A. Zimmerling A. Zimmerling IMLR MSPU / Institute of linguistics RAS 1P ORDERS IN 2P LANGUAGES1 1. Research problem I discuss the phenomenon of Information Structure (IS) motivated clausalinitial placement of diagnostic categories conventionally labelled clausal second (2P) elements, notably, finite verbs in Verb Second languages (V2) and clustering clause-level clitics in Clitic Second Languages (CL2). The main research question is whether V1 orders in V2 languages and CL1 orders in CL2 languages have identical IS-triggers or not. If the answer is positive, one gets additional support for the hypothesis that V2 languages and CL2 languages form together one class in terms of shared constraints on clause structure. 2. Framework, data and methodology I use the apparatus of formal typology and apply the notion of type-specific i.e. shallow constraints on clause structure, in terms of Roberts [10]. The diagnostics for 2P languages is based on the so called bottleneck condition predicting two features: a) that a combination of two (or more) phrasal categories X, Y preceding the finite verb in a V2 language / the clustering clitics in a CL2 language should be ungrammatical; b) that the XP-position in the diagnostic type of V2 declaratives / any clause type with CL2 is not reserved for any particular syntactic category (e.g. NP) and does not express any particular grammatical relation (e.g. subject). The bottleneck condition gives us a large class of CL2 languages attested in different areas and a small of class of Germanic V2 languages, the language of Kashmiri, cf. Bhatt [1], and several Romance idioms influenced by neighbor V2 Germanic languages. I adhere to classical definitions of V2 based on the bottleneck condition, cf. Vikner [12] and Zimmerling & Lyutikova [21] and deny the existence of the so called residual V2 languages which license a variation of V2 ~ V > 2 orders in the same diagnostic type of declarative clauses. Unlike Holmberg [5], Wolfe [13], Bech 1 The paper is written with financial support from the Russian Science Foundation, project RFH 14-18-03270. I am grateful to the anomymous reviewers, to the audience of Typology of Morphosyntactic Parameters 2015 (Moscow, 14-16 Ocʕober, 2015Ŋ and, perʔonally, ʕo Aʕle GrЛnn and Liza Kuʔhnir for ʕhe helpful comments. I am also grateful to Tatiana Yanko for the discussion of information structure issues and to Loren Billings for the discussion of Rikavung data.The sole responsibility for all shortcomings is on the author. 459 1p orders in 2p languages & Salvesen [3], I do not classify Old and Modern English and Old Romance languages with the 2P class and exclude them from my sample. A group of CL2 languages licenses derived orders with late placement of clitics [X] Y CL, cf. Zimmerling & Kosta [20: 197]; this does not count as 2P violation, insofar CL2 placement is blocked by some feature of the clause-initial element. Instances of late finite verb placement in the same diagnostic type of declarative clauses in the in true V2 languages are extremely rare, cf. but Swedish, where the modal adverbial kanske competes with the finite verb for the 2P slot, cf. Platzack [9]. The main differences between V2 languages and CL2 languages, apart from late placement constructions, are due to two features: a) clitics clusterize in clausal 2P, finite verbs do not; b) V2 is a root phenomenon primarily characteristic of main clause declaratives, while CL2 is usually generalized in all types of clauses, see Zimmerling [19] for details. A descriptive schema of 2P needs four symbols a symbol of clausal (left) border (#), a symbol of the preverbal or preclitic constituent (XP) and symbols for the pivotal category the finite verb (Vfin) or clustering clitic (CL), as represented by (G)eneralization 1. (G1) #XP Vfin, /CL *#XP Y Vfin/CL The bottleneck condition is crucial for 2P diagnostics. It predicts two features of 2P syntax: a) that a combination of two (or more) phrasal categories X, Y preceding the finite verb in a V2 language / the clustering clitics in a CL2 language should be ungrammatical; b) that the XP-position in the diagnostic type of V2 declaratives / any clause type with CL2 is not reserved for any particular syntactic category (e.g. noun phrase) and does not express any particular grammatical relation (e.g. subject). In other words, the XP-position in a 2P language can be filled by any element in an OR-expression {Cat1  Cat2  … ɤaʕn}, but simultaneous spell-out of two or more hierarchically independent categories in XP is blocked: *#{XP Cat1 & Cat2 &… ɤaʕn} Vfin/CL. Parsing of well-formed 2P structures is licensed by a combination of an OR-expression filter, which lists sentence categories that fill XP in language L, and an &-expression filter which determines which types of expressions count as single constituents when filling XP in language L, and which do not [19]. 3. Hypothesis I assume that the bottleneck condition G1 results from a combination of head movement of the diagnostic category to 2P & accompanying XP-movement of phrasal categories to clausal-initial position and that, with 1P orders, the diagnostic category moves further than its canonic position. V1 orders in V2 languages always 460 A. Zimmerling have IS-triggers, while CL1 orders in some cases have IS-triggers and in some other cases can be motivated by prosody/interface mechanisms. 4. Analysis At the first step, I classify all V2 languages into two groups V1/V2 languages, where V1 declaratives are regularly used in standard narrative texts and V1 orders are linked with a wide number of different IS-readings rheme dislocation, verb focalization, verb topicalization, theticity vs. rigid V2 languages, where V1 sentences are marginally possible in substandard oral narratives and have fixed IS-readings. The majority of V2 languages, cf. Old Icelandic, Old Swedish, Old Saxon, Old High German, Modern Icelandic, Faroese, Yiddish and Kashmiri are V1/V2 languages, while Danish, Swedish, Norwegian, German, Dutch, Flemish and Afrikaans are rigid V2 languages. 4.1. V1/V2 languages Old Icelandic example (1) shows rheme dislocation in a categorical sentence, while Old Swedish example (2) is a thetic sentence. The symbol # stands for clausal left border, the symbol t shows the original position of the moved category in a generative framework with movement. The basic word order in Old Icelandic and Old Swedish is SVO or XP V, cf. Zimmerling [17], while V > 2 orders in main clause declaratives do not occur. The example (1) is taken from a Saga narrative and hightlights an episode linked to a particular individual, whose name is Sigmundr, while (2) comes from a law code, where a hypothetical situation is analyzed in terms of legal obligations for having left a dead body on a battlefield. (1) Old Icelandic, V1/V2: #{Kemri } Sigmundr ti {til dura}. Come-Prs.3Sg.Ind Sigmundr-Nom.Sg. to door-Gen.Pl. And Sigmundr komeʔ up ʕo ʕhe doorʔ (2) Old Swedish, V1/V2 #Ligghaeri lik ti at wigwalli. Lie-Prs.3Sg.Ind corpse-Nom.Sg. at battlefield-Dat.Sg. There lieʔ a corpʔe on ʕhe baʕʕlefield . V1 orders in V1/V2 languages can also be mapped to other IS types. The verb fronting can also involve verb focalization in contexts where the postverbal part of the sentence contains already activatived information or verb topicalization. Cf. Zimmerling [17: 362-366] for Old Icelandic, Zimmerling [17: 457-459] for Middle Norwegian, Bhatt [1: 117] and Zimmerling [18: 193] for Kashmiri1. Arguably the 1 Bhatt (op. cit., 117) interpretes the V1 data in Kashmiri differently and claims that all V1 Kashmiri declaratives are Topic-Comment structures without a topic. 461 1p orders in 2p languages same verb fronting conditions are attested in non-V2 SVO and SOV languages as e.g. Latin, Hungarian, Old Greek, Portuguese, Alyawarra, Archi, cf. Yanko [14: 222223], Ossetic and Russian, cf. Zimmerling & Lyutikova [21]. If V1 sentences are frequent and occur in texts of a different type, the prediction is that the V1 order is communicatively opaque and can be mapped to different IS types. In this respect, V1/V2 languages do not differ from non-V2 languages with IS-motivated verb fronting, in any case, the opposite hypothesis must be proved on an individual basis. This statement is captured in the generalizion (G2) below. (G2) If a language is a V1/V2 system conforming to (G1), V1 declaratives can be linked with at least two different IS-readings. If a language is a rigid V2 system conforming to (G1), V1 declaratives are associated with one shared IS-reading. 4.2. Rigid V2 languages In rigid V2 languages V1 orders are occasionally licensed in special pragmatic contexts e.g. in thetic sentences with inferential semantics, cf. the German examples (3ab) and the Swedish example (4): (3) German, V2 a. #<Hans Zender?> #Kenni ich ti H.Z. know-Prs.1Sg.Ind ihn nicht. Pron-3Sg.M. not <Hanʔ ZenderŠ> I <really> don ʕ know <him> . b. #<Noch ein Sʕűck Kuchen?> #Habei Further Det.3 piece pie have-Prs.1Sg.Ind ich ti nichts da gegen. 1Sg not against <One more piece pie?> I do not object. (4) Swedish, V2 #Såi jag ti den nyss på bordet. See-Prt. 1Sg now on table-Def. <Where iʔ iʕŠ> I have <juʔʕ> ʔeen iʕ on ʕhe ʕable . Sʕill hiʔ example (53bŊ <X firʔʕ wanʕed ʕo by Y, ʕhen he wanʕed ʕo buy Z,> bought that junk piece of U, eventually> is most likely a Topic Comment structure with a topicalized verb and different from examples like (52c), where the verb is part of a Rheme. Therefore, we prefer classify Kashmiri with V1/V2 languages, not with rigid V2 languages. 462 A. Zimmerling V1 structures in (3) (4) can be regarded elliptical from the viewpoint of codified German and Swedish 1. In (3aŊ, ʔubjecʕ pronoun ich I iʔ dropped. In (3bŊ, the postverbal subject is present but the deictic prefix da- is dropped. In (4) the object pronoun den is dropped. It is unclear whether ellipsis of arguments is a regular condition for licensing of V1 structures like (3)-(4) as Reis [11: 218] has suggested for German, or just a side effect of less formal speech. I analyze sentences like (3), (4) as indirect speech acts rather than true declaratives: verb fronting amounts here to adding an overt discourse particle. Some rigid V2 languages like modern German marginally license V1 categorical sentences with rheme dislocation in lively oral narratives. In (5a), the rheme {F sitzen {FProperbeim Stammtisch}} is split due to verb movement to V1 across the subject position. The same communicative strategy with text-initial V1 categorical sentences with rheme dislocation is common in oral narratives attested in non-V2 languages, cf. Yanko [14: 201] for Russian. Meanwhile, Russian, unlike German, normally avoids V1 declaratives in subsequent sentences introducing direct speech, cf. example (5b) placed in the same text right after (5a). (5) German, V2 a. #{FSitzeni} vier Ärzʕe ti {FProperbeim Sit-Prs.3Pl. four doctor-Nom.Pl. by-Det.Dat.Sg.M. Stammtisch}. wood.table.Dat.Sg.M. <Introduction of a story about doctors>. Four doctors are sitting in a pub aʕ ʕhe ʕable b. #{FSteht} der Augenarzt {FProperauf} Stands-Prs.3Sg. Det.Nom.Sg. eye.doctor.Nom.Sg. up und {FPropersagt}: and say-Prs.3Sg. Ich gehe jetzt. Man sieht ʔich . I go-Prs.1Sg. now. Indef see-Prs.3Sg Refl The oculiʔʕ ʔʕandʔ up and ʔayʔ: <I am leaving. We ll ʔee each oʕher laʕer. V1 declaratives with rheme dislocation are also attested in Flemish oral narratives, cf. (6a) and (6b) which come from one and the same text. 1 In the notation of elided categories, we proceed with the assumption that postverbal subjects always precede postverbal objects in Swedish and German with the basic word order. For German, this is a simplification since this language has subject-object scrambling in the middle field, cf. Zimmerling [18: 176]. 463 1p orders in 2p languages (6) Flemish, V2. a. #{FVerhuis} ik {FPropernaar een Move-Pres.1.Sg. 1Sg to a <Imagine> I move ʕo a differenʕ room... b. #{FZitten} daar {FProperkakkerlakken}. Sit-Prs.3Pl. there cockroach-3Pl. …And cockroacheʔ are ʔiʕʕing ʕhere. andere different kamur}. room. 4.3. Diagnosing V1/V2 languages and V1 declaratives The diagnostics of V1/V2 languages vs V2 languages on the basis of (G2) is by and large unproblematic given the small number of V2 languages conforming to (G1). However, there are borderline cases, where identifying a language as a rigid V2 system vs V1/V2 systems partly depends on linguistic tradition. E.g., Modern German is recognized as a rigid V2 system because V1 declaratives like (3a-b) and (5a-b) are not accepted in written codified texts. Otherwise, German would pattern with V1/V2 systems since V1 orders are attested in informal texts both in categorical and in thetic sentences. V1 orders in German have a broader distribution than in Kashmiri, they can be both text-initial, cf. (5a) and non-initial, cf. (5b). Meanwhile, we identified Kashmiri as a V1/V2 system, despite the fact V1 declaratives in this language do not occur text-initially according to Bhatt [1: 120]. Still, non-initial V1 declaratives in Kashmiri can be mapped to IS types. The fronted verb can boʕh be parʕ of ʕhe rheme, cf. ɣhaʕʕ ʔ example (52cŊ reproduced below aʔ (7Ŋ, or be ʕhe ʕheme, af ɣhaʕʕ ʔ example (53bŊ reproduced below aʔ (8Ŋ ʔuggeʔʕʔ. (7Ŋ is best analyzed as an IS with rheme dislocation, while (8) is best analyzed as an IS with verb topicalization. (7) Kashmiri, V1/V2. <A. Az gatsh-na> ba skuul.> Today go-Neg I school Aʔ for ʕoday, I don ʕ wanʕ ʕo go ʕo ʔchool. B. {Fner} tsI {Fproper goD razaayi-tal}, pat karav Leave you first blanket-under later will-do ath-pyaTH mashvar. this-on discussion Firʔʕ you come ouʕ of ʕhe blankeʕ and ʕhen we ll diʔcuʔʔ iʕ. 464 A. Zimmerling (8) Kashmiri, V1/V2. <goD oos yetshaan su skuuTar anun, pat vonun first was wanting he scooter get then said akyaa vanyi OK now Hyam moTorsaykal> {Tonnun} {Fmoksarkaryith yi Buy-Fut mobike bought(3, m,s) eventually this khaTaar pat}. junk then. Firʔʕ he wanʕed ʕo buy a ʔcooʕer, ʕhen he ʔaid he ll buy a mobike, evenʕually (iʕ ʕurned ouʕŊ he boughʕ ʕhiʔ piece of junk. 4.4. Functional vs formal explanation of V1 declaratives in V2 languages Reis [11: 124] explains the distribution of V1 declaratives vs V2 declaratives in German in term of the event vs fact opposition but this is not a fitting pair of values: since V2 is a default order in German and other rigid V2 languages, V2 declaratives are predictably heterogeneous, some of them having the status of facts, some of them the status of events. Besides, the event vs fact opposition is neutralized in many contexts. The weaker claim that part of V1 declaratives share some semantic value related to non-factivity/narrative mode has more empiric adequacy, since V1 sentences with rheme dislocation, like (1), (5a), (6a-b), (7) are primarily associated with a narrative mode and represent inner vision and recollection rather than direct description and on-line reportage, cf. Yanko [14: 198]. At the same time, other types of V1 declaratives, such as inferential thetic sentences, cf. (3a-b), (4), verb topicalization, cf. (8), verb focalization cannot be explained that way since they either refer to observable facts, cf. OIcel. LeiРr Б varit Spring comeʔ , or explicate causal or temporal relations. Cf. typical contexts with verb focalization, where the fronted verb becomes the rheme since all other elements have already been acʕivaʕed in diʔcourʔe: Ruʔ. #<Priʔhel bol noj k vraаu.#{FOsmotrel} {Tvraа bol nogo}. <A paʕienʕ came ʕo a docʕor>. The docʕor examined ʕhe paʕienʕ , liʕ. Examined ʕhe docʕor ʕhe paʕienʕ , cf. Yanko [14: 206ž1. Finally, encoding of communicative meanings related to mental reconstruction seems to be associated in European languages with a different intonation, lexical and syntactic cues, cf. Yanko [15: 113-117], so the distribution of V1 vs V2 declaratives with rheme dislocation in V2 languages at best only partly overlaps with the opposition of narrative vs non-narrative communicative meanings. 1 The formal aspect of deriving Russian sentences with verb focalization is discussed in Zimmerling [18 : 282]. 465 1p orders in 2p languages Reis [11], within a functional approach, and Platzack [9], in a non-cartographic version of the minimalist framework, attempt at analyzing V1 declaratives in German and Swedish as defective V2 structures with a silent operator element in front of the finite verb. This logic is straightforward. If L is a rigid V2 system, the preverbal position (XP) must be filled in 100 % of diagnostic declarative clauses. The once chance to escape this vicious circle and explain the existence of V1 declaratives in V2 languages without postulating silent operators in XP is to assume that all V1 structures in this class of languages lack the illocutionary force of declaratives. This, as we argued above, is an attractive solution for V1 thetic sentences like (2), (3a-b) and (4), but not for other V1 structures. Holmberg [5], in a cartographic version of the minimalist framework, assumes that V2 effects arise because of two partly independent features a uPhi-feature triggering movement of finite verbs to some functional head in the C-domain and a [+ EF] feature triggering movement of phrasal categories to SpecFinP, SpecForceP or any other projection in the C-domain. Empirically, Holmberg ʔ aʔʔumpʕion ʕhaʕ uPhi-features and [+ EF] features can be independent amounts to taking V1 declaratives in V2 languages and other apparent deviations from V2 at face value. Along these lines, the finite verb still may take FinP, ForceP or other functional heads in the C-domain if no XP-movement takes place and V1 orders like those in (1) (8) arise. Likewise, if verb movement to FinP, ForceP etc does not depend on the number of spelled-out constituents in front of the moved verb, V > 3 may arise. The problem wiʕh Holmberg ʔ analyʔiʔ iʔ ʕhaʕ lifʕing ʕhe reʔʕricʕive boʕʕleneck condition (G1) and accepting V > 2 main clause declaratives ruins the diagnostics of V2 languages, while accepting V1 clause declaratives does not. I conclude that Holmberg is on the right track in assuming that V1 declaratives and V2 declaratives in V2 languages have different syntax but the analysis in terms of [+ EF] features is misleading. V1 and V2 declaratives differ not in the presence of overt vs covert categories in XP1 but in the position of the moved verb. In V1 declaratives, the finite verb raises higher than in V2 declaratives, V1 movement is always triggered by IS mechanisms that block the movement of phrasal categories to XP (SpecFinP, SpecForceP etc.). The typological evidence for this solution is that V1/V2 languages apparently show the same conditions for verb-fronting as those non-V2 SVO / SVO languages where V1 declaratives are frequent and show a diversity of IS-readings. This is captured by generalization (G2) proposed for the 1 The idea that finite verbs invariably sit in 2P slot in all diagnostic declarative clauses in V2 languages, while the filling of XP is optional is much older than modern generaʕive ʕheorieʔ. Iʕ iʔ fully developed in Diderichʔen ʔ (1946Ŋ structuralist analysis of Danish word order, cf. Zimmerling [17: 136] for an overview. 466 A. Zimmerling diagnostics of V1/V2 systems. Rigid V2 languages in the perspective of the syntax-to-IS interface are a degraded class of V1/V2 languages where V1 declaratives have a limited distribution and are mapped with fixed IS types. Otherwise, IS-triggers of V1 declaratives in V1/V2 systems and rigid V2 systems are similar. 4.5. V1 and clause type shifting V2 languages and V1/V2 languages make use of V1 orders in non-declarative clauses, generally in yes-no questions, imperatives and headless relative clauses. The data from Germanic languages are well-known, for Kashmiri yes-no questions see Bhatt [1: 120]. V2 is a phenomenon characteristic of a diagnostic group of main clause declaratives (German, Swedish) or for all declaratives plus some types of questions (Kashmiri, Afrikaans). V > 3 orders are possible or mandatory in non-diagnostic declarative clauses [12; 21]. The shift from V2 and V > 2 to V1 in yes-no questions, imperatives and headless relatives has a different function than IS-triggered variation within the same clause type it signals that one is dealing not with declaratives but with a clause with a different illocutionary force. In Germanic linguistics, V1 orders with clause type shifting as well as V2 orders in wh-questions have been widely discussed as aspects of V2 syntax. E.g. Platzack [9] postulates silent imperative and question operators for corresponding Swedish V1 clauses, along the same lines he postulates a null topic operator for V1 Swedish declaratives like (4). However, V1 orders with clause type shifting and V2 orders in Wh-question are also characteristic of non-V2 SVO and SOV languages. 5. CL1/CL2 languages At the second step, I apply the same procedure to CL2 languages and divide them into CL1/CL2 and rigid CL2 subgroups. CL1/CL2 languages proved to be extremely rare. This group includes Macedonian (South Slavic), where clustering pronominal clitics behave as strict 2P enclitics in non-verbal clauses and as 1P/2P universal clitics in verbal clauses, Rikavung Puyuma (Austronesian), where clustering clitics behave as strict 2P enclitics in intransitive clauses and as 1P/2P universal clitics in transitive clauses, Kashibo-Kakataibo (Panoan), where some clustering clitics behave as 1P proclitics in imperative clauses and interrogative clauses but as strict 2P enclitics in other clause types and, probably, Slovene (South Slavic), where clustering clitics can be fronted in declarative clauses. This tentative classification in terms of CL2 vs CL1/CL2 systems overlaps with a classification of CL2 languages proposed in Zimmerling [18], Zimmerling & Kosta [20]. The crucial feature is the contrast of standard systems with clustering 2P clitics that they lack clitic-and-verb adjacency and conform to the formula #XP ɤL…V ~ #V CL vs. modified systems with clustering 2P clitics that add a condition on clitic-and-verb adjacency and license orders like #XP CL V ~ #V CL but not orders like 467 1p orders in 2p languages *XP CL Y V, *V Y ɤL, ŋ#…ɤL Y V, *#...V Y CL. Word order systems of the first type are called W-ʔyʔʕemʔ ( W ʔʕandʔ for word or Wackernagel Ŋ, while word order systems of the second type are called W+-systems. In our selection, Kashibo-Kakataibo and Slovene, where clitic fronting is mostly optional, pattern with W-systems, while Macedonian and Rikavung Puyuma belong to a small subgroup of W+-systems where clitic fronting is obligatory in a group of clauses. I stick to the hypothesis previously put forward in [17: 89; 20: 201-202; 18: 71; 19] and independently, in [4; 2] that the features {+verb-adjacent clitic} and {+ 2P clitics} are not mutually exclusive and 2P effects in such languages vith verbadjacent clitics, as Tagalog, Cebuano, Bulgarian, Macedonian, Rikavung etc., are not fake or epiphenominal. Note however that the assumption that W +-systems and languages with vP-internal clitics like French, Italian, Spanish, Albanian, Greek etc. have different syntax does not play any substantial role in the discussion of language data in the following sections since I focus on clitic placement strategies in different type of clauses, not on general characteristics of word order systems with clitics. Empirically, the diagnostic criterion for systems with vP-internal clitics of the Romance type is based on the idea that clause-level (clustering) clitics in such systems lack a fixed position regarded clausal (left or right) border, while the clitics invariably remain verb-adjacent regardless of the linear position of the verb in its clause [17: 85; 18: 76-81; 20: 203]. This criterion is captured in G(eneralization) 3. (G3Ŋ #....V….#, # ….[V + ɤLž… ŋ# … V Y ɤL…#, ŋ#...ɤL Y V…#. (G3) is obviously not compatible with the general condition for 2P languages, (G1). For the sake of simplicity, I assume here that a language with clustering clause-level clitics conforms to (G1) if it has 2P effects at least in one type of clauses. 5.1.Slovene Slovene is a W-system where clustering 2P clitics attach to different types of clitic hosts maximal projections, complementizers, initial proclitic conjunctions and even to sententional constituents which is atypical for Slavic languages, cf. Zimmerling [18: 381]. The placement of Slovene 2P clitics cannot be explained in purely phonetic terms since Slovene clitics can occur after a pause, e.g. after direct speech or a sententional constituent. Another remarkable feature of Slovene clitics is that they allow fronting in declarative clauses. (9a) exemplifies the basic CL2 order, while (9b) with clitic fronting is a marked option for thetic sentences with inferential semantics. 468 A. Zimmerling (9) Slovene, CL2. a. Videl sem ga. See-Past.Prt. CL-Be.Prs.Ind. 1 Sg. CL-3Sg.Acc.M/N I have ʔeen him/iʕ. b. #Sem ga videl. CL-Be.Prs.Ind.1Sg. CL-3Sg.Acc.M/N See-Past.Prt. I have <indeed> ʔeen him/iʕ Getting the inverted order (9b) with fronted clitics in Slovene brings about an effect similar to adding an overt discourse particle. The IS-conditions for clitic fronting in Slovene are similar to IS-conditions triggering V1 thetic sentences with inferential semantics like (3a-b) and (4) in colloquial German and Swedish. In all three languages, the fronting of the diagnostic category (verb vs clitic) remains optional. 5.2. Kashibo-Kakataibo Kashibo-Kakataibo is a Panoan CL2 language where the distribution of clustering 2P clitics is strikingly similar to Slovene. Clustering 2P clitics are described by Zariquey Biondi [16] as elements retaining stress and high tone. They attach to maximal projections and to sententional constitutents and can / must be fronted. The clitic template has 6 slots. Fig. 1. Clitic template in Kashibo-Kakataibo, after Zariquey Biondi (2011). 1 Modality 2 Register 3 Mood 4 Evidentiality kuni cerʕiʕudinal ka narraʕive ra inʕerrogaʕive is reporʕaʕive sapi dubiʕaʕive kaia conʕraʔʕive ri conʕroverʔial 5 Subject cross-reference a 3 6 Mirativity/participant agreement mënë miraʕive mina 2 na 1Sing (na)nuna 1Pl. pa cerʕiʕudinal, nonproximal to the addreʔʔee , pënë miraʕive, nonproximal to the addreʔʔee The possibility of fronting partly depends on the choice of particular clitic, partly on the clause type. From clitics taking slot 1, kuni cerʕiʕudinal doeʔ noʕ move out from clausal 2P, while sapi dubiʕaʕive and kaia conʕraʔʕive do. 469 1p orders in 2p languages (10) (11) Kashibo-Kakataibo, CL1/CL2 a. ë=x =kaia=ka=na Aguaytia=nu kwan-ti ain 1sg=S CONTR.NAR.1Sg Aguaitia=DIR go=NOM be1/2 p. Differenʕly from oʕher people, I will go ʕo Aguaʕiya . b. #kaia=ka=na Aguaytia=nu kwan-ti 1sg=S CONTR.NAR.1Sg Aguaitia=DIR go=NOM ain be1/2 p.1 Kashibo-Kakataibo, CL1/CL2 a. ë=x =kuni=ka=na Aguaytia=nu kwan-ʕi ain 1sg=S DUB.NAR.1Sg Aguaitia=DIR go=NOM be1/2 p. I will cerʕainly go ʕo Aguaʕiya . b. *#kuni=ka=na Aguaytia=nu kwan-ʕi ain DUB.NAR.1Sg Aguaitia=DIR go=NOM be1/2 p.2 In declarative clauses, fronting of clustering 2P clitics remains optional and is similar to the situation in Slovene, although the descriptive grammar of KashiboKakataibo does not specify the exact IS-conditions for CL1 orders in declaratives. In yes-no interrogative clauses the fronting of clustering 2P clitics is obligatory, cf. Zariquey Biondi [16: 491]. (12) Kashibo-Kakataibo, CL1/CL2 [#Ka=ra ain bashi NARR.INT.3p. 3Sg.GEN mountain.ABS Are ʕhere mounʕainʔŠ ikën.ž be.3p. Clitic fronting in yes-no questions lacks a counterpart in Slovene but provides an exact parallel to V1 orders in Germanic V2 languages. The diagnostic 2P category Kashibo-Kakataibo clustering clitics and Germanic finite verbs leave out 2P, take clausal-initial position and mark the shift of clause type by shifting from 2P to 1P. The same mechanism is attested in Kashibo-Kakashibo imperative clauses they require fronting of clitics that normally take 2P in declaratives and other types of clauses, cf. Zariquey Biondi [16: 498]. (13) 1 2 Kashibo-Kakataibo, CL1/CL2 #Ka = ux ! NARR. sleep ʔleep! The example (10) is from Zariquey Biondi, op.cit.,ex. 716. Example (12) is from Zariquey Biondi, op. cit., ex. 711. 470 A. Zimmerling (14) Kashibo-Kakataibo, CL1/CL2 #ka = pi-kan! NARR. eat-PL.IMP Eaʕ you all! (15) Kashibo-Kakataibo, CL1/CL2 #mi-na nipakëʕ-i-n =ka 2sg-lest fall.down-IMPF-1/2 NARR. ɤome down, leʔʕ you will fall. 1 'ibu' descend.IMP The shift in the prosodic orientation of clitics from 2P enclitics in declaratives clause to 1P proclitics in imperatives is just a mirror image of the distribution of clitics in Romance and Balcanic languages (cf. French, Italian, Greek, Macedonian), where pronominal clitics behave as verbal proclitics in finite declaratives and as strict enclitics in imperatives. But shocking as this distribution may seem to European linguists, it only proves that the underlying mechanism of clitic shift is the same and there is no mystic feature of clitic heads requiring enclisis or proclisis in finite declaratives or in imperative clauses. Note that Kashibo-Kakataibo is a W-system and, unlike French, Macedonian or Rikavung, lacks clitic-and-verb adjacency. What matters is that clitic fronting is just a cue of clause type shift and in order to allow this kind of marker, languages with clitics must license them at the clausal left border. Along the same lines, finite verbs in V2 systems get the fronting option if verb fronting serves as marker of clause type shift in imperatives, yes-no questions, headless relative clauses etc. If this obligatory verb fronting is combined with optional V1 fronting in declaratives, one gets an exact counterpart of the Kashibo-Kakataibo situation. To sum up, the data from Kashibo-Kakataibo confirms the hypothesis about the parallelism of clause structure in V2 and CL2 languages. Compared to Slovene, Kashibo-Kakataibo has made a few more steps towards the grammaticalization of clause type shift. 5.3. Macedonian Macedonian (South Slavic) is a W+-system licensing #XP CL V ~ #CL V ~ #V CL orders in finite declaratives. It is sometimes analyzed as a language with vP-internal clitics of the Romance type, where clustering clitics allegedly lack a fixed position in clause. This can not be true for Macedonian on several reasons. Firstly, there are configurations (complex verbal forms) like *AUX Y CL V, where clustering clitics do not attach to a string of two preceding categories AUX, 1 Examples (13-15) are from Zariquey Biondi, op. cit., exx. 618, 619, 626. 471 1p orders in 2p languages Y, so the generalization (G1) is satisfied, cf. examples (20) and (21) below1. Secondly, in a large group of clauses nominal clauses and imperative clauses clustering clitics invariably take 2P and behave as strict encliʕicʔ, cf. Mi eʔka TomiЪ [8: 226ž, Zimmerling [18: 107]. In terms of this paper, that means that Macedonian is a W+-system with clause type shift markers. Most clustering clitics in Macedonian behave as 1P/2P ~ 2P elements depending on the clause type but the question marker =li is a strict enclitic and is never fronted 2. In non-negative finite declaratives, clustering clitics usually precede the verb with the order (X) [CL V] and are fronted if no overt category precedes them, cf. (15a-b). (16) Macedonian, CL1/CL2 a. #Se skina fustanot. Acc.Refl.CL. tear.3Sg.Aor. dress. +D.M.Sg. The dreʔʔ goʕ ʕorn. 1 It is however unclear whether (G1) is satisfied in passive and middle sentences with the order NP/DP + Adv CL V. cf. examples (i) and (iiŊ from Mi eʔka TomiЪ [7: 173-174]. (i) (ii) Ovaa ku a dolgo se grade e. This.F.Sg. house long.Adv. Acc.Refl.CL. build.3Sg.Imperf. Thiʔ houʔe waʔ being builʕ for a long ʕime . Slamenite pokrivi lesno se palat. Straw.Adj.+D.F.Pl. roofs easily Acc.Refl.CL set-on-fire.3Pl. Sʕraw roofʔ are eaʔily ʔeʕ on fire . In a similar Bulgarian W+-system where clustering clitics cannot be fronted, sequences of clause-internal categories NP/DP + Adv are occasionally licenced before the clitics. The same holds for the Czech W-system [20]. Zimmerling [18: 380] and Zimmerling & Lyutikova [21] explain this effect as multiple XP-fronting to a single head. Multiple XP-fronting is possible both in languages with and without clitic-and-verb adjacency. Therefore, Macedonian examples (i), (ii) neither falsify nor verify that this language conforms to (G1). 2 In generative Slavic linguistics, Mac. =li is usually not acknowledged as a clustering clitic since it can be split from the remaining part of the cluster [7: 228]. This is not a valid argument since splitting does not falsify the existence of clusters. Moreover, clustering clitics frequently have non-identical phonetic and syntactic features within one and the same language, cf. Zimmerling [18: 128-153] for a general discussion. Cf. also the discussion of Mac. li in Zimmerling [18: 422-426] and the discussion of Kashibo-Kakataibo data in section 5.2. above. 472 A. Zimmerling b. Na Marijai ii se skina To Marija 3Sg.F.Dat. Acc.Refl.CL. tear.3Sg.Aor. fustanot. dress. +D.M.Sg. Marija ʔ dreʔʔ goʕ ʕorn , liʕ. To Marija, her dreʔʔ goʕ ʕorn. (16b) shows clitic doubling i.e. indexing of nominal expressions by pronominal (dative and accusative) enclitics. If an argument is added, indexing is in many contexts obligatory, e.g. in (16b) one can not drop the dative clitic ii coindexed with the initial phrase [Na Marija]i. Therefore, indexed phrases like [Na Marija] in (16b) are often analyzed as extraclausal in the Slavic linguistic tradition. However, at least in one case, with doubling of wh-words in wh-questions and relative clauses, the indexed category that precedes clustering clitics is not extraclausal. (17) Macedonian, CL1/CL2 a. Kogo vide? Whom.Acc see.2/3Sg.Aor. Who did you/(ʔŊhe ʔeeŠ b. Kogo go vide? Whom.Acc 3Sg.M.Acc.CL see 2/3Sg.Aor Who (of perʔonʔ we knowŊ did you/(ʔŊhe ʔeeŠ c. Komu mu gi dade Whom.Acc 3Sg.M.Dat.CL 3Pl.Acc.CL give cve nijata? 2/3Sg.Aor flowers+D.Pl. To whom (of ʕhe perʔonʔ we knowŊ did you/(ʔŊhe give ʕhe flowerʔŠ 1 The most natural explanation of (17b-c) is that clustering clitics in wh-clauses take a clause-internal position XP (Wh) CL V and are neither strict proclitics nor strict 2P enclitics. The enclisis configuration (X) [V CL] is not typical for Macedonian finite verbal clauses, #V CL orders are usually not accepted in Macedonian normative grammars, at least, in communicatively unmarked sentences, cf. (18). (18) 1 Macedonian, CL1/CL2 a. #Mu gi pokazvala. Dat.3Sg.M.Sg.CL. Acc.3Pl. show-Perf.-l-part. Sg.F. I/you/ʔhe reporʕedly ʔhowed ʕhem ʕo him. The examples (17a-cŊ are from Mi eʔca TomiЪ [7: 257ž, ex. 72. 473 1p orders in 2p languages b. ?? #pokazvala mu gi. This distribution is likely explained by the fact that Macedonian tends to grammaticalize the placement of the verb and to restrict verb movement rather than by constraints on clitic shifting since Macedonian clustering clitics are not specified as strict proclitics (see below). The [CL V] order is the default option, therefore verb fronting (regardless of it is analyzed as XP-movement #Vi ɤL…ʕi or as Long Head Movement across the clitic head #.... Vi CL ti) in sentences like (17b) lacks both IS-motivation and prosodic triggers since Macedonian pronominal and auxiliary clitics do not need a prosodic host to their left 1. In negative finite clauses (except for negative imperatives, which show the derived order #Neg V CL with pronominal clitics) clitic clusters attach to the initial negative proclitic ne and take clause-internal position with the order #Neg CL V, cf. (19). (19) Macedonian, CL1/CL2 Ne e si im gi Not Fut.CL. be.2Sg.CL 3Pl.Dat.CL 3Pl.Acc.CL dadela knigite. given.M.Sg.-l-Part books.+D.Pl. Reporʕedly, you would noʕ give ʕhem ʕhe bookʔ. 2 In non-negative finite clauses with an interrogative clitic li, clitic clusters must take clause-internal positions as well. (20) Macedonian, CL1/CL2 Po Marija li e ti go To Marija Q.CL Fut.CL 2Sg.Dat.CL 3Sg.Acc.N.CL prati (pismoto)? send.3Aor. letter.+Det.N.Sg. Iʕ iʔ wiʕh Maria ʕhaʕ (ʔŊhe will ʔend ʕhe leʕʕer/iʕŠ Mi eʔca TomiЪ [7: 228ž tries to expel li from the cluster by claiming that li in (20) is a 2P enclitic, while the subsequent elements = e=ti=go are proclitics. This claim is impossible to prove without non-circular arguments. It only masks the basic fact that Macedonian W+-system lacks generalized clitic placement and different types of clauses use their own strategies for clitic shifting. I analyze the 1 In the closely related Bulgarian W+-system verb fronting has prosodic triggers since Bulgarian clustering clitics, unlike Macedonian clustering clitics are specified as {+ strict enclitics} and do not front. 2 The example (19Ŋ iʔ from Mi eʔca TomiЪ [7: 229ž, ex. 20a. 474 A. Zimmerling clitic clusters in (19) and (20) as standing in 2P. That 2P/clause-internal placement remains an option in finite clauses without negation and without li, is proved by examples like (21), where the verbal complex of the past perfect construction can not be fronted as a whole. This gives a 2P configuration in (21b). The variation #CL AUX.PST V ~ #AUX.PST CL V in (21a-b) hints that the past perfect auxiliary bea is not part of the cluster. (21) Macedonian, CL1/CL2 a. #Si go bea Refl.Dat.Cl 3Sg.M.Acc.CL be.3Pl.Imperf zavr ile proektot. finished.Pl.l-part. project+ D.M.Sg. They have finiʔhed ʕhe projecʕ. b. Bea si go be.3Pl.Imperf Refl.Dat.Cl 3Sg.M.Acc.CL zavr ile proektot. finished.Pl.l-part. project+ D.M.Sg. ʕhe ʔame. c. *Bea zavr ile si go proektot. (a violation of (G1)) Thus, clauses with a past perfect auxiliary allow both clitic fronting (21a) and auxiliary fronting (21b) but if the auxiliary is fronted, no sentence category can intervene between the host and the clitics, hence (21c) is ill-formed. That in (21b) the clitics really attach to the fronted auxiliary, is proved by another test provided by Mi eʔka TomiЪ [7: 321ž. Cf. the ill-formed sentence (22b), where the subject intervenes between the fronted auxiliary and the clitics. (22) Macedonian, CL1/CL2 a. Beshe mu go Be.2Pl.Imperf M.Dat.CL 3Sg.M.Acc.Cl. dal. 3Sg. given.M.Sg.Pf. l-part. Iʕ waʔ YOU who had given iʕ ʕo him . b. *Beshe TI mu go dal. TI you.2Sg The puzzling thing about (21) (22) is that the same Macedonian clustering clitics in one and the same clause type can either be fronted and behave as proclitics or attach to the fronted category and behave as strict 2P enclitics if the past perfect auxiliary is fronted. The same alternation is attested in one more clause type with n-participleʔ, cf. Mi eʔka TomiЪ [8: 226ž. The clustering clitics here enclitize to n-participle if the participle is fronted and takes XP, cf. (23a). Otherwise, the 475 1p orders in 2p languages clitics are fronted themselves and the participle ends up to the right from them, cf. (23b). (23) Macedonian, CL1/CL2 a. Dojdena sum ti Come.F.Sg.Pf.Pass.Part be.1Sg.CL 2Sg.Dat.CL I have come ʕo pay you a viʔiʕ. b. #sum ti dojdena be.1Sg.CL 2Sg.Dat.CL Come.F.Sg.Pf.Pass.Part ʕhe ʔame .1 vo in poseta. visit vo in poseta. visit In imperative clauses, Macedonian clustering clitics behave as strict 2P enclitics, clitic fronting is blocked: #V CL, *#CL V2. In gerundive clauses, Macedonian clustering clitics normally behave as 2P enclitics but the alternation is possible, cf. Zimmerling [18: 106]. Finally, in nominal clauses, Macedonian clustering clitics behave as strict 2P enclitics and can not be fronted, cf. (24). (24) Macedonian, CL1/CL2 a. #Utre mu e Tomorrow 3Sg.Dat.CL be.3Sg.CL Tomorrow, he haʔ birʕhday. b. #Rodenden mu e Birthday 3Sg.Dat.CL be.3Sg.CL Hiʔ birthday is tomorrow. c. *#Mu e rodenden utre. rodenden. birthday utre. tomorrow To sum up, Macedonian is a W+-system with reduced possibilities for verb fronting: optional alternation of 1P ~ 2P clitics within one and the same clause type is preserved exactly where verbal heads (past perfect auxiliaries, n-participles and gerunds) may be fronted. In other cases, clitic orientation is a marker of clause type shift. Different types of Macedonian clauses use a wide variety of clitic linearization strategies, but the overall trend is that clitics in verbal clauses are fronted or take clause-internal position. 1 2 Examples (23a-bŊ are from Mi eʔka TomiЪ [7: 328ž. On negative imperatives in Macedonian, cf. Zimmerling [18: 391]. 476 A. Zimmerling Fig. 2. Clitic shifting in Macedonian clauses Clause type Non-negative non-imperative finite clauses without past perfect auxiliary and yes-no operator Nominal clauses Imperative clauses Negative finite clauses Finite clauses with a yes-no operator Wh-clauses Finite clauses with a past perfect auxiliary Finite clauses with n-participle Finite clauses with a gerund Orientation of clustering clitics 1P (proclitics)/ clause-internal universal clitics; *Strict 2P enclitics; (16ab); (18) Strict 2P enclitics; *1P proclitics; (24ab) Strict 2P enclitics; *1P proclitics Clause-internal universal clitics; *1P proclitics; (19) Clause-internal universal clitics; *1P proclitics; (20) Clause-internal universal clitics; *1P proclitics; (17bc) 1P (proclitics) ~ Strict 2P enclitics; (21ac); (22ab) 1P (proclitics) ~ Strict 2P enclitics; (23ab) 1P (proclitics) ~ Strict 2P enclitics 5.4. Rikavung Puyuma Rikavung Puyuma is an Austronesian language of Taiwan recently described by Jiang & Billings [6]. It is a W+-system with 1P/2P clustering clitics of Macedonian type but clitic orientation is used as marker of clause type shift more consistently. Optional alternation of CL1~CL2 orders in the same clause type is eliminated. Rikavung has an extremely limited inventory of clitic hosts: in 2P configurations, clustering clitics may attach either to a fronted verb or to a short negative proclitic ъa(ʐi), if it is present1. Intransitive and transitive clauses have different syntax (cf. the contrast of nominal vs verbal clauses in Macedonian). The sole pronoun of an intransitive clause invariably takes 2P and is never fronted. It is either enclicized to the fronted verb, as in (25a) with the order #Vintr CL, or to ʕhe negaʕive procliʕic ъa(ʐi), as in (25b) with the order #Neg Vintr CL2. 1W-systems where 2P clitics attach to just two types of clause-initial hosts verbs/open class predicates vs operator elements have been described earlier in Lummi (Salish) and Kabyle (Afro-Asiatic), cf. Zimmerling [18 : 90]. 2 An exact structural parallel is provided by imperative clauses in standard Macedonian: non-negative imperatives require the order #V CL, while negative imperatives require the order #Neg V CL. As Jiang & Billings (op. cit, 90, footn. 4) mention, imperative Rikavung clauʔeʔ have inʕranʔiʕive ʔynʕax ʕoo, even wiʕh ʕranʔiʕive verbʔ. Tranʔiʕive imperaʕive (including prohibiʕiveŊ clauʔeʔ, ʕhough paradigmatically attesting transitive verbal morphology, nonetheless position a lone overt first-person Undergoer clitic pronoun after the lexical verb (with the second-person Actor unexpressed) /sukun-anaj={ku/*ka}/ (push-TR=1SGŊ Puʔh me. 477 1p orders in 2p languages (25) Rikavung Puyuma, CL1/CL2 a. s<əm>ənaj =ta <INTR>sing =INCL1PL Did we ʔingŠ b. ъa(ʐi) =ta s<əm>ənaj NEG =INCL1PL <INTR>sing Didn ʕ we ʔingŠ 1 Both pronouns of a transitive verb immediately precede the lexical verb regardless of whether Neg is present [6: 90]. Rikavung has only 1-2 p. pronominal clitics, which are not case-marked. The order of clitic pronouns encodes the person hierarchy 1p > 2p. and the Actor > Undergoer hierarchy. If the Actor is 1 p. and the Undergoer is 2 p., both clitic pronouns of a transitive verb are visualized, as shown in (26). (26) Rikavung Puyuma, CL1/CL2 a. {ku/ka}= {*u/nu}= sukun-anaj2 1SG = =2SG= push-TR I puʔhed you. b. ъa(ʐi) ={ku/ka} ={*u/nu} sukun-an NEG =1SG =2SG push-TR.DEP I didn ʕ puʔh you. If the Undergoer is 3 p., only the Actor is expressed by a clitic pronoun. If both the Actor and the Undergoer are 3 p., neither of them is expressed by a clitic pronoun [6: 99] and an inverse marker taw= is added, cf. (27). (27) Rikavung Puyuma, CL1/CL2 a. taw= sukun-anaj INV= push-TR {She/He/They}i pushed {her/him/them}j. b. ъa(ʐi) =taw sukun-an NEG = =INV push-TR.DEP {She/He/They}i didn ʕ puʔh {her/him/ʕhem}j. The inverse marker taw, as Jiang & Billings argue, is not a pronoun. Still, it is a clustering clitic compatible with overt clitic pronouns. The combination taw + 1 All glossed examples from Rikavung Puyuma are from [6]. We stick to Jiang & ɣillingʔ gloʔʔeʔ, where pronounʔ wiʕh ʕhe Undergoer role are marked with boldface. The expression in curly brackets {*xy/xyz} show clitics allomorphs that are possible or impossible in this combination. 2 478 A. Zimmerling single overt pronominal clitic is attested in clauses where the Actor is 3 p. and the Undergoer is 1-2 p., cf. (28). The combination taw + two overt pronominal clitics is attested in clauses where the Actor is 2 p. and the Undergoer is 1 p., cf. (29). (28) (29) Rikavung Puyuma, CL1/CL2 a. ta(w)= {mi/niam}= sukun-anaj INV= =EXCL1PL= push-TR {She/He/They} puʔhed uʔ. b. ъa(ʐi) =ta(w) ={mi/niam} sukun-an NEG =INV = =EXCL1PL push-TR.DEP {She/He/They} didn ʕ puʔh uʔ. Rikavung Puyuma, CL1/CL2 a. ta(w)= {ku/ka}= {*u/nu}= sukun-anaj INV= =1SG= = 2SG= push-TR Did you puʔh meŠ b. ъa(ʐi) =ta(w) ={ku/ka} ={*u/nu} sukun-an NEG = =INV = 1SG = =2SG push-TR.DEP Didn ʕ you puʔh meŠ Jiang & Billings (op.cit., 101) list one more non-pronominal clitic that seems to be part of the Rikavung clitic cluster the irrealis marker a. It is inserted between the inverse marker taw (slot 1) and clitic pronouns (slots 3 and 4). (30) Rikavung Puyuma, CL1/CL2 a. taw= a= {ku/*ka} = sukun-anaj INV = IRR = 1SG = push-TR {She/He/They} would like ʕo puʔh me. b. taw= a= {mi/niam} = sukun-anaj INV = IRR= EXCL1PL = push-TR {She/He/They} would like ʕo puʔh uʔ. It is plauʔible ʕhaʕ Jiang & ɣilling ʔ analyʔiʔ of exʕernal cliʕic pronoun placement in intransitive and transitive clauses can be generalized for the whole Rikavung clitic cluster, albeit it is unclear from the data whether the irrealis clitic is possible in intransitive clauses or intransitive clauses may have sequences consisting of 2 or more clitics1. Meanwhile, the authors state explicitly that 1 According to the description of Rikavung, the inverse marker and combinations of two clitic pronouns should be impossible in intransitive clauses. Note that in Macedonian nominal clauses that exhibit 2P orders with strict enclisis, the inventory of clustering clitics is more limited than in pure verbal clauses with 1P/2P clitics. 479 1p orders in 2p languages transitive and intransitive clauses have the same inventory of 1-2 clitic pronouns and that clitic pronouns are not case-marked and do not encode semantic roles: information about semantic roles comes from syntax and is signaled by the linear order of clitic pronouns and by the presence/absence of an inverse marker. Fig. 3. Rikavung clitic cluster, after [6] Inverse marker 1 ta(w) Irrealis 2 a Pronominal clitics 3 More Person-prominent 1 Sg: ku ~ ka 1 Excl.Pl: mi ~niam 2 Incl.Pl: ta 2 Sg: u ~ nu 2 Pl: mu ~nmu 4 Less Person-prominent 5.5. Interim summary CL1 orders in languages with 2P clitics are triggered by two different underlying factors 1) IS-motivated alternations of word order within one and the same type of clauses; 2) marking of clause type shift by imposing different conditions for clitic placement in different types of clauses. A necessary requirement for clitic fronting and clitic shifting is that clitics lack a phonetic feature preventing them from being positioned at a clausal margin and from changing orientation towards their hosts. Clustering clitics in Slovene, Macedonian, Rikavung and Kashibo-Kakataibo are not specified as {+ strict proclitics} or {+ strict enclitics} regardless of clause type, at least, no such characteristics can be projected to the level of clitic cluster, cf. but KashiboKakataibo enclitic =kana discussed in 5.2. and Macedonian enclitic =li discussed in 5.3. Therefore, clustering clitics in these languages can move and show up in different positions. Slovene and Rikavung are placed at the different ends of a grammaticalization scale, with Kashibo-Kakataibo and Macedonian taking intermediate positions. Slovene is a rigid CL2 language, which only has optional CL1 ~ CL2 alternation in declarative clauses: Slovene CL1 declaratives have inferential semantics and provide a parallel to Swedish or German thetic V1 declaratives. Kashibo-Kakataibo is a CL1/CL2 language, which has both optional CL2~ CL1 alternation and obligatory shifting from CL2 to CL1 in two types of clauses yes-no questions and imperatives. CL2 must be recognized as a default order in Kashibo-Kakataibo, and CL1 as the marked order. Macedonian has optional alternation of CL1 ~ CL2 in three types of clauses where verbal heads of a special type can be fronted. In all other cases, clitic orientation is a marker of clause type shift. Rikavung does not have optional alternation of CL1 ~ CL2 orders and marks clause-type shift (transitive vs intransitive clauses) more consistently than Macedonian. The overall tendency towards CL1 orders in Rikavung might be due 480 A. Zimmerling to a limited inventory of clitic hosts Rikavung clitics only attach to verbs or to a special clitic base negation ъa(ʐi). The overall tendency towards CL1 orders in Macedonian is likely a parasitic effect of restrictions imposed on verb fronting, cf. (18b). At the synchronic level, CL1 orders in Rikavung and Macedonian can not be explained as derived, though these languages preserve 2P orders with strict enclitics, which may reflect the original word order in a diachronic perspective. 6. Conclusions and perspectives The analysis of 1P orders in two classes of 2P languages V2 languages and CL2 languages supports the hypothesis that V2 languages and CL2 languages form a class definable in terms of shared constraints on clause structure. 1P orders in 2P languages are best explained as overt syntactic movement of an operator category (finite verb or clustering clitic) to clause-initial position. 1P orders in 2P languages arise due to two different mechanisms 1) IS-triggered optional alternation of 1P ~ 2P orders in the same clause type or 2) clause type shifting. V1/V2 languages have optional IS-triggered verb fronting in declarative clauses, while V1 orders in declaratives may encode a number of different IS types including rheme dislocation and thetic structures. CL1/CL2 and rigid CL2 languages may have optional IS-triggered alternation of CL1 ~ CL2 orders in some types of clauses. Clause type shifting may feature declarative vs non-declarative split (Germanic languages), nominal-type clause vs verbal-type clause split (Macedonian), transitive vs intransitive split (Rikavung). The data from CL1/CL2 languages with clause type shifting undermines the analysis in terms of edge features [+ EF] and brings a negative evidence against the practice of ascribing the characteristics of clause structure to syntactic heads in languages where different clause types have different positioning of the same fixed positioned operator elements. Kashibo-Kakataibo person markers and Macedonian clitic pronouns do not have any mystical edge feature requiring proclisis or enclisis in imperative clauses but a different configuration in declarative clauses: what matters is that some of these clauses require IS-motivated fronting of non-clitic categories while some other clauses block it. Rikavung clitic pronouns in transitive and intransitive clauses have different distribution not because they get some vague [+EF] feature in intransitive non-negative clauses and lose it in transitive clauses but because Rikavung intransitive verbs must be fronted in non-negative clauses. CL1/CL2 positioning is to some extent an interface phenomenon, since clitics must have non-syntactic characteristics allowing them to change prosodic orientation and to be fronted. The standard feature-driven movement analysis of V2 and CL2 effects explains them by the need to check an uninterpretable uPhi-feature, cf. Roberts [10]. This theory still may be useful for parametric typology. But then, one has to assume that, in 1P/2P languages, the operator element (clustering clitic or finite verb) raises 481 1p orders in 2p languages higher in 1P clauses than in 2P clauses, so that their target positions not only look but really are different. Then one can straightforwardly explain the puzzling alternation of 1P ~ 2P orders in contexts like (21a-b), (10), (12) and 2P effects in contexts like (22) etc.: complementary distribution of the fronted operator and some other fronted categories hints that they compete for one and the same slot. This makes ad hoc stipulations to [+EF] redundant, unless one is concerned by saving framework-internal postulates. References 1. Bhatt, Rajesh M. (1999). Verb movement and the syntax of Kashmiri. Dordrecht. 2. Billings, Loren & Abigail Konopaski. (2002). The Role of Morphology in Ordering Verb-Adjacent Clitics: from Syntax to Prosody in Bulgarian and Tagalog. In: A.Alexiadou, S.Fischer and M.Stavrou (eds.). Linguistics in Potsdam, 19. University of Potsdam. 1-26. 3. Bech, Kristin & Salvesen, Christine Meklenborg (2014). Preverbal word order in Old English and Old French. In: Information Structure and Syntactic Change in Germanic and Romance Languages. P. 233-269. 4. Dimitrova-Vulchanova, Mila. (1999). Clitics in the Slavic languages. In: H. van Riemsdijk (ed.), Clitics in the Languages of Europe (Eurotype 20-5). Berlin- New York: Mouton de Gruyter. P. 83-121. 5. Holmberg, Anders. Verb second. In: Syntax Theory and Analysis. An International Handbook. T. Kiss & A. Alexiadou (eds.). Vol.1. Berlin: Mouton de Gruyter, 242-283. 6. Jiang, Haowen & Billings, Loren. (2015). Person-based ordering of pronominal clitics in Rikavung Puyuma: An inverse analysis. Forthcoming in: Amber CAMP, Claire STABILE, Nozomi TANAKA, & Yuko OTSUKA (eds.), AFLA XXI: Proceedings of the 21st annual meeting of the Austronesian Formal Linguistics Association (Asia-Pacific Linguistics series). Canberra: College of Asia and the Pacific, Australian National University. 87-106. 7. Mi eʔka Tomic, Olga. (2012Ŋ. A Grammar of Macedonian. Bloomington. 8. Mi eʔka Tomic, Olga. (2004Ŋ. The Souʕh Slavic Pronominal ɤliʕicʔ. In: Journal of Slavic Linguistics, 12 (1-2), 213-248. 9. Platzack, Christer. The Edge Feature on C. Lund, University of Lund, 2008. 10. Roberʕʔ, Ian. Phaʔeʔ, Head-movement and Second-Poʔiʕion Effecʕʔ . In: Phases developing the framework / Angel J. Gallego (ed.). Berlin, Boston: Mouton de Gruyter, 2012, 385-440. 11. Reis, Marga. Anmerkungen zu Verb-erst-Satz-Typen im Deutschen. In: Deutsche Grammatik in Theorie und Praxis. Hg. v. R. Thieroff eʕ al. TПbingen: Niemeyer. 2000, 215-227. 482 A. Zimmerling 12. Vikner, Sten. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford Studies in Comparative Syntax, New York, Oxford: Oxford University Press, 1995. 13. Wolfe, Sam (2015). A New Perspective on V2 and the Evolution of Romance Clausal Structure // Traces of History Conference. The University of Oslo, March 9-10, 2015. 14. Yanko, Tatiana (2001). Communicative Strategies of Russian Speech. Moscow. [ . . . . ., 2001.ž (In Russian). 15. Yanko, Tatiana (2008). Intonation Strategies of Russian Speech. Moscow. [ . . . . ., 2008ž (In RuʔʔianŊ. 16. Zariquiey Biondi, R. (2011) A grammar of Kashibo-Kakataibo. La Trobe University. 17. Zimmerling, Anton (2002). Typological Scandinavian Syntax. Moscow, 2002. [ . . . . , 2002]. (In Russian). 18. Zimmerling, Anton (2013). Slavic word order systems from the viewpoint of formal typology. Moscow. [ . . . . , 2013] (In Russian). 19. Zimmerling, Anton (2015). Parametrizing Verb Second Languages and Clitic Second Languages. // Proceedings of the 2015th International Conference on Artificial Intelligence. Vol. 1. 2015, CSREA Press. P. 281-287. 20. Zimmerling, Anton & Kosta, Peter. Slavic clitics: a typology. In: STUF Language Typology and Universals. Vol. 66. No 2, 2013, 178-214. 21. Zimmerling, Anton & Lyutikova, Ekaterina (2015). Approaching V2: verb second and verb movement. In: Computational Linguistics and Intellectual Technologies. Issue 14 (21) in two volumes, Volume 1, Moscow, RSUH, 2015, 663-675. 483