Complementizer Agreement in Kipsigis

Complementizer Agreement in Kipsigis Meghana Rao Pomona College April 22, 2016 Advised by Michael Diercks Submitted to the Department of Linguistics and Cognitive Science at Pomona College In partial fulfillment of the requirements for the degree of Bachelor of Arts

1

Contents 1 Introduction 4 1.1 Upward Agreeing Subject-Oriented Complementizers in Kipsigis............... 4 1.2 Analytical Possibilities............................ 6 1.3 Upward Agreeing Object-Oriented Complementizers in Kipsigis............... 6 2 RelevantBackgroundon KipsigisPhraseStructure 8 2.1 Flexibility of Word Order............................. 8 2.2 Tone As A Case Marker.................................. 9 2.3 Basic Word Order..................................... 10 2.3.1 Context Dependence and Focus Positions....................... 10 2.3.2 Out-of-the-Blue Diagnostic........................ 12 2.4 Basic Analysis of Phrase Structure..................... 13 2.4.1 Evidence and Arguments For Head Movement.............. 14 2.4.2 Verb Movement to TP................................. 15 2.4.3 Binding Data Suggests A Movement......................... 16 2.5 Summary of Kipsigis Phrase Structure Conclusions........ 17 3 PropertiesofPrefixedComplementizerAgreement(Subj-CA) 18 3.1 Partial Complementizer Inventory..................... 18 3.2 Prefixed Complementizer Agreement Forms................ 19 3.3 Prefixed CA Agrees with the Most Local Matrix Subject................... 20 3.4 Subj-CA in Impersonal Constructions................... 21 3.5 (Non-)Locality Effects for Subj-CA.................... 22 3.5.1 CPs at the Right Edge................................. 23 3.5.2 Subj-CA Possible Over an Intervening Object........... 23 3.5.3 Subj-CA out of Noun Complement Clauses................ 24 3.6 Intermediate Conclusions: Prefixed (Subj-) CA........... 25 4 InterpretationofSubj-CA 26 4.1 Information Source Effect on Subj-CA................... 26 4.2 Common Ground Distinguishes Subj-CA........................... 27 4.3 Rejection of an Embedded Assertion Hypothesis....................... 29 4.4 CP as the Main Point of the Utterance (MPU)................ 30 4.5 Conclusions........................................ 31 5 SuffixedComplementizerAgreement(Obj-CA) 32 5.1 Suffixed Complementizer Agreement Forms (Obj-CA).................... 32 5.2 Suffixed CA Targets the Most Local Matrix Object........... 32 5.3 No Obj-CA on Non-agreeing Complementizer............... 33 5.4 Suffixed (Obj-) CA in NCCs.............................. 34 5.5 Suffixed (Obj-) CA in Impersonal Constructions........... 34 5.6 Interpretation of Obj-CA............................. 34 5.7 Intermediate Conclusions: Suffixed (Obj-) CA............ 36 6 TheoreticalDiscussion and Conclusions 37 6.1 Summary of Findings.................................. 37 6.2 Theoretical Implications.................................... 37 6.2.1 Subj-CA as a Widespread Phenomenon........................ 37 6.2.2 Obj-CA and Evidence for Raising Complementizers....... 38 6.3 Issues for Future Research............................ 41 2

6.3.1 Kipsigis Phrase Structure.......................... 41 6.3.2 Degradation of Subj-CA with DP Objects................ 42 6.3.3 Obj-CA: Further Looks.............................. 42 6.3.4 Effect of Verb Type on Complementizer Acceptability................ 42 6.4 Conclusion......................................... 43 3

1 Introduction 1.1 Upward Agreeing Subject-Oriented Complementizers in Kipsigis The Agree operation is a central part of Minimalist syntactictheory.chomsky s(2000,2001)agreerelation posits that unvalued features serve as Probes on heads, whichsearchtheirc-commanddomainsforagoalto value their features. (1) YP Y [Probe] XP... Goal... Complementizer agreement (CA) is one of many operations that havetraditionallyinvolvedagreeandca patterns will be the focus of this paper. Several well-known complementizer agreement relations accord relatively well with this conception of Agree (see Carstens (2003) and van Koppen (2005) for West Germanic CA, and Deal (2015) for Nez Perce CA). The examples in (2) from Carstens (2003) clearly demonstrate a downward agreeing complementizer. (2) a. Kpeinzen dan-k (ik) morgen goan. I-think that-i (I) tomorrow go I think that I ll go tomorrow. b. Kpeinzen da-j (gie) morgen goat. I-think that-you (you) tomorrow go I think that you ll go tomorrow. [West Flemish] Deal (2015) shows that Nez Perce CA appears on a variety of complementizers (yes/no question particles, an inferential evidential marker, a conditional, a negator, andana -particle). Thoughdistinctfromthe declarative-embedding complementizers of West Germanic, Lubukusu, and Kipsigis, these nonetheless show agreement properties, as shown in (3). In this example from Deal (2015), the complementizer bears plural agreement with the lower subject and 2nd person agreement with the lower object. 1 (3) ke-pe-m kaa A.-nim kaa T.-nm hi-cewcew-tée nix pro obj C-PL-2 then A.-ERG and T.-ERG 3SUBJ-telephone-TAM PRO.2SG 3pl/2sg: whena.andt.callyou(sg) [Nez Perce] 1 Though this Nez Perce data is certainly intriguing, the concerns it raises are not central to the patterns presented in this paper and so will not be explored here. 4

Following the traditional Agree relation mechanism, these attested examples all show agreement from a structurally higher probe (on C) with a c-commanded goal. (4) CP C [φ: ] TP... Goal... Kipsigis is a Nilotic language of the Kalenjin subgroup, spoken in western Kenya by roughly 2 million people (Lewis et al., 2016). 2 It is a verb initial language, with both VSO and VOS word orders, a phenomenon that is discussed further in section 2. In contrast to the more well-known forms of agreement presented above, Kipsigis shows an upward-oriented pattern of agreement where complementizers agree not with the subject of the embedded clause 3 but with that of the main clause 4 : (5) ko-a-mwaa A-lE ko-ruuja tuga amut PST-1sg-say 1sg-C PST.3-sleep cows yesterday I said that the cows slept yesterday. [Kipsigis] This pattern of complementizer agreement has been documented for relatively few languages. It has been most systematically investigated in Lubukusu (Bantu, Kenya), though it has also been documented in Chokwe, Luchazi, Lunda, and Luvale (central Bantu languages), and some Mande languages (Diercks, 2013) (Kawasha, 2007) (Idiatov, 2010). An example from Lubukusu appears in (6) (Diercks, 2013). (6) a. baba-ndu ba-bol-el-a Alfredi ba-li a-kha-khil-e 2-people 2S-said-AP-FV 1Alfred 2-that 1S-FUT-conquer The people told Alfred that he will win. b. Alfredi ka-bol-el-a baba-ndu a-li ba-kha-khil-e 1Alfred 1S-said-AP-FV 2-person 1-that 2S-FUT-conquer Alfred told the people that they will win. [Lubukusu] Asimilarupward-orientedcomplementizerpatternhasalsobeen recently documented in Ibibio, a Lower Cross language of Nigera (Torrence, 2016). 5 (7) m-ma ng-kop (m-bo/n-te) ke 1sg-PAST 1sg-hear 1sg-COMP that I heard that Koko bought a book. Koko koko a-ma 3sg-PAST a-dep ngwet 3sg-buy book [Ibibio] While research on the Ibibio pattern is still underway, its existence (coupled with the Kipsigis pattern described in this paper) demonstrates the widespread nature of this linguistic phenomenon. Upward-facing subject-oriented complementizer agreement is present in Bantu, Nilo-Saharan, and Benue-Congo language families, spanning East to West Africa. Clearly, it does not seem to be a strictly genetic or areal phenomenon. 2 The data were collected by the Field Methods class, by Professor Michael Diercks, and by me over 2015-2016 at Pomona College. Speakers during these sessions were Sammy Bor and Robert Langat. Grammaticality judgments are based entirely on their interpretations. 3 Subjects will be glossed with only the number and person of the DP(andnotadistinctsubjectgloss).Markersofobjectswill be explicitly indicated as designating objects. 4 Tone plays a syntactic role in Kipsigis, a fact that is elaborated on in section 2.2. However, tone transcription is omitted from this paper unless directly relevant to a particular example because there is no existing analysis of tone patterns in KipsigisandIdonotwant to unintentionally misrepresent the data. 5 Ihavealteredtheboldingtoemphasizecertainpartsofthesentence 5

1.2 Analytical Possibilities While these upward-oriented complementizer patterns pose significant questions for linguistic theory, this paper will offer a systematic description of the syntactic and interpretive properties of Kipsigis CA without systematically arguing for any particular analysis or constructing a new theory to account for these phenomenon. Instead, I will briefly lay out two analytical possibilities that can account for the data presented in this paper, to be elaborated on in section 6. Given the similarities between the Kipsigis and Lubukusu subject-oriented prefixed complementizer agreement patterns (discussed further in section??), one plausible analysis to account for the Kipsigis data comes from Diercks (2013). This paper analyzes upward-facing subject-oriented complementizer agreement pattern as an anaphoric relation. Briefly, the analysis laid out in Diercks (2013) posits a null operator (specifically a subject-oriented anaphor) in speccp of the embedded clause, which is accordingly bound by the matrix subject. Agreement features appear on the complementizer through a local agree relation between the operator in speccp and the complementizer in C position. DierckstermsthisoperationIndirect Agree, indicating an apparent Agree relation that is in fact mediated by another syntactic element, in this case a null anaphor. While there are some differences between the Kipsigis and Lubukusu patterns, this analysis fits well with the Kipsigis prefixed CA pattern and I will not attempt to further this account in this regard. Another more recent analytical possibility that, interestingly, also arises from the Lubukusu pattern is that of Delayed Valuation, as proposed by Carstens (2015). Carstens posits that if a set of unvalued features is not valued immediately after it has been merged onto the structure, this lack of valuation will not cause the sentence to fail unless it is at the point of Transfer, at the edge of a phase. She terms this Delayed Valuation.ShesuggeststhattheagreeingcomplementizerfoundinLubukusu begin in ForceP and that there is a phase boundary between ForceP and FinP (following the articulated left edge proposed by Rizzi (1997)). Due to Chomsky (2000) s Phase Impenetrability Condition, the complementizer cannot probe the embedded subject as it has passed the point of Transfer and instead, delays its valuation until the matrix clause vp is merged, whereupon ForceP features raise to specvp and then are successfully valued by a downward probing operation. Like the analysis from Diercks (2013), this analysis does for the most part account for the subject-oriented CA data presented in this paper. IwillnotbeconsideringanalysesinvolvingupwardprobingAgree relations, such as those described by Zeijlstra (2012) and Baker (2008), in which goals with interpretable features move above probes with uninterpretable features, and probing and agreement occur unidirectionally upwards. These do not fit the patterns present in the data found in Lubukusu and here in Kipsigis, for reasons more thoroughly discussed in Diercks et al. (2016a). To summarize the arguments presented there, one would expect intervention effects from an indirect object in the matrix clause given an upward Agree relation. This is not found in either Lubukusu or Kipsigis complementizer agreement. Therefore, an upward Agree relation cannot be a viable analysis of these patterns. 1.3 Upward Agreeing Object-Oriented Complementizers in Kipsigis As mentioned above, Kipsigis also demonstrates an upward facing agreement relation similar to that described in 1.1 except with the matrix object as the agreement target, rather than the matrix subject. (8) ko-a-mwaa-un A-lE-ndZin ko-it tuga amut PST-1sg-tell-2sg.OBJ 1sg-C-2sg.OBJ PST-arrive cows yesterday I told you (sg) that the cows arrived yesterday. [Verum Focus-like effect] 6

In contrast to the CA pattern in the previous section, this object-oriented agreement form is realized as asuffixonthecomplementizerratherthanaprefix.thispattern seems to be a novel one, not documented (to my knowledge) formally in other languages in the current literature. Significantly, this suffixed CA pattern is not well-accounted for by most current theories addressing upward-facing subject-oriented CA, for reasons more thoroughly covered in section 6. In order to account for this, I draw from the ideas in the more recent work of Diercks et al. (2016a) and from Carstens (2015). Briefly, Diercks et al. (2016a) s analysis posits that anaphoric φ features generated higher up in ForceP move to the edge of the main-clause vp, agreeing with the main clause subject in its base position at the edge of vpthroughatraditionaldownwardagree operation (Diercks et al., 2016a). Note that this analysis is nottoodifferentinabroadertheoreticalsense from that of Carstens (2015). Given a range of facts surrounding suffixed object-oriented CA, which will be discussed in section 5, I posit that the suffixed object-oriented pattern is in fact a clitic-doubling pattern on the complementizer. Instead of φ features in raising up, I suggest that the entire complementizer raises into the matrix clause, thus making it available for a clitic-doubling operation with the matrix object. This analysis will not be systematically defended, but I suggest it as a preliminary account of these puzzling facts and elaborate on this in section 6. My primary focus in this paper will be a description of Kipsigis complementizer agreement patterns, but I will first lay out a framework from which to analyze basic Kipsigis phrase structure. Section 2 begins with an overview of the flexibility of Kipsigis word order and proceeds to lay out some basic assumptions of Kipsigis phrase structure that I will hold throughout the paper. I then describe the properties of the upwardoriented subject complementizer agreement relation in section 3, demonstrating broad similarity between the Kipsigis pattern and previously-documented patterns, particularly that in Lubukusu (as described by Diercks (2013). Section 4 explores some of the interpretive differences between the agreeing and nonagreeing prefixed complementizer. In section 5, I describe the novel agreement pattern of upward-oriented object agreement on complementizers, offering only a few speculations as to the mechanism by which this could be achieved. Finally, in section 6, I discuss the theoretical relevance of both of these complementizer agreement patterns and offer a possible analysis to account for them. 7

2 Relevant Background onkipsigis PhraseStructure 2.1 Flexibility of Word Order As mentioned in section 1, Kipsigis is a verb-initial language that allows for VSO and VOS word orders. These word orders are both accepted as basic configurations for Kipsigis sentences. 6 (9) a. ko- -GeeR tsito n daaret PST-3-see person snake " The person saw the snake (recently). b. ko- -GeeR n daaret tsito PST-3-see snake " person The person saw the snake (recently). VSO VOS These two word orders are generally interchangeable, though theyhaveslightinterpretivedifferences,a point that is relevant in section 3.4. Word orders in which the verb is not the first constituent is ungrammatical. 7 (10) a. * tsito ko-geer n daaret person PST.3-see snake " The person saw the snake (recently). b. * tsito n daaret ko-geer person snake " PST.3-see The person saw the snake (recently). c. * n daaret ko-geer tsito snake " PST.3-see person The person saw the snake (recently). d. * n daaret tsito ko-geer snake " person PST.3-see The person saw the snake (recently). SVO SOV OVS OSV As long as the verb precedes the rest of the constituents in the sentence,allvariationsinwordorderwithin that framework appear to be grammatical. This same flexibility in word order is present in ditransitive sentences, as well. 6 Sentences with third person subjects do not have overt subject markers, here marked as a null morpheme. For ease of readability, I will include the third person subject marker with the tense marker and will not gloss a null morpheme for the remainder of this paper. 7 It is possible to place constituents before the verb in topic/focus-like constructions in which a constituent is moved to theleft periphery of a clause. This, however, requires an extra particle to be placed between the fronted constituent and the rest of the clause. I assume these are movements into the left periphery of the clause to CP-level projections, and are not directly relevant to the fundamental verb-initial word order. 8

(11) a. ko-goo-tsi kiruoogiindet tsepkoets artet PST.3-give-3.OBJ chief Chepkoech goat The chief gave Chepkoech a goat (recently). b. ko-goo-tsi kiruoogiindet artet tsepkoets PST-3-give-3.OBJ chief goat Chepkoech The chief gave Chepkoech a goat (recently). c. ko-goo-tsi tsepkoets kiruoogiindet artet PST.3-give-3OBJ Chepkoech chief goat The chief gave Chepkoech a goat (recently). d. ko-goo-tsi tsepkoets artet kiruoogiindet PST.3-give-3.OBJ Chepkoech goat chief The chief gave Chepkoech a goat (recently). e. ko-goo-tsi artet kiruoogiindet tsepkoets PST.3-give-3.OBJ goat chief Chepkoech The chief gave Chepkoech a goat (recently). f. ko-goo-tsi artet tsepkoets kiruoogiindet PST.3-give-3.OBJ goat Chepkoech chief The chief gave Chepkoech a goat (recently). V S IO DO V S DO IO VIOSDO VIODO S V DO S IO V DO IO S When other constituents are added into a sentence, the extentofthisflexibilitybecomesincreasinglyapparent. In (12), a temporal adverb is grammatical in any postverbal position. (12) (*amut) ko-al (amut) artet (amut) girwogindet (amut) (*yesterday) PST.3-buy (yest) goat (yest) chief (yest) The chief bought a goat yesterday. 8 Just like with the DP constituents, the only ungrammatical sentence is one with a constituent appearing in a pre-verbal position. 2.2 Tone As A Case Marker Tone appears to have a role in case-marking in Kipsigis. Non-extracted subject DPs are marked distinctly from the tone of the same DP in a non-subject role (See Jake and Odden, 1979). 9 (13) a. kò-gé ē R tsì t ò n dāārēt PST.3-see person(sbj) snake " The person saw the snake. b. kò-gé ē R tsì t ō ǹ dāārèt PST.3-see person snake(sbj) " The snake saw the person. 8 Though there may be slight variations in precise meaning, each word ordering roughly translates to the same sentence. Also note that this example does not intend to suggest that it is possible to have multiple copies of the temporal adverb in the same clause. 9 Acompletedescriptionandanalysisofthetonesystemisoutside the scope of this paper. 9

Generally, nouns in nominative case positions display a lower tone pattern overall. A consistent subject tone pattern, however, has not yet been discovered and is a topic for future research. While tone markings could be characterized as markers of Case, DP constituents in object position have the same tone patterns as those in citation form, as shown by (14). (14) a. kò-gé ē R tsì t ō ǹ dāārèt PST.3-see person snake " The snake saw the person (recently). b. tsì t ō person c. kò-gé ē R tsì t ò n dāārēt PST.3-see person snake " The person saw the snake (recently) d. n dāārēt snake " This phenomenon certainly bears further investigation, but Iwillnotgrapplefurtherwiththetonepuzzle here. 2.3 Basic Word Order The main focus of this paper is on complementizer agreement, so it is not possible to fully explore analyses of Kipsigis phrase structure. This section provides some background on the salient aspects of Kipsigis phrase sructure and offers a working analysis that can serve as a basis to move forward into the investigation of complementizer agreement. Though I have indicated that there are two most natural word orders (VOS, VSO), it is difficult to determine which of these is the basic word order, as both have often been deemed interchangeable by native speakers. Through an out-of-the-blue diagnostic (laid out in 2.3.2 ), I conclude that VSO is the most natural word order for Kipsigis. 2.3.1 Context Dependence and Focus Positions AkeyaspectofKipsigiswordorderisthatitisheavilycontext-dependent. Specifically, there is an immediately post-verbal focus position, where high-prominence (focused) constituents in the sentence are preferred. As demonstrated by (15), focused DPs appear immediately post-verbally. (15) Q: ko-al Noo artet amut PST.3-buy who goat yesterday Who bought the/a goat yesterday? A1: ko-al girwogindet artet amut PST.3-buy chief goat yesterday The CHIEF bought the/a goat yesterday. A2: # ko-al artet girwogindet amut PST.3-buy goat chief yesterday The chief bought the/a GOAT yesterday. 10

One of our consultants judgment on the answer in (A2) was that itfeelslikeyou reansweringthewrong question. In other words, that answer was placing focus on a constituent that did not answer the question being asked. While (15) showed an example focusing a subject DP, the example in (16) demonstrates that this judgment holds for questions regarding object DPs, as well. (16) Q: ko-al nee girwogindet amut PST.3-buy what chief yesterday What did the chief buy yesterday? A1: ko-al artet girwogindet amut PST.3-buy goat chief yesterday The chief bought a/the GOAT yesterday. A2: # ko-al girwogindet artet amut PST.3-buy chief goat yesterday The CHIEF bought a/the goat yesterday. Clearly, there appears to be some sort of relevance to the constituent in the immediate post-verbal position. This emphasis is not restricted to DPs and can be applied to other constituents such as yesterday in (17). 10 (17) Q: ko-al au girwogindet artet PST.3-buy when chief goat When did the chief buy a goat? A1: ko-al amut girwogindet artet PST.3-buy yesterday chief goat The chief bought the goat YESTERDAY. A2: # ko-al artet girwogindet amut PST.3-buy goat chief yesterday The chief bought the GOAT yesterday. A3: # ko-al girwogindet artet amut PST.3-buy chief goat yesterday The CHIEF bought the goat yesterday. Given this pattern, it is possible to suggest that these data are in fact the result of echoing the word order of the question and are not actually a sign that constituents are focused in the immediate post-verbal position. The examples in (18) ask the same question as the one in (17), but place the wh- word ( who ) in a different spot within the sentence. Each version of the question is followed by the answer word ordering that was designated most natural for the speaker. 10 Though (A1) was the most preferred word order out of the options in (17), the temporal adverb amut is slightly dispreferred in the post-verbal position overall. There seems to be a preference for DPs in the immediate post-verbal spot,but this is a topic for future research. 11

(18) a. Q: ko-al Noo artet amut PST.3-buy who goat yesterday Who bought the goat yesterday? A: ko-al girwogindet artet amut PST.3-buy chief goat yesterday The CHIEF bought the goat yesterday. b. Q: ko-al artet Noo amut PST.3-buy goat who yesterday Who bought the goat yesterday? A: ko-al girwogindet artet amut PST.3-buy chief goat yesterday The CHIEF bought the goat yesterday c. Q: ko-al artet amut Noo PST.3-buy goat yesterday who Who bought the goat yesterday? A: ko-al girwogindet artet amut PST.3-buy chief goat yesterday The CHIEF bought the goat yesterday. V wh- OAdv VOwh- Adv VOAdvwh- The preferred position of the constituent that directly answered the question (in this case, girwogindet) remained the same regardless of the word order of the questions: the immediate post-verbal position. Therefore, I conclude that there is in fact a focus position directly after the verb and that the word order of the previous answers is not merely the result of echoing the word order in the original question. 2.3.2 Out-of-the-Blue Diagnostic In light of the highly context-dependent word orders discussed in the previous section, it is important to account for the possibility of a bias in word order preference due to assumed common ground. One diagnostic used to determine basic word order is to elicit sentences in out-of-the-blue contexts. Thisattemptsto remove any biases in discourse contexts by assuming no previous context at all. An example of this kind of situation is a scenario where a person simply walks into a room and says the utterance in question to the listener. In (19), this context-free situation is induced through a newscaster scenario. Anewscasterina Kipsigis broadcast gets on, and simply wants to announce someinformationinanewsituation. Given this scenario: (19) a. VSOAdvword order sounded most natural: ko-min lagok bandek komie PST.3-plant children maize well The children planted the maize well. b. VOAdvSsounded next best, but not as good as (19a): komin bandek komie lagok PST.3-plant maize well children The children planted the maize well. Consultants even reported that (19b) would have been more appropriate if there was maize in the background shot, indicating that a VOS construction is used in contexts where the object is more immediately relevant. However, in a discourse-neutral context, VSO constructions come most naturally. Additionally, speakers sometimes translated the VOS word order as an English passive construction. This suggests that VSO word order is the basic word order of Kipsigis. 12

2.4 Basic Analysis of Phrase Structure I nowprovidea possible mechanismto explainthe wide variation in Kipsigis word order. This structure will serve as the assumed phrase structure for Kipsigis in the rest of this paper. While this analysis is not the core contribution of this paper, this section will provide a preliminary framework in which to analyze other aspects of Kipsigis syntax. Most analyses that account for verb-initiality in languages fallunderthecategoriesofeitherphrasal movement of the VP or head movement of the verb itself (Otsuka, 2005;McCloskey,2005). Inorderto explain the word order facts in Kipsigis, I propose a head-raising analysis of the verb to the head of TP and the existence of a Focus Phrase (FocP) projection directly below TP. This accounts for the information structure data in section 2.3.1 above, which demonstrated that a focus position exists immediately after the verb. I further posit that depending on the context and focus of the sentence, different constituents raise up through A movement into the specifier of FocP to derive either VOS or VSO word order. Both of these movements are drawn out in (20). (20) a. VSO CP C TP T Verb DP FocP FocP Sbj F vp DP Sbj v Verb vp V Verb VP DP Obj 13

b. VOS CP C TP T Verb DP FocP FocP Obj F vp DP Sbj v Verb vp V Verb VP DP Obj This also allows for a number of adjunct positions in which adverbs can appear between the verb, subject, and object, thus leaving room to analyze sentences like thosein(21)(copiedfrom(12)above). (21) (*amut) ko-al (amut) artet (amut) (*yesterday) PST.3-buy (yest) goat (yest) The chief bought a goat yesterday. girwogindet (amut) chief (yest) In the following subsections, I defend this analysis by showing that VP-raising is unlikely and that binding data suggests that word order is a result of A movement. 2.4.1 Evidence and Arguments For Head Movement While initially it may seem reasonable to suggest a phrasal movement analysis for VOS order, there are a number of reasons to settle on a head movement analysis. If VOS wordorderwerederivedthroughvp movement, the verb and its object complement would be one constituent, as diagrammed in (22). (22) CP C TP V Verb VP DP Obj TP...vP... 14

Given a VP raising account, it should not be possible to find a constituent between the verb and object when the entire VP has been moved up intact (VOS word order). However, as shown in (21) above and produced specifically here in (23), constituents can appear between the verb and object in VOS constructions. (23) ko-al amut artet girwogindet PST.3-buy yesterday goat chief The chief bought a goat yesterday. In (24) below, innei ( definitely ) is a high adverb, referring to its position in the structure. Cinque (1999) characterized evaluative adverbs (such as honestly or definitely ) as occurring higher up in the structure, presumably around TP or CP, so as to take scope over the entire clause. This positioning is relative to lower manner or frequency adverbs. Here we see that the high adverb innei, whichcouldnothaveoriginatedin the VP, occurs between the verb and the direct object object. (24) ko-al innei artet girwogindet PST.3-buy definitely goat chief The chief definitely bought a goat. Therefore, head movement must necessarily exist in Kipsigis inorderfortheverbtoescapethevpand allow room for constituents between the verb and its complement. This suggests that the verb moves up to T through head movement. At this point, it is worth noting that falsifying a VP-fronting analysis is very difficult, if not impossible. The very same flexibility that allows it to account for a wide range of word orders also prevents the theory from making concrete and testable predictions. An analysis in which most or all constituents evacuate the VP before it is fronted is entirely possible, but not necessary helpful. Positing head movement of the verb post-vp-raising could violate the Freezing Principle (Clemens and Polinsky, 2014; Wexler and Culicover, 1977), where moved constituents are islands to extraction. The presence of head movement in a language has traditionally lowered the probability of phrasal movement occurring in the same language (Otsuka, 2005). The most convincing argument justifying the use of less economical 11 phrasal movement in an analysis of averb-initiallanguageistheabsenceofheadmovement.theunderlying assumption is that in order to justify movement of a much bulkier VP instead of choosing the much less costly optionofraisingonlythevhead, the language must use VP raising as a last resort to move the verb up to the front of a sentence because it cannot use head movement (Coon, 2010). Therefore, because headmovementisnecessarytoexplainthe word order facts discussed earlier in this section, a VP fronting analysis is theoretically dispreferred. 2.4.2 Verb Movement to TP Though there is enough evidence to reasonably suggest a head movement analysis, there is little evidence in the current research that serves to definitively determine the precise final position of the verb. For the purposes of this paper, I conclude that TP is the most likely position for Kipsigis verbs to appear. Given the fact that constituents are rarely allowed to appearbeforetheverb,thispositionislikelytobe higher up in the structure. The two most plausible positions for the landing site of the verb are the heads of TP and CP. Looking at the examples in (25), we see that verb initiality and word order flexibility are both preserved in embedded clauses in Kipsigis (word orders of theembeddedclauseareprovidedalongsideeach example). 11 The currency of this economy has not been clearly defined, whether it deals in phonological weight or a more abstract measure. I use the term economical to denote the movement of more or less phonological material, under the assumption that moving all the material in the VP would be more costly than only moving a smaller sub-section of it (i.e. verb). 15

(25) a. ko-geer kiproono kole ko-min lagok bandek komie PST.3-see Kiprono that PST.3-plant children maize well Kiprono saw that the children planted maize well. b. ko-geer kiproono kole ko-min lagok komie bandek PST.3-see Kiprono that PST.3-plant children well maize Kiprono saw that the children planted maize well. c. ko-geer kiproono kole ko-min bandek lagok komie PST.3-see Kiprono that PST.3-plant maize children well Kiprono saw that the children planted maize well. d. ko-geer kiproono kole ko-min bandek komie lagok PST.3-see Kiprono that PST.3-plant maize well children Kiprono saw that the children planted maize well. e. ko-geer kiproono kole ko-min komie bandek lagok PST.3-see Kiprono that PST.3-plant well maize children Kiprono saw that the children planted maize well. f. ko-geer kiproono kole ko-min komie lagok bandek PST.3-see Kiprono that PST.3-plant well children maize Kiprono saw that the children planted maize well. VSOAdv VSAdvO VOSAdv VOAdvS VAdvOS VAdvSO Because all of these word orders can co-occur with an overt complementizer, I conclude that the verb can t appear in C, and must appear in T. 2.4.3 Binding Data Suggests A Movement Having settled on the movement pattern of the verb, the next question is how to derive the alternate VOS/VSO word orders. Data involving binding facts reveal that movementof the object or subject to create either VSO or VOS word order is most likely a form of A movement. In order for a quantifier to bind to a variable, the variable must be located within the c-command domain of the quantifier. In instances where the subject is a quantifier and the object is a variable like (26), bound readings are acceptable in both VSO and VOS constructions. (26) a. ko-sus NOkt@ aga tugul k dzito-ñin k/i PST.3-bite dog every person-its Every dog k bit its k/i person. b. ko-sus dzito-ñin k/i NOkt@ aga tugul k VOS PST.3-bite person-its dog every Every dog k bit its k/i person. VSO VOS However, binding is not possible if the quantifier is the object and the variable is the subject, regardless of word order. (27) a. ko-mwet dzito-ñin k/i Nokt@ aga tugul k PST.3-wash person-its dog every Its k/i owner washed every dog k. b. komwet Nokt@ aga tugul k dzito-ñin k/i PST.3-wash dog every person-its Its k/i owner washed every dog k. VSO VOS 16

Looking back at the trees in (20), we see that objects are within the c-command domain of subjects in base position. Therefore, it follows that a construction with a quantifier DP subject allows binding of a DP object variable lower down in the structure. However, because the same sentencewith VOS word order also allows binding, despite the fact that the object must have moved above the subject in order to obtain the different word order, we must conclude that the movement which resulted in VOS order is an A type movement. The same logic holds for the ungrammatical binding judgments in (27). A movement is known to reconstruct its binding properties to base positions in LF, which results in the patterns like the ones we see in(26) and (27). 12 This fits well with what we know of focus movements, thus matching the information structure evidence pointing to a post-verbal focus position (Otsuka, 2005). To account for the focused reading,i posit a Focus projection underneath TP(where I concluded the verb lands) to which the appropriate constituent moves, resulting in the proper word order. 2.5 Summary of Kipsigis Phrase Structure Conclusions Kipsigis phrase structure is certainly an intriguing puzzleinitselfandisundergoingmoreextensiveresearch by Diercks et al. (2016b). In this section, I have laid out a foundation from which to analyze complementizer agreement in Kipsgis. Though both VSO and VOS word orders are the most natural for Kipsigis speakers, the basic word order is VSO. Verbs raise to a clause-initial position, presumably T. Word order is flexible post-verbally, a phenomenon which is achieved by movement of constituentstoanimmediatepost-verbal focus position. There is evidence that this movement is an A -movement, which is unsurprising given that this is a focus movement (McCloskey, 2005). It is important to notethatthismechanismcannotexplainthe full extent of Kipsigis flexible word order, but it will suffice forthepurposesofanalyzingcomplementizer agreement patterns. For a more thorough analysis, see Diercks et al. (2016b), currently in progress. 12 Recent work from (Diercks et al., 2016b) has uncovered movements with A-properties, suggesting this position may have mixed effects. Research is ongoing on this issue. 17

3 Properties ofprefixedcomplementizer Agreement (Subj-CA) 3.1 Partial Complementizer Inventory Before examining prefixed CA in Kipsigis, I provide a partial inventory of complementizers that I have identified in Kipsigis, laid out in (28). Only the (non-)agreeing complementizer forms will be discussed in the rest of this paper. (28) Kipsigis Gloss AGR-lE that (agreeing) kole that (non-agreeing) ke-le that (default agreement) amun because koti if ne focus head/relativizer ko topic head Examples of each of these complementizers is provided in (29). 13 (29) a. A-bOOti kole ko-laal mat tseepto 1sg-remember that PST.3-light fire girl I remember that a girl lit the fire. b. A-bOOti A-lE ko-laal mat tseepto 1sg-remember 1sg-C PST.3-light fire girl I remember that a girl lit the fire. c. ko-ge-mwoo-tsi ke-le ko-ruuja tuga amut PST.3-IMP-tell-3.OBJ DEF-C PST.3-sleep cows yesterday He/she/they was/were told that the cows slept yesterday. d. A-baibai amun ko-it lagok 1sg-be.happy because PST-arrive children I am happy because the children arrived. e. ko-a-teeb koti ko-laal mat tseepto PST-1sg-ask if PST.3-light fire girl I asked if the girl lit the fire. f. A-bOOti tseepto ne ko-laal mat 1sg-remember girl who PST.3-light fire I remember a girl who lit the fire. g. neetsiriet ko ko-jega laakwet sheep KO PST.3-look.after child The child looked after the sheep. (Intuition: The sheep was looked after the child. ) [kole] [AGR-lE] [kele] [amun] [koti] [ne] [ko] 13 The properties and roles of ne and ko is not currently fully understood, but they have been used for topicalization,clefting,andas relativizers. Research on these particles is ongoing. 18

There is no null complementizer available, as shown below in (30), for both speech and non-speech verbs. (30) a. A-NgEn *(A-lE) ko-ruuja tuga amut 1sg-know 1sg-C PST.3-sleep cows yesterday I know (that) the cows slept. b. ko-a-mwaa *(kole) ko-ruuja tuga amut PST-1sg-say that PST.3-sleep cows yesterday I said (that) the cows slept. Only the AGR-lE declarative-embedding complementizer shows agreement in Kipsigis (to my knowledge), though further research may uncover another such construction. 3.2 Prefixed Complementizer Agreement Forms The agreeing forms of the upward-oriented prefixed complementizer agreement pattern are listed in the table in (31). (31) Prefixed Complementizer Agreement Forms (Subj-CA) SG PL 1st A-lE ke-le 2nd i-le o-le 3rd ko-le ko-le Examples for each of these are shown in (32) below. (32) a. ko-a-mwaa A-lE ko-ruuja tuga amut PST-1sg-say 1sg-C PST.3-sleep cows yesterday I said that the cows slept yesterday. b. ko-i-mwaa i-le ko-ruuja tuga amut PST-2sg-say 2sg-C PST.3-sleep cows yesterday You (sg) said that the cows slept yesterday. c. ko-mwaa ko-le ko-ruuja tuga amut PST.3-say 3-C PST.3-sleep cows yesterday He/She/They said that the cows slept yesterday. d. ko-ge-mwaa ke-le ko-ruuja tuga amut PST-1pl-say 1pl-C PST.3-sleep cows yesterday We said that the cows slept yesterday. e. ko-o-mwaa o-le ko-ruuja tuga amut PST-2pl-say 2pl-C PST.3-sleep cows yesterday You (pl) said that the cows slept yesterday. 19

There is no distinction between the singular and plural agreeing forms for constructions with a third person matrix subject. 14 This is standard for the language, which does not usually distinguish between the singular and plural form of the third person in many aspects of the grammar, including adjective formation or verbal marking (Jake and Odden, 1979). The non-agreeing form of the complementizer that (kole)canappearwithallpersonsandnumbers. 15 (33) a. ko-a-mwaa kole ko-ruuja tuga amut PST-1sg-say that PST.3-sleep cows yesterday I said that the cows slept yesterday. b. ko-i-mwaa kole ko-ruuja tuga amut PST-2sg-say that PST.3-sleep cows yesterday You (sg) said that the cows slept yesterday. c. ko-mwaa kole ko-ruuja tuga amut PST.3-say that PST.3-sleep cows yesterday He/She/They said that the cows slept yesterday. d. ko-ge-mwaa kole ko-ruuja tuga amut PST-1pl-say that PST.3-sleep cows yesterday We said that the cows slept yesterday. e. ko-o-mwaa kole ko-ruuja tuga amut PST-2pl-say that PST.3-sleep cows yesterday You (pl) said that the cows slept yesterday. Crucially, the complementizer forms in (32c) and (33c) are analytically ambiguous. For this reason, for the rest of this paper, I will be dealing primarily with examples in the first and second person to demonstrate complementizer agreement. The agreeing complementizer forms, though similar in meaning to the non-agreeing form, are available in certain contexts and lead to interpretational differences when used in place of the non-agreeing form. This will be discussed further in section 4. As of now, evidence for the default agreement form of the complementizer has only been found in impersonal constructions and takes the form kele. Thispatternwillbediscussedinsection3.4below. Thefollowing subsections describe the various syntactic properties of Kipsigis prefixed CA (later referred to as Subj- CA) and demonstrate the multiple similarities between the prefixed subject-oriented CA patterns found in Kipsigis and Lubukusu. 3.3 Prefixed CA Agrees with the Most Local Matrix Subject Kipsigis prefixed CA has a strict matrix subject orientation. The Germanic complementizer agreement pattern in which the complementizer displays agreement with the embedded subject is ungrammatical in Kipsigis, as demonstrated by the example in (34). 14 There is also no identifiable distinction between the non-agreeing form and the agreeing form for the third person. For this paper, I make this distinction for the purpose of consistency of form and to account for the distinct contexts in which the agreeing complementizer is grammatical or not. The interpretive difference between the agreeing and non-agreeing forms of the complementizer are discussed in section 4. 15 Note that though the translations provided in these examples arethesameasthosefortheagreeingcomplementizerexamples, there is an interpretive difference between the two forms. However, this difference involves discourse contexts rather than semantic truths and so will be discussed more thoroughly later in this section. 20

(34) A-NgEn kole/a-le/*i-le ko-i-amisje amut 1sg-know that/1sg-c/*2sg-c PST.3-2sg-eat yesterday I know that you ate yesterday. Complementizers cannot agree with subjects of the embedded clause and are restricted to agreement in the matrix clause. The prefixed agreement pattern is also subject-oriented and cannot target objects in the main clause. (35) ko-a-mwaa-wuun kole/a-le/*i-le ko-ruuja tuga amut PST-1sg-tell-2sg.OBJ that/1sg-c/*2sg-c PST.3-sleep cows yesterday I told you (sg) (that) the cows slept yesterday. Prefixed CA is locally constrained in that only the most local superordinate subject may trigger agreement, as is evident in (36). (36) ko-a-mwaa A-lE ko-i-bwot i-le/*a-le ko-ruuja tuga amut PST-1sg-say 1sg-C PST-2sg-think 2sg-C/1sg-C PST.3-sleep cows yesterday I said that you thought that the cows slept yesterday. The pattern in (34)-(36) is exactly what is reported in Lubukusu (Diercks, 2013), Ibibio (Torrence, 2016), Chokwe, Luchazi, Lunda, and Luvale (Kawasha, 2007), demonstrating similarities in target properties crosslinguistically. 3.4 Subj-CA in Impersonal Constructions AfeatureoftheLubukusuCAconstructionisthatmanyspeakers readily accept the agreement pattern with aderivedsubjectinapassiveconstruction(diercks,2013). (37) Ba-sasi ba-bol-el-wa nende Sammy mbo (*a-li) ba-keni ba-a-rekukha. 2-parents 2S-say-AP-PASS by 1Sammy that (*1-that) 2-guests 2S-PST-leave The parents were told by Sammy that the guests left. [Lubukusu] To my knowledge, there is no passive construction in Kipsigis. Similar interpretations are achieved either via a VOS construction or by the impersonal construction (Payne, 2011, cf.). This is formed by adding a Geprefix to the verb in the place of a subject agreement marker. 16 Despite its passive-like interpretation, the impersonal construction does not allow for prefixed agreement with the matrix clause object: (38) ko-ge-mwaa-an kole/*a-le ko-ruuja tuga amut PST-IMP-tell-1sg.OBJ that/1sg-c PST.3-sleep cows yesterday I was told that the cows slept yesterday. [Kipsigis] 16 Impersonal constructions appear segmentally identical to an active sentence with a first person plural subject, but the constructions are distinguishable by their differing tonal patterns. (1) a. kò-gè - m w ā à - ù n kè-lè kò-rú ú j á tùgà à m ù t PST-IMP-tell-2sg.OBJ DEF-C PST.3-sleep cows yesterday You (sg) were told that the cows slept yesterday. b. kò-gē - m w á à - ú n kè-lè kò-rú ú j á tùgà à m ù t PST-1pl.SBJ-tell-2sg.OBJ 1pl-C PST.3-sleep cows yesterday We told you that the cows slept yesterday. 21

This is not surprising, however, as the object in these instances is not promoted to subject, instead being marked as an object clitic on the verb. Therefore, rather than acommentaryonthepossibilityofagreeing with derived subjects, this serves as another instance of non-subjects being unable to trigger prefixed complementizer agreement. Instead, a default prefixed complementizer agreement marker, kele 17,isavailableinimpersonalconstructions for all persons and numbers of matrix clause objects. 18 (39) a. ko-ge-mwaa-an ke-le ko-ruuja tuga amut PST-IMP-tell-1sg.OBJ DEF-C PST.3-sleep cows yesterday I was told that the cows slept yesterday. b. ko-ge-mwaa-wuun ke-le Go-Ruuja tuga amut PST-IMP-tell-2sg.OBJ DEF-C PST.3-sleep cows yesterday You were told that the cows slept yesterday. c. ko-ge-mwoo-tsi ke-le Go-Ruuja tuga amut PST-IMP-tell-3OBJ DEF-C PST.3-sleep cows yesterday He/she/they was/were told that the cows slept yesterday. d. ko-ge-mwaa-wets ke-le Go-Ruuja tuga amut PST-IMP-tell-1pl.OBJ DEF-C PST.3-sleep cows yesterday We were told that the cows slept yesterday. e. ko-ge-mwaa-woog ke-le Go-Ruuja tuga amut PST-IMP-tell-2pl.OBJ DEF-C PST.3-sleep cows yesterday You (pl) were told that the cows slept yesterday. This marker is very likely an instance of the subject-agreeing complementizer occurring in a construction where there is no subject marked on the verb, and so default agreement features surface as a morphological stand-in. Further reasons to interpret kele as another version of the subject-agreeing complementizer are explored in section 4. 3.5 (Non-)Locality Effects for Subj-CA AstandardfeatureoftheAgreeoperation(andagreementphenomena crosslinguistically) is that it is subject to locality effects (Chomsky, 2000, 2001). In these next subsections, I describe some of the (non-) locality effects for Kipsigis Subj-CA. 17 Like the impersonal subject marker, the default agreement marker is morphologically identical to the 1pl agreement marker in active declarative sentences. A tonal distinction between the two complementizer forms has not yet been identified, but research on this topic is ongoing. 18 Interestingly, the default agreeing complementizer form kele, similar to the first person plural agreeing form of Subj-CA,invokes a more indirect meaning. One of our speakers described this distinction by explaining that the use of kele is interpreted as It was mentioned to me... whereas kole signifies Iwastold....Thedetailsofthisphenomenonarenotfullyunderstoodasofyet,andwill hopefully be the subject of future research. 22

3.5.1 CPs at the Right Edge Speakers prefer to place CPs at the right edge of the clause: (40) a. ko-i-geer iñe i-le ko-ruuja tuga amut PST-2sg-see 2sg 2sg-C PST.3-sleep cows yesterday YOU saw that the cows slept yesterday. b. *?ko-i-geer i-le ko-ruuja tuga amut iñe PST-2sg-see 2sg-C PST.3-sleep cows yesterday 2sg YOU saw that the cows slept yesterday. Recall from section 2.1 that Kipsigis readily allows VSO and VOS word orders, with focused material preferably following the verb. Yet, even in a focused context liketheonein(41),thecppreferstobe clause-final, following the recipient object, rather than occurring immediately post-verbally. (41) a. Q: ko-i-mwoo-tsi nee Kibeet? PST-2sg-tell-3.OBJ what Kibeet What did you tell Kibeet? b. A1:??ko-A-mwOO-tSi A-lE/kOlE ko-it tuga amut Kibeet PST-1sg-tell-3.OBJ 1sg-C/C PST.3-arrive cows yesterday Kibeet I told Kibeet that the cows arrived yesterday. c. A2: ko-a-mwoo-tsi Kibeet A-lE ko-it tuga amut PST-1sg-tell-3.OBJ Kibeet 1sg-C PST.3-arrive cows yesterday I told Kibeet that the cows arrived yesterday. d. A3: ko-a-mwoo-tsi Kibeet kole ko-it tuga amut PST-1sg-tell-3.OBJ Kibeet that PST.3-arrive cows yesterday I told Kibeet that the cows arrived yesterday. This tendency is not surprising given the well-known cross-linguistic tendency for CP-complements to occur at the right edge of clauses (Moulton, 2015; Stowell, 1981). The choice of an agreeing or non-agreeing complementizer doesn t have any effect on this word order preference. 3.5.2 Subj-CA Possible Over an Intervening Object Another property of the Lubukusu pattern of CA is that non-subjects in the matrix clause do not affect the viability of agreement on the complementizer (Diercks, 2013, pp. 368). (42) (Ese) n-a-bol-el-a Nelsoni n-di ba-keni ba-a-cha. (I) 1sgS-PST-say-AP-FV 1Nelson 1sg-that 2-guests 2S-PST-go I told Nelson that the guests left. [Lubukusu] Similarly in Kipsigis, the Subj-CA pattern is not disrupted by overt objects in the matrix clause: (43) ko-i-mwoo-tsi laakwet i-le ko-ruuja tuga amut PST-2sg-tell-3.OBJ child 2sg-C PST.3-sleep cows yesterday You (sg) told the child that the cows slept yesterday. [Kipsigis] 23

Kipsigis also has optional overt pronouns that can be used in addition to the verbal suffixes. These do not disrupt Subj-CA either 19 : (44) ko-a-mwaa-wuun iñe A-lE ko-ruuja tuga amut PST-1sg-tell-2sg.OBJ 2sg.OBJ 1sg-C PST.3-sleep cows yesterday I told you (sg) that the cows slept yesterday. Thus, Kipsigis again follows the Lubukusu CA pattern. This non-intervention effect is also documented in Ibibio (Torrence, 2016). 3.5.3 Subj-CA out of Noun Complement Clauses Aperhapssurprisingfeaturethathasbeenpreviouslydocumented in Lubukusu complementizer agreement is that a complementizer inside a noun complement clause (NCC) can agree with the main-clause subject (Diercks, 2013, pp. 378): (45) n-a-ulila li-khuwa n-di Sammy ka-a-kula li-tunda. 1sgS-PST-hear 5-word 1sg-that 1Sammy 1S-PST-buy 5-fruit I heard the rumor that Sammy bought the fruit. [Lubukusu] In Lubukusu, however, this is constrained by the presence of an intervening possessor of that noun phrase, which cannotitself triggerca but preventsca with the main clause subject (Diercks, 2013, pp. 378). (46) M-bona bu-ng ali bw-a Alfredi mbo/*a-li/*n-di ba-ba-ana b-ewe ba-kha-khil-e 1sgS.see 14-certainty 14-of 1Alfred that/*1-that/*1sg-that 2-2-children 2-his 2S-FUT-win-FUT I see Alfred s certainty that his children will win. The same pattern occurs in Kipsigis, though consultants differed in their judgments on the acceptability of agreeing forms of the complementizer in NCCs. One did not find these constructions grammatical, while the other provided them readily and robustly. 20 AcomplementizerinaNCCmayagreewiththemainclause subject in appropriate contexts: (47) a. ko-a-ibut logujuwek%a-le ko-ruuja tuga amut PST-1sg-bringnews %1sg-C PST.3-sleep cows yesterday I brought news that cows slept yesterday. b. A-tIñE kajenet %A-lE/kOlE/*kE-lE ko-it lagok 1sg-have belief/faith 1sg-C/C/*DEF-C PST.3-arrive children I have belief/faith that the children arrived. c. ko-a-mwaa Atindoniot %A-lE/kOlE/*kE-lE ko-it lagok PST-1sg-tell story %1sg-C/C/*DEF-C PST.3-arrive children I told the story that the children arrived. [Kipsigis] Likewise, as in Lubukusu, the presence of a possessor inside the noun phrase degrades Subj-CA. Example (48) is the equivalent of (47c), with the difference that a possessor is added to the noun phrase in (48), resulting in a ruling out of the agreeing complementizer now for both consultants. 19 We have seen some variability in judgments from our two consultants on this issue, where at times the agreeing complementizer was degraded in the presence of an object. We have numerous instances of agreement occurring naturally, however, which have led us to believe that the troublesome aspect in those instances had moreto do with information structure constraints on CA rather than actual intervention of an object. More on the interpretive properties of agreeing complementizers is discussed in section??. 20 Iannotatethisinterspeakervariationontheexampleswitha%symbol. 24