility in support of the above analysis is the first goal of the next section on alternate analyses.

4.1 The case For detransitivization

Some authors in their descriptions of Shona explicitly group reflexive zvi- with such affixes as applicative, causative, etc. (Bellusci Reference Bellusci1991, Brauner Reference Brauner1995), thus making at least an implicit claim that the reflexive is an operation on valence rather than anything related to object marking. As noted above, the invariant form, combined with subject orientation and clause boundedness do suggest such an analysis. However, treating the reflexive as an operation on the predicate valence suffers one drawback: all other such affixes are suffixes in this language. Maintaining that the reflexive is derived by a change in valence would come with the explanatory burden of accounting for this anomaly.

The question of whether or not a given reflexive affix should be treated as indicating a binding relationship or a detransitivizing operation has been addressed previously in the literature. One explicit test is presented in Hirose (Reference Hirose2003) in a discussion of Plains Cree. For Hirose, the crucial case is one where a transitive verb bears both applicative and reflexive marking, as in (42). His proposed analysis is given in (43):

1. (42) ni-pakam-ah-amâ-so-n John.

   1.sg-hit-by.tool-appl-refl-lcal John

2. ‘I hit John for myself.’ (Hirose Reference Hirose2003: 224)

3. (43)

   (Hirose Reference Hirose2003: 225)

According to Hirose's analysis, pro _i, corresponding to the first person, is initially merged as the applied object. Then, merging the -iso head effects the detransitivization by allowing pro _i to re-merge in a second argument position, that of the agent. Crucially, this is an unambiguous sentence in Plains Cree. The reflexive can only map the applied object to the subject; it is not possible to get a reading of this sentence in which the speaker hit himself for John. Under this view where applied objects are added between the external argument and the theme, it should never be possible for a verbal reflexive in such a ditransitive construction to map an external argument to a theme. Hirose claims that if a language allows such readings, it derives reflexivity through a mechanism of binding rather than valence manipulation.

The Shona equivalent to (42) appears above in (30), repeated below as (44):

1. (44) Bill_i a-$\emptyset $-zvi-rov-er-a (refl _1i) John (refl _2i).

   Bill sub.1-past-obj.8-hit-appl-fv refl John refl

   ‘Bill_i hit John_j for himself_i.’

   ‘Bill_i hit himself_i for John_j.’

If zvi- were a detransitivizing reflexive, one would expect the same result as in Plains Cree, with a mapping between the subject and the highest internal argument, the applied object. Instead, ambiguity emerges reliably with all consultants. This diagnostic thus strengthens the present analysis: Shona zvi- behaves like a binding-based reflexive rather than a detransitivizing one.

Further traits of reflexive verbs are enumerated by Reinhart and Siloni (Reference Reinhart and Siloni2005), who provide a list of other types of predicates on which reflexive morphology may appear. This list includes passives, which have already been shown not to use the zvi- prefix. Reinhart and Siloni, using examples of Italian si, also identify impersonals and reciprocals as potentially sharing a morphological form with reflexives when reflexivity derives from a change in valence. This is easily demonstrated not to hold of Shona:

1. (45) Kwa-$\emptyset $-pis-w-a zvi-garo.

   sub.17-past-burn-pass-fv 8-chair

   ‘There was burning of chairs.’

2. (46)

   1. a. Va-rume va-ka-won-an-a.

      2-man sub.2-rpst-see-recip-fv

      ‘The men saw each other.’

   2. b. Ta-ka-won-an-a.

      sub.1^st.pl-rpst-see-recip-fv

      ‘We saw each other.’

The impersonal construction in (45) makes use of the passive suffix, not a reflexive. As shown in (46), Shona reciprocals use the dedicated morpheme -an, which does appear as a suffix, the traditional locus of morphological marking of valence-changing operations in Bantu. Of course, the facts in (45) and (46) are not in and of themselves conclusive proof that the Shona reflexive is not a valence-reducing reflexive on the Reinhart and Siloni analysis, as these other uses of reflexive morphology are not described as necessary conditions, but merely cross-linguistic traits of a valence-reducing reflexive. However, the facts do fall out exactly as the current analysis predicts: reflexive morphology is not used in non-reflexive contexts. Beyond this, the behaviour of the Shona reflexive does not accord with either of the Reinhart and Siloni analyses of reflexivization as a lexical or a syntactic process.Footnote ¹⁴

The suggested contrast between the reflexive and the reciprocal (one being binding-related, and the other looking more like a valence reduction) suggests an interesting divergence in this language, where two phenomena that are usually quite similar within a language have quite different analyses. Exploration of this issue, and the reciprocal in general, is held over for future work. Despite this one puzzle, the diagnostics in this section, combined with the difficulty of maintaining an analysis that the reflexive stands as a prefixal valence-related suffix, eliminate the possibility that the Shona reflexive involves a reduction in valence.

4.2 Dual marking

In Sikuku (to appear), the reflexive morpheme of Lubukusu is discussed in the context of treating it as one of the object markers in that language. The basic pattern in Lubukusu is similar to that in Shona: the reflexive morpheme is an invariant prefix appearing in the verbal template between tense marking and the verbal root. Furthermore, the object markers and the reflexive prefix in Lubukusu share several properties already discussed, including parallel constraints on co-occurrence with a full argument, and largely similar distributions. A key distinction though is that Sikuku treats the object markers in that language as incorporated. While an incorporation analysis for Shona has already been rejected, Sikuku's proposal represents a departure from those already considered, as he argues that while both the reflexive prefix and the object markers are derived via incorporation, they have distinct syntactic positions, with the reflexive incorporating via the voice head, while object markers use an immediately dominating functional head associated with case:

1. (47)

A crucial piece of evidence in Sikuku's analysis is that while two object markers cannot co-occur in Lubukusu, a reflexive prefix can appear in the same predicate as an object marker, though only in the observed order:

1. (48)

   1. a. Khalayi a-a-mu-i-siim-isy-a.

      Khalayi sub.1-past-obj.1-refl-like-caus-fv

      ‘Khalayi made him like herself.’

   2. b. *Khalayi a-a-mu-ba-siim-isy-a.

      Khalayi sub.1-past-obj.1-obj.2-like-caus-fv

      ‘Khalayi made him like them.’ (Sikuku to appear: 10)

As has already been shown in (28), repeated below, Shona does not allow zvi- to co-occur with another object marker, in either order:

1. 49.

   1. a. * A-ka-ri-zvi-bik-ir-a refl.

      sub.1-rpst-obj.5-obj.8-cook-appl-fv refl

      Intended: ‘He cooked it for himself.’

   2. b. * A-ka-zvi-ri-bik-ir-a refl.

      sub.1-rpst-obj.8-obj.5-cook-appl-fv refl

      Intended: ‘He cooked it for himself.’

While Sikuku's analysis convincingly demonstrates that the reflexive prefix is derived through the same operation as object markers in Lubukusu, though at a different position, the facts in (49) show that such an analysis is not possible for Shona. Furthermore, in Lubukusu, the reflexive prefix is not homophonous with class eight or any other object marker. There is, however, some limited similarity between the Lubukusu reflexive and Shona: Sikuku also proposes that the reflexive in Lubukusu derives from a pronominal anaphor which is dedicated to that function. Evidence for the existence of such an anaphor can be found in the limited doubling property of Lubukusu object marking. Recall from the initial discussion of incorporation that Sikuku makes the claim that full doubling of object DPs and object markers constitutes evidence for agreement-based object marking. In Lubukusu, object markers may only be doubled with a class-matching pronoun or, for the reflexive, an anaphor:

1. (50)

   1. a. Wekesa a-a-i-siim-a o-mu-eene.

      Wekesa sub.1-past-refl-like-fv 1-1-own

      ‘Wekesa likes himself.’

   2. b. Wekesa a-a-mu-siim-a ni-ye.

      Wekesa sub.1-past-obj.1-like-fv foc-him

      ‘Wekesa likes him.’ (Sikuku to appear: 7)

To keep these facts consistent with the incorporation analysis, Sikuku proposes that this limited doubling derives from optional articulation of the trace left behind by the incorporation operation, spelled out as a pronoun.

Shona does not readily allow such doubling:

1. (51)

   1. a. Shingi_i a-ka-zvi-pis-a refl _i, Shingi_i.

      Shingi sub.1-rpst-obj.8-burn-fv refl Shingi

      ‘Shingi burnt herself.’

   2. b. Shingi_i a-ka-zvi-pis-a refl _i, i-ye_i.

      Shingi sub.1-rpst-obj.8-burn-fv refl pron-1

      ‘Shingi burnt herself.’

These attempts at doubling in Shona are felicitous only with a very heavy intonational break following the inflected verb, suggesting an extraposed position. Such a pause is not required when object markers are doubled, making this one more distinction between the reflexive and other object markers in the language. Furthermore, no dedicated form parallel to the Lubukusu eene from (50a) is attested in such constructions. The refl anaphor I am proposing is similar in character to this morpheme, though covert, with differences in distribution arising from parametric differences between the languages in terms of the number of object marking positions, and the nature of the underlying mechanism.