2.1 General remarks

The conditional marker ki in Kazym Khanty differs from e.g. English complementizers in that it can occupy various positions in a clause. First of all, it can both precede and follow the clause-final finite verb in the conditional clause. In Kazym Khanty, conditional clauses generally precede the main clause and are verb-final. (1)–(2) illustrate these two positions of ki. The penultimate and ultimate positions are most common ones (cf. Nevis 1990). In (1)–(2), main clauses are identical. This does not mean, however, that these two positions of ki are in free variation. The preverbal (penultimate) position is generally less marked and hence the most common one. I will discuss the distribution of the two positions below in more detail.

In all examples in the paper, the conditional clause is given in square brackets and the host is underlined for convenience. Sometimes, there are multiple instances of ki given in parenthesis. This means that either of these positions, but always strictly one is available in the given clause. ki cannot be omitted in conditional clauses.

As (3)–(5) make evident, ki can also occupy other positions in the middle of a conditional clause, e.g. after the subject (3), the object (4), and an adverbial (5).

Note that the position of ki in (1) and (4) can be distinguished by the presence of a preverb. Preverbs are originally adverbial elements, which now are quite similar to verbal particles in Germanic (e.g. Wurmbrand 2000; Dehé 2015). As with Germanic particles, they can be semantically compositional or opaque. In the latter case they add aspectual, perfectivizing semantics to the verb (Zakirova & Muravjev 2019). They usually occupy a position before the verb, separating it from a direct object. I will address the problem of their syntax in the Section 3.1. For the present purposes, it is enough to say that they allow to distinguish a position after a direct object from a penultimate one. As (1) and (4) show, both are available for ki.

Based on the relatively free placement of ki in a clause, it is commonly accepted that ki is a clitic. Indeed, as (6) shows, ki is an enclitic and requires a constituent to its left (but not to its right) to prosodically attach to.¹

Summing up, ki can be placed in any position in the clause, as long as this position satisfies the requirement that enclitic is attached to an item to the right. However, “free” distribution of the conditional enclitic ki in Kazym Khanty is not unrestricted.

2.2 Default position

A position before the finite verb is a default, i.e. the most common position for ki in Khanty (see e.g. Nikolaeva 1999; Nevis 1990). “Penultimate” ki has the widest distribution: it follows a wide VoiceP-focus in (7), but is also compatible with e.g. contrastive subjects ((8): wɵntər ‘otter’ vs worš ‘eagle’. Hence, the ambiguous term “default” is used on purpose, and means (i) a position with the widest distribution, and (ii) a position compatible with the most neutral, wide-focus reading of a clause.

As example (8) shows, ki can often choose out of two positions in a clause. There are, however, cases, when penultimate position is not available, as in (9).

The difference between (9) and (8) is in the value of non-contrasted part of the clause. In (9), everything, except the subject is given in the preceding question. But this is not the case in (8). Here, there is no pre-given information. Even repetition in the second sentence is not necessarily expected. Hence, a correct generalization is that ki cannot be attached to a background information. But if a clause has both wide-focus and contrastive topic, ki can attach to either of them.

2.3 Marked positions

As mentioned in the previous subsection, distribution of ki is restricted by information structure. Namely, ki is attached to an emphasized constituent.³ The term emphasis is used in this paper as an umbrella term for contrastive topics and new-information/contrastive foci.⁴

First, ki cannot be hosted by any item, which is a background information. In (10), the object in the conditional clause (χośəm jiŋk-əλ ‘fish soup’) is contrasted to the object in the matrix clause (λitəp jiŋk-en ‘caviar soup’). Hence, ki cannot be hosted by the non-emphasized, topical subject (pirəś iki ‘old man’).

Secondly, any position, which is not immediately adjacent to the emphasis and not default, is ungrammatical. If an emphasized adverbial precedes a subject, ki cannot be attached to the subject (11). Similarly, ki cannot follow the [Sub Adv]-sequence, if only the subject is emphasized (12).

Similarly, in (13), contrastive emphasis on the subject (imi-λeŋk-en ‘little woman’) is incompatible with ki hosted by direct object (λitəp jiŋk ‘caviar soup’). Here, ki can occupy the preverbal and the subject-adjecent positions in this clause, but cannot be object-adjacent.

Finally, the clause-final position of ki is allowed in two cases.⁶ (i) In contexts with a narrow focus on verb (14-a). Note that [V_fin + ki] order is not possible, if a direct object is emphasized (14-b).

(ii) When a finite verb is the single overt element in a clause, ki can only be postverbal (15). Whether a clause has wide focus or the verb is narrowly focused, does not play any role. This is due to the enclitic nature of ki and can be seen as a repair operation.

This subsection shows that ki is sensitive to information structure. Its host should be emphasized, i.e. be either a focus or a contrastive topic. Crucially, ki cannot be hosted by any background element.

2.4 Phrase-internal position

Another restriction to the placement of ki is illustrated in (16)–(17). As shown in Section 2.3, generally ki can directly follow an emphasized item. But not every emphasized item can host ki, e.g. it cannot be attached to a demonstrative (16), or to a complement of postposition (17) even if these have a clearly contrastive interpretation.

This restriction, however, cannot be generalized as a ban of any phrase-internal position. ki can be directly attached to a possessor (18), to a nominal modifier (19), and to an adjective (20). Importantly, this only becomes possible under contrastive emphasis on the hosts. The same seems to be true for numerals though not all native speakers allow ki-hosting numerals as in (21).

Importantly, emphasized modifiers stay DP-internal. Though topicalization of a nominal modifier from an object-DP is possible in a conditional clause (22-a), scrambling of a modifier to a position between an indirect object and a subject is ungrammatical (22-c). Hence, I assume that karti in (22-b) does not move out of the DP.⁷ ki can be hosted by a modifier both in a noun-adjacent position (22-b), and in a topicalized one (22-a).

This data allows for two generalizations: first, ki, and not the host undergoes displacement, since modifiers seem to stay DP-internal, when hosting it. Secondly, a contrastive interpretation of a modifier helps ki to be attached only to possessors, nominal/adjectival modifiers and – for some speakers – numerals, but not to demonstratives or complements.⁸ From a syntactic point of view, this looks arbitrary: demonstratives, numerals, adjectives and possessors are all considered to be phrasal modifiers (e.g. Svenonius 2008; Dékány 2011; Danon 2012). However, demonstratives pattern together with complements and not with other modifiers. One can argue that demonstratives are structurally different and are not phrasal modifiers, in contrast to adjectives and possessors (consider Dékány (2011)’s analysis of Hungarian non-inflecting demonstratives). However, Khanty demonstratives can be used independently as pronouns (23), which suggests that they should be phrasal (Dékány 2011).

In Section 4 I suggest an analysis, based on syntax-prosody mapping, as a way to capture this pattern. I argue that the crucial difference between demonstratives vs numerals, adjectival/nominal modifiers and possessors comes from syntactic positions and different prosodic phrasing options.

2.5 Interim summary

This section summarizes the main facts about the distribution of the conditional marker ki in the Kazym dialect of Khanty.

• ki behaves as an enclitic, and always attaches to the right to a prosodically stronger element. If a clause has a single overt item, ki can only follow, but not precede it.

• The position of ki is dependent on the information structure of the clause. Namely, ki is required to occupy a position adjoined to an emphasized (focused/contrasted) XP, e.g. position of ki after a direct object strictly requires a contrastive emphasis on the object, except for cases when it is simultaneously the preverbal position. ki can never be hosted by the background. If a clause has both contrastive emphasis and wide focus, ki can be attached to any of them.

• The default position of ki is the “penultimate” position, i.e. the position between a preverb and a finite verb. In the next section, I argue that this is not a “penultimate position”, and ki is attached to the VoiceP. This position is available, when VoiceP is new-information focus, and is compatible with narrow/contrastive topic on any of the preceding phrases. This leads to apparent optionality of ki-placement in a clause. Another case of such apparent optionality arises in clauses with emphasis on NP-modifiers. If an NP-modifier (e.g. an adjective) is contrastively emphasized, ki can directly attach to this modifier or follow the whole DP. In Section 4, I argue that in both cases the optionality is resolved before ki is displaced.

In a nutshell, the conditional clitic ki in Kazym Khanty must be right-adjoined to a focused or contrastively topical constituent. When several constituents satisfy this condition, the most prosodically prominent is chosen as a host for ki (more on that in the Section 4).

In the next section, I review properties of Khanty syntax which will be important for the analysis. In Section 4, I propose a prosodic analysis of the placement of ki, that is able to capture the pattern presented in the current section.

3.1 Penultimate position is VoiceP-adjacent

In Section 2.2 it was shown that the penultimate position of ki (=before the finite verb) is a default one. The core argument comes from its compatibility with almost all information-structural configurations. If this position is analyzed in linear terms, two problems arise. First, the ultimate position of ki is only compatible with a narrow focus on the verb. It is unexpected, if we analyse ki-placement in linear terms: when in the ultimate position, ki follows a bigger string of words, however, this position is more restricted. Secondly, under closer investigation, it turns out that the default position of ki is in fact not penultimate, as is evident from contexts with negation (see details below).

Both problems can be solved, if we assume (i) that the finite verb moves out of its base-position and (ii) that a “penultimate” ki is attached to the VoiceP. Hence, when in the ultimate position, ki should be attached to the verb alone, which leads to the requirement on narrow focus on the verb in this case.

As a first step of analysis, the position of the verb in the clause should be defined. The linear position of the finite verb does not provide any information about its position in the tree-structure in a head-final language. Therefore, under possibility of head-movement, verbs cannot be used to track the VP. Preverbs are more reliable as VP-markers.

It was noticed that Germanic verbal particles behave differently, depending on whether they build a semantically compositional or an idiomatic unit with a verb (e.g. Wurmbrand 2000). Semantically non-compositional particles are more restricted in their syntactic behavior, stay in situ and cannot be modified separately from the verb. This is also true for Khanty preverbs, which add perfective interpretation to verbs (Zakirova & Muravjev 2019). (24)–(26) sum up the main tests for the mobility of preverbs and these are all more or less failed.

The test in (24) suggests that aspectual preverbs are hesitant to move to the C-domain. The tests in (25)–(26) show that aspectual preverbs cannot move to shorter distances either, e.g. they cannot scramble over an object or a low adverbial. Based on these tests, it can be assumed that aspectual preverbs stay in situ inside the VP, at least when not fronted.⁹

Since aspectual preverbs stay VP-internally, an element between a preverb and a finite verb, if not base-generated inside the VP, provides an argument for the verb to move out of the VP. There are indeed several elements possible in this position:

Negation and discourse adverbials can separate the a preverb(=VP) from the overt copy of a finite verb in the linear order. None of these elements is expected to be base-generated inside the VP, or even VoiceP, for semantic reasons. I also assume them to not be clitics and to be displaced in syntax.¹⁰

Since preverbs are most likely to stay VP-internal, it should be the verb, that moves out of the VP. Moreover, the verb moves higher than the VoiceP, because clausal negation and discourse adverbails are not expected to lower from their base-position across a phase-boundary to a position inside the VoiceP.

The second necessary step of the analysis is to show that penultimate ki is better defined as VoiceP-adjoined, rather than as penultimate/adjoined to a V_fin. An argument for such analysis comes from negation. (29) shows that ki can surface both before and after negation marker ănt. If unmarked position of ki would be truly penultimate, it would correspond to (29-b) in negative clauses. This is, however, not the case. Native speakers, when asked explicitly about the difference between these two sentences, judge the […prev ki ănt verb] order to be a neutral one. Hence, (29-a) has wide focus interpretation, i.e. the whole VoiceP is focused and ki follows the VoiceP. On the contrary, (29-b), where ki occupies the penultimate position and directly precedes the finite verb, is judged as marked and sounding “like a threat”. I interpret this as an emphasis (narrow focus) on negation.

Since both the preverb and the negation marker can be absent in a clause, ki in the neutral position can end up adjacent to an object and a verb, as illustrated in (30).

Given that the verb moves out of its base position, the order in (30) can be formalized as in (31), with ki being adjacent to a VoiceP-internal material, rather than to the clause-final V_fin.

This allows to contrast “penultimate” and ultimate ki more explicitly. Contexts with negation allow to argue that “penultimate” ki does not care about the clause-final verb, it rather follows the VoiceP.¹¹ The ultimate ki, on the contrary, looks for a finite verb and attaches to it. This analysis automatically predicts the the ultimate position of ki is restricted to the narrow focus on the verb (see Section 2.2), while VoiceP-adjacent position corresponds to wide focus.

3.2 Direct objects

As was mentioned several times before, ki can immediately follow a direct object, but precede other VoiceP/VP-internal items, e.g. example (4) repeated as (33-b).

Non-agreeing direct objects in Khanty are usually treated as staying in situ (e.g. Nikolaeva 1999). However, Smith (2020) and É. Kiss (2021) show that direct objects can move out of their base-position, when they trigger object agreement and have a non-new-information status, i.e. are (secondary) topics. This is not a unique feature of Khanty that scrambling, and short object-scrambling in particular, is related to topicality (e.g. Diesing 1992; Choi 1998; 1999 and references therein for German and Korean, É. Kiss 2003 and references therein for Hungarian, Biskup 2006 for Czech, Bailyn 2001 for Russian, Hinterhölzl 2012 for German, Neeleman & Van de Koot 2008; Schoenmakers et al. 2022 and references therein for Dutch, and many others). The exact landing position of the raised objects in Khanty is not important for the purposes of this paper. Crucially, it is either Spec,VoiceP or higher.

When a direct object hosts ki, it obligatorily moves out of its base position: (33) shows that direct object χośəm jiŋk-əλ ‘fish soup’ obligatorily precedes manner adverbial sora ‘fast’. In clauses where direct object does not host ki, both orders would be possible.

Tree in (34) illustrates the grammatical example in (33-b). When hosting ki, object move from its base position to Spec,VoiceP (any maybe higher).

As (33) and a tree in (34) show, only scrambled objects are able to host ki. Nikolaeva (1999) and É. Kiss (2021) argue that only topical objects can and should scramble out of VP. From this follows that only topical objects are possible hosts for ki. And indeed, as (35) shows, an object is interpreted as contrastive topic, when it hosts ki.

Summing up, the order [DO+ki] have specific requirements for information-structural value of the DO. This is parallel to the requirement on NP-modifiers to be contrastively emphasized, when hosting ki (compare Section 2.4). In addition, contrastively topical direct objects differ syntactically from new-information direct objects: topical objects move out of the VoiceP (Smith 2020; É. Kiss 2021). Hence, ki adjoins to an object not inside the VP, but in the scrambled position.

3.3 DP-internal ki

In Section 2.5 I argued that adjectives, numerals and nominal modifiers stay DP-internal, when hosting ki. Here I argue that they additionally move to an outer specifier of the DP.¹²

As (36-a) shows, an emphasized nominal modifier karti ‘iron’ can be separated from the head noun by a focus particle tɵp ‘only’.

I assume that tɵp is DP-adjunct, and not a separate projection in the clausal spine (compare similar proposal in Smeets & Wagner (2018) for focus particles in German and Dutch). (i) As shown in Section 2.4, a nominal modifier can be topicalized, but cannot be scrambled out of a DP, and (ii) no NP-modifier can be raised over an adjunct (37).

Since tɵp ‘only’ is a DP-modifier and scopes over the nominal modifier karti ‘iron’ in (36-a), I assume that tɵp is merged above this modifier. After that, karti moves to an outer specifier of the DP, over the only-projection, as shown in (38). In this position a modifier can host the conditional clitic. This explains why the order [modifier only if noun] in (36-b) is not possible. I suggest that such movement of an emphasized modifier to the left-edge of the DP happens every time, when it hosts ki.

A question arises at this point, why a demonstrative cannot do the same, namely, why it cannot rise to a position, where it would host ki. Apparently, demonstratives are base-generated too high and this movement step is banned as a too short one.

Note that modifier raising is, apparently, syntactically optional. Native speakers allow two positions for ki in clauses with a contrastive NP-modifier: after the modifier and after the whole DP. I suggest that in cases, when ki is attached to the whole DP, modifier stays in situ, and vice versa, when a modifier moves, ki can be only attached to this modifier, and not to the whole DP.

3.4 Interim summary

At this point a conclusion can be drawn that ki is at least partially sensitive to syntax. The arguments in favor are the following.

First, a “penultimate” position is better defined as VoiceP-adjacent, while the verb moves higher. This is supported by the fact, that DP-internal position (Section 2.4) is restricted by very specific requirements on information structure, while the position between the verb and the preverb (=“penultimate”) has the widest distribution and is compatible with almost any information structure.

Secondly, the ultimate, postverbal position has a strict requirement on narrow focus on the finite verb. In a flat linearized sequence of elements this is rather inexplicable. However, if to assume that placement of ki is determined with the access to hierarchical structure, this pattern can be explained via verb-movement.

Thirdly, position after a direct object, if it does not equal the post-VoiceP (=”penultimate”) position, is only available, if the object is emphasized. Before the linerization, direct object is either in situ and non-emphasized or raised out of the VoiceP. Crucially, objects can only be emphasized in a raised position. Since ki adjoines to emphasized hosts and since in situ objects cannot be emphasized, a direct object is a possible host for ki only when it is raised out of VoiceP.

Fourthly, DP-internal position of ki seem to be connected to raising of low modifiers to an outer specifier of DP. A high modifier, a demonstrative, cannot be raised, because movement step would be too short, and hence is not able to host ki.

At the same time, it should be noticed that a complication for a purely syntactic analysis comes e.g. from phrase-internal position and a position after two separate XPs, where only one is emphasized. If ki is base-generated as a complementizer, these positions cannot be derived with syntactic movement. For the position after two XPs, one would have to postulate a separate movement operation for each XP. However, only movement of the ki-adjacent XP could be properly motivated. DP-internal positions cannot be properly captured either. A model, where we postulate movement of the hosting modifier across ki, overgenerates: it is unable to explain why demonstratives behave differently from other modifiers. In addition, this analysis leads to an island violation: when ki is hosted by a modifier of a DP in subject position, we would need to postulate extraction from a subject island to a position above C^o, which is undesirable.

4.1 Prosodic phrasing in Khanty

Crucially for the present purposes, prosody has a hierarchical organization (e.g. Truckenbrodt 1995; Selkirk 2009; Selkirk 2011; Samek-Lodovici 2015; Elfner 2018 for discussion). There are four main types of prosodic constituents: prosodic words (ω), prosodic phrases (ϕ), intonational phrase (ι), and utterance phrase (U). They are organized in a hierarchical structure, where prosodic words are included into prosodic phrases and these are included into an intonational phrase and an utterance phrase. Importantly, boundaries of each prosodic and intonational phrase are dependent on syntactic structure. In (40), I provide core mapping principles (after Bennett et al. 2016, adopted from Selkirk 2009; 2011 and Elfner 2012).¹³

These rules predict that prosodic structure mirrors syntactic structure to a certain extent (41-a). I also adopt the view that prosodic phrasing is blind to traces (41-b), as well as binary branching of the prosodic structure (Bennett et al. 2016; Elfner 2018).

Apart from phrase boundaries, both prosodic and intonational phrases are assigned a head, i.e. the most prosodically prominent element inside each phrase. As argued by e.g. Truckenbrodt (1995), Samek-Lodovici (2015) and Ishihara (2011), head alignment is language-specific. Kazym Khanty seems to group with Japanese in this respect. All prosodic units smaller than intonational phrase are left-headed (42), i.e. for each prosodic phrase, the most prominent, high-pitched element is located at the left boundary (attest H*+L contour of the ϕ χošəm ńań ‘fish pie’ in Figures 1 and 2, and Filchenko 2011 for Eastern Khanty).¹⁴

Figure 1

Default intonation in a conditional clause with wide focus.

Intonational phrases have been assumed to be left-headed as well (Sosa 2020 for Surgut Khanty). However, my data do not show the characteristic descending pitch contour in wide-focus clauses (attest H* tone on the clause-final verb in Figures 1 and 2). Moreover, Sosa (2020) never controls for pro-drop, when topical subjects and objects are phonologically zero. His left-headed descending intonation seems to be an epiphenomenon of pro-drop, combined with narrow focus on direct objects. Alternatively, high pitch on the finite verb in Figure 1 can be characteristic for conditional clauses, as opposed to main clauses. For now, I assume that phonological phrases in Kazym Khanty are left-headed and intonational phrases are right-headed.

In wide focus clauses, as in Figure 1, each ϕ has descending H*+L or L+H*+L pitch contour (high tone on ńań can be attributed to a tracking error, because it is absent in Figure 2). Aligh-ι-Right ensures that the rightmost ϕ has pitch raising after L-tone at the right-boundary of the previous ϕ.

Another part of the theory, required to model the distribution of ki, is prosodic focus-marking (or emphasis-marking). Féry & Ishihara (2010) and Ishihara (2011) argue in detail that a focused ϕ-phrase is universally marked by a high(er) pitch, or pitch range expansion (Burdin et al. 2015). This holds for different dialects of Khanty (Filchenko 2011; Plotnikov 2021). Compare this with the focus-prominence constraint, introduced by Truckenbrodt (1995) ((43)).

Figure 2

Contrastive focus on the direct object.

Sahkai & Mihkla (2017) show that Estonian contrastive topics are marked by an emphatic realization of the pitch accent on the constituent together with a reduced emphasis on the narrow focus. Crucially, prosodic realisation of contrastive topics clearly differs from that of aboutness (background) topics. This is consistent with Khanty data, where contrastive topics, but not background topics, can host ki. Hence, ki-displacement to foci/contrastive topics can be linked to one common prosodic feature, namely prosodic prominence.

The prosodic phrasing constraints, combined with the FocusProminence constraint are enough to predict the host of ki. Moreover, the prosody-based distribution pattern of the Kazym Khanty conditional clitic can be captured in one rule both for focal and topical hosts, as in (44).

Indeed, as in Figure 2, the contrastively focused ϕ ‘χošəm ńań’ does host ki and is most prominent.¹⁵ As Figure 1 shows, the ϕ which hosts ki have a radical pitch lowering on ki. The pitch range expansion seems to be obligatory, unless the next unit begins with [i] or [j]. In the latter case, pitch only descends to the medium level. Presumably, this is due to linear adjacency of two phonetically identical segments and is motivated by articulation. Crucially, pitch assimilation is a consequence of clitic displacement, since it happens after ki is adjoined.

4.2 Theoretical model

In the current model, displacement of ki is not syntactic and happens in the prosodic component. This assumption is made for the following reasons: (i) lowering in syntax violates the Strict Cycle Condition; (ii) any movement of the conditional marker in syntax affects semantic interpretation, which is not the case for ki.

I assume that the placement of ki is regulated by the constraint in (45). At this stage of research, it is not quite clear whether all prosody-oriented clitics form one natural class and whether (45) can be kept as a universal constraint. For this reason, (45) is formulated as a language-specific constraint. Ideally, it would be true cross-linguistically for all prosody-oriented clitics.

An example for a prosodic displacement of ki to an emphasized object is given in (46-a). If the DP χošəm ńań ‘fish pie’ is emphasized, i.e. contrastively topical, the sentence in (46-a) has underlyingly two structures in (46-b). Recall from Section 3 that topical objects move, and that V is also not in its base position. This is why, when the mapping rules apply, the prosodic phrasing is such that the subject DP, the verb, the object DP and the preverb all form separate ϕs. Finally, the object DP is emphasized and the corresponding ϕ₆ is the most prominent ϕ in the clause. Combined with the displacement rule for ki, this prosodic structure requires ki to adjoin to fish pie. In case of a wide focus reading, the topical object still moves and the prosodic structure stays the same. However, now it is the VoiceP-corresponding ϕ₇ that contains the highest prominence. Hence, ki would adjoin to ϕ₇ and surface in default position after the preverb.

The rule in (44) captures not only the post-DO position of ki, as opposed to the “default” position, it also predicts the correlation of intonation and the ki-position after the subject vs default one. In Section 3.1, I have argued at length that the default, “penultimate” position of ki is a post-VoiceP position. This allows to explain (i) why ki precedes a finite verb in a neutral case, and (ii) why this “preverbal” position of ki corresponds to the most neutral, wide-focus interpretation of a clause. Wide focus is universally marked with default (prosodic) prominence (e.g. Féry & Ishihara 2010; Ishihara 2011; Samek-Lodovici 2015 etc.) and is, therefore, able to attract ki, as e.g. in (47). In (47) turn ‘grass’ is nonreferential and indefinite and stays in situ (Nikolaeva 1999; Smith 2020; É. Kiss 2021). This means that ϕ₂ corresponds to VoiceP. As an indefinite object, ‘grass’ cannot bear narrow focus/topic interpretation. Hence, the clause has a wide focus interpretation and ϕ₂ (=VoiceP) hosts ki.¹⁶

In clauses with contrastive subjects and wide focus on VoiceP, as in (48), there are two ϕs competing in prominence. And indeed, ki can be hosted by any of these ϕs. Though this looks as an optionality in ki-placement, I strongly argue that competition in prominence between the two ϕs is resolved before ki is displaced. Namely, at some early point in the derivation, a speaker can choose, which one wins, i.e. becomes more prominent in the final prosodic structure. If subject has a contrastive focus intonation, i.e. it has a higher pitch and is followed by a pause and a deaccenting of the next ϕ, ki is hosted by the subject as in (48-a). But if ki is hosted by VoiceP as in (48-b), a neutral intonation is expected, namely the ϕ corresponding to VoiceP is the most prominent one in the clause and there is no prosodic emphasis on the subject DP. This is indeed borne out in my data.

Relative placement of ki and clausal negation is another example of how position of ki depends on prosodic phrasing and prominence. In Section 3.1, I have argued that the order […neg ki …] requires emphasis on negation. I assume negation to be merged below AspP.¹⁷ If negation is not emphasized, it forms one prosodic unit with V_fin (e.g. as a purely phonological clitic, a ‘leaner’ in Embick & Noyer (1999)’s terms). Since, ϕ₃ does not contain emphasis in (49), there are two options left – either the clause has wide focus, or an emphasis lies on the direct object. In the both cases, ki is attached to ϕ₂. This corresponds to the surface word order as in (49). However, if negation is emphasized, it is not a ‘leaner’ anymore and forms a separate prosodic phrase, which in turn hosts ki, as (50) shows.

I follow Féry & Ishihara (2010) and Ishihara (2011) in that information structure cannot influence prosodic structure pre se. In particular, it has no effect on prosodic phrasing, which is defined by syntactic phrasing only. This plays a role in analysis of ki, adjoined to NP-modifiers and to direct objects. However, this view could be a potential problem for the analysis of […neg ki …] order, because there is no evidence for neg to change its position in syntax. I argue that in its present form, the analysis does not violate the generalization. The contrast in (49) and (50) is between a phonologically defective and full forms of ănt. The phrasing alternation emerges as a byproduct of this contrast (compare similar conclusion in Bennett et al. (2016)).

Finally, one further argument in favor of prosody-based analysis comes from the phrase-internal position of ki. For a syntactic analysis, it is highly problematic to capture the fact that ki is able to follow possessors, adjectives and nominal modifiers, separating it from the rest of the DP, but the same is not possible for demonstratives and complements of postpositions. In Section 3.3 I have shown that adjectives and nominal modifiers move to a higher position inside DP, where they can host ki. Demonstratives, on the other hand, cannot do so, because they are merged too high and the movement would be too local. At this point, it can be suggested that adjectival/nominal modifiers and – for some speakers – numerals are able to form a separate prosodic phrase, because they move to an outer Specifier of the DP. When staying in situ, these NP-modifiers form one ϕ-phrase with the head noun. Demonstratives cannot move and, hence, cannot form a separate prosodic phrase. This goes in line with Ishihara (2011)’s and Féry & Ishihara (2010)’s analysis that prosodic phrasing depends on syntax. Information structure (emphasis) per se cannot change prosodic phasing, e.g. an X cannot form a separate ϕ-phrase only because it is focused, a syntactic movement step is always required for it.

Intriguingly, possessors can host ki, even when embedded in another possessor DP, e.g. in [[[mᵾŋ] wᵾλij-ew] pant] ‘our reindeer’s trail’ (51), which makes it even more complicated for a syntactic movement analysis.

As for possessors, two analyses are possible: (i) either they are base-generated low enough to be able to move to an outer Specifier of the DP (compare Dékány 2011 for Hungarian), or (ii) they are generated as high as demonstratives (e.g. Pleshak 2018 for Kazym Khanty, Assmann et al. 2013 for Udmurt, Deal 2013 for Nez Perce), but always form a separate ϕ-phrase, as adjuncts (Bárány & Nikolaeva 2021). If to follow the analysis, suggested by Bárány & Nikolaeva (2021) for Tundra Nenets (Samoyedic, Uralic), possessors which trigger a possessive marker on a head-noun are DP-level adjuncts, and not modifiers, while in absence of a possessive suffix on a noun, possessors are merged as modifiers lower in the structure. In both cases a possessor is expected to be able to host ki: (i) as an adjunct it already forms a separate ϕ-phrase in situ, (ii) and as a modifier, it is merged low enough to be able to raise to Spec,DP and form a ϕ there (compare suggestions for possessor-raising to/through a Spec,DP in e.g. Szabolcsi 1983; 1994; É. Kiss 2014; Dékány 2015). This is indeed the case, as (52) shows.

For now I leave an indepth investigation of possessor behaviour in Kazym Khanty for further research and assume that the analysis for Tundra Nenets in Bárány & Nikolaeva (2021) holds in Kazym Khanty as well. This predicts that possessors always have a possibility to form a separate prosodic phrase and, hence, can always host ki. The prediction is indeed borne out.

4.3 OT-analysis

So far I have treated the ki-displacement as a rule-based operation. However, most of the studies, which make use of post-syntactic prosody-triggered clitic displacement are done in the framework of Optimality Theory (OT) (e.g. Truckenbrodt 1995; Selkirk 2011; Bennett et al. 2016; Dawson 2017 etc.). This section intends to illustrate that the distribution of the Kazym Khanty conditional clitic can be successfully modeled in OT. As an example, I take a clause with a contrastively focused possessor in (54). A list of relevant constraints is given in (53).

The Tableau in (55) shows that candidate (a) is indeed the most optimal one, despite the violation of align-ɩ-right. It is the candidate where the embedded possessor builds a separate prosodic phrase because it is raised to a DP-adjoined position. It is the most prominent element in the ι-phrase and hosts ki. If the whole DP, including the raised possessor builds one ϕ, as (c), a violation of MatchPhrase makes it less optimal than (a). If we try to keep the prominence, but attach ki to any other host, as in (d), the structure would have a fatal violation of Align-ki constraint. If a candidate satisfies MatchPhrase and both head-alignment constraints, as in (f), it violates the higher ranked focus. Candidate (b) violates matching- and alignment-constraints and candidate (e) violates all constraints.

Potentially problematic for an OT anaylsis are cases of “optional” placement of ki, e.g. (48) repeated in (56).¹⁸ As discussed before, this optionality is resolved before ki is displaced: in order to successfully form a prosodic structure, the prominence hierarchy should be defined, i.e. a single most prominent element should be chosen from several candidates. If competition between two elements is not resolved until the very end, the speaker should make a choice based potentially on some extralinguistic factors, e.g. on the level of emotionality of their speech. Irrespective of how the choice is made, the result is the same – a well-formed prosodic structure with a single most prominent element. Only when this is defined, ki can be displaced.

Another example of the apparent optionality is a DP-internal ki, alternating with a DP-adjacent ki. In (54), ki is hosted by the contrastive possessor mᵾŋ ‘we’, but the same information structure is available, if ki is attached to the whole DP [[[mᵾŋ] wᵾλij-ew] pant] ‘trail of our reindeer’ or to the second level possessor [[mᵾŋ] wᵾλij-ew] ‘our reindeer’, as (57) illustrates.

An example as in (57) looks like true optionality and, hence, is potentially problematic for the OT-model. Nevertheless, I argue that, if the optionality exists, it is resolved before ki is displaced. More concretely, in Section 3.3, it has been shown that whenever an NP-modifier hosts ki, it moves to the outer specifier of the DP. This allows the modifier to form a separate ϕ later in the derivation (see Tableau (55)). However, this movement is not obligatory. Therefore, a DP with a contrastively focused modifier can have two different structures, when it is sent to PF: (i) [_DP [_NP Mod_i [_NP N]] (D)] and (ii) [_DP Mod_i [_NP t_i [_NP N]] (D)]. In the second case, the modifier moves to an outer Spec,DP. This creates a necessary condition for it to form a separate ϕ and allows it to host ki. In the first case, the modifier does not move, hence, it is not able to form a separate ϕ and cannot host ki separately from the rest of the DP. Since the whole DP corresponds to one ϕ, ki can be attached only to the whole DP. Hence, optionality is resolved in syntax and the OT-model can capture the data without severe problems.

To sum up, prosodic phrasing rules are able to capture the distributional pattern of the conditional enclitic ki in Kazym Khanty. Nevertheless, I want to point out that more empirical research on this topic is required. I leave testing the predictions made by the model in this paper for future work as more detailed research of prosodic phrasing in Khanty is needed.

5.1 A second-to-last position clitic (Nevis 1990)

The first formal analysis of Ob-Ugric (Khanty and Mansi) conditional clitic has been suggested by Nevis (1990). He bases his analysis on Klavans’ (1985) theory of cliticization. Therefore, I first give a short summary of Klavans’ theory and then go over to Nevis’ analysis of Khanty and Mansi conditional clitics.

Klavans’ theory decomposes cliticization into three parameters:

There is an immediate consequence: a phonological host of a clitic can differ from a syntactic one, i.e. it predicts existence of ditropic clitics. Said that, the three parameters give rise to eight possible clitic types, represented in Table 1.

Nevis (1990) adopts this classification for a study of Finno-Ugric sentential clitics. Conditional clitic in Khanty and Mansi (both Ob-Ugric) is classified as Type 5/Type 7, depending on its penultimate/ultimate position respectively.

In an ultimate position, ki (=Mansi ke/t’e) is introduced in C. Since Nevis assumes CP to be head-final, C is the last element in the flattened structure. Thus, ki attaches to whatever happens to be linearized before C (parameter 1). It is an enclitic both in syntactic (parameter 2) and in prosodic sense (parameter 3). Hence, the result of cliticization looks as in (59):

As for penultimate ki, it behaves as a type 5, ditropic clitic: its syntactic host is a final constituent, dominated by C (parameter 1), it attaches to the host on the left (parameter 2), but crucially it is an enclitic, and hence, phonologically it attaches to whatever precedes its syntactic host (parameter 3):

This analysis has several problems. First, it is merely descriptive and does not provide any information about mechanics of (syntactic) attachment. Secondly, penultimate clitics are in general controversial and argued to be absent (e.g. Embick & Noyer 1999; Anderson 2005). As Anderson (2005: 79) points out: (i) the term ‘penultimate position’ does not give any information about the placement of ki in the syntax; (ii) all known examples of penultimate clitics are in fact before-the-verb clitics. Hence there is an alternative to analyze them as a pre-verbal clitics and not as penultimate ones. Thirdly, “… syntactically -ke [Mansi; =ki in Khanty] is a left-sister to the (inflected) verb, but prosodically is parsed as a suffix with the preceding phonological word. If correct, this analysis requires that -ke be treated as a truly ditropic clitic.” (Embick & Noyer 1999: 300). And the last but most severe problem of this analysis is that it does not take into account non-(pen)ultimate positions of ki, as well as completely ignores its dependence from information structure (emphasis/contrastive topic).

An attempt to solve these problems is suggested by (Embick & Noyer 1999) for the conditional clitic in Mansi (a closest related Ob-Ugric language).

5.2 Phonological cliticization of a proper complementizer (Embick & Noyer 1999)

An alternative analysis of the Ob-Ugric conditional clitic was proposed by Embick & Noyer (1999). This analysis attempted to solve two main problems of Nevis’ analysis: (i) definition of ki as a ditropic clitic, and (ii) existence of alternative, non- (pen)ultimate positions of ki in some specific contexts. In their paper, Embick & Noyer look at conditional clitic in Mansi ke (=t’e = t’i), which has a distribution, pretty similar that to Kazym Khanty ki.

5.3 Morphological analysis

Another alternative to the prosody-based analysis is a postsyntactic displacement in the morphological component (e.g. Embick & Noyer 2001). There is no existing analysis of Ob-Urgic conditional clitic in morphological terms, and I argue that there cannot be any such analysis. This theory allows only two displacement operations, available for clitics in postsyntax: (i) Lowering, and (ii) Local Dislocation. Both of them are incompatible with Kazym Khanty data.

Lowering is a postsyntactic dislocation operation, happening prior to Vocabulary Insertion and flattening of the structure. Hence, it operates on a hierarchical structure. (71) shows that an item X can be lowered from its base-position as a head of a XP to the head of its complement:

Embick & Noyer (2001) argue that Lowering (i) should be local, and (ii) is blind to adjuncts. Both these properties are incompatible with the data from Section 2. First, ki can be hosted by adjuncts, by both low and high adverbials. Secondly, ki in DO-adjacent position, as well as in DP-internal position is dependent on movement of a host to some higher position. This is problematic for Lowering-analysis, because (i) moved element is in a Spec-position, i.e. should be ignored as an adjunct, and (ii) theoretically, no movement is required for lowering to apply, so this dependency would be redundant. Thirdly, if to assume for semantic reasons that conditional clitic is base-generated in C⁰, it violates locality, if it skips V_fin in Asp (or T) in order to attach to VoiceP.

Local Dislocation takes place later in the derivation, namely simultaneously with Vocabulary Insertion (VI). Embick & Noyer (2001) base their analysis on Late Linearization Hypothesis. Namely, they assume that linearization takes place at VI. Consequently, Local Dislocation operates on a flat structure. Another crucial property of Local Dislocation is that it is local, i.e. can only affect string-adjacent elements. If ki is base-generated in C⁰, it can never be string adjacent to DO in the following clause:

Hence, the crucial argument against Local Dislocation analysis of ki comes from the fact, that its displacement is not local. Furthermore, I have shown in Section 3 that position of ki seem to be sensitive to hierarchical structure, contra to what is argued for Local Dislocation.

The morphological theory of clitics (Embick & Noyer 2001) explicitly allows for one further purely phonological displacement operation, Prosodic Inversion (Halpern 1995).

As a prosodic operation, PI is blind to the category of its host, as is ki. However, similarly to Local Dislocation, PI can be only attached to the closest element to its right or to its left. If CP in Khanty is head-final, ki would never be able to escape the ultimate position (74). If CP in Khanty is head-initial, PI can derive a ban on clause-initial position of ki, but it only allows appearance in the second position (75).

As was discussed throughout this paper, the Khanty conditional clitic can in fact occur in all positions, banned by Prosodic Inversion rule. Prosodic Inversion has the same disadvantage as the Local Dislocation – it is too local. If a clitic is base-generated in C⁰, it can never occur in the middle of a clause, skipping possible hosts closer to a clause-edge. Hence, PI is insufficient for a description of the behaviour of ki, as well.

5.4 Interim summary

The two existing analyses of the conditional marker in Ob-Ugric (Nevis 1990 and Embick & Noyer 1999) cannot account straightforwardly for all the data in Kazym Khanty. They both are able to account for ultimate and penultimate positions, Embick & Noyer (1999)’ analysis additionally captures second position of ki. However, neither of them can handle ki in third or forth position from the beginning of a clause. And neither of these analyses can handle phrase-internal position of ki. The same is true for the PF-model of clitic displacement in Embick & Noyer (2001). The three operations, available in this analysis, are too restrictive to capture the data. Lowering is unable to skip heads and is blind to adjuncts, unlike ki. Local dislocation and Prosodic Inversion are too local for ki.