Mande languages exhibit a typologically unusual S-Aux-O-V-X word order. As shown below for Mandinka (Creissels & Sambou 2013; Creissels in press), the subject appears clause-initially.1 With the exception of the completive, tense/aspect information is consistently realized as an auxiliary following the subject (sometimes in combination with a postposition following the verb), which is followed by the direct object (if the verb is transitive). The progressive aspect, for example, is expressed with a locational copula auxiliary, in combination with the postposition kaŋ following the verb, as shown in (1). The completive is realized as a suffix on an intransitive verb (see 2), but if the verb is transitive, an auxiliary form of the completive surfaces instead, as shown in (3).2 Everything else must follow the verb. This includes adjuncts, such as n na buŋo kono (‘in my house’) in (2), as well as CP complements (which require a dummy 3SG pronoun in preverbal object position) (4), locative PP complements (5), and indirect object complements (6). Note that it is impossible for both objects to precede the verb (6b), or for both objects to follow the verb (6c).

Given that ascribing an SOV word order to a language has often been synonymous with describing it as “verb-final,” rigidly SOVX languages present a puzzle for the generative literature on word order, since NP direct objects precede the verb, but the verb does not appear clause-finally in the presence of other arguments. Furthermore, SOVX word order raises the question of whether the postverbal material should be analyzed as structurally low or high in these languages. The answer to this question is controversial. Koopman (1992) argues for Bambara that the direct object, along with any postverbal material, is generated within a head-initial VP (Kayne 1994; for similar proposals for SOVX word order in other West African languages, see also Koopman 1984 on Mahou and Travis 1989 on Kpelle). The direct object then raises to a preverbal position in order to receive Case, while everything else remains in its base-generated position within the VP, as shown in (7) (adapted from Koopman 1992: (2)).3

Nikitina (2009; 2019), however, argues from a range of phenomena in Wan and other Mande languages that postverbal material is not VP-internal. She instead proposes that postverbal elements, including arguments, are located in a high, clause-adjoined position, as shown in (8). In Nikitina’s view, the only possible analysis for this position within a generative-transformational framework is to assume that the postverbal elements undergo obligatory rightward extraposition. She argues that obligatory rightward movement is conceptually undesirable, and that the facts therefore cannot receive a satisfactory explanation in a generative-transformational approach. Instead, she adopts the Lexical-Functional Grammar (LFG) framework, where she accounts for the high, clause-adjoined position of the postverbal arguments by proposing that they are part of the verb’s argument structure at one level of the grammar (the functional level, or “f-structure”), and adjoined to the clause at another level of the grammar (the surface structural level, or “c-structure”) (Nikitina 2019).

However, as I will show, neither an obligatory rightward extraposition account nor abandoning a generative-transformational approach is necessary to account for this word order phenomenon. In fact, I provide robust syntactic evidence from Mandinka which favors a VP-internal analysis of postverbal indirect objects (like those in 6a) within a generative-transformational framework, like the one proposed by Koopman (1992), and I will argue that Nikitina’s (2009; 2019) findings from Wan can be accommodated within this analysis. The specific structure I adopt is an updated version of (7), where I assume binary branching (Kayne 1984), and accordingly represent the base position of the direct object as a specifier, as shown in (9). I continue to follow Koopman in assuming a head-initial structure given the head-initial position of T and C elements, namely auxiliaries (see examples 1, 3–6) and complementizers (see example 4).

In section 2, I show that postverbal indirect objects must immediately follow the verb, while adjuncts are more freely ordered. In section 3, I show that postverbal indirect objects must also be included in the material replaced by the verbal pro-form a ke (‘do it’), while postverbal adjuncts are not necessarily included. In section 4, I provide binding data that show that postverbal NPs must be lower than preverbal NPs: preverbal objects can serve as antecedents for reciprocals in postverbal argument position, and quantified expressions in preverbal objects can bind postverbal pronouns. Together, these facts strongly suggest that postverbal indirect objects are located in a fixed, low position in the VP.4 In section 5, I thus argue that a VP-internal analysis is the best fit for these findings, in contrast with a clausal-adjunction analysis like the one proposed by Nikitina (2009; 2019). Section 6 concludes the paper.

Postverbal indirect objects are rigidly ordered preceding various adjuncts, including adverbs (10–11), as well as benefactive (12), locative (13), and instrumental (14) PPs, all of which must follow the postverbal indirect object; none of these adjuncts may precede it.

The fact that the indirect object and adjuncts must appear in this order is expected under the traditional assumption that arguments are generated closer to the verb than adjuncts, and is therefore compatible with an analysis in which the indirect object is located internal to the VP.

Within a clausal adjunction analysis, however, there is no straightforward account for this word order restriction. Under Nikitina’s LFG analysis, in fact, the mapping between f-structure and c-structure does not determine the order of the postverbal elements, so it is predicted that postverbal arguments and adjuncts may appear in any order. While Nikitina claims that this is the case in Wan, deriving the order found in Mandinka would require an ordering stipulation.5 A generative-transformational alternative, in which postverbal elements are obligatorily extraposed would require not just unmotivated obligatory movement, but also a stipulation about the order in which these elements attach in their postverbal positions. These kinds of stipulations are particularly unexplanatory given that other postverbal elements, including CP and locative PP arguments, do not adhere to this order. CP complements must follow adverbs and PP adjuncts, as shown in (15) and (16).

Note that this is consistent with English, where CP arguments typically undergo heavy shift to the right: while the adverb loudly follows the NP argument in (17), it must precede a CP argument (18–19).

In addition, locative PP complements may optionally follow some PP adjuncts (20–21) and adverbs (23) (though not all; see examples (22) and (24), where the locative PP complement must precede the adjunct). Also note that adjuncts themselves, such as the adverbs saama (‘tomorrow’) and koteŋke (‘again’) can be flexible in their word order with respect to one another, as shown in (24), where saama can appear either before or after koteŋke (but crucially not before the PP complement).

If all postverbal material is clause-adjoined, as Nikitina (2019) claims, why should only indirect objects have a fixed position that precedes adjuncts, while other arguments and adjuncts can or must appear in a different order? This problem is easily avoided if postverbal indirect objects originate and remain in VP-internal position, while other postverbal elements are located higher in the structure, and can potentially be locally reordered.

While this paper is primarily focused on postverbal indirect objects, I would like to briefly elaborate on the reason why I suggest that other postverbal elements undergo local reordering, and are crucially not located in a high, clause-adjoined position, focusing in particular on the locative PP complement of ‘put.’ As shown in examples (22) and (24), these PP arguments do not have the option to follow all adjuncts. This lack of systematicity, supported by the very fact that the reordering is optional even when it is possible, may be more easily explained if the relevant reordering operation is a relatively superficial one (see section 4.3 for additional discussion).

In addition, note that the element in (20) that optionally precedes the PP argument is a benefactive. As shown in (25), the benefactive also appears closest to the verb in the presence of multiple postverbal adjuncts.

It therefore appears that benefactives have a preference for appearing close to the verb and reordering may occur in order to preserve this order. Crucially, however, recall that the benefactive may not precede a postverbal indirect object, as shown in (12). The fact that this adjunct seems to be located in a position relatively close to the verb and still must be preceded by the postverbal indirect object therefore provides further support for an analysis where the postverbal indirect object is located in a low VP-internal position.6

‘Do-so’ replacement is a well-known constituency test in English: verbal constituents may be substituted by the verbal pro-form ‘do-so’ (Hankamer & Sag 1976; Hauser 2010), as shown in (26). Mandinka has a similar construction in which verbal constituents may be replaced with a pro-form composed of the dummy verb ke (‘do’) preceded by the 3sg pronoun a (27).

I argue that the a ke pro-form replaces vP-level constituents (Stroik 2001; Haddican 2007), and can therefore be used to diagnose what postverbal material is part of the vP in Mandinka. Both the postverbal argument and postverbal adjuncts can be replaced by a ke, as shown in (28).

The fact that both the argument and adjunct are low enough to be included in the a ke pro-form lends additional support to the conclusion from section 2 that the postverbal argument must be located within the vP. If both postverbal arguments and adjuncts are adjoined at the clausal level, neither would be expected to be replaced by the vP pro-form.

Interestingly, while both adjuncts and arguments can be included in the a ke pro-form, an asymmetry arises between adjuncts and arguments with respect to which ones must be included. Adjuncts, including adverbs (29), benefactive PPs (30), and locative PPs (31), may follow a ke.

Arguments, on the other hand, cannot be excluded from the material replaced by a ke. This is true for the preverbal direct object, regardless of where the preverbal direct object is placed relative to a and ke, and whether the oblique case marker is present (32). It is also true for the postverbal indirect object (33). Instead, both objects must be included in the material substituted by a ke.

This asymmetry suggests that while both postverbal arguments and adjuncts are low enough to be included within the a ke pro-form, postverbal adjuncts are high enough that they need not be included, while postverbal indirect objects are more deeply embedded and must be included. These results are expected if postverbal indirect objects are VP-internal while adjuncts are adjoined at a higher vP-level: a ke minimally replaces the smallest vP-constituent containing the verb and its arguments, but may optionally target a higher vP-segment containing adjuncts. However, these results are unexpected in a clausal-adjunction analysis where there is no structural difference between postverbal arguments and adjuncts, and both would be expected to be able to appear outside of the a ke pro-form.

Nikitina (2019) provides binding data from Wan which demonstrate that a postverbal reflexive pronoun may be bound by a preverbal object, as shown in (34).

Nikitina argues that the binding data show that postverbal material cannot be base-generated in a clause-adjoined position within a transformational grammar framework, since it would not be c-commanded by the preverbal object. However, she also argues that the alternative, namely obligatory rightward extraposition, is conceptually undesirable (see Nikitina 2019 for details), so she uses these data to motivate an LFG approach.

In this section, I show that binding facts from Mandinka also suggest that postverbal arguments cannot be base-generated in a clause-adjoined position, and I similarly argue against a rightward extraposition account. However, I depart from Nikitina in arguing that the postverbal material is not in a clause-adjoined position at all: instead, I argue that these binding facts are expected within an analysis in which the postverbal material is located within VP and c-commanded by the preverbal object, as the word order and a ke facts indicate.

I argue that the evidence from the ordering of postverbal arguments and adjuncts, a ke replacement, and binding lead to the conclusion that the structure in (48), where the postverbal object remains within VP, is to be preferred over that in (49), where it is adjoined at the clausal level (regardless of whether clausal-adjunction is the result of base-generation or rightward extraposition).

Recall that I adopt a head-initial analysis of the OV word order given the head-initial position of C elements and T elements. In (48), since the subject precedes the auxiliary, I assume that it raises to Spec,TP in order to receive Case. Following Koopman (1992), I assume that the direct object also raises in order to receive Case, specifically to Spec,vP. I also adopt the standard assumption that the verb raises from V to v.7 Note that while we cannot detect the raising of the verb and direct object from word order alone, evidence for verb movement comes from the placement of the causative suffix -ndi, as shown in (50).

Assuming that causative morphology is located on v (e.g., Folli & Harley 2007; Pylkkänen 2008), the suffixal nature of this causative morpheme can be taken to suggest that the verb undergoes head movement to v. Furthermore, the fact that the word order is still OV, with the direct object preceding the verb, suggests that the direct object must raise as well. Additional support for raising of the direct object comes from Kandybowicz & Baker (2003), who similarly argue that in the Kwa language Nupe, which has both OV and VO word orders, the OV order is derived from a head-initial structure via movement of the object from Spec,VP to a vP-internal Case-licensing position. Note, however, that nothing rests on these assumptions: as long as the direct object is base-generated higher than the indirect object, an analysis in which the verb and direct object do not raise and the Case of the direct object is valued in situ is equally compatible with the facts, and with the central claim that the indirect object is VP-internal. Finally, the indirect object receives Case from the preposition, and therefore does not need to raise to Spec,vP, instead remaining within the prepositional phrase.8 The resulting order is S-Aux-O-V-X, with the subject preceding the aspectual auxiliary located in T, the direct object preceding the verb, and the indirect object appearing after the verb.

This structure in (48) captures the fact that the postverbal indirect object is closer to the verb than postverbal adjuncts, since it is located directly adjacent to the verb as its complement. It also captures the fact that the verb, preverbal direct object, and postverbal indirect object are replaced by the vP pro-form a ke, since together they form a vP-constituent. Adjuncts (not depicted) are adjoined at a higher vP-level, which allows them to be optionally replaced by a ke. Finally, this structure also captures the fact that an indirect object anaphor may be bound by a preverbal antecedent, and a pronoun in a postverbal indirect object may be bound by a preverbal quantified expression, since the postverbal indirect object is located low enough to be c-commanded by the preverbal arguments.

Recall that Nikitina (2009; 2019) argues against a VP-internal analysis of SOVX word order in Wan. For example, Nikitina (2019) shows that the IO does not necessarily need to appear immediately after the verb in Wan, in contrast with the Mandinka examples in (10–14), where the IO must immediately follow the verb. As shown in (51), the IO may appear either before or after another PP. For Nikitina, the fact that other postverbal material may intervene between the verb and IO suggests that the IO is not part of the VP. Instead, she argues that it is located in a high, clause-adjoined position.

I will show that the head-initial approach to Mande VP and vP that I adopt easily derives the separation of a PP complement or modifier such as ‘with dogs’ from its associated preverbal nominal. That the PP does not appear within that nominal is in fact consistent with Nikitina’s (2009) generalization that PPs are ruled out inside Wan nominals (see also Goertzel 2022, who identifies the same restriction against PP modifiers in Mandinka nominals). Nikitina shows that nominals in Wan may only be modified by bare NPs, and never by PPs:

While Nikitina takes this restriction on PPs to mean that they are clause-adjoined, the restriction can be met simply by stranding the PP in the base position of the nominal complement when it raises to preverbal position (assuming that internal arguments which precede the verb raise to that position, as in (48)), as shown in (53).9

Note that this same approach can be extended to account for another set of examples with non-finite complements discussed by Nikitina (2009). She shows that in Wan, some verbs may select a non-finite VP as their complement, and that when that VP complement is transitive, its preverbal object appears adjacent to it, as shown in (54), where the direct object ‘work’ appears directly before the complement verb ‘do.’ However, when the non-finite complement verb has a postverbal PP complement, the postverbal complement does not appear adjacent to the complement verb, as shown in (55), where the PP ‘in the tree’ appears clause-finally, rather than immediately following the complement verb ‘climb’ that selects it.

According to Nikitina, this result is unexpected if the postverbal PP complement is part of the VP constituent, since one would expect the postverbal complement to immediately follow the verb that selects it, so she argues that these findings suggest that the postverbal complement is located in a high, clause-adjoined position. Crucially, however, this result is only unexpected under a head-final analysis in which the non-finite complement VP is base-generated to the left of the verb that selects it. On the other hand, the position of the postverbal complement can easily be explained within a head-initial analysis like the one proposed here. One way to account for the postverbal location of the PP within a head-initial analysis is to assume that, like in the analysis suggested above for (51a), the non-finite VPs in (54)–(55) are also nominal, which requires raising to Spec,vP for Case, thus stranding the PP in (55) like in (51a).

In addition, note that (54) and (55) involve the aspectual verb ‘begin,’ which is among what Wurmbrand (1998; 2001) calls “core” restructuring verbs, meaning that they are generally consistent across different languages in displaying transparency effects. I therefore suggest that another possible explanation for (54) and (55) is to analyze them as involving restructuring. According to Wurmbrand (see also Cinque 2006), transparency effects arise with restructuring verbs because the sentences are monoclausal: restructuring verbs head a functional projection and take a nonfinite VP as their complement. Assuming a head-initial structure for this functional projection yields the structures in (56) and (57) for the sentences in (54) and (55), respectively. Given the linear order of the restructuring verb relative to the verb in its complement, I assume that the verbs form a verb cluster in the sense of Keine & Bhatt (2016), who propose that the complement verb undergoes head-movement to incorporate with the restructuring verb (for an overview of verb cluster phenomena, see Wurmbrand 2006). Crucially, since the restructuring verb and the verb in its complement are not in different clauses, and since the structure is head-initial, the complements appear in their usual positions relative to the entire verbal complex: just as before, the direct object raises from Spec,VP to Spec,vP, as shown in (56), while the PP complement remains in situ, as shown in (57).

Admittedly, the above discussion provides an outline for an analysis, while a full exploration into constructions like (51) and (54–55) lies beyond the scope of this paper, and would require in-depth research into these aspects of Wan. But the adherence to the generalization that PPs do not appear in nominals, as well as the resemblance to the cross-linguistic phenomena of restructuring and verb clustering are striking, and the hypotheses sketched here yield principled accounts of otherwise puzzling word order phenomena. Lacking speakers of Wan to consult, I must leave it to future research to confirm the approach or provide an alternative consistent with the Chomskyian assumption that movement derivations underlie word order puzzles like this one. Crucially, given Nikitina’s (2009) report that Wan disallows PPs within nominals, and lacking evidence that that verbs like ‘begin’ do not behave as restructuring verbs in Mande languages, examples like (51) and (54–55) do not provide conclusive evidence that postverbal elements must surface in a high clause-adjoined position, particularly in light of the robust syntactic evidence presented above, which supports a VP-internal analysis of postverbal indirect objects.10

In this paper I have argued that the internal argument that appears postverbally in Mandinka is base-generated within the VP, where it remains throughout the course of the derivation. This VP-internal position has been proposed by previous authors (e.g., Koopman 1984; 1992; Travis 1989), but empirical support for this position is lacking in the literature on SOVX word order. The Mandinka data I present thus contribute to advancing our understanding of this typologically unusual word order. I have shown that postverbal indirect objects consistently precede postverbal adjuncts, while other postverbal arguments like CPs and locative PPs can or must follow them, an order that is consistent with a VP-internal analysis. I have also shown that postverbal objects must be replaced by the pro-form a ke, indicating that postverbal objects are low within the verbal domain, unlike postverbal adjuncts, which can but need not be replaced by a ke. Additionally, I have provided evidence that postverbal objects must be in a position low enough to be c-commanded by preverbal arguments by showing that reciprocal ɲoo inside a postverbal object can be bound by a preverbal antecedent, and pronouns inside a postverbal object can be bound by a quantified expression in a preverbal object, but not vice versa.

I have also shown that this evidence presents serious challenges for an analysis like Nikitina’s (2009; 2019), where all postverbal elements, including indirect objects, are located in a clause-adjoined position. For example, the ordering of postverbal arguments and adjuncts suggests that CP and locative PP arguments may optionally or obligatorily undergo local rightward movement, but this appears to be unavailable to indirect objects. While this order is expected in a VP-internal approach, a clausal adjunction approach would require additional stipulations about the order in which postverbal material can adjoin to the right of the clause. Furthermore, the a ke replacement facts reveal an asymmetry between postverbal objects and adjuncts. This suggests that they are not located in the same position, and that the postverbal object in particular is located low within the verbal domain. Finally, the binding data suggest that the postverbal object must be in a position low enough to be c-commanded by the preverbal object, which is consistent with a VP-internal approach.

In conclusion, the syntactic evidence that I have provided from Mandinka favors an analysis of SOVX word order in which postverbal objects remain in-situ within the VP domain, rather than one in which they are adjoined to a higher position to the right of the clause.