2.1 Verbal forms

Hittite verbal forms can be divided into synthetic and analytic.² Synthetic verbal forms consist of the verbal stem and one or several markers: inflectional marker (cumulative marker of mood, tense, voice and person-number agreement), causative marker, imperfective/durative marker (Hoffner & Melchert 2008: passim; Boley 1984; 1992; Shatskov 2012). The shortest and longest word-level strings are represented in (1).

Analytic verbal forms are represented by the perfect and the passive. Both perfect and passive use a participle with the suffix -ant-. Auxiliary verbs ēš- ‘be’ and ḫark- ‘have’ form the perfect, and ēš- ‘be’ is also used in the formation of the passive (Hoffner & Melchert 2008; Boley 1984; 1992; Shatskov 2012). Thus the participle itself is voice-neutral. The auxiliary verb does not undergo recursion, as the analytic form is either an (active) present/past perfect (with the auxiliary verb in the present or past form), or a passive which expresses tense and mood on the auxiliary verb. The participle precedes the auxiliary verb (2a–b). Auxiliary verbs bear mood and tense markers, whereas derivational morphemes of causative and imperfective are found on the participle (2c).

2.2 Sequence of functional heads

The finite verbal form usually occupies final position in the clause. To its left the following functional elements are found.

— Preverbs. Preverbs in Hittite are separate words with originally locational or directional semantics. Not only single preverbs, but also combinations of two preverbs denoting direction and location (e.g. ‘from behind’, ‘up to’) are attested; in the latter case, the two are strictly adjacent.

— ‘Low’ adverbs. Low adverbs are a closed set of manner adverbs such as kiššan ‘thus’, apeniššan ‘likewise’ which are consistently placed immediately in front of the verb; other adverbs are freely distributed over the clause, scene-setting and temporal adverbs appearing at the left edge of the clause.

— Negation markers. In Hittite, negation markers are sensitive to mood: indicative clauses employ the particle natta ‘not’, often represented by the Akkadogram UL; prohibitive clauses employ the particle lē ‘do not’.

— Modal particle (IRR). The modal particle man/mān can be either a Wackernagel second position enclitic or an independent form in its own linear position. It can be combined with present-future or past indicative finite verb forms.

In (6.1), we see the irrealis particle following the initial complementizer mān ‘if’. In (6.2), the irrealis particle is clause-initial whereas the complementizer kuit ‘as’ is preverbal.⁵

— Subordinators. Subordinators are elements signaling the dependent status of the clause, as well as its type (declarative, interrogative, relative, adverbial). Subordinators are instantiated by elements of various categories: they may be either heads (complementizers) or phrases located in the projection of such heads (e.g. relative wh-phrases). Thus, they do not form a natural class taxonomically, but rather belong to the same functional cluster, and are often complementarily distributed and diachronically related. In Hittite, the distinction between complementizers and wh-pronouns is clear-cut in some cases (e.g. maḫḫan ‘as’ can only be a complementizer, and kuiš ‘who’ a wh-pronoun) and rather controversial in others (e.g. kuit ‘which’, ‘that’ can function as a wh-pronoun or as a complementizer); however, this is of no immediate concern for us in this paper, since they never co-occur (see below for details).

The constituents mentioned above display some variability in their position. For instance, preverbs can be positioned either adjacent to the verb, following the negation marker and adverbs (8a–b), or further leftward in front of the negation marker and adverbs (8c–d) (Salisbury 2005). Complementizers appear both in the left periphery of the clause (9) and considerably closer to the verb, in front of the negation marker, preverbs and low adverbs (10) (Sideltsev 2015; 2017b). In the case of complementizers their position in the clause largely correlates with their class. For example, subordinators mān ‘if’, takku ‘if’ belong to the clause-initial subordinators (С_IN), whereas subordinators kuit ‘because’ (Huggard 2013) and kuwapi ‘when, where’ (Hoffner & Melchert 2008: 417) belong to the preverbal subordinators (C_PREV). It must be observed, however, that a given subordinator can vary somewhat in its positioning, e.g., maḫḫan ‘when, as’ can behave like an initial or a preverbal subordinator (Sideltsev 2017b). The combination of the complementizer with the irrealis marker is revealing: clause-initial subordinators appear to the left of it, and preverbal subordinators to the right, cf. (11). As for preverbs, by contrast, the two positions identified in (8) are not associated with specific preverbs; any preverb can be verb-adjacent or not. The only restriction is that a preverb cannot appear to the left of a subordinator, see Salisbury (2005).

In Sideltsev (2015; 2017a), the sequence of functional elements preceding the finite verbal form is characterized for Hittite as the preverbal position. It includes the following constituents: (a) focused phrases⁶ and interrogative wh-phrases; (b) preverbal complementizers, relative pronouns, indefinite pronouns; (c) preverbs; (d) negation markers; (e) low adverbs. These elements form a strictly ordered array to the left of the verb. If only the sequence of heads is considered and phrases are excluded, it looks like (12):

Introducing phrasal categories into this image we observe, firstly, that indefinite pronouns in the scope of negation immediately follow the negation marker (cf. (8a), (8c)). So it is quite likely that to the immediate right of the negation marker position there is a head that attracts them. As Hittite indefinite pronouns precede low adverbs irrespective of the presence of the negative operator and their scope with respect to it, we suggest that indefinite pronouns occupy the same structural position in whatever configuration and are attracted there by the same functional head F, cf. Sideltsev (2017a) for a similar idea. Alternatively, they may be supposed to adjoin at the edge of the verbal domain, or, as Huggard (2015: 60) suggests, to be placed in Spec, vP, if multiple specifiers are allowed.⁷

Secondly, the linear position of relative pronouns in correlative constructions, interrogative pronouns and focused constituents is to be discussed.

Relative and interrogative pronouns are complementarily distributed with preverbal complementizers and, as expected, with each other. We conclude that they are attracted by a relative/interrogative complementizer in the preverbal position, and the ban on their co-occurrence should be treated as an instantiation of the familiar spell-out constraint known as the Doubly Filled Comp Filter (Chomsky & Lasnik 1977).

Focused constituents, on the other hand, may co-occur with (and invariably precede) both preverbal complementizers (13a) and wh-phrases (13b–c).

Many researchers agree that constituents bearing identificational focus (Kiss 1998) occur in the preverbal position in Hittite (Goedegebuure 2013; 2014; Huggard 2015). Thus, Goedegebuure (2014) treats the OSV word order in (14) as a consequence of the preverbal placement of the constituent zik ‘you’, which belongs to a specific type of identificational (contrastive) focus, so-called replacing focus.⁸ Following Goedegebuure (2014: 401), in this example the actual offender is replaced with another person, who might take his sin upon himself.

On the basis of these data, we conclude that focused constituents are associated with the leftmost head in the array of preverbal elements. They may be attracted by a separate head Foc c-commanding the preverbal complementizer or, alternatively, adjoin to the maximal projection of the preverbal complementizer. While not opting for any specific analysis, in what follows we assume a more economical variant with only one functional head C_PREV.

It is significant that all the remaining arguments and adjuncts of the verb which are not instantiated by constituents attracted to the preverbal position appear to the left of it. Sideltsev (2017a) suggests that they obligatorily raise to the specifier positions of functional projections within the СР layer; other possible explanations involving remnant movement of the verbal domain to TP will be provided below. In any case, the movement produces the effect of indefinite pronouns, wh-pronouns and focused constituents being situated in the preverbal position, to the right of other argumental and adverbial phrases. Incorporating these data, as well as the variable position of the preverb, into the linear sequence of heads in the preverbal position (12) produces the representation in (15):

Finally, combining the data of the linear structure of the preverbal position with the information about the structure of verbal forms and the left periphery of the clause we obtain the resulting linear sequence (16):

2.3 Bi-verbal clauses and verb clusters

Another type of monoclausal configuration involving linear ordering of heads is represented by bi-verbal constructions. They can be divided into two types: the inchoative construction, consisting of an inchoative aspectual verb and its supine⁹ complement, and the complementation construction, comprising a lexical matrix verb and its infinitival complement. In these constructions, the embedded verb projects only a reduced functional structure: both predications share just one Wackernagel position, preverbal position, negation marker, and set of tense and mood markers.

The inchoative and complementation constructions show different grammatical and interpretational properties, as well as a different linear position of the verbal heads. The inchoative construction is characterized by the following properties: the main predicate is used not in its lexical meaning (dai- ‘put’, tiye- ‘step’, epp- ‘take’), but rather as a light verb (LV) (Hoffner & Melchert 2008). The inchoative light verb does not project its own argument and thus does not impose specific requirements on its grammatical subject; it can be instantiated by an inanimate noun phrase, and its theta-role need not be agentive, see examples (17a–c):

In the inchoative construction, the lexical verb is adjacent to the light verb and all the elements of the preverbal position appear in front of the lexical verb. Nothing can intervene between the lexical verb form and the light verb (Hoffner & Melchert 2008: 338). In (18a) we see that the low adverb is placed immediately in front of the supine, and in (18b) we see the indefinite pronoun:

Therefore in Lyutikova & Sideltsev (2021) the inchoative construction is argued to involve functional restructuring (Wurmbrand 2001). The idea of restructuring is in line with the earlier analysis of Hittite inchoative constructions by Koller (2013), although the amount of functional structure Koller associates with the embedded clause is much larger than we suppose it to be.

In the complementation construction the dependent infinitive also projects a reduced structure, but the matrix verb is used in its lexical meaning and has its own argument structure. As a result, semantic restrictions on the compatibility of the matrix and embedded predicates are attested. These restrictions are even more severe in the lexical restructuring configurations available with several matrix verbs, such as ḫantaye- ‘establish’, zinna- ‘finish’: the embedded verb must belong to the class of agentive controllable verbs. For the complementation construction, the linear position of the embedded verb is different from that in the inchoative construction: the embedded infinitive and the matrix lexical verb are separated by elements of the preverbal position, such as negation markers and indefinite pronouns. This is demonstrated in (19).

We observe that the linear order of bi-verbal constructions fits nicely into the linear sequence of the monopredicative construction presented above. In the aspectual construction, the supine is found in the position of the lexical verb (V) and the inchoative light verb assumes the position of the auxiliary verb. In the complementation construction, the matrix predicate occupies the position of the lexical verb whereas the embedded infinitive is located in the position of the preverb. This generalization allows us to conclude that the inchoative light verb and the auxiliary verb, on one hand, and the embedded infinitive and the preverb, on the other, occupy identical positions in the syntactic structure. This generalization is particularly important in identifying the structural position of the preverb. As the embedded infinitive construction is clearly the complement of the matrix verb, the preverb is also likely to project a phrase embedded under the lexical verb as its complement.

A possible alternative suggested by a reviewer is to consider the combination of the verb and preverb as a complex head. Although this analysis can presumably be justified for other languages, we do not think that it is well suited to Hittite preverbs. First of all, preverbs can be separated from the verb by other constituents, cf. (8c–d), which can only be obtained by excorporating either of the heads in the alleged Prev+V complex head. Secondly, preverbs often contribute to the argument structure of the verbal predicate, providing a new telos associated with its own argument. The following examples illustrate the point. The verb šarnink- without preverb is transitive with the meaning ‘compensate’ and is used with accusative of the object compensated for (as in (20a)) and with dative of the beneficiary. When used with the preverb anda ‘in’ (illustrated by (20b)) šarnink- has the same meaning ‘compensate’, but with accusative of the person compensated.

Other examples of this kind involve motion and position verbs. As seen in (21a), iya- ‘go’ is intransitive without preverbs. However, when used with the preverb āppan ‘behind’, it can be transitive ‘follow (somebody)’, as in (21b) (Salisbury 2005: 62):

Likewise, the verb ar- ‘stand’ is intransitive without a preverb, as in (22a), but when used with the preverb āppan ‘behind’ it is transitive with the meaning ‘attend to (something)’,¹⁰ as seen in (22b) (Salisbury 2005: 62–63):

Given that Hittite preverbs commonly specify the resulting state acquired by the internal argument, and can affect the argument structure of the verb by introducing their own argument not selected by the verb, it is reasonable to identify the preverb with the head of ResP (Kratzer 2005; Folli & Ramchand 2005; Ramchand 2008).

In (23) it is shown how the components of the aspectual and complementation constructions are placed in the linear sequence of heads (23a): in (23b) we see the aspectual construction and in (23c) the complementation construction.

Our conclusion is supported by the following evidence. First, in the aspectual construction, the light verb does not attest analytic forms, just like the auxiliary verb, whereas in the complementation construction, the matrix predicate can have analytic forms, just like the lexical verb. In (24) we show complementation constructions where the matrix predicate is passive or perfect. Thus, (24a) is a typical example of the analytic passive (participle of the transitive verb plus a zero present form of the auxiliary eš- ‘be’), whereas (24b) shows the analytic perfect form of the matrix predicate (participle of the transitive verb plus auxiliary ḫark- ‘have’).

Secondly, it follows from (23) that the preverbal position in the complementation construction is expected to split: the negation marker, indefinite pronouns and low adverbs are located to the right of the infinitive, between the infinitive and the matrix verb, whereas preverbal complementizers, relative and interrogative pronouns as well as focused constituents occur to the left of the infinitive. This expectation is met by the linguistic data. We observe this rule in (25): in (25a–b) the focused phrase and the complementizer are to the left of the infinitive, whereas in (25c–d) the negation marker, indefinite pronouns and low adverbs are to the right:

This split in the array of preverbal elements revealed in complementation constructions is highly important for modeling the linearization of the Hittite clause. It indicates that “preverbal position” is not homogeneous, but contains two subarrays of functional elements: the immediate preverbal position encompasses negation markers, indefinite pronouns and low adverbs, whereas the remaining preverbal elements constitute a separate domain. In what follows we refer to the whole array of preverbal elements as occupying preverbal position, and to the subset of them found between the infinitive and the finite verb as being in immediate preverbal position.

2.4 Structural hierarchy of the Hittite clause

In the previous section we established that the order of major clausal constituents can be described as the linear sequence (23), repeated here as (26). In this section we confront the linear sequence (26) with the alleged hierarchical structure of the Hittite clause. The latter emerges as a result of identifying specific lexical items and functional elements of Hittite with (a subarray of) a fixed hierarchy of functional heads constituting the clausal spine (Pollock 1989; Rizzi 1997; Cinque 1999). We assume that structural relations between the functional heads implied by this hierarchy are universal, which allows us, for given heads X and Y, to determine their structural relation unambiguously. As to the question whether all the functional heads are equally represented in whatever clause of whatever language, we refrain from taking a specific position. We limit ourselves to discussing those functional heads which are necessary to accommodate all the elements involved in the derivation of the Hittite clause. Accordingly, identifying a lexical item with a specific head and an XP with a specific projection results in structural ordering of these elements. The resulting hierarchy, based on the identification procedure argued for in Sideltsev (2015) and Sideltsev (2017a), is represented with some simplifications and corrections in (27).

Let us discuss several correspondences which are not straightforward.

The first problematic issue concerns complementizers. It is clear that preverbal and clause-initial complementizers should be hosted by different functional heads. Unfortunately, the distribution of complementizers over these two positions cannot provide us with a cue about the specific heads they instantiate. Conventional functions of complementizers, such as clause typing and illocutionary force encoding, are equally available for preverbal and initial complementizers; no difference between them can be drawn in terms of the finiteness of the clause (since higher functional projections are only found in finite clauses). The only property that appears to distinguish between them is that a subset of preverbal complementizers attract wh-constituents to their specifier and, accordingly, possess the strong corresponding feature (or the EPP feature), whereas clause-initial complementizers never trigger movement.

There is also a weak correlation between the diachronic source of the complementizer and its position: complementizers originating from wh-pronouns (e.g. kuit ‘that’, kuwapi ‘when’) are preverbal, whereas other complementizers are found in both preverbal and clause-initial position. The association of preverbal complementizers with wh-pronouns is further manifested in their complementary distribution. By contrast, clause-initial complementizers are compatible with wh-elements in the preverbal position, as in (28a–b). (28a) is a subordinate temporal clause with complementizer kuitman ‘while’ in the initial position and wh-word (relative pronoun) kuiēš ‘which’ in the preverbal position. (28b) is a conditional clause containing clause-initial complementizer našma ‘or if’ and kuiēš ‘which’.

The association of preverbal complementizers with wh-pronouns is instructive as important evidence about the emergence of preverbal complementizers (see Erschler 2012 for a specific case study), but does not help us to establish a synchronic identification of preverbal complementizers with a specific functional head.

Interestingly, a number of modern and ancient Indo-European languages attest two different positions for complementizers. Thus, a very similar situation obtains in Ossetic, where complementizers are lexically distributed over preverbal and clause-initial positions (Lyutikova & Tatevosov 2009). Moreover, preverbal complementizers are complementarily distributed with wh-constituents attracted to their specifier, but clause-initial complementizers are compatible with them. Lyutikova & Tatevosov (2009) take Fin as the base position of preverbal complementizers in Ossetic and assume that clause-initial complementizers derive through their obligatory raising to Force.

Another well-documented case of preverbal complementizers is provided by Latin. Danckaert (2012: 107ff) argues that the Latin complementizers cum, si and ut are associated with the lowest projection of the split-CP, namely FinP. Similar evidence comes from Modern Greek, where the clause-initial relative complementizer pou contrasts with lower declarative or interrogative complementizers like óti and an (Alexiadou 1997: 76). Two complementizer positions – Force and Fin – are commonly posited for Romance languages, see for Italian and Italian dialects (Ledgeway 2005: 380–389; Paoli 2007; D’Alessandro & Ledgeway 2010), for European Portuguese see (Mascarenhas 2014). The system is also attested in British English journalistic prose (Haegeman 2012: 25–26 fn. 21), and in Gungbe and Saramaccan (Aboh 2006).

We believe that this analysis can be extended to the Hittite data as well. Accordingly, in (27) we identify preverbal complementizers with the Fin head, and clause-initial subordinators with Force.

The specialized topic and focus positions between Fin and Force are not invoked in the further argumentation and are omitted in (27). The Mood head, situated between Fin and Force in (27), seems to be somehow misplaced, since it is usually analyzed as a higher functional element in the split IP domain, and not the split CP domain (Cinque & Rizzi 2008). However, there is evidence that Mood can be bundled with heads belonging to the higher domain: specifically, Jayaseelan (2014) argues that in Dravidian, Mood is a realization of finiteness, and that it competes with typical left-edge elements, such as a relativizer and a (clausal) coordination marker, for the same structural position in the CP domain. It may also be the case that the irrealis particle is merely an exponent of agreement of some higher head in the CP domain in the relevant formal feature. Taking this into account, we prefer to place Mood between Fin and Force, as its linear position suggests.

We do not introduce separate projections for the passive voice or the perfect, since the same participle is used to form both. This is why we associate these categories (passive and perfect) not with the participle, but with the auxiliary verb (Aux). Low adverbs are provisionally associated with the AspP projection, although some of them, namely manner adverbs, are semantically compatible with the verbal phrase; we are not in a position to distinguish between the various semantic classes of low adverbs, or between various treatments of preverbal adverbs (heads of their own functional projection, specifiers of a dedicated functional projection or adjuncts to a semantically compatible functional projection), and will delve into this complicated issue elsewhere.

It is easy to see that (26) and (27) do not map onto each other: the structural hierarchy does not translate either to straight order (hence, consistent right branching) or to inverse order (hence, consistent left branching). In fact, for a given hierarchical structure [1 [2 [3]]] (i.e. where 3 is embedded under 2, and 2 is embedded under 1), various linear orders are attested in Hittite. Throughout this paper, we will use Svenonius’ (2007) notation, represented in (29):¹¹

Our goal in the rest of the paper is to consider various models of linearization of the syntactic structure and to choose those which are compatible with Hittite patterns.

3.1 Narrowing down the range of options

There exist various theoretical options for deriving structures with a required branching direction in syntax¹² — base generation, head movement and phrasal movement.

Let us start with base generation accounts. Setting the branching direction in syntax amounts to establishing a specific value of the Head Directionality Parameter (HDP, Chomsky 1981: 171). Initially, it was supposed that this parameter setting embraced all types of heads uniformly, which was in compliance with typological data. Large-scale cross-linguistic studies from the second half of the 20^th century have established the prevalence of consistent branching patterns across various categories (Greenberg 1963; Dryer 1992; 2009¹³), and deviations from this rule are usually connected with elements which unambiguously identify the syntactic category of a constituent without being its head, e.g., a possessor inside a noun phrase (Hawkins 1994). At the same time, attempts to accommodate deviant word order patterns led to the idea that HDP can be loosened in order to allow different types of phrases to be right- or left-branching. It has been proposed that branching direction can be connected with the syntactic class of heads: thus, e.g., in German and Dutch, [+V] (verbal and adjectival) heads follow, and [–V] heads (nouns and prepositions) precede their complements (Haider 2015). Not infrequently, the branching direction is different for lexical and functional heads: in Tzotzil, for instance, lexical heads taking a rightward specifier contrast with functional heads taking a leftward specifier (Song 2012: 204). Accordingly, assuming the possibility of variable branching direction is often accompanied by the requirement that left (or right) branching heads form a natural class. The obvious reason for this lies in language acquisition and parsing pressure. The assumption that branching direction varies randomly from head to head entails a scenario where a child acquiring a language has to set the head directionality parameter separately for every head, without any cues from previously acquired patterns. Inconsistent branching direction imposes an additional burden on the parser, since structural relations — and specifically c-command, which translates into scope and binding — cannot be read off directly from the linear order.

Abels & Neeleman (2012) argue for an even less restrictive theory of branching where rules of linearization can mention not only the head’s syntactic category and the bar level, but also the category of the dependent. On this proposal, it is possible to base-generate, say, a DP complement to the left of the verb and a CP complement to the right of the verb. Importantly, even in such a system, there are still attested orders that cannot be base-generated and require an extra adjustment of ordering through movement. An additional concern is that this approach creates several homonymous analyses of a phrase which can differ not only as to the movement involved, but also as to the base-generated structure.

In section 2.4, we reached a point where comparing the linear order of heads and the structural hierarchy of the clause led to the conclusion that neither head-initial nor head-final order is consistently observed in clausal constituents in Hittite. Indeed, the uniform HDP setting produces the following linear orders:

It is obvious that neither of the linear orders in (30a–b) fits the linear sequence (26), repeated here as (30c). Right branching fails to derive a verb-final clause and a left-branching complex verbal form. Left branching, by contrast, is accurate in deriving the verbal form and positioning the finite verb at the right edge of the clause, but fails to create the preverbal complex of functional elements represented by both heads and phrases.¹⁴

Now we will consider the suggestion that the head directionality parameter is inconsistent for different types of heads in Hittite. In (31) the syntactic structure of the Hittite clause is shown, which is supplemented with the information about the branching direction of every head¹⁵ (denoted by an arrow → or ← directed from the head), such that it would produce the required linear ordering. It is easy to see that the heads projecting their complement to the right (C_IN, Mood, C_PREV, Neg, F) and the heads projecting their complement to the left (Prev, V, Caus, Asp, Aux, T) do not form natural classes in terms of either category type or morpho-phonological status (separate words vs. affixes). Besides, it is evident that this model is unable to derive variable linear position of the preverb and embedded infinitive. Indeed, if the preverb or the embedded infinitive is positioned between the preverbal complementizer and the negation marker, as in (8c–d) or (19a–b), this order cannot be obtained by varying the branching direction exclusively, but should involve phrasal movement as well; moreover, this movement should target NegP or TP, which is highly unlikely if it is feature-driven. Interestingly, deriving the linear position of the Hittite infinitive, Koller (2013) proposes movement of the infinitival complement CP to an FP situated between NegP and TP; he assumes that all arguments except for indefinites are evacuated out of the VP via this movement. Although this assumption raises many objections — e.g. how come a single functional projection can evacuate more than one XP out of the verbal domain if the verb itself remains in situ and, therefore, remnant movement of VP is excluded? — it illustrates the need for movement in order to obtain the attested position of the infinitive.

Consequently, we abandon the idea that word order in the Hittite clause reflects the base-generated syntactic structure and focus on accounts involving syntactic movement.

The obvious question here is whether we should consider inconsistent branching direction as a starting point for movement accounts, along the lines of Abels & Neeleman’s (2012) proposal, or confine these accounts to consistent branching. We believe that the answer to this question is as follows. From a theoretical perspective, assuming inconsistent branching direction, especially if it is not motivated by some characteristic property of a head, amounts to positing a number of additional stipulations in associating every head with its own branching parameter setting. Since the set of assumptions producing inconsistent branching direction is not sufficient to derive basic word order patterns, and phrasal movement must be introduced into the analysis anyway, economy considerations lead us to limit movement accounts to a base structure with consistent branching. From an empirical perspective, assuming branching direction like that in (31) leads to further problems in deriving the linear positions of argumental and adjunct XPs. Indeed, argument-projecting heads of the verbal domain should be left-branching; therefore, all the XPs they project should appear to the left of them, but to the right of the first head projecting its complement to the right, which is F or Neg. In this way, we obtain the ordering where arguments and adjuncts — except for those attracted by F (i.e. indefinites), C_PREV (wh and focused XPs) and probably T (subject DP) — appear between the lower element of the preverbal complex and the verb. Since this ordering is inconsistent with what we observe in Hittite, we have to evacuate all these constituents to the left, presumably to the specifiers of information-structural heads, or resort to remnant movement of the verbal phrase, having extracted the verb via head movement beforehand. Both operations are highly unlikely: information-structural movements generally lack the degree of obligatoriness which is required to create the rigidly adjacent preverbal array of elements where arguments or adjuncts never intervene; remnant movement of the verbal domain to the left periphery of the clause is difficult to motivate in a feature-driven syntactic system. In view of the above considerations, we see no advantages in assuming the structure created by inconsistent branching as a base for developing movement accounts, and stick to the universally set HDP.

We start with head movement. In this paper, we assume head movement to be an operation present in core syntax, although the issue is highly controversial, and even the unified character of the phenomenon is disputable (see Dékány 2018 for a recent survey and references therein as evidence for a broad spectrum of approaches). However, even if head movement is ultimately to be identified as a PF (or mixed) phenomenon, this will apparently not affect the discussion in this section, since the linear order we are interested in may as well be formed post-syntactically.

Head movement within a left-branching structure is either PF-vacuous or yields permutation of adjacent heads, if adjunction to the right is allowed. This means that although it is possible to derive chunks of straight order within a reverse-ordered sequence of heads, the head array would still be situated on the right edge of the clause, and no phrasal constituents could intervene between heads. Head movement within a right-branching structure seems more promising. It can easily derive the verbal form corresponding to the hierarchical sequence of heads [AspP [CausP [VP [PrevP…]]]] as Prev+V+Caus+Asp, which fits the structure of non-finite verbal forms in Hittite. However, further derivation faces a number of difficulties. Firstly, since head movement is heavily constrained with respect to locality (Travis 1984; Baker 1988), intervening heads disrupt the successive head movement. Thus, the resulting complex head (or the higher head Aux) cannot move to T to build the finite form Prev+V+Caus+Asp+T/Aux+T because of the intervening heads F and Neg, which block the movement. Secondly, head movement within right-branching configurations leaves all the dependents of the lower head in situ, behind its target position. This implies that if Hittite verbal forms were derived by head movement, all the components of the clause except the functional projections and phrases they attracted (indefinite, relative and interrogative pronouns as well as focused and topicalized phrases) would be located behind the verb. As this is not the case, we do not pursue this line of reasoning either.

Now we will turn to phrasal movement as a potential source of the linear hierarchy of heads and morphological word structure. It has been convincingly demonstrated that effects usually associated with head movement, such as head displacement accompanied by order inversion, can result from phrasal movement of the constituent containing this head (see Rackowski & Travis 2000; Kayne & Pollock 2001; Mahajan 2003). Moreover, Julien (2002; 2007) and Svenonius (2007) argue that adjacent positioning of heads (as a single morphologically complex word or as an uninterruptable word sequence) can correspond not only to the structural relation of minimal asymmetrical c-command (32a) or complex heads obtained via head movement under the same conditions (32b), but also to configurations resulting from phrasal base generation or movement (32c–d). Specifically, (32c) assumes that if a constituent in the specifier of X has a head Y on its right edge, Y and X will be linearly adjacent and can form a morphological word. Similarly, in (32d) a head Y on the left edge of a constituent in the specifier of Z will be linearly adjacent to the higher head X and can form a morphological word with it.

(32)	a.
	b.
	c.
	d.

It is important to stress that phrasal movement can create head adjacency only under right branching. Within left-branching configurations, phrasal movement cannot create new head-adjacent combinations which would otherwise be unavailable in the base-generated structure or a structure produced by head movement.

Thus, the less controversial means of obtaining the observed linear sequence of heads based on the syntactic hierarchy above turns out to be consistent right branching supplemented by phrasal movement. The additional advantage of this option is that it belongs to the widely assumed family of antisymmetric approaches compatible with the Linear Correspondence Axiom (LCA, Kayne 1994), which unambiguously translates the asymmetric c-command into linear precedence. The LCA defines a uniform linear structure of the phrase where complements are always to the right and specifiers/adjuncts to the maximal projection¹⁶ to the left of the head ([_XP YP [_XP X ZP]]). Derivation of left-branching structures in the antisymmetric syntax involves multiple movements of complements in right-branching structures to the left, to the position of the specifier/adjunct.¹⁷

Obviously, the substantial question involved in modeling word order via phrasal movement concerns the motivation of this movement. Svenonius (2007) develops a theory which is able, on one hand, to substantially restrict the theoretically possible linear orders arising with phrasal movement in right-branching structures and, on the other hand, to motivate phrasal movement. The idea is that the driving force behind phrasal movement can be the selective features of functional heads (e.g. the auxiliary requires a vP, the causative a VP/vP/PredP, the T head any verbal constituent). If these features are strong, the head which is selected is required to be linearly adjacent to the selecting one. Correspondingly, two options are possible: either the phrase with the selected head at its right edge moves to the (lowest) specifier of the selecting head (cf. (32c)), or the phrase with the selected head at its left edge moves into the (highest) specifier of the phrase which is the complement of the selecting head (cf. (32d)).

In the next section, we present a sketch of the derivation of the Hittite clause that relies on the idea that phrasal movement is motivated by the checking of selectional features under head adjacency.

3.2 Developing the model

The derivation of the clause proceeds in four stages. Firstly, the verbal domain and the dedicated functional shells as high as AuxP are projected. At this stage, successive roll-up applies: every newly merged head attracts its complement to the specifier position, and ends up at the right edge of its own constituent, thus ensuring adjacency with the higher head. This provides preconditions for strong selectional feature checking, as well as for the morphological merger of adjacent heads. The successive roll-up process derives the structure in (33). Note that (33) is identical to the consistent left-branching structure as to the linear order it yields: the indivisible chain of heads comes in the reverse order at the right edge, whereas phrasal constituents reside to the left of it.

The next step is the merger of the lower functional elements of the clause: the functional head F, which attracts phrasal movement of indefinite XPs and NPIs, and the negation head Neg. This is shown in (34). At this stage, PrevP (or the infinitival phrase in complementation constructions) optionally raises to NegP, thus occupying the higher of the two positions available for Prev/V_INF. This is how the immediate preverbal position of these functional elements arises.

At the third stage, T merges. Apparently, T has a strong selectional feature [+Verbal] that ensures that TP dominates a verbal constituent (AuxP/AspP/CausP/VP). By hypothesis, this feature can be checked in a configuration like (35), where movement of NegP to Spec, TP has created the adjacency configuration for the higher verbal head (Aux) and T.

It should be pointed out that phrasal movement is the only option available for deriving a finite verbal form when the verb and T are separated by a number of functional projections that are spelled out independently. Indeed, in the configuration [T [Neg [F [V]]]], T and V cannot form a complex head either in syntax, through local head movement, or in morphology, through morphological operations equally requiring adjacency.

At the last stage, higher functional projections of the clause merge. The preverbal complementizer endowed with relevant features attracts wh and focused constituents, then MoodP and ForceP are projected (36).¹⁸

An anonymous reviewer raises an interesting issue by expressing concern about freezing effects which may prevent extraction of wh- and focused phrases out of constituents which have been moved previously. The term freezing refers to the phenomenon that a constituent becomes an island for subextraction when that constituent has undergone syntactic movement (Wexler & Culicover 1980; Collins 1994; 1997; Müller 1998) — though sometimes freezing affects the moved constituent itself, preventing it from further movement, as in the case of criterial freezing (Rizzi 2006; 2007; Rizzi & Shlonsky 2007). Indeed, if freezing effects are supposed to be universal, the proposed extraction of wh-constituents, focused and topicalized phrases out of the NegP (as well as extraction of indefinites out of the AuxP) is excluded. Obviously, this problem has a much wider scope than merely the analysis argued for here, since any antisymmetric account of word order would face a ban on extraction out of left-sided complements, e.g. preverbal objects (which is not the case in many OV languages, e.g. German). Besides, there are a number of counterexamples to freezing involving widely acknowledged types of movement, e.g. A-movement of the vP/VP-internal subject, raising, and short scrambling (Corver 2017). Therefore, it is reasonable to suggest that the standard formulation of the freezing principle is too strong and should be replaced by a more empirically adequate principle.

In this regard, we find the approach advocated for in Abels (2007) highly promising. Abels proposes that constraints created by movement, such as traditional freezing, criterial freezing and restrictions on remnant movement, should be replaced by a Generalized Prohibition against Improper Movement (GenPIM), which regulates the order of operations a given constituent (or its subcomponents) can undergo. Specifically, with respect to subextraction, GenPIM predicts that subextraction of type i is only possible out of a constituent which has undergone movement of type j if i³j, where the order of operation types is universally defined; roughly, it can be identified with the hierarchy (37), with further detailed elaboration involving scrambling, wh-movement, and topicalization as special cases.

The GenPIM restricts freezing configurations with respect to the type of intended subextraction relativized to the type of the preceding movement. Thus, it is possible to subextract a wh-constituent out of an A-moved constituent, but not vice versa.

Turning back to our proposal, we observe that the analysis involves subextraction of operators (and short movement of indefinites, which might potentially prove to be of A-type); the question is, therefore, what type the successive roll-up movement in the TP domain belongs to. We believe that movement motivated by the checking of selectional features under head adjacency is a natural analogue of θ-relations: functional heads select their complements just as lexical heads select their arguments. If so, the roll-up movement producing head-final constituents is of the lowest type, and subsequent subextraction involving an operation of a higher type (A-movement or Operator movement) is licit.

In this way, the proposed derivation captures all the key features of the Hittite clause identified in section 2. The consistent head-final verbal form is derived as the result of successive roll-ups of verbal projections which locate all the verbal heads adjacently. The clause-final position of the finite verb follows from the two characteristics of clausal projections: first, the strong selective feature of T which is checked by moving the verb-final complement to the left of it to meet the adjacency requirement on checking, and second, the absence of strong selectional features in higher functional heads. The preverbal position emerges as a result of several processes: roll-up derivations in the verbal domain (lower position of the preverb), left-adjunction of lower adverbs (adverb), merger of right-branching functional heads F, Neg and C_PREV and attraction of relevant phrasal constituents (indefinite XPs, wh- and focused constituents) by F and C_PREV. Since selectional features of heads forming the skeleton of the preverbal position are not strong, it exhibits a straight order of functional elements. Finally, the split of the preverbal array by preverbs or embedded infinitives comes as a result of the following combination of properties: intervention of T between the higher and the lower subarrays of heads forming the preverbal complex; the strong selectional feature of T, requiring movement of the verb-final constituent in the position ensuring feature checking under adjacency; and the optional raising of PrevP/InfP to the left edge of NegP, the higher position in the subarray of immediately preverbal elements.

In the last part of this section we provide additional arguments supporting the analysis outlined above.

3.3 More arguments for the analysis

In the presentation above we limited ourselves to the following characteristics of the phenomena under modeling. Firstly, we focused primarily on the linear order of functional heads forming the clause’s skeleton, paying less attention to the position of other clausal constituents. Secondly, we assumed the prevalent word order pattern where the finite verbal form occupies the clause-final position; however, clauses with postverbal constituents are attested as well, though much less frequently. In what follows we address these issues and argue that they provide additional arguments for our account.