2.1 Verbs come first

Kipsigis is a verb-initial language, as illustrated in (2):¹

Any word order in which the verb is not clause-initial is consistently ungrammatical:

Although standard declarative sentences are V1 in Kipsigis, there are two left-peripheral particles—nɛ and ko—which allow for the fronting of lexical material before the verb:

The particle nɛ leads to a focused interpretation of the fronted constituent, generally best translated as an it-cleft. The particle ko, on the other hand, generates a topicalized and sometimes generic interpretation of the fronted constituent.

2.2 VSO/VOS alternation

Like many other V1 languages, Kipsigis exhibits a VSO/VOS alternation, as seen in (2a) and (2b). This word order “flexibility” is far-reaching in the language, extending to ditransitives and adjuncts.² In ditransitive sentences, all six V1 word orders containing a verb, subject, direct object, and indirect object are grammatical:³

The addition of a temporal adverb to a sentence containing a verb, subject, and object likewise allows for all six possible word orders in which the verb comes first:

Although our consultants deem each sentence in (5) and (6) grammatical, there are constraints on the discourse contexts in which the different word orders may be used (see §2.8).

2.3 Pragmatically neutral word order

“Out of the blue” diagnostics show that VSO is the most pragmatically neutral word order. In the context below, a newscaster for a Kipsigis broadcast gets on the air and wants to announce new information to the audience. This diagnostic assumes that the newscaster does not know the audience’s familiarity with the situation under discussion and thus cannot assume any shared knowledge (Bailyn 2012, see also Gundel 1974/1988; Chafe 1976; Givón 1976). As a result, we expect the most appropriate response to display the non-context-dependent word order:

VSO-Adv word order surfaces most naturally here, showing that it is the most pragmatically neutral word order of the language.⁴

2.4 Non-verbal predicate constructions

As in many other languages, Kipsigis verbs represent just one type of sentence-initial predicate (see Clemens & Polinsky 2017 and references therein). Non-verbal predicates—NP (8a), AP (8b), and locations (8c)—all appear in the clause-initial position, though for convenience we continue to label the Kipsigis word order pattern verb-initial (V1):

2.5 (Lack of) grammatical factors in VSO/VOS alternation

In many VSO/VOS alternating languages, there are grammatical differences between the object of a VSO sentence and that of a VOS sentence. For instance, England (1991) notes that in Mayan VSO tends to occur when objects are animate, specific, definite, or phonologically heavy. Chung (1998) observes that VSO/VOS alternations in Maori are affected by the agency and (pro)nominal status of the object. Massam (2001) and Coon (2010) argue that in Niuean and Chol, respectively, the object in VSO constructions is a full DP, while the object in VOS constructions is a bare NP.⁵

No such phonological or morphosyntactic factors dictate the VSO/VOS alternation in Kipsigis. Phonological weight does not determine VSO vs. VOS word order, as phonologically heavy constituents—lɑgojɑtɑɑp ʧeeptoo ‘Cheptoo’s orange’ in (9)—can surface in either postverbal position:

In example (10) from a spoken narrative, we see a similar outcome; the heavy NP koita ne kimi ngweny ‘stone that was on the ground’ precedes bare noun subject moset ‘monkey.’ This is the opposite word order as would be predicted if phonologically heavy phrases move rightward to create VSO/VOS word order alternations (see (59) for an example that makes the same point):⁶

Likewise, we have not observed any definiteness restrictions; both VSO and VOS word orders are grammatical with objects that are bare nouns (11) and objects that are unambiguously DPs, containing demonstratives (12):

The animacy facts paint a similar picture. When the subject is animate and the object inanimate, VSO and VOS are acceptable (13) and both remain equally viable in sentences with inanimate subjects and animate objects (14):

These patterns suggest that there are no purely phonological or morphosyntactic factors motivating the VSO/VOS alternation in Kipsigis.

2.6 Word order patterns with CP-level phenomena

V1 word order and postverbal word order flexibility are preserved under different kinds of CP-material. Example (15) shows the VSO/VOS alternation under declarative-embedding complementizers:⁷

Example (16) shows the VSO/VOS alternation inside of relative clauses:

Finally, (17)–(18) show that this word order flexibility persists inside topicalization (KO) and focus (NE) cleft structures:

V1 and word order flexibility also persist in yes-no questions, another presumably CP-level phenomenon. Yes-no question formation in Kipsigis has two morphophonological effects: the addition of an epenthetic vowel at the end of the clause (when the clause-final word ends in a consonant) and the addition of a high tone to the final syllable of the clause:⁸

We assume that the epenthetic vowel is in a C head, which is either right- or left-facing. A right-facing C head straightforwardly accounts for the clause-finality of the epenthetic vowel, while a left-facing C head requires its complement to be raised in a roll-up style movement to derive the surface word order. For our current purposes these analyses are equivalent; V1 and word order flexibility are retained regardless of CP-level phenomena, suggesting that the position of the verb and all other syntactic movements (that derive postverbal word order) are below the CP domain.

2.7 The position of wh-words

Wh-words surface most naturally in the immediately postverbal position (IPP). The examples in (21) show the most pragmatically neutral word orders for three interrogatives in which different constituents are questioned:

Although there are other grammatical wh-question word orders (§5), those in (21) are most natural and most discourse-neutral.

2.8 Word order patterns in response to wh-questions

2.9 Interpretation of the IPP

We have seen that focused material (e.g. wh-phrases and new information focus) surfaces naturally in IPP. This section previews §5, which shows that the interpretations compatible with IPP are more complex—and numerous—than simply focus. Here we focus on two core patterns: one directly built from the Q&A congruence patterns from above and the other showing that aboutness topics also occupy the IPP.

3.1 Possible accounts of V1 derivation

For a general overview of approaches to verb-initial syntax, we refer readers to Carnie & Guilfoyle (2000), Carnie et al. (2005), Chung (2017), and Clemens & Polinsky (2017). There are two common movement-driven analyses of V1: those that rely on head movement and those that rely on phrasal movement.

Under a head movement analysis, V1 word order is derived from a base-generated SVO structure via head movement of the verb to a position above the subject. Head movement has been proposed for Celtic languages like Irish, Afro-Asiatic languages, and Austronesian languages like Tongan and Tagalog (Ouhalla 1994; Richards 1996; Otsuka 2000; 2005b; McCloskey 2005; a.o.). VOS word order can then be explained through postverbal scrambling (Rackowski 2002; Otsuka 2005a; Richards 2013 for Austronesian), postsyntactic prosodic re-ordering of constituents (Sabbagh 2014 and Clemens 2019 for Austronesian; Clemens 2014; Clemens & Coon 2018 for Mayan; Bennett et al. 2016 for Irish; Richards 2016 for a variety of languages), heavy-NP shift (McCloskey 1983; Clemens & Coon 2018 for Irish and Mayan, respectively), and subject movement to a right-edge topic position (Clemens & Coon 2018 for Mayan; Polinsky 2016 for Austronesian).

Under predicate-fronting analyses, VOS word order results from raising the entire VP to a position above the subject.¹² Predicate-fronting analyses have been adopted for a variety of Austronesian and Mayan languages (Massam 2001; 2005; Pearson 2001; Aldridge 2004; Coon 2010; a.o.). Predicate-fronting derives VOS word order quite directly, since the verb and its object raise together as a constituent. VSO can be derived via remnant movement of the VP after the object evacuates the VP.

3.2 V1 languages with properties similar to Kipsigis

There is little syntactic work on languages from the Kalenjin subgroup of southern Nilo-Saharan. The only accounts of word order patterns in Kalenjin are Jake & Odden’s (1979) discussion of raising in Kipsigis and Creider’s (1977; 1983) descriptive work on the syntax and information structure of Nandi, a language closely related to Kipsigis. The only generative syntactic analysis that we are aware of comes from Creider (1987), who, in an overview of various southern and eastern Nilotic languages, claims that VSO is generated by movement of the verb to C and that VOS is generated by VP movement, though without a precise proposal about the landing site of that movement.¹³

Nilo-Saharan languages cover a broad geographic and typological space. They include V1 languages spoken in Kenya and Uganda like: Kipsigis, Nandi (Creider & Creider 1989), Turkana (Dimmendaal 1983a; b; Finer 2013), and Ateso (Barasa 2017). They also include SVO languages like Dholuo (Tucker 1994; Okoth-Okombo 1997; Cable 2012) and V2 languages such as Dinka (Van Urk & Richards 2015; Van Urk 2015 and references therein). In the generative syntactic tradition, there is some work on V1 in the Nilotic languages Turkana and Maasai. Turkana is a strongly VSO language, but it displays a VSO/VOS alternation influenced by the prominence of the object. Finer (2013) argues that this prominence is largely morphosyntactic (i.e. determined by the pronominal status, definiteness, etc. of the object), but it does have discourse correlates. Finer concludes that V1 in Turkana results from head movement of the verb to T, while the VSO/VOS alternation is dictated by a series of ranked constraints related to topicality. These constraints interact to determine the linear order of constituents, which generally progresses from “more topical” (e.g. subjects, pronominals) to “less topical” (e.g. objects, descriptive NPs). On Finer’s proposal, more topical elements move to a low topic position between TP and vP.

Maasai—much like Turkana—displays rigid VSO word order, though VOS is also acceptable; speakers usually offer VSO word order first, but they accept VOS as grammatical (Payne 1995). This VSO/VOS alternation is largely driven by information structure, with postverbal constituents generally progressing from “more topical” to “less topical” just as in Turkana (Payne 1995). Under Koopman’s (2005) generative analysis of word order in Maasai, nominative DPs raise to at least Spec,TP for case checking and often continue up the syntactic structure to a low topic position in the left periphery. Verbs then raise to a position high in the left periphery above the topic via remnant movement of the verbal predicate.¹⁴

There are several non-Nilotic languages that show similar properties to Kipsigis. Tongan is neither genetically nor geographically related to Kipsigis, yet it displays a similar VSO/VOS alternation dictated in large part by information structure (Otsuka 2005b; but cf. Polinsky 2016). Otsuka (2005b) claims that Tongan V1 is generated through head movement of the verb to C and that the VSO/VOS alternation results from information focus-motivated movement to Spec,TP. As we will argue below, the Kipsigis facts are best explained by a similar analysis, involving head movement of the predicate and scrambling of postverbal constituents. In a similar fashion, Richards (1996) proposes that the Tagalog verb raises to π, a position immediately above TP. Richards (1996; 2013) proposes that topicalization is covert movement and shows that there is extensive scrambling in the language. In the course of discussing focus in Tagalog, Kaufman (2005) argues that scrambling is not semantically vacuous, showing that focus contexts distinguish word order possibilities. Therefore, we see many parallels between these languages and Kipsigis, though Austronesian languages differ in important ways (e.g. ergativity and voice systems) that complicate direct comparison.

3.3 V1 in Kipsigis is derived through head movement

The ready acceptability of VOS sentences might suggest a predicate-fronting analysis of V1 in Kipsigis; the sentence-initial position of non-verbal predicates might also tempt us in the same direction. However, there is good evidence from both non-verbal and verbal predication that head movement is involved in deriving the predicate-initial word order of Kipsigis.

4.1 The Tongan parallel

Both Tongan and Kipsigis are verb-initial, exhibiting a basic VSO word order:

In order to generate Tongan V1, Otsuka (2005b) posits that verbs raise to C via head movement, followed by subject movement to Spec,TP to check the EPP (i.e. D-) feature on T. Tongan also shows a VSO/VOS alternation that is affected by information structure, exhibiting a discourse-based restriction on what may appear in the postverbal position:

Inspired by work in Miyagawa (2001; 2003), Otsuka (2005a) proposes that focused elements A-scramble to Spec,TP in a movement operation that checks features for information focus and EPP. Per Otsuka, the information focus feature motivates this scrambling movement, while EPP features are checked as a by-product.

Miyagawa (2010) describes Japanese A-scrambling to Spec,TP, driven by an EPP-like effect. The Kumamoto dialect makes use of two different forms of nominative case marking: the suffix -ga for constituents outside vP and the suffix -no for constituents within vP:

Miyagawa notes that in SOV word order the subject must be marked with -ga, whereas in OSV word order the subject can be marked with -no. This indicates that the subject must be outside of vP when in clause-initial position and inside of vP when preceded by the object. He then suggests that the landing site for the clause-initial element is Spec,TP.

Miyagawa (2010) claims that EPP features do not exist independent of other grammatical features. Instead, scrambling to Spec,TP is motivated by the grammatical features of topic and focus—or, more precisely, [–focus] and [+focus]—which originate on C and are inherited by T. The SOV/OSV alternation in Japanese arises through topic ([–focus]) movement of either the subject or object to Spec,TP. However, [–focus] movement does not operate under the traditional Probe-Goal pattern (i.e. raising the most local [–focus] constituent); instead the [–focus] feature on T can target any constituent for movement to Spec,TP where it receives a topical interpretation, in a process Miyagawa terms free movement (Miyagawa 2010: 87). In this way, movement of the object, rather than the more local subject, does not cause a violation of the Minimal Link Condition (Chomsky 1995). In contexts with multiple distinct topic and focus constituents, Miyagawa proposes the existence of an additional functional projection between CP and TP, which he terms αP. When a single sentence includes both a topic and a focus constituent, αP inherits the [+focus] feature from C, while TP inherits [–focus], or vice versa. In this way, separate topic and focus constituents can both surface at the left-edge of the clause in distinct functional projections.

4.2 Analysis of Kipsigis V1

Recall the patterns of word order felicity presented in §2.8, summarized in Table 3. As concluded previously, the target constituent of a wh-question usually surfaces most felicitously in the IPP (with exception of the manner adverb). Aboutness topics also surface in this position. VSO word order is always acceptable. We claim there is a position for discourse-prominent constituents immediately following the verb, as schematized generally in (51):

Following both Otsuka (2005a) and Miyagawa (2010), movement to this position is so-called “scrambling” to Spec,TP. Assuming with Miyagawa that TP may inherit discourse (δ) features from C, we assume that in Kipsigis TP inherits [uδ] as an instance of a [uδ] feature originating on C.

Based on the dispreference for manner adverb material in the IPP we conclude that postverbal scrambling for discourse prominence in Kipsigis also involves the EPP, which we formalize as checking a D-feature.¹⁶ Cross-linguistically, the EPP is assumed to be associated with the T head and we find no evidence here to suggest otherwise (Chomsky 1981; Bobaljik & Jonas 1996; see Svenonius 2002 for a survey of EPP effects cross-linguistically).

We propose that [uδ] features initiate an Agree relation, finding the discourse-prominent XP (i.e. the [δ]-bearing constituent) in the c-command domain of T and raising it to Spec,TP. Rather than Agreeing in situ, the [δ]-marked constituent must move to Spec,TP in order to satisfy local Probe-Goal union along the lines of Miyagawa (2010), which claims that goals must raise into proximity with their probes. In most instances, this [δ]-driven movement satisfies the EPP, since D-features on the raised constituent can check T’s D-feature. In this way, EPP in Kipsigis differs from prominence movement, since it is typically checked as a by-product of independently-motivated movement and does not act as a probe itself.

Consider the derivation of a simple VOS-Adv sentence like (52), which has a structure as in (53):

In (53) the [uδ] feature on T probes its c-command domain, identifying the [δ] (prominence) feature on the DP object bɑɑndeek ‘maize.’ This object then raises to Spec,TP to satisfy Miyagawa’s (2010) Probe-Goal union. Since the prominent constituent bears D-features, the D-feature on T is checked and TP’s EPP quality is satisfied. The verb raises via head movement to a functional projection above the D-featuring-bearing prominent constituent in order to generate V1.¹⁷ VOS-Adv word order is achieved through these two movements. The following sections address consequences of—and further evidence for—this account, while §5 addresses the prominence analysis and the nature of the [δ] feature.

4.3 On the position of the verb

The derivation in (53) shows that the verb moves through each successive phrasal projection before ultimately stopping at α.¹⁸ In §2.6, we showed that Kipsigis V1 and postverbal word order flexibility are independent of CP-level phenomena. The persistence of verb-initiality and word order flexibility within various types of embedded structures suggests that the verb’s final position and the syntactic mechanisms controlling scrambling occur below the complementizer domain. But, given our conclusions about discourse-driven scrambling to Spec,TP, the verb must also land in the head of a functional projection above Spec,TP in order to derive V1.

We conclude that V sits in the head of Miyagawa’s (2010) αP.¹⁹ In Kipsigis, the verb head-moves to α—the highest inflectional projection—to generate V1. Our conclusion about the relatively high position of the verb contrasts with several existing analyses of V1 languages, in which the verb surfaces at T or below (see Richards 2000; Massam 2001; Rackowski 2002; Aldridge 2004; Coon 2010; Clemens & Polinsky 2017; Clemens & Coon 2018; a.o.). There are, however, many exceptions; for instance, Richards (1996) puts the Tagalog verb above TP, as does McCloskey (2017) for Irish. We suggest that verb movement is motivated by a [PRED] feature on α—rather than [V] feature—as non-verbal predicates also raise (for similar proposals see Sproat 1985; Guilfoyle et al. 1992; Branigan 1996; McCloskey 1996; Carnie et al. 2000; Pearson 2001; Aldridge 2002; Rizzi 2004; Shlonsky 2004, a.o. Some researchers’ use of FinP mirrors our use of αP, e.g. Roberts 2004; 2005).

4.4 Satisfying PROM and EPP separately

But what happens when the discourse-prominent element is not capable of checking EPP, as was the case with answers to manner wh-questions? We saw in (32) and (33) that the manner adverb does not surface naturally in the IPP, despite bearing new information focus. Instead, the most appropriate answer to the manner question in (32) involves VS-Adv-O word order, where komie ‘well’ occurs in the second position after the verb and the subject is immediately postverbal:

The strong preference for VS-Adv-O word order in answers to manner questions is particularly notable, since it contrasts with the general dispreference towards VS-Adv-O word order (with manner adverbs) in all other contexts that we encountered.

In deriving (54a), the [uδ] feature on T probes its c-command domain and finds the relevant [δ] feature on the manner adverb, which bears [δ,FOC] features as a response to a manner wh-question (see §5 for more details). Following the expected procedure, the manner adverb komie ‘well’ raises to Spec,TP. However, it cannot satisfy the EPP as it does not bear D-features. We assume, therefore, that the D-feature on T is checked via Last Resort movement whereby the most local nominal (i.e. the DP subject) raises to Spec,TP as well, satisfying the EPP in a higher specifier position:²⁰

Crucially, on this approach, the EPP is only ever satisfied independently as a Last Resort when discourse-driven movement targets a constituent without D-features. While scrambling for discourse purposes is relatively non-local (as is often the case for discourse-driven movements; see Saito 1985; 1994; King 1993; Müller & Sternefeld 1993; Erteschik-Shir 2007 for a review of existing theories of discourse scrambling), EPP satisfaction is strictly local, resulting in the observed VS-Adv-O word order.

4.5 Are PROM and EPP a fused probe?

This analysis necessitates that the [uδ] and [D] features on T not be a fused probe in the sense of Coon & Bale (2014). On their account, multiple distinct features on a head probe as a unit and must (if possible) find a single goal that checks both features. Specifically, Coon & Bale propose that the probes for person and number agreement in Mi’gmaq are fused, searching for the appropriate person and number features simultaneously. Although this fused probe allows searches for person and number to be separate, the conditions of a successful match between probe and goal are fused together (i.e. failure in one search results in failure for the entire probe). Per Coon & Bale’s analysis, each potential goal is assigned a conjoined match of rank based on its ability to satisfy the components of the fused probe (see Coon & Bale 2014: 99 for detailed discussion of rank assignment). Most importantly, though, the one highest ranking constituent is ultimately targeted for agreement.

In Kipsigis, however, this is clearly not the case, since the discourse-prominent XP is not always immediately postverbal. Since prominence and EPP features are not always satisfied by the same phrase, these features need not act as a unit (though in most instances they appear to do so, since most discourse-prominent XPs that raise to Spec,TP are capable of satisfying T’s EPP quality). In this way, the Kipsigis facts diverge from those presented for Mi’gmaq.²¹

The Kipsigis facts are not unique, however. In his analysis of movement in the Nilotic language Dinka Bor, van Urk (2015) relies on a fused probe (or a composite probe, in his terms) consisting of φ-features and an unspecified A’-feature. He adopts a modified version of Coon & Bale’s conjoined match of rank calculus, in which “an active probe P enters into an Agree relation with the closest syntactic object that matches the most features” (van Urk 2015: 173). Crucially, though, van Urk’s implementation of fused probes diverges from Coon & Bale’s in one key way; if there is no single constituent that matches both fused features, two distinct partial match constituents can raise instead.

Why Dinka Bor’s composite probe would behave differently than the fused probe in Mi’gmaq is not entirely clear to us (it may have to do with the nature of the particular features in each fused/composite probe). Regardless, our analysis is essentially equivalent to van Urk’s split-able composite probe. Across Mi’gmaq, Dinka Bor, and Kipsigis, we see instances of Probe-Goal relationships in which a single Agree or movement operation often satisfies two distinct requirements on the probe. When this does not happen, however, Dinka Bor and Kipsigis allow those two distinct requirements to be satisfied by two phrases, whereas Mi’gmaq settles for the best-matching single phrase.

4.6 On pragmatically neutral word order

van Urk (2015) assumes that A’-features are optional as a definitional criterion and we assume something similar; there are sentences in Kipsigis with no [uδ] feature to be inherited by T. In these instances, it is only the EPP feature on T that triggers raising to Spec,TP. We have seen that when the EPP acts independently of prominence movement, it attracts the most local DP constituent (i.e. the subject). [uδ]-less sentences in Kipsigis will, therefore, display VS word order. This accounts for the acceptability of the pragmatically-neutral word order (VSO-Adv) in all discourse contexts in Kipsigis; it is the resultant word order in a derivation without a [uδ] feature on T.

5.1 Contrast is insufficient to characterize the IPP

Neeleman et al. (2009) argue that there is a three-way typology of information structure concepts encoded syntactically: topic, focus, and contrast. Contrast applies to topic and focus, yielding a four-way distinction between contrastive topic and aboutness topic vs. contrastive focus and new information focus.

Looking first at contrast, when a speaker’s response is intended to contrast with information conveyed in a previous utterance (i.e. contrastive focus scenarios), the contrastive constituent surfaces most naturally immediately postverbally, but need not surface in that position:

The response in (57a) is preferred without additional context, but the use of VOS in (57b) is natural if, for example, the speaker is trying to separate the rice from a variety of other foods: while Chepkoech ate the rice, other people ate other foods like beans or roasted maize. The potential acceptability of (57b)—in which the contrastively focused constituent surfaces clause-finally—shows that such constituents do not obligatorily occur in the IPP.

Along similar lines, many phrases which clearly do not create contrast are also preferred in IPP; this includes new information focus (§2.8.1), wh-words (§2.7), and aboutness topic (§2.9.2). This suggests that while certain contrastive phrases are preferred in the IPP, the notion of contrast itself is not an accurate generalization of the interpretation of IPP constituents.

5.2 Givenness is insufficient to characterize the IPP

If contrast is not the appropriate generalization, the typology from Neeleman et al. (2009) suggests it may instead be topic or focus. Yet definitions of topic and focus are notoriously difficult to pin down. It is commonly assumed that topics are discourse familiar/given, while foci are discourse unfamiliar/novel. However, many have argued that relative familiarity in discourse must be separated conceptually from both topic (Reinhart 1981) and focus (Krifka 2007). As seen in §2.8 and §2.9, discourse-new constituents (e.g. answers to wh-questions) are preferred in IPP, but at the same time, previously-mentioned aboutness topics are also preferred in IPP. In contrast, §6.1 shows that discourse-familiar topics instead tend to appear at the right edge of the clause (and aboutness topics are prohibited in that position). These patterns show that the interpretation of IPP constituents is orthogonal to discourse familiarity/givenness, as both types of material can surface there.

5.3 Aboutness topic is insufficient to characterize the IPP

Perhaps, then, aboutness is a better generalization of IPP interpretation. Reinhart (1981) conceives of pragmatic aboutness as a means of organizing propositions in the common ground; instead of being stored in an unordered fashion, propositions in the common ground are ordered based on sentence topics (i.e. discourse referents that each proposition is about).

We saw in §2.9 that aboutness topics are preferred in the IPP and that supplanting a continuing aboutness topic with a different phrase yields clear intuitions of a topic shift, such that the unexpected phrase in the IPP is now the aboutness topic. This indicates an aboutness quality of the IPP. A similar effect is evident in the ditransitive example in (58):

Here the prompt establishes the indirect object lɑɑgook ‘children’ as the aboutness topic. As expected, the pragmatically neutral word order VS-IO-DO (58b) is acceptable, but the preferred word order in (58a) places the indirect object in the IPP. Other responses are less acceptable, with two particularly incompatible responses exemplified here; lɑɑgook ‘children’ cannot appear clause-finally in this context (58c) and promoting some other phrase to the IPP is infelicitous (e.g. the direct object zɑwɑdi ‘gift’ in (58d)).

There are, however, multiple reasons to think that aboutness topic is not the correct generalization for the IPP. Frey (2004: 158) notes that quantificational phrases should be unable to serve as aboutness topics because they are unable to “point to a referent in the context set as would be needed for the aboutness relation” and shows that such phrases cannot serve as German middlefield aboutness topics. We see no such restriction on the Kipsigis IPP:

Likewise, non-referential phrases like wh-phrases should be inappropriate as aboutness topics, but recall from §2.7 that such phrases are preferred in IPP. Related, the pervasive assumption is that topic and focus are incompatible. If this assumption holds, the notion of aboutness topic must be insufficient to characterize a general interpretation of the IPP, given the preference for wh-phrases and new information focus in this position (§2.8). Altogether, despite the preference for aboutness topics in the IPP, the notion of aboutness topic does not capture all the IPP patterns in Kipsigis.

5.4 Focus is insufficient to characterize the IPP

Given the data presented so far, focus might seem like a viable IPP characterization; wh-phrases (§2.7), constituents with new information focus (§2.8), and contrastively focused phrases (§5.1) prefer IPP. Yet we have already encountered challenges to this generalization (e.g. aboutness topics in IPP) and this section introduces additional difficulties.

5.5 Empirical parallels to the Kipsigis prominence facts

5.6 Parallels between prominence and semantic highlighting

Notions like prominence, salience, and discourse-differentiation are an important part of a recent tradition of semantic work known as inquisitive semantics (Ciardelli et al. 2013). In traditional semantic frameworks, declaratives denote a set of possible worlds, while interrogatives denote a set of sets of possible worlds. In inquisitive semantics, both types of sentences denote a set of sets of possible worlds, known as possibilities (Coppock & Brochhagen 2013: 361).

In this approach, polar questions like “Is the door open?” and “Is the door closed?” have identical denotations, namely a set of two possibilities (i.e. the possibility that the door is open and the possibility that the door is closed). But as Roelofsen & Van Gool (2010) point out, there is an obvious empirical difference between these two questions in how they can be answered by a polarity particle:

Despite (79) and (80) having the same semantic denotation, polarity particles have opposite meanings in response to the different questions. This pattern is explained by the notion of highlighting, or making a possibility particularly salient. The questions in (79) and (80) highlight different possibilities via their explicit mention of one or the other, which then determines the inferences created by the response polarity particles. Highlighting has been used to explain a range of semantic and pragmatic phenomena (Roelofsen & Van Gool 2010; AnderBois 2013; Coppock & Brochhagen 2013; Roelofsen & Farkas 2015; Roelofsen 2016).

Roelofsen (2016) also points out that various pragmatic effects may arise on account of highlighting (e.g. the conclusion that the highlighted element is of particular relevance to the exchange). It is important to note that while highlighting itself has largely been taken to apply to possibilities (i.e. sets of possible worlds), Roelofsen & Van Gool (2010: 390) also give a semantics for highlighting individual entities. We see therefore that there is not only empirical precedent from Danish and Korean for prominence/salience as a grammatical phenomenon, but there is also a range of research on highlighting as a semantic phenomenon. We argue, then, that prominence (or salience, discourse-differentiation, or highlighting) is not merely a cognitive phenomenon but is a linguistic one as well, and we can expect to see it grammaticalized in a syntactic analysis.

5.7 Speculative proposal: The features of prominence

Returning to Kipsigis, recall that we propose that postverbal word order flexibility results from a scrambling operation to Spec,TP driven by discourse considerations. We have described this discourse effect as prominence, which captures speaker intuitions that the scrambled constituent is more important or especially relevant to the exchange. In the preceding subsections, we showed that this pattern crosscuts familiar information structure notions like specificity, contrast, givenness, and topic/focus (on various definitions). As such, we continue to use the term prominence to describe the empirical circumstance in which a particular constituent receives a salient/highlighted/discourse-differentiated interpretation. Speakers’ tendency to attribute aboutness interpretations to the IPP constituent can be attributed to relevancy inferences generated by highlighting a particular constituent (as suggested by Roelofsen 2016).

Yet how exactly should a notion like prominence be implemented in a syntactic analysis? Rather than assuming a de novo [PROMINENCE] feature, we assume that it is grammaticalized syntactically via an underspecified [δ] (discourse) feature that can be satisfied by phrases of any information structure designation or—even more generally—by any phrase that is sufficiently discourse-differentiated (in the sense developed in the discussion of Mikkelsen 2015). We assume a rudimentary feature hierarchy of discourse-features in syntax, in which all information structure features (e.g. topic, focus, etc.) are also designated as [δ] (i.e. differentiated from the rest of a sentence with respect to a discourse property). But if a constituent or head contains a lone [δ] feature without a more specific information structure designation, it is simply discourse-differentiated; that is, it is semantically highlighted without further syntactic or semantic specification. The T head in Kipsigis, therefore, bears a [uδ] feature that can be satisfied by any discourse-differentiated phrase, regardless of its particular designation.

We assume that discourse-driven movement to Spec,TP in Kipsigis follows the type of free movement described in Miyagawa (2010). Recall that Miyagawa claims that in Japanese the [–focus] probe inherited by T can trigger free movement of either the subject or the object regardless of locality, as it simply requires Spec,TP to be filled. Whichever constituent surfaces in this position receives a topical interpretation. Although Miyagawa restricts free movement to [–focus] elements, we extend his reasoning to all discourse-differentiated elements in Kipsigis; that is, any constituent with a [δ] feature—be it a lone [δ] or an information structure specified [δ]—can raise to Spec,TP to satisfy Probe-Goal union regardless of locality. We saw many instances above in which a phrase expected to be in the IPP is supplanted by another phrase deemed more important. We claim that this prominence interpretation of the IPP—the “more important” supplanting effect—arises via pragmatic implicature. This occurs when there are multiple discourse-differentiated elements in a single sentence and though the grammar allows the speaker to move a more specific discourse-differentiated constituent (e.g. [δ,foc]) to the IPP, they instead place a less specific discourse-differentiated constituent (i.e. [δ]) there. Choosing the underspecified feature set (vs. the more specific feature set) creates the implicature of comparatively larger importance of the constituent raised to the IPP.

Therefore, wh-phrases, responses to wh-questions, and contrastively focused constituents typically surface in the IPP because they bear specified [δ,FOC] features and raise to Spec,TP to satisfy Probe-Goal union with the [uδ] feature on T. Yet when speakers choose to make use of unexpected or marked word orders in these contexts, movement of a distinct [δ]-marked constituent creates an implicature and generates a prominent interpretation of comparative importance for that phrase. Similar reasoning applies to the multiple-wh facts. This account also explains the lack of focus movement of ɪʧɛɛgɛn ‘only’ constituents to the IPP. Semantic focus—signaled by ‘only’—is distinct from discourse differentiation; only discourse-differentiated constituents surface in the IPP, so movement of semantically focused ‘only’ constituents is not predicted. Aboutness topics, on the other hand, generally surface in Spec,TP, as they are specified as [δ,TOP]. Both focus and topic constituents are discourse-differentiated and, in this way, form a natural class.

Our analysis draws heavily on the notion of discourse differentiation from Mikkelsen (2015), though it differs from her account in some important ways. Recall that Mikkelsen claims that Danish grammar makes a consistent distinction between clause-initial XPs that are information structure differentiated and those that are not. Mikkelsen assumes that all C heads in Danish bear some kind of discourse-related feature that triggers movement of an XP with the corresponding discourse function to Spec,CP.²⁹ Standard information structure features on C and standard feature-checking operations drive Mikkelsen’s analysis. We could, in principle, implement this kind of analysis in Kipsigis; instead of T bearing a universal [uδ] feature, Kipsigis T would have many instantiations in the lexicon, each with distinct discourse features. Yet this faces empirical and theoretical challenges. If, for instance, Kipsigis T is sometimes lexically specified as [uFOC], how can we account for the supplanting effect where some other (non-focused) XP can raise to that position if the XP is considered more important? This approach also seems to miss an important generalization; the fact that IPP constituents comprise a natural class of discourse-differentiated XPs is simply a coincidence under this approach, since there is nothing explicitly preventing T from having countless other lexical specifications.³⁰ Given this, we adopt Mikkelsen’s idea of discourse differentiation, but we formalize it as an underspecified [δ] feature that can account for a single discourse-prominence movement to Spec,TP despite that movement accommodating many different sorts of information structure distinctions.

In this section, we have offered a speculative proposal of how to formalize prominence movement to Spec,TP. For the sake of space, we necessarily leave several questions unanswered. We have not yet spelled out or defended an entire discourse feature hierarchy, nor have we explored a specific semantics of prominence/highlighting in Kipsigis. Yet the analysis here—based upon solid empirical and theoretical precedent—is a step in understanding the discourse role of the IPP in Kipsigis and the syntactic mechanisms underlying movement to this position.

6.1 Rightward extraposition of topical subjects

Currently, our analysis relies on a single scrambling process—namely, [δ]-motivated movement to Spec,TP—which usually satisfies the EPP quality on T. As it stands, our analysis allows for the generation of the vast majority of word order patterns that we have observed in Kipsigis. However, the derivation of word orders like V-Adjunct-OS necessitates an additional process; scrambling of the adjunct to the IPP alone cannot give rise to V-Adjunct-OS word order.

In discussing the derivation of VOS word order in Tongan, Polinsky (2016) argues for base-generation of topical subjects in a high right-side specifier position. There is ample evidence that the content of right-side subjects in Tongan is backgrounded or presupposed; for instance, subject pronouns often surface clause-finally when their antecedents have already been introduced in discourse, focused elements are impossible in clause-final position, and if a wh-phrase occurs clause-finally, then it must be interpreted as an echo-question with the wh-phrase taking the widest scope. Polinsky (2016) argues that right-side subjects are base-generated in that position based on patterns of accentuation and binding.

Clemens & Coon (2018) argue that VOS word order in Mayan languages may arise from multiple distinct operations including heavy-NP shift, prosodically-motivated re-ordering of constituents, and right- and left-peripheral topic positions. These different paths to VOS word order are each independently motivated and may coexist in a single language. Clemens & Coon (2018: 269) show that topics may raise to a clause-peripheral position (internal topics), or be base-generated in that position (external topics):

(81)	Internal Topics
	a.		b.

(82)	External Topics
	a.		b.

We can borrow the term “rightward extraposition” from Polinsky (2016) as an umbrella term encompassing both displacement and base-generation of a constituent on the right edge of the clause.

Returning now to the derivation of V-Adjunct-OS word order in Kipsigis, we note the findings of Creider & Creider (1983) for Nandi, a Kalenjin language closely related to Kipsigis. They demonstrate that there is a tendency in Nandi for discourse-familiar arguments to appear at the right edge of the clause. This comment-topic pragmatic ordering generally aligns with the Kipsigis facts described in this paper; while discourse-prominent constituents tend to surface immediately postverbally, discourse-familiar constituents tend to surface clause-finally. Although these two observations might seem like two sides of the same coin, this is complicated by the fact that aboutness topics are interpreted as prominent and are preferred in the IPP. We suggest that only discourse-familiar topics occur in this right-edge position.

There are select examples in our texts where a clearly discourse-familiar topic appears clause-finally and—just as crucially—no contradictions to this generalization.³¹ In (83) the subject cheplanget ‘leopard’ is the familiar topic and appears sentence-finally in the VOS sentence:³²

It also seems that these right-side topics are necessarily discourse-given topics and cannot be salient, sentence-level aboutness topics. This is illustrated by one of the aboutness topic prompts from above in (84a), where the aboutness topic was placed at the right edge.³³ A consultant explicitly offered that (84b) is problematic because “children should not be an afterthought:”

This fits with the conclusion that there is a right-side topic position, but that it is explicitly for discourse-familiar topics and not for aboutness topics (which are instead attracted to the IPP). This assumes a bifurcation among so-called topics where discourse-given topics are backgrounded, while aboutness forms a natural class with focus, contrast, and simple emphasis as highlighted constituents (in the sense discussed in §5.6).

Returning to the leopard example in (83), a consultant—who was not present when the story was originally told—considered a rephrasing of the relevant sentence that instead adopted canonical VSO word order:

This third party consultant actually preferred this word order to the original, but for an interesting reason; the VOS word order in (83) was judged to convey the relevant information too straightforwardly (i.e. was almost too neutral in the narrative context). The VSO word order in (85), on the other hand, was judged a more powerful, dramatic story-telling choice, largely because it suspended the listener to the last minute; the speaker is hiding the relevant information, creating an unexpected word order. Assuming that these interpretations derive from implicatures created by not choosing the expected order, we see not only that prominent material is expected in the IPP, but also that backgrounded, familiar information is expected at the right edge and that putting the discourse-familiar topic in non-clause-final position is surprising.

Given these observations, it is possible to account for V-Adjunct-OS word order in Kipsigis by positing that discourse-familiar subjects can appear as right-edge topics. The specific structures and derivations underlying these right-side topics are secondary to our present concerns, but the existing analyses from Polinsky (2016) and Clemens & Coon (2018) provide a framework for probing this question in the future.

The resulting analysis in effect overdetermines the facts; that is, in a given VOS sentence, it is not transparent whether this structure is derived through object scrambling to Spec,TP or through rightward extraposition of the subject. However, this overdetermination is neither unprecedented in the V1 literature, nor necessarily a shortcoming of our analysis. Clemens & Coon (2018) explicitly propose three different paths to VOS word order in Mayan, all of which can coexist in a single language. In addition, appealing to common sense notions, the willingness of speakers to accept any possible postverbal word order as grammatical—and even pragmatically viable given an imagined scenario—may fit well with a grammatical situation in which there are multiple means to any particular word order.

6.2 A- vs. A’-scrambling to the IPP

This section offers preliminary evidence that scrambling to the immediately postverbal prominence position in Kipsigis displays mixed A-/A’-effects, similar to other languages with parallel scrambling operations. Movement to the postverbal position in Kipsigis has A’-effects with respect to quantifier-variable binding, as the moved element reconstructs to its base position. A quantified DP subject can bind a pronominal variable in the DP object regardless of the surface word order:

The object pronominal variable in (86b) cannot be be bound by the quantifier in its surface position, yet a bound reading is nonetheless natural in this context. This suggests that movement to the prominence position has not created a new binding position, a reconstruction effect characteristic of A’-movement. Likewise, when the subject contains the variable and the object the quantifier, scrambling the object over the subject does not change the binding relationships:

Movement of the object over the subject in the VOS sentence in (87b) does not remedy the unacceptable binding configuration; the bound reading remains unavailable. This once again suggests that base positions of arguments are respected for interpretation, a characteristic of A’-movement.

However, this same scrambling process is able to generate a Principle C violation, suggesting that movement creates new binding possibilities, which is typical of A-movement. Principle C of Chomsky’s Binding Theory states that R-expressions can never be bound (Chomsky 1981). In (88a), Principle C is not violated, since the R-expression Kipɾoono ‘Kiproono’ is not c-commanded by the coreferential pronoun ɪnɛɛndɛt ‘him.’ However, movement of the object pronoun to the IPP in (88b) renders the coreferential reading ungrammatical:

In (88b) the object pronoun scrambles to a higher syntactic position (argued to be Spec,TP) and, from this new position, c-commands the R-expression, rendering it unacceptably bound. Scrambling the object over the subject, therefore, results in a change in binding relations.³⁴ In this way, there is preliminary evidence that movement to the IPP in Kipsigis displays mixed A/A’-effects.

Kipsigis is not alone in showing mixed mixed A-/A’-properties. Miyagawa (2010) shows that scrambling to Spec,TP in Japanese displays properties of both A- and A’-movement. Holmberg & Nikanne (2002: 23) show that a similar topicalization operation in Finnish has similarly mixed effects. Van Urk (2015) observes that movement in Dinka Bor is long-distance like A’-movement, but behaves like A-movement with respect to binding. The Kipsigis properties are particularly complicated, since the prominence-based scrambling to Spec,TP does not behave uniformly across all binding diagnostics, though the Kipsigis patterns are not alone in this respect. Richards (2013) observes that it is not uncommon for languages to have a movement operation that only affects certain kinds of binding relationships; German scrambling has been proposed to have no effect on Principle A despite repairing weak crossover, while Scandinavian object shift has been argued to create a new binding position for Principle C despite not doing so for Principle A (see Holmberg & Platzack 1995 and Grewendorf & Sabel 1999). Although we do not offer a complete analysis of these mixed A-/A’-effects in Kipsigis here, we simply contribute the Kipsigis scrambling facts to the growing literature on mixed A-/A’-movements, recognizing that there is more work to be done both to clarify these properties empirically and to develop an analytic framework that can capture such mixed effects.

6.3 Conclusions

This paper provides the first empirical description and analysis of the basic word order properties of Kipsigis, showing both the prevalence of predicate-initial word order and the role of information structure in determining postverbal word order. To account for the observed patterns, we proposed a head movement account of Kipsigis V1, in which the verb raises to a functional projection above Spec,TP but below CP, termed αP by Miyagawa (2010). On this analysis, discourse-prominent material moves to Spec,TP to satisfy a [uδ] feature on T (following Otsuka 2005a; Miyagawa 2010). Because this postverbal prominence position was shown to have an EPP quality—preferring D-feature-bearing constituents—we proposed that [δ]-driven movement typically checks T’s D-feature as a by-product; yet when the [δ]-marked constituent is unable to satisfy the EPP quality on T, the most local DP constituent (i.e. the subject) moves to a second TP specifier as a Last Resort. In extended discussion of the interpretation of prominence in Kipsigis, we showed that no widely-accepted information structure concept captures the generalization well (e.g. topic, focus, contrast, givenness, etc.); prominent constituents can be focused phrases, topics, wh-words, and even otherwise unmarked constituents that the speaker deems most important. This motivates our adoption of an underspecified [δ] feature driving movement to the IPP, capturing a natural class within sentential information structure of discourse-differentiated phrases that subsumes topic, focus, and contrast.