Mismatch between syntax and prosody and complex sentence structure in Hittite

Hittite attests a set of complex sentences in which (mostly) relative clauses (other subordinate clauses are also attested) appear in linear syntax to be within another subordinate (usually conditional, rarely temporal) clause or main clause. The relative clause can be preceded by a very limited array of constituents from the matrix clause, e.g.,mān ‘if ’ or the quotative particle ⸗wa(r). There are also examples which attest the conditional subordinator and the irrealis particle in two positions simultaneously, at the left edge of the whole sentence and in the conditional clause (=main clause for the relative clause). I provide a fine-grained descriptive and structural analysis of the structure and show that the type cannot be explained as bare indefinites, embedded relative clauses or parenthetic clauses. It is shown that, structurally, the sentences containing subordinate (mostly relative) clauses within other clauses are heterogenous and of three types. The difference lies in the elements that precede the relative clause and in the structure of the sentences, whereby a standard Hittite relative clause adjoins at different heights. All three types display the same mismatch between prosodic domains and syntax/semantics—the constituent that is part of themain clause from the semantic and syntactic perspective is prosodically part of the relative clause. Since there is a clause boundary delimiting the end of the subordinate clause within the main clause, itmakes sense to treat the surface structure as a distinct taxonomic unit, which is correspondingly labelled amismatch sentence. The new evidence allows us to obtain a fuller understanding of the Hittite left periphery than was previously possible. It thus offers an important window into hitherto unrecognized aspects of the underlying structure of the complex sentence in Hittite. Downloaded from Brill.com08/01/2023 10:52:16AM via free access


Introduction
This paper, based on the corpus outlined below in section 9, deals with the syntax/prosody interface in Hittite. It does so by exploring a mismatch between prosody, semantics, and syntax in the Hittite left periphery. The other closely related issue, the distribution of clitics, is systematically taken notice of, but will be analyzed in detail in a different paper. The mismatch is seen in the following example, which served as a starting point for our research: (1) NH/NS (CTH 255.1.A) KUB 21.42 + rev. iv 33-35, cf. Miller (2013: 290- ēš-ten be-2pl.imp '(1) Or if (2) you who are courtiers, (1) you approach the undefiled person of the king, (3) be mindful […] of (your) purity.' Here the relative clause (cl. 2) immediately follows the clause-initial conditional subordinator [našm]a 'or if ' (1a), whereas all the rest of the subordinate conditional clause (without the subordinator, (1b)), follows the relative clause.
Thus the structure consists of a subordinate relative clause within another subordinate (conditional) clause introduced by [našm]a 'or if' . More explicitly, the relative clause is preceded by part of the conditional clause (the conditional subordinator, marked as (1a) and followed by the rest of the conditional Indo-European Linguistics 11 (2023) 1-69 clause (marked as (1b)). The conditional clause is the main clause to the relative one; nothing resumes the relative clause as its subject is 2pl (and there is no enclitic resumptive available for this person in Hittite). There is no clause boundary marked by the clause connective between the relative and conditional clauses. However, the second part of the conditional clause (1b) has a sentential second-position clitic -kan after its first word, which implies a clause boundary immediately in front of ana lugal 'to the king' . At the same time, there is no clause boundary between the conditional subordinator [našm]a 'or if' in (1a) and the relative clause that follows it (cl. 2); this is shown by the fact that the second-position enclitic pronoun -šmaš 'you' (functioning as reflexive in a nominal sentence and thus belonging to the relative clause) follows not the first word of the relative clause (šumeš 'you' in (cl. 2)), but rather the subordinator [našm]a 'or if ' (in (1a)) which is syntactically part of the conditional clause. We will see later that both of these properties are important and recurrent. Thus the use of enclitics indicates that part of the conditional clause (1a) and the relative clause (2) form one prosodic unit for the placement of clitics, whereas the rest of the conditional clause (1b) is set off from this unit by a clause boundary. Seen from another perspective, the relative clause splits the conditional clause and thus necessitates marking of the part of the conditional clause to the left of the relative clause as (1a) and the part of the conditional clause to the right of the relative clause as (1b). Thus, we witness a mismatch between the prosodic domain relevant for the placement of enclitics and syntactic structures: in syntactic terms, (1a) is the initial part of the syntactic structure (1b)-the conditional clause, as marked by the same shading in (1'), whereas in prosodic terms (1a) is the initial part of (2), as marked by the same shading in (1''): (1') syntax NH/NS (CTH 255 ēš-ten be-2pl.imp '(1) Or if (2) you who are courtiers, (1) you approach the undefiled person of the king, (3) be mindful […] of (your) purity.' Such cases will be the topic of this paper. The phenomenon is at first sight marginal and is not acknowledged in Hoffner & Melchert (2008) or in any other research on Hittite syntax that I am aware of (the only exception being Lyutikova & Sideltsev (2021) in Russian, who provide a formal analysis of one of the subtypes of the structure that I am going to discuss), but it recurs in texts of different periods and genres and cannot be written off as a scribal slip.
Subordinate clauses that appear in purely descriptive terms in linear syntax to be within other subordinate clauses are infrequent. I will provide a list of cases that come up in my database.
The paper is arranged as follows. In section 2, I provide a brief overview of the syntax of the Hittite complex sentence as the background to the discussion that will follow. Then, in sections 3 and 4, I will list and taxonomize all "mismatch sentences" that I have come across in my corpus. First, they will be described from a pre-theoretical, inner-Hittite perspective. I will do so by carefully confronting them with structures that appear to be virtually identical, but that will be shown to be different, namely bare indefinites (section 4.1) and embedded relative clauses (section 4.2). Then, in section 5, I will sketch their theory-dependent structural construal and argue that mismatch sentences are very heterogenous in structural terms. The structural analysis will be done in the Minimalist framework, assuming a split CP hypothesis and supraCP projec-Indo-European Linguistics 11 (2023) 1-69 tions hosting vocatives and clitic-doubled left dislocations. It will be shown that the only common property of mismatch sentences is that they all attest a mismatch between the syntactic domain and the prosodic domain. The paper will then proceed in section 6 to confront mismatch sentences, now firmly established both descriptively and structurally, with yet another type of complex sentence in Hittite-parenthetical clauses. The contribution of this section is that it will confirm the conclusions of sections 3-5 and crucially highlight another important property of the mismatch between prosody and syntax, that the first part of the main clause in mismatch sentences has to be a one-word constituent. In section 7, I then explore the subtype of mismatch sentences that involves doubling of the subordinator and show this subtype to stem from one of the mismatch sentence subtypes previously seen (in section 4.1). Section 8 presents the conclusions, and section 9 describes the corpus of the texts used for the study.
It is important that in exploring every issue I will start from linear facts described pre-theoretically and only then proceed to theory-dependent structural analysis. Sections 1-4 and 6-8 are pre-theoretical, whereas section 5 is couched in the Minimalist paradigm. The heterogeneity is a very conscious attempt to produce a description of Hittite primary data accessible to specialists in Hittite without much theoretical background and at the same time to argue a cross-linguistically interesting case of prosody/syntax mismatch for theoretically informed scholars both with and without prior knowledge of Hittite. Thus, rather than being conceptually inconsistent, the paper aims to be accessible to both potentially interested audiences.

Complex sentence syntax: multiple subordinate clauses
First, I will provide a brief sketch of the syntax of Hittite complex sentences with multiple subordinate clauses, one of which is a conditional clause. The conditional clause in Hittite always precedes the main clause, as is illustrated by (2) MH/MS (CTH 262) IBoT 1.36 rev. iv 15-16, cf. Miller (2013: 118- If the conditional clause is one of several dependent clauses, it may follow other dependent clauses, including relative clauses: (3) a. NH/NS (CTH 577) KBo 2.2 obv. i 1-3, cf. van den Hout (1998: 124- wemiya-z find-3sg.prs '(1) As long as His Majesty (will be) within the country of Nerik (2) until he comes up (home), (3) if then fever will not befall His Majesty, […].' b. MH/MS (CTH 262) IBoT 1.36 rev. iv 18-20, cf. Miller (2013: 118- ḫarnink-andu destroy-3pl.imp '(2) If you do not observe (1) these matters that we have put under oath for you, (3) and violate them, (4) these oath deities will destroy you.' In the last example, the relative clause 'these matters that we have put under oath for you' is resumed by the anaphoric pronoun in the two conditional clauses (-at 'them' in cls. 2 and 3) which thus are the main clauses to it. Cls. 1-3 thus constitute a typical correlative structure, lit. '(1) [these matters that]i we have put under oath for you, (2) if you do not observe themi (3) and violate themi' . This example cannot be understood in any other way than the conditional clause functioning as the main clause for the relative clause, because the clause that follows the two conditional clauses (2 and 3)-clause 4, main clause for the conditional clause-does not resume the relative phrase, either explicitly or implicitly.
In other cases other dependent clauses follow the conditional clause: (4) MH/MS (CTH 262) IBoT 1.36 rev. iv 22-23, cf. Miller (2013: 118- In (4), the relative clause is independent of the conditional clause; it is resumed by the anaphoric pronoun in the main clause that follows it.
In the following case, the conditional clause is followed by a temporal clause: (1) If, however, he (the king) goes into some place by chariot, (2) then as soon as the king steps down from the chariot, (3) the chief of the bodyguard along with the bodyguards bow behind the king.' Thus, in linear syntax subordinate and main clauses follow each other, with clear demarcation of clause boundaries by clause connectives (most commonly nu) and enclitics, which are after the first word that follows the clause boundary (clause connectives also count as first words). The use of clause connectives exhibits more variation at the older stages of Hittite and is also governed by a complex set of discourse factors, see (CHD L-N: 446-448;Hoffner 2007: 387-388;Widmer 2009Widmer : 332 & 2016 esp. section 6.3), but still, if viewed across a set of examples, it is a reliable indication of clause boundaries. It is significant for the discussion that will follow that commonly a clause connective simultaneously marks both the end of one cause and the beginning of another.

Relative and temporal clauses within conditional clauses
However, in yet other cases other subordinate clauses-mostly relative clauses-are in the position that appears in linear syntax within the conditional clause, not preceding or following the conditional clause. In other words, some of the material of another subordinate clause precedes the relative clause and some follows it.
The first example has already been provided in the Introduction. All of its properties hold for all the following examples, which I will term "mismatch sentences" for brevity in the remainder of the paper.  Here the structure is literally as follows: '(1a) until-(2) the years which are determined for him-(1b) he exhausts them.' The sentence is a common correlative structure where the relative clause (mu.kam ḫi . a kuiēš daranteš 'the years which were decreed (for him)') precedes its main clause and the relative phrase is resumed by the anaphoric pronoun (-aš 'them') in its main clause. The relative clause (2) and its main clause (1b) are clearly demarcated by the clause connective nu between cls. (2) and (1b) and by the enclitic chain in clause-second position in its main clause (1b), i.e., -war-aš-za. However, the clause which is the main one for the relative clause (1a-b) and resumes with an anaphoric pronoun the relative phrase in the relative clause (2) is simultaneously the temporal subordinate clause to the following main clause (cl. 3). The subordinator from the temporal clause (kuitman in (1a)) precedes the relative clause (2) in the surface syntax. This is unexpected from what we currently know about the Hittite syntax, but I will show that it is quite commonly attested.
However, an important feature of the example is that the subordinator not only precedes the relative pronoun, but is prosodically part of the relative clause, even though in syntactic and semantic terms it belongs not to the relative clause itself, but to the temporal clause which is the main clause for the relative clause. As a prosodic part of the relative clause, it hosts the enclitics of the relative clause, -šši as well as the direct speech enclitic -wa, which belongs to the complex sentence.
Thus, the relative clauses in exx.
(1 & 6) appear in purely linear descriptive terms to split conditional and temporal clauses, i.e., they can be understood as being 'nested' within conditional and temporal clauses. I stress that by 'splitting' , 'being within' , or 'being nested' I have in mind purely linear phenomena in the surface position of the two clauses relative to each other. The relative clauses themselves seen in exx. (1 & 6) are not different by their own properties Indo-European Linguistics 11 (2023) 1-69 in any significant respect from 'regular' relative clauses which either precede or follow other subordinate or main clauses. Thus they do not constitute any particular taxonomical class of relative clauses. It is their position within another clause, as well as the overall properties of the resulting complex sentence (mismatch between prosody and syntax) that are special.

Analysis of mismatch sentences
In what follows I will tackle several potential problems for the analysis of mismatch sentences and show that they are illusory: a. Are relative2 clauses in mismatch sentences bare indefinites (relative pronouns functioning as indefinites in conditional clauses)? b. Can the relative clauses in mismatch sentences be assessed as embedded relative clauses? c. Can the relative clauses in mismatch sentences be assessed as parenthetical clauses? In answering these three questions, I will list and assess other cases of mismatch sentences.

4.1
Relative clauses in mismatch sentences or bare indefinites? When assessing relative clauses in mismatch sentences, the question that immediately comes to mind is whether we really have a relative clause within a conditional clause and not what is labelled 'a bare indefinite' .
The relative clauses within conditional clauses are indeed superficially identical to relative pronouns used as bare indefinites, i.e., functioning as indefinite pronouns in conditional clauses. I will illustrate the bare indefinite with the following example: (1) Further, if one of the forward men allows something in (to the procession), (2) either horses or aggressive cattle, (3) then it is the forward man's fault; (4) if, however, one of the rear men allows something in (to the procession), (5) then it is the rear man's fault.' This example attests bare indefinites twice, in clauses (1) and (4). Both times, what are formally relative pronouns (kuiš 'who' and kuit 'what') are used as indefinites (in lieu of kuiški 'someone' and kuitki 'something' , which are also attested in such contexts elsewhere). This happens-in this context and elsewhere-when there is a conditional operator mān 'if' in the clause. However, examples (1) and (6) from the previous sections are different from bare indefinites. Ex.
(1) definitely belongs to the mismatch sentences: its relative clause cannot be assessed as a bare indefinite in a conditional clause 'if some courtiers' because the verb form is 2pl (a bare indefinite would require a 3sg or pl verb form). What is more, the use of a 2nd pl pronoun (functioning as reflexive) is attested in nominal sentences when the subject is 2nd person, not 3rd person. It also provides evidence that the relative clause is a separate nominal clause, and not a bare indefinite phrase within the conditional clause.
Ex. (6) also unambiguously contains a relative clause in a mismatch sentence and not a bare indefinite in a conditional clause for two reasons: (a) the relative clause is resumed in the following clause as a regular relative clause; (b) it is within a temporal clause, and not within a conditional clause as is the case with bare indefinites.
The following example is also likelier to be understood as a relative clause in a mismatch sentence than as a bare indefinite because if it were a bare indefinite, a singular form of the indefinite pronoun would be expected agreeing with the singular verb of the main clause (išt[a(mašzi)] in (1b)), which is not the case. As an anonymous reviewer correctly observes, the verb is restored, but it is restored after duplicates, so we can be secure that it is 3sg indeed. What we have is the following structure: '(1a) if-(2) which (pl) are a brother of the king, a lord, a prince, a courtier, (1b) (if someone) hears of an evil matter regarding the king':  The same is basically applicable to the following case, although there is no resumption of the relative clause: (9) NS (CTH 712.A) KUB 27.1 obv. i 20-22, cf. Wegner (1995: 32), CHD (L-N: 4b-5a): (If the years passed, for the Istar of old they resume the old ritual, but for Istar the mighty of Mursilis they do not resume the old ritual […] The campaigns that the king regularly fought, as many campaigns as he has fought in the years that went by, while he celebrates the deity, for these field ( ēš-zi be-3sg.prs '(1) But if the king has gone nowhere on a campaign (2) in the years which have gone (by), (3) there is no ritual' ,3 lit. '(1a) if (2) which years have gone (by), (1b) the king does not go to any land (3) there is no ritual.' Here there is a very explicit clause boundary between the main clause and the subordinate clause: it is marked by the clause connective nu before the main part of the conditional clause (1b), which is the main clause for the relative clause (2).
The position of the adversative particle -(m)a here is noteworthy. It has wide scope over the whole sentence and contrasts the two stretches of discourse: the campaigns that the king has gone on and consequently celebrated with the ritual (the previous context given here in translation), and the campaigns that the king has not gone on and consequently did not celebrate with the ritual (the present context). More concretely, the fact that the king went on a campaign which entails the ritual versus the fact that the king did not go on a campaign which does not entail the ritual are contrasted. Thus, obviously the adversative marker -ma scopes over the main clause (1a-b), yet it is hosted by the first nominal in the subordinate relative clause (cl. 2). This is not at all surprising as this is the common pattern in Hittite, not only for wide scope -(m)a, for which see Meacham (2000), but also for the quotative particle -wa(r) introducing direct speech; see Sideltsev (2020). This is illustrated by the following two examples for -(m)a: (10) a. NH/NS (CTH 81.A) KUB 1.1+ rev. iv 53-55, cf. Otten (1981: 26-27); Meacham (2000: 130, 140 [(up)]p-ir send-3pl.pst '(1) They began sending me gifts, (2) but the gifts that they send me, (3) they never sent to any (of my) fathers and grandfathers.' In this example the contrast is between cls. (1) and (2-3) on the whole, but -(m)a has a prosodic host within the first relative clause (cl. 2) of the second sentence (2-3). There is nothing in cl.
(2) itself that would be contrastive to the previous context. (If a male slave runs away and goes to the land of Luwiya, (his owner) shall pay six shekels of silver to whoever brings him back. If a male slave runs away and goes into an enemy country,) 1. kui-š-an who-nom.sg.c-him āppa-ma back-but uwate-zzi bring-3sg.prs 2. n-an-za conn-him-refl apā-š-pat that-nom.sg.c-foc

dā-[i]
take-3sg.prs '(1) but whoever brings him back, (2) he (and not the owner) shall take him instead (of receiving a financial reward).' Meacham (2000: 94) convincingly argues that here there is no contrast in the relative clause to the previous context. All the contrast lies in the main clause, yet -(m)a is placed in the relative clause (1) that linearly precedes the main clause (2). See also for a very careful analysis of the context Goedegebuure (2014: 359 & 400). We see the same for the direct speech particle -wa(r) in the following example:  (1) Just as the whorl of the spindle turns, (2) may bad innards likewise turn away from the child (3) As a shaft of an arrow does not find another shaft, (4) may likewise bad innards not find the child.' Here we see that -wa(r) is present once in each of the two independent syntactic units (1-2 and 3-4, respectively). Each sentence has the identical structure of dependent clause-main clause. -wa(r) marks each sentence and is used once at its leftmost clause, which is in this case the dependent clause (cl. 1 and 3, respectively).
What is more interesting, though, is that in ex. (9) above the conditional mān 'if' does not host -(m)a, even though it also scopes over the whole sentence, not just over the relative clause whose prosodic domain it belongs to (hosting the enclitic of the relative clause). Instead, it precedes the nominal of the embedded relative clause that hosts -(m)a. This placement of the conditional mān 'if' vis-à-vis -(m)a is typical of Old and Middle Hittite texts and also occurs in New Hittite texts; see Meacham (2000); Kloekhorst (2014); Sideltsev & Molina (2015). Thus both mān 'if' and -(m)a scope over the whole complex sentence made up of clauses (3-4), but, according to common rules of placement, are positioned in front of the first clause of the sentence. As -(m)a cannot cliticize to mān 'if' , it cliticizes to the first word of the relative clause.
The following examples are likely to belong to the same type, even though they can be alternatively analyzed as 'or if there is some Hittite border commander and you request infantry and chariotry from him' . However, the parallelism of the clause which functions as the main clause for the relative clause (2) with the first clause (1: 'or if you request infantry and chariotry from My Indo-European Linguistics 11 (2023) 1-69 Majesty') makes it likely that here we have another example of a mismatch sentence-'or if you request from a Hittite border commander who is in your vicinity': (12) NH/lNS (CTH 67) KBo 5.4 rev. 25-26, cf. Friedrich (1926: 64-

2b) you request infantry and chariotry from a Hittite border commander (3) who is in your vicinity.'
Here the relative clause (3) is resumed by the stressed pronoun apēdani in the main part of the conditional clause (cl. 2b). The clause boundary between the relative clause (cl. 3) and its main clause (cl. 2b) is again marked by nu. The clause boundary between the first part of the conditional clause (2a) and the relative clause (3) is unmarked. The enclitic pronoun -tta is after the first word of the main clause (2a), yet it belongs to the relative clause (3).
The following example is analogous to the previous mismatch sentences: In this sentence, the relative clause (cl. 3) is not resumed in its main clause (cl. 2b) by an enclitic personal pronoun, because there may be a coreferential indefinite pronoun in the main clause (although it is restored and thus not certain), but the clause boundary between the relative clause (cl. 3) and the main part of its main clause (cl. 2b) is again marked by nu. The relative clause is also not nominal, differently from all the previous examples, but contains a finite verb form ešir 'were' . There is again no clause boundary between the initial part of the conditional clause (2a) and the relative clause (3); the enclitic of the relative clause -tta is in the first part of the conditional clause on the subordinator (2a).
Here the alternative understanding of cl. (2) 'if there were formerly some of your sworn allies' is inferior to the relative clause assessment above as follows from the clumsiness of the rendering involving bare indefinite understanding of ku[iēš]: *'If some Hittite comes to you on that matter or if there were formerly some of your sworn allies and some (of them) speaks to you as follows' .
Finally, I will list yet another example which cannot be interpreted as containing a bare indefinite because what functions as the main clause for the relative clause here has temporal semantics 'when' and the situation is episodic in the past, not habitual. Thus, the context cannot license a bare indefinite: get_evil.3sg.pst '(1a) When (2) Urhi-Tessub, who [kept pursuing(?)] me because of the lordship, (1b) became alienated from me over the land of Nerik, (my friends and associates kept intimidating me saying: "For Nerik you will perish.").' Yet another reason for attributing this example to relative clauses within conditional clauses, and not to bare indefinites, is the fact that a definite description (mUrḫi-d10-upaš 'Urhi-Tessub') is meant, and not 'some person' , as would be the case with the bare indefinite. In all other respects the example is also a typical Indo-European Linguistics 11 (2023) 1-69 mismatch structure: the enclitic -mu of the relative clause (2) is within the first part of its main clause (1a), and there is no clause boundary between the first part of the main clause (1a) and the relative clause (2), but there is a clause boundary between the relative clause (2) and the main part of the main clause (1b). The enclitics of the main clause are within the second part of the main clause (1b).
Thus far, I have presented evidence that the structure labelled here as mismatch sentences cannot be assessed as involving a bare indefinite phrase within a conditional clause as: (a) at least some of the nominals in the relative clauses within conditional clauses are definite descriptions; (b) some of the contexts involve temporal and not conditional clauses as the main clause for the relative clause; (c) some of the contexts involve a finite verb form in the relative clause, thus invalidating their analysis as a bare indefinite phrase; (d) the relative clause in a mismatch sentence is demonstrably a different clause, and not a nominal in the conditional clause-there is a clause boundary between the clauses, and resumption of the relative phrase in the main clause. I hold the sum of the evidence as irrefutable, even though not all the four properties are attested in every clause.

4.1.1
namma preceding relative pronouns The following cases also argue against the analysis of relative clauses in mismatch sentences as bare indefinites for the simple reason that they occur not within subordinate clauses, but rather within main clauses. They are thus different from the examples assessed in the previous section, which involved relative clauses within temporal and conditional clauses. Nevertheless, they are identical to the examples above in the mismatch between syntax and prosody.
The following cases contain namma followed directly by the relative pronoun. This namma belongs syntactically and semantically to the main clause and not to the relative clause. In ex. (15a), it follows from the broader context, given here in translation: in the preceding context, namma 'then' (bolded in the translation) is used to order different actions within the ritual; thus in the context under discussion namma introduces yet another ritual action (1b), and not a relative clause (2) gín sekel dāi take.3sg.prs '(1) Then (2) a man, who (belongs) to his family, (1) takes a silver axe (weighing) twenty shekels (and he cut[s] the grapevine).' In this case yet again, the clause boundary between the relative clause (2) and its main clause (1b) is marked by the clause connective nu. There is no resumption of the relative phrase as the verb of the main clause is transitive, thus no resumption by an enclitic pronoun is possible in Hittite. There is also once again no clause boundary marked between the first part of the main clause (1a) and the relative clause (2). The reflexive enclitic -za which is part of the main clause occurs not after the first word of the first part of the main clause (1a), but rather after the first word of its second part (1b). This contradicts the common pattern according to which clitics of the main clause are hosted by the first word of the main clause that follows the relative (or another subordinate) clause, as very clearly seen in (5) of the following example (15b). Thus, if ex. (15a) were a regular complex sentence, -za would be expected to be hosted after the first word of the main clause, i.e., in (1a) of ex. (15a), but (15a) is not arranged as a common complex sentence, so all we can say is that it deviates in clitic placement from regular complex sentences.  (2) your divine images which are of silver and gold, (3) whatever had grown old on any god's body, (4) which objects of the gods had grown old, (1b) no one had ever renewed them as we have. # (5) Furthermore, no one had established such respect in the matter of the purity of the rituals (var.: recitations) for you […].' Here we see a very neat contrast between two sentences with namma. The first sentence comprises cls. (1-4). In (1a), namma precedes the series of relative clauses (2-4) and does not host the enclitic -at of the main clause (1b), although it scopes over it. Namma in (1a) is also separated from (1b), of which it is syntactically and semantically part, by a clause boundary nu. There is no clitic in the relative clause (2), thus there is nothing to cliticize to namma in (1a); nevertheless, the available material indicates that the structure (1a-2-1b) is a mismatch sentence. The mismatch is determined in all unambiguous cases by combination of syntax, semantics, and prosody. Whereas not all its components may be present in a given case, scope and clause devision warrant the taxonomic attribution in this particular case. The sentence (1-4) is first in a series of sentences, each introduced by namma. In the second sentence of the series, constituted by cl. (5), namma is within the clause it scopes over and hosts its enclitics (-šmaš-ša[n]). This latter case proves beyond any doubt that namma can easily host enclitics in independent main clauses and that namma in (1a) scopes over (1b), which subordinates clauses (2-4).
Prosodically, just as in the cases where relative clauses were within another subordinate clause, when namma precedes the relative clause it hosts the enclitics of the relative clause. This was not seen in exx. (15a-b) because there are no enclitics in the relative clause, but is seen in: Here namma scopes over the entire sentence (including (1b), where it syntactically belongs), but it hosts the enclitic -za of the subordinate clause (2). Thus, syntactically and semantically it is part of the main clause (1b), whereas prosodically it is part of the relative clause (2) and is available as the host for its clitics. That this is not the only option is demonstrated by cases like cl. (5) of ex. (15b), where namma is inside the main clause and hosts the enclitics of the main clause. The only clause boundary that is marked in ex. (15c) is between the relative clause (2) and the main part of the matrix clause (1b).
This example shows that the clause within the main clause need not be necessarily relative. It can be temporal or, as in the following case, conditional: (2) if one of the forward men allows something in (to the procession), (3) either horses or aggressive cattle, (1b) then it is the forward man's fault; (4) if, however, one of the rear men allows something in (to the procession), (5) then it is the rear man's fault.' All the properties characteristic of mismatch sentences are available here too: no clause boundary between the first part of the main clause (1a) and the conditional clause (2), but a clear clause boundary between the conditional clause and the main part of its main clause (1b), marked by the clause connective nu and second-position enclitic -at.
The clause within the main clause in the same construction can also be one of manner: (17)  It is only to be expected that there can be several subordinate clauses within the main clause, in linear word order between the sentence-initial namma (syntactically belonging to the main clause, but prosodically part of the first subordinate clause) and the rest of the main clause, as is the case in:  (2) whatever temples there are, (3) as soon as the priest and the diviner go (there) early in the morning, (1b) they must take a look around outside the temples. (4) Further, they must sweep them out (5) (and) sprinkle them, (6) then they pull them shut.' In this context two actions are arranged temporally/logically: 'they must take a look around outside the temples' (cl. 1b) and 'Further, they must sweep them out (and) sprinkle them, then they pull them shut' (cls. 4-6). Whereas the placement of namma is straightforward in cls. (4-6), it is not at all so obvious in cls. (1-3). Here namma precedes both subordinate clauses which are dependent on main clause (1b), namely the relative (cl. 2) and temporal (cl. 3) clauses. However, it does not narrowly scope only over them; in cl. (2) it is not 'what further temples there are' , and in cl. (3) it is not 'as soon as the priest and the diviner further go there early' . So namma is not part of either of the clauses, but instead belongs to the main clause (1b) syntactically and semantically. Yet it precedes the leftmost subordinate clause (2).
Namma is not the only adverbial that can precede relative (and other subordinate) clauses in mismatch sentences. Virtually any other adverbial with analogous temporal semantics (and occupying the same structural position, see below) can be attested in the same construction:  '(2) When in Kaška there had never been a single ruler, (1a) suddenly (1b) this Piḫḫuniya ruled like a king.' The following example attests still another adverbial anda-ma 'furthermore' and simultaneously yet again shows what happens if there are multiple clauses within the main clause between the sentence-initial adverbial and the rest of the main clause and if the second subordinate clause of the two has enclitics: the enclitics of the second subordinate clause remain within this clause, but the enclitics of the first subordinate clause, coordinated with the second, are hosted by the sentence-initial adverbial, which syntactically belongs to the main clause: (2) If you select at some point a selection (of the animals), (3) and they drive them to the deities, your lords, (1b) then the cowherds and the shepherds shall go along with the selection.' Here the enclitic of the first conditional subordinate clause (2) (-ašta) is hosted by the first word of the main clause (1a), whereas the enclitic of the second conditional clause (3) (-at) is within that clause (3).
Finally, I believe that a semantically slightly divergent type also belongs here: kī this.acc.sg.n danzi take.3pl.prs '(1a) (Then) on the morning, on the second day, (2) while the sun still stands, (1b) they take these (things) from the house of the ritual patron: …' Still, it is not completely assured that the adverbial cannot semantically belong to the relative clause, see, e.g., the different interpretation of Collins (1997: 174): 'Then on the morning of the second day the Sun God has not yet risen' , even though her translation is rather liberal with respect to the Hittite text. Actually, there are contexts when adverbials are in the scope of the temporal clause: ui-zzi come-3sg.prs '(1) And when a star appears on the evening of the 2nd day, (2) the ritual patron comes into the temple (, and he bows to the deity).' But a possible argument for understanding the initial adverbial in ex. (21) as belonging to the main clause is the fact that kuitman is commonly clause-initial (HED K: 227; Hoffner & Melchert 2008: 416-417) with one exception. Referring lukkatta and ina ud 2 kam syntactically to the main clause will make kuitman first in its own clause, as is normal for it. Ex. (21) deviates from other mismatch sentences in that the first part of the main clause (1a) is not a one-word constituent, but rather two constituents. We will see later that this is important.
To conclude, it is important to note that this subtype dominates both among mismatch sentences and in cases when both namma (or another adverbial) and relative (or another subordinate) clause are simultaneously attested within a complex sentence.

4.1.2
Left dislocations preceding relative pronouns? The material that can precede the relative clause is not limited to conditional or temporal subordinators or namma and other adverbials. There are also Here uzu muḫrain 'muḫrai-bodypart' in (1a) is left-dislocated out of the main clause (1b), n-an a[p]ēdani dāi. As is regular in left dislocation structures, it is separated from the main clause by a clause boundary (marked by the clause connective nu and the second-position enclitic pronoun -an at the beginning of (1b)) and is resumed in the main clause by the anaphoric enclitic pronoun -an. Unusually, however, nothing separates the left-dislocated phrase from the relative clause (2). This is seen in the lack of clause boundary between the left-dislocated noun phrase and the relative clause; there is no marking of it either by a clause connective nu or by enclitics. Curiously, the relative phrase is also resumed by the anaphoric pronoun a[p]ēdani 'to him' in the main clause.4 -(m)a on the left-dislocated UZUmuḫrain 'muḫrai-bodypart' (1a) contrasts the event described in clauses (1-3) of the context with the previous context, given here in translation. There are two bearers of narrow contrast here, the recipient of the sacrifice and the sacrificed body part; thus -(m)a has wide scope and scopes over the whole sentence consisting of clauses (1-3). It is thus placed in the first part of the sentence, as expected. However, here we face a problem. Normally, left-dislocated nominals do not host -(m)a, as the collection in Vai (2011) clearly shows. Instead, -(m)a is hosted by the main clause. All these differences set ex. (23) apart from prototypical left dislocations. Thus the marked clause boundary at the beginning of cl. (1b) may in reality separate not the left-dislocated nominal from its main clause (1b), but rather it may separate the relative clause (2) from its main clause (1b). In any case, the enclitic pronoun -an in cl. (1b) resumes the left-dislocated nominal in cl. (1a), just as in regular left dislocations. Thus the placement of -(m)a on the left-dislocated nominal contradicts common left dislocation patterns and testifies to the fact that the structure is rather amenable to the syntax of relative clauses inside other clauses, where UZUmuḫrain 'muḫrai-bodypart' forms one prosodic domain with the relative clause (2) and thus nothing prevents placing -(m)a after it. We saw something similar in many cases above, where a subordinator that belonged semantically to the following clause was prosodically part of the relative clause. Most notably, it was revealed by the placement of second-position sentential clitics of the relative clause after the subordinator of the main clause and the clause boundary after the relative clause that effectively separated not only the relative clause from its main clause (a regular phenomenon), but also separated the subordinator from the clause it was part of! What is also highly significant is that, differently from prototypical left dislocations that never preserve the case that is assigned by the verb of the main clause and are in the nominative, this example preserves the accusative case of what appears at first sight as a left dislocation. Nevertheless, even though the material to the left of the relative clause is difficult to taxonomize, the sentence definitely falls into the mismatch type.
The structure of ex. (23) is all the more evident in the light of a regular relative clause from the same text: (24) NS (CTH 391.1.A) KUB 9.25+ rev. iv 13'-14': 1. Again, there is no clause boundary marked between the left dislocation (1a) and the relative clause (2), but there is an explicit clause boundary between the relative clause (2) and its main clause (1b), marked by nu. The left-dislocated phrase is not resumed in the main clause by anything, if one follows Hazenbos (2003: 40), whose translation I follow. This implies reading apuš as they.nom.pl.c, a regular possibility for late New Hittite (Hoffner & Melchert 2008: 143).5 The data from this section furnishes additional arguments against the analysis of relative clauses in mismatch sentences as bare indefinite noun phrases, as the contexts do not involve conditional clauses that can licence bare indefinites. However, it has to be borne in mind that the cases are extremely rare and one of them, ex. (23), is extraordinary in many respects.

4.1.3
Material preceding subordinate clauses and types of subordinate clauses The data from the preceding sections is important in that it not only convincingly shows that relative clauses in mismatch sentences cannot be assessed as bare indefinites, but also in broadening the array of words that can occur to the left of the relative clause while belonging to the clause that functions as the main clause for that relative clause. Nevertheless, based on the position of clitics, these words are prosodically part of the relative clause and are separated from the rest of the subordinate clause (= main clause for the relative clause) by a clause boundary, together with the relative clause. It was originally suggested in section 3 that these are commonly conditional subordinators. However, it follows from the previous sections that the class is not limited to subordinators, but also includes sentential adverbials like namma 'then' and even left dislocations.6 It is also important that relative clauses do not exhaust the array of subordinate clauses that can occur in mismatch sentences. Virtually any other type of 5 It is suggested by an anonymous reviewer that it is in principle possible to interpret the stressed pronoun apuš as they.acc.pl.c. In this case apuš will resume the left dislocation. However, I think this is unlikely, as left dislocations are not resumed by stressed pronouns. 6 This uncommon behavior of adverbs behaving like namma was already observed by Garrett (1994), who termed them s-adverbs; see his discussion of his ex. (35b) and see also Probert (2006: 72-73). He did not, however, realize that the class of constituents with analogous syntax was much broader.
Indo-European Linguistics 11 (2023) 1-69 subordinate clause can appear within the structure (with the possible exception of complement clauses, but these are notoriously rare in Hittite in any case).
Having laid out the data that I interpret as mismatch sentences, I would like to stress a very important property of the type, namely that it cannot really be defined by exactly what subordinate clause is inside exactly what main clause. Any subordinate clause can be in a position that is descriptively within any other type of clause that serves as the main clause for it. The only constraint is that there is a one-word constituent syntactically and semantically belonging to the main clause that stands to the left of the subordinate clause, while the rest of the main clause follows the subordinate clause. The second important property of mismatch sentences is that clitics of the subordinate clause (if available) cliticize to the part of the main clause to its left, whereas clitics of the main clause cliticize within the part of the main clause that stands to the right of the subordinate clause. Both of these characteristics determine the prosody/syntax mismatch observed in this structure. Although I will deal with the structural position occupied by clitics in such sentences in another paper, their evidence is crucial for the descriptive adequacy of mismatch sentences as a distinct taxonomic unit of Hittite syntax.

4.2
Mismatch sentences or embedded clauses? The question that I posit in this section is whether the cases treated above as a special taxonomical entity, namely relative clauses within another subordinate or main clause (mismatch sentences), can simply be subsumed into the category of relative clauses embedded in other dependent clauses or in the main clause. Such embedded relative clauses have been previously described for Hittite: Very rarely, forms of relative kui-stand for unexpressed clauses of the type 'who (is/are there)' , in which case kui-stands in the main clause itself: The most large-scale analysis of some relative clauses as embedded clauses is that of Probert (2006); see also Huggard (2015: 160 & 172-173 give.3sg.prs '(The one) who is impure gives three shekels of silver.' The relative clause here is paprezzi kuiš 'who is impure' , and the main clause is 3 gín kù.babbar pāi 'gives three shekels of silver' . Here there is no marker of the clause boundary between the relative clause and the main clause, nor resumption of the relative noun phrase in the main clause; the relative clause functions as a phrase within the main clause, directly serving the function of subject in this particular case.
My dataset supplements and expands Probert's data from NH/NS texts. In what follows, I provide examples.

4.2.1
Embedded relative clauses Outside the Old Hittite examples collected by Probert (2006), the clearest type is the one where there is no explicit clause boundary between the finite verb of the relative clause and that of the main clause, marked either by clause connectives like nu or clause-second enclitics: In ex. (27a), the relative pronoun does not attest the case which is required by the clause it is embedded in, although this is completely expected in an embedded clause. This is particularly clear in the light of a relative clause from the same text which is not embedded, but is part of the regular correlative structure: This relative clause shows a clear clause boundary marked by the clause connective nu between it and the main clause that follows it.
The following case is identical to the embedded relative clause in ex. (27a): Ex. (27c) is a clear case of an embedded relative clause, as the relative clause kuiē [š] peran ueḫanda stands between two parts of the main clause, preceded by lú ur.gi7-aš lú-aš and followed by 1 ninda.sig miyanit eme paršiya. It is significant that there is no clause boundary between the two finite verbs, the one in the relative clause and the one in the main clause. The sign that it is a separate clause is that the case of the relative pronoun is not the case which is required by the verb of the main clause paršiya 'break' , but rather the case required by the verb of the relative clause ueḫanda 'turn' . This case marking is expected of an embedded clause, but not of a phrase.
The following cases also unambiguously attest an embedded relative clause: The clause mun-an kuedani ul išḫuwān is likely to be a finite clause with passive verb form and zero copula. This follows from the word order, which is typical of finite clauses and not of phrases with participles. The participle išḫuwān agrees with the subject of the clause in number and gender. As is the norm for Indo-European Linguistics 11 (2023) 1-69 the analytical passive in Hittite, it is marked for nominative, which is identical with the accusative case in neuter gender.
The last example (27e) is so clumsy, however, that a pure performance error suggests itself; in other words, the structure is very likely to be not an embedded relative clause, but rather a false start structure. This is also independently favored by the fact that mḪūtupianzan-ma dumu.lugal dumu mZidā gal lú mešedi 'Hutupianza, a prince, son of Zida chief of the bodyguard' is actually repeated within the main clause (2) and not referred to by a resumptive anaphoric pronoun, as left dislocations usually are. It is also noteworthy that unlike prototypical left dislocations, which never preserve the case that is assigned by the verb of the main clause and are in the nominative, this example preserves the accusative case. This makes this example very different from prototypical left dislocations, despite its qualification as such by Huggard (2015: 172). However, this does not concern the relative clause, which might indeed be embedded within the false start mḪūtupianzan-ma dumu.lugal dumu mZidā gal lú mešedi 'Hutupianza, a prince, son of Zida chief of the bodyguard' , although the resulting structure is very uncommon.
The following cases are similar, but here the embedded relative clause follows the main clause and is not within it. Once again, there is no marker of clause boundary between the main and the embedded clauses: These cases need not be embedded, as it is known that in a minority of cases relative clauses which are regular adjoined clauses do follow the main clause. However, the fact that there are no explicitly marked clause boundaries argues in favor of the embedded type.
Summing up the evidence of the embedded clauses presented so far, it should be observed that none of the embedded clauses has any enclitics of their own.
Embedded non-relative clauses are also attested, if only very rarely. They are close to and obviously derived from relative clauses by sporadic analogy. It is also possible to assess them as bona fide relative clauses introduced by a locatival correlative pronoun. This is the first case where the embedded clause has enclitics of its own. Unfortunately, the example is not very revealing, as the embedded relative clause is first in a complex sentence.
The clauses that I have assessed in this section provide an exact parallel to the Old Hittite data mentioned above and described by Probert (2006). Curiously, Probert (2006: 71-78) originally suggested that such clauses fall out of use in Middle Hittite. However, later in Probert (2014) she retracted the idea and suggested that embedded relative clauses might be attested in post-OH times as well, referring to work by Huggard which was still unpublished at the time. Huggard (2015: 172-173) provides two examples (also occurring in my corpus; see above, (27d) and (27e)) from MH and NH texts.

4.2.2
Are relative clauses within other clauses simply embedded? After reviewing the evidence for embedded relative clauses, I now return to the question to what extent the material assembled above in sections 1, 3, and 4 as (in purely descriptive terms) clauses within other clauses can also be classified as embedded relative clauses.
Indo-European Linguistics 11 (2023) 1-69 I argue that they cannot. I will start from the descriptive linear evidence. The first difference is that in the majority of cases there is a clear demarcation of clause boundary between the relative clause and the main clause in the case of relative clauses in mismatch sentences, but the part of the main clause which precedes the relative clause is never set off by clause boundaries. In contrast, no marking of clause boundary is ever attested either before or after embedded relative clauses. Another clear difference between relative clauses in mismatch sentences and embedded relative clauses is that the former are resumed in the main clause by an anaphoric pronoun (as regular relative clauses in Hittite are), whereas the latter are never resumed. The third difference is that relative clauses in mismatch sentences are highly restricted in their distribution: they occur only after one constituent, which semantically and syntactically belongs to the clause that otherwise follows the relative clause (either a main clause or a subordinate, mostly conditional clause, but always the main clause to the relative clause). These constituents include the conditional and temporal subordinators, a dislocated noun phrase, or namma 'then' (and other temporal adverbials). Embedded relative clauses do not show such a restricted distribution; their distribution is identical to that of noun phrases with the same syntactic function, whether subject, object, or adjunct. Finally, in mismatch sentences the clitics of the main clause occur within the part of the main clause that follows the relative clause, whereas the clitics of the relative clause are hosted by the portion of the main clause which precedes the relative clause. Consequently, this first portion of the main clause and the relative clause form a single prosodic domain. The same pattern is never attested in embedded relative clauses: we simply lack data on how clitics of the relative clause behave vis-à-vis the main clause, but clitics of the main clause are not sensitive to the embedded relative clause and are hosted by the first word of the main clause, although the only case attested in the corpus involves -(m)a.
Thus, even from a descriptive viewpoint the mismatch sentences examined above in sections 1, 3, and 4 cannot be subsumed under embedded clauses and must therefore represent a different taxonomical unit.7 Instead, we have in sections 1, 3, and 4 a very curious structure which is obviously similar to prototypical embedded relative clauses, but attests 1) a very clear demarcation of clause boundaries between a subordinate (relative) clause and another Indo-European Linguistics 11 (2023) 1-69 subordinate (mostly conditional) clause or the main clause, and 2) resumption of the relative phrase in the relative clause by a pronoun in the main clause.
Consequently, it is obvious that the peculiarity of the construction under discussion in the present study does not lie in the syntactic structure of the relative clause, which is a regular relative clause. The uncommon property of this structure lies exclusively in the fact that there is some material that semantically and syntactically pertains to the clause that follows the relative clause and which serves as the main clause for it, but which is placed not within that clause, but rather in front of the relative clause. This material is not just in front of the relative clause, it is also prosodically part of the relative clause: it can host enclitics of the relative clause and is not separated by a clause boundary from the relative clause.
Schematically, the differences between mismatch sentences and embedded relative clauses can be represented as follows: a. (relative) clauses in mismatch sentences (as seen in exx. 1, 6, 8-9, 12-20, 23, 25 above): subordinator of clause 1/adverb/left dislocation-clause 2 # (clause boundary)-clause 1 (with resumption of the relative phrase); b. embedded relative clauses (as seen in exx. 26-28 above): clause 1 (no clause boundary)-clause 2 (no clause boundary)-clause 1 (no resumption of the relative phrase); This type allows some variation: clause 2 (no clause boundary)-clause 1 (no resumption of the relative phrase); clause 1 (no clause boundary)-clause 2 (no clause boundary). Until now, comparisons of relative clauses in mismatch sentences and embedded relative clauses have suffered from terminological confusion, as the labels "relative clauses within other clauses" or "relative clauses in mismatch sentences" or "mismatch sentences" were defined in terms of purely linear surface syntax, whereas "embedded relative clause" is a structurally defined term. It has been argued above on the basis of purely surface structures that mismatch sentences are in fact distinct from embedded relative clauses. The following section will proceed to define the two classes structurally.

Structural properties of mismatch sentences
Embedded relative clauses are CPs adjoining a zero NP which is in the position and case required by the verb of the main clause; i.e., they adjoin to TP in the main clause if they are subjects and are in other positions which correspond to their syntactic role if they are not subjects, e.g., complement of VP if they are objects, etc. They thus simply adjoin to the syntactic position reserved for their syntactic function (subject, direct object, indirect object, adjunct) and are not in this respect different from any other NP.
In comparison with embedded relative clauses, mismatch sentences are structurally more heterogenous.

5.1
Three structural types of mismatch sentences The most common subtype of mismatch sentences-subordinate (mostly relative) clauses following namma and other temporal adverbials in the main clause-are CPs adjoining to the CP of the main clause. They are thus simply regular relative clauses for Hittite (see Huggard 2015, Lyutikova & Sideltsev 2021. Their particularity lies in the syntactic context in which they occur: whereas relative clauses are usually the highest adjunct and so linearly precede all the material of their main clause, in this case there is a yet another higher adjunct, namely namma or a temporal adverbial. Thus adjoining relative clauses are in surface syntax within the main clause, following this higher adjunct and preceding the rest of the main clause: (29) a. CP namma CP CP rel./cond. clause CP Their syntactic position makes them distinct from embedded relative clauses, whose position is identical to that of any other case-marked NP. Relative clauses in mismatch sentences are distinguished from regular relative clauses only by the fact that there is a still higher adjunct to their left.
To take an example of the type, sentence (15a) above would structurally look as follows, with [] marking relevant syntactic constituents and () prosodic domains: Indo-European Linguistics 11 (2023)  Another type of mismatch sentence, however, cannot be structurally analyzed along the same lines. When relative clauses follow the conditional or temporal subordinator, they cannot be construed as adjoining to CP at the same height as the previous type, as they would then precede in linear word order the subordinator in C, which is not the case. The relative clauses following the clause-initial subordinator require a complex reanalysis of the original structure where the relative clause is in Spec,FinP;8 for a sketch, see Lyutikova & Sideltsev (2021). Here I will just reproduce the outline of their analysis: because the relative clause is located low enough in the structure of the conditional clause and due to the fact that conditional clauses contain an overt clause-initial subordinator C in , whereas the relative clause contains overt constituents in the position of C prev , there is an opportunity to reanalyze the adjoining relative clause as the complement of the initial subordinator, as represented schematically in To take (9) as an example, Indo-European Linguistics 11 (2023)   If this happens, the rest of the conditional clause may be seen as a separate fullscale clause (CP) (Lyutikova & Sideltsev 2021). The result is the type in which relative clauses are linearly preceded by the initial subordinator and followed by a clause boundary.
There is yet another structural type of mismatch sentence, which involves elements before relative clauses that are supraCP. Of the material presented above, left dislocations belong here, as they are obviously above CP in Hittite; see Sideltsev (2021) (23), is so difficult of interpretation that this part of the analysis must remain very provisional; it will be dealt with at greater length in my paper on clitic climbing (Sideltsev 2023).
The most important consequence of this structural analysis is that it is impossible to characterize mismatch sentences as a single class in structural terms in view of their heterogeneity, so one might have to continue using purely descriptive terms such as "relative clauses within other clauses." Yet as all the structures share a number of surface properties-the most conspicuous being the mismatch between prosodic domain and syntax/semantics with respect to clitic and clause boundary placement-I believe a common label is justified.
It was suggested by an anonymous reviewer that some of the examples I taxonomize as relative clauses in mismatch sentences are actually parenthetical clauses. To see if this is really so, I will first look at some prototypical examples of parenthetical clauses. There are two common types of parentheticals attested in Hittite texts: the type inserted between clauses (ex. 30) and the type inserted inside a clause in lists (ex. 31): (30) OH/NS (CTH 321.A) rev. iv 11'-13' , cf. E. Rieken et al. (ed.), hethiter.net/: CTH 321 (TX 2012-06-08, TRde 2012-06-08); differently Hoffner (1998: 14) regularizing the Hittite text: eš-āri sit-3sg.prs.med '(1) Whichever GUDU-priest holds (the image of) Zaliyanu-(2) a basalt throne is set above the spring/basin-(3) he will be seated there.' Here the parenthetical clause (2) is inserted within the correlative structure (1, 3), but it is inserted between the two sentences, not within one of them. 1-nutim 1 giš tarmalla tarmalla '(1a) 2 offering-tables, 2 tables, 2 stands of wood, 1 throne 6 sekan in height-(2) they are double-pazzanant; (1b) 1 throne for sitting, 1 footstool and 1 set of tarmalla.' Here the parenthetical clause (2) is inserted within the enumeration of ritual paraphernalia. Differently from the mismatch sentences that I have treated above, the clause boundary is marked to the left of the parenthetical clause. I keep such lists separate from other contexts, as it is shown by Sideltsev (2010) that their syntax is very peculiar and that they do not easily fit the definition of a clause in Hittite.
As both types of parenthetical clauses are demonstrably different from the structure that I am dealing with in this paper, there is no danger of confusion between them. It is the much rarer parenthetical clauses inserted in another clause, not in a list, that may appear to be identical to mismatch sentences. In the remainder of this section, I will list the examples and discuss them vis-à-vis mismatch sentences.
The first example is: '(1a) The waters of the mountains, the gardens, the meadow(s)-(2) let your refreshment go-(1b) but let it (i.e., ḫaḫḫima-) not paralyze them (i.e., waters of the mountains, the gardens, the meadow(s))' (essentially following E. Rieken et al. (ed.), hethiter.net/: CTH 323.1 (TX 2009-08-26, TRde 2009).9 The analysis of (1a) ḫur.sag meš -aš widār giš kiri6 ḫi . a wēl [l]u 'the waters of the mountains, the gardens, the meadow(s)' as a left dislocation (out of clause 1b) and not as a vocative is independently strengthened by the fact that no direct address is available in the context, whereas the nominals are resumed in the main clause, a typical property of left dislocations never attested with vocatives. The possessive second person pronoun tuel in clause (2) is not coreferential with the dislocated nouns in cl. (1a). Clause (2) is thus parenthetical within clause (1). This parenthetical clause is very different from the relative clauses in mismatch sentences assessed in the paper, however: its clause boundaries are marked by nu on both sides of the clause. The same applies to the following example, although here only the initial clause boundary is marked: (32) b. OH/NS (CTH 429.1) KBo 10.37 obv. ii 7'-9' , cf. Christiansen (2006: 192-193 Similarly 'Die Gewässer der Berge, die Gärten, die Weiden! Deine Gnade soll (weiter-) gehen, damit sie nicht (länger) erstarren!' (Haas 2006: 118). Often cl.
(2) is understood as the main clause to (1a). This implies introducing a pronoun into cl.
(2) which is not there in Hittite: 'The waters of the mountains, the gardens, the meadow(s)-let your refreshing go (through) the lands-but let it (i.e., ḫaḫḫima-) not paralyze them' (Hoffner 1998: 27); 'L'eau des montagnes, les jardins, la prairie, que ton haleine les traverse et que (le Gel) ne les immobilise pas' (Mazoyer 2003: 167 & 178); 'La tua essenza divina spiri sulle acque delle montagne, sui frutteti e sui prati, cosicché il gelo non possa paralizzarli!' (Pecchioli Daddi & Polvani 1990: 64). Cf. very differently HEG (T 375). 10 Restored following CHD (Š 18). Cf. Christiansen (2006: 192-193 '(1a) All you big and small mountains, (1b) why did I come in the inaccessible valleys? (2) Why did I exert myself? (3) They threw a human child behind the corral like a bull. (4) And also you, mountains, join me!' Ex. (32f) makes it likely that in ex. (32e) it is clause (2a-b) that is parenthetical. This conclusion is also supported by the fact that there is a nu between the vocative 'clause' (1a) and (2a) in ex. (32e)-a very unusual property for sentences with vocatives (see Sideltsev 2021)-making it likely that cl. (2a-b) is parenthetical. However, this clause is itself split by the coordinated clauses (1b) and (3), which are thus good candidates for being parenthetical themselves. If understood thus, this case is similar to mismatch sentences in that it marks the clause boundary after the parenthetical clause (in 2b), but it is different from mismatch sentences in that the clitics of the main clause are to the left of the parenthetical clause (in 2a). However, the context in ex. (32e) is so complex that a definitive conclusion is hard to reach on its basis.  c '(1a) This little puppy-(2) big as to (its) limb, (3) big as to its heart-(1b) then it is the carrier (of evil) (like) an ass.' This case is similar to mismatch sentences in that the enclitics of the main clause (1b) come after the parenthetical clauses (2-3), but it differs once again from mismatch sentences in that the NP (kāš ur.tur 'this puppy') in the main clause to the left of the parenthetical clause (in 1a) is resumed by a coreferent enclitic pronoun (-aš 'it') in the main clause (1b)-something never attested with mismatch sentences. The resumption can occur with left dislocations, but the semantics of (1a) here is not that of a typical left dislocation. Besides, left dislocations are never preceded by a marked clause boundary, as is the case here. Thus this case is more likely to be a real false start.11 Indo-European Linguistics 11 (2023) 1-69 '(1) If it (i.e., the law case) is a matter of blood(shed), though, (2a) (and) a man (3)-either his legal opponent (4) or his avenger-(2b) that (man) impedes the king, …' Here the clause (2a) is not simply continued by (2b), as there is a clause boundary marked after the parenthetical clauses (3) and (4) by a sentential enclitic -kan (in 2b). This is seemingly identical to mismatch sentences, but the NP antuwaḫḫaš 'a man' from the first part of the clause (2a) is resumed by the demonstrative pronoun apāš 'that' in the 'continuation' of the main clause (2b) after the parenthetical clauses (3) and (4). This again sets the case apart from mismatch sentences. (2a) would appear to be left-dislocated, but left dislocations are not resumed by stressed pronouns. Thus this case is more like a false start. The parenthetical clauses (3-4) immediately follow the noun they refer to and are a typical 'either … or …' parenthetical, otherwise very commonly attested between clauses. We observe yet another case of 'either … or …' parenthetical in the following example: (32) j. NH/NS (CTH 68.C) KBo 5.13 obv. ii 30-32, cf. Friedrich (1926: 124-125); Beckman (1996:  iya-zi do-3sg.prs '(1a) But if domestically someone (1b) carries out a revolt against My Majesty, whether a nobleman, a unit of the infantry or chariotry (2) or whatever sort of person it might be, (if I, My Majesty, put things right, then I will capture that person or that unit of the infantry or chariotry).' In (32j), the parenthetical clause is (2) našma-aš kuiš imma kuiš antuḫḫaš 'or whatever sort of person it might be' . It is inserted quite low down in the main clause, following the nominal that it modifies. There is no clause boundary marked either before or after the clause, but it has its own enclitic (-aš Summing up this section, I conclude that parenthetical clauses are demonstrably different from mismatch sentences with respect to the marking of clause boundaries and the position of enclitics. There is exactly one context which by both of these parameters is identical to mismatch sentences. I put this sole context into the same category as mismatch sentences and consequently set it apart from the rest of the parenthetical clauses. The results of this analysis once again support the claim that mismatch sentences are a separate taxonomic class, not amenable to any other type of complex sentences in Hittite. Another important contribution of this section is that it shows that clause boundaries can be marked not only after the clause within another clause, which we already saw above in sections 2 and 3, but also before the clause within another clause, as is the case with several parenthetical clauses. This is important new information. Typically, clause connectives in Hittite simultaneously mark the beginning of one clause and the end of another. However, parenthetical clauses show that only the beginning of a clause can be marked by a clause connective, even if the previous clause is not yet over. Quite importantly, there is also a conspicuous lack of contexts testifying to nu marking only the end of a clause.14 This supplements the material from mismatch sentences and clearly shows that in mismatch sentences the first part of the main clause and the subordinate clause do form a single prosodic domain (differently from parenthetical clauses), whereas the second part of the main clause is prosodically a domain of its own. This section has also shown that parenthetical clauses can contain their own enclitics within them in the part of the main clause that stands to the left of the parenthetical clause. Moreover, if there is a parenthetical clause within a main clause, the clitics of the main clause are hosted by the first part of the main clause (although they may be doubled by the second part of the main clause). Both of these properties show that the behavior of Wackernagel clitics in mismatch sentences is nontrivial.
Finally, it is significant that the part of the main clause to the left of the parenthetical clause is longer than the part of the main clause to the left of the relative clause in mismatch sentences. In the case of parentheticals, the first part of the main clause always consists of several constituents, whereas in the case of mismatch sentences there is always just a one-word constituent in the first part of the main clause. The only case of a mismatch sentence from my corpus where there are two constituents in the part of the main clause to the left of the relative clause is: 14 In exx. (23 & 25) above nu in front of the second part of the main clause may follow the left dislocation (which it optionally does elsewhere). It must be admitted that, due to the absence of nu between (1a) and (2) or clitics belonging to (2), it is also thinkable to refer abu-ya 'my father' in (1a) to the relative clause (2). However, this would put kui-'which' in the relative clause (2) into third position rather than second, which would be unexpected; thus I suppose that abu-ya should belong to the first part of the main clause (1a). However, since this case lacks enclitics in the relative clause, we simply do not know what would happen to enclitics in such cases. We will return to this question in the following section.

7
Other types of mismatch sentences The material I have analyzed above does not exhaust the available types of mismatch sentences. Besides the complex sentence I outlined above as: (34) b. mān relative clause mān conditional clause (= main clause for the relative clause).

7.1
The sequence relative clause-main clause with doubling of 'if' There are several subtypes of the doubling pattern, the most common of which involves repetition of the conditional subordinator mān 'if' .
The doubling of mān has been observed before. It was described in CHD (L-N 144 & 157 sub 7j) 'as temporal and relative clauses inserted in conditional clauses, with repetition of mān "if"': ištamaš-teni listen-2pl.prs '(1a) And if for me you two, o 2 Sarrumanni-s and 1 Allanzunni, (2) you who from the womb of the god are sprung, (1b) if for me you listen to this matter.' The repetition of the subordinator is observed by the editor of the text (de Roos 2007: 100137): "With nu-mu mān the mān of line 28 is taken up once more after the relative subordinate clause." Unlike the examples above, where the enclitics on the subordinator were of the relative clause, the enclitic in (1a) cannot belong to the relative clause.
This example is particularly interesting as, apart from the repetition of the conditional subordinator, the address to the gods (marked by the nominative case) 2 dlugal-manniš 1 dAllanzunni-šš-a '2 Sarrumanni's and 1 Allanzunni' is separated from the rest of the conditional clause (1b) by the relative clause (2). Such addresses are not supraCP (see Hoffner & Melchert 2008;Sideltsev 2021), but an appositive phrase to the zero 2pl pronoun in (1b). These appositive phrases have free distribution in a clause, but it is important that they are intraCP, differently from morphologically marked vocatives, which are supraCP. Thus ex. (35b) is a rare mismatch sentence with some material other than the conditional or temporal subordinator in the part of the sentence to the left of the relative clause (there is just one other secure case, for which see ex. (33) above). This indirectly supports the structural analysis of this type of mismatch sentence put forward in Lyutikova & Sideltsev (2021), which implies that there is space for other constituents between the subordinator and the relative clause and that its absence in the majority of cases is accidental.
But what is most significant about this example is that it complements the analysis of ex. (33), which also attested two constituents in the first part of the main clause, but no enclitics of the relative clause. Ex. (35b) not only also displays two constituents in the first part of the main clause, it finally shows what happens to enclitics of the subordinate clause in such cases. In contrast to mismatch sentences with just one constituent to the left of the subordinate clause, and similarly to the parenthetical clauses in the previous sections with more than two constituents preceding the parenthetical clause, the enclitics remain within the subordinate clause and do not surface in the first part of the main Indo-European Linguistics 11 (2023) 1-69 clause. I will discuss how to capture this strange distribution in a separate paper on clitic climbing (Sideltsev 2023 (2) while My Majesty cone back from a fieldtrip to Assur, (1b) those border lands will not defect (from us, and the situation will not become threatening for Nerik).' This example is instructive as to the material that can be repeated. Whereas mān 'if' (spelt be-an in the example) is repeated together with -(m)a in (1a), the other clitic in (1b), locatival -kan, is not repeated in (1a). The explanation is simple: only material that scopes over an entire complex sentence can be repeated at its left edge, whereas material that scopes only over the conditional clause is never repeated at the beginning of a complex sentence.
(36) b. NH/NS (CTH 582) KUB 18.36 13 '-16' , cf. van den Hout (1998: 113-114 The double use of irrealis man in ex. (41) was already observed by Hoffner (2009: 20233): The ma-an is merely anticipating the ma-an in line 34 and need not be translated in the first clause. Literally: "If only-how it is there-if only you would keep writing …" Interrupted thoughts in the wording are usually a sign that the letter was not drafted by a scribe from preliminary notes, but was either dictated directly or-if the writer is himself a scribe-composed as it was being written.
Differently from Hoffner, I rather consider the structure planned, for which see the arguments above. The locative enclitic -kan in cl. (1a) is syntactically the enclitic of cl. (2).

7.3
Double vs single position of subordinator: diachrony There are thus two distinct structures. In the first, some constituents semantically and syntactically belonging to the conditional clause (= main clause for the relative clause) occur within the relative clause: (41) a. mān namma LD -(m)a -wa(r) relative clause conditional clause (= main clause for the relative clause)/ main clause.
Indo-European Linguistics 11 (2023) 1-69 The second construction involves doubling of the conditional subordinator at the left edge of the whole sentence and in the conditional clause (= main clause for the relative clause). The same pattern is attested for the irrealis marker: (41) b. mān relative clause mān conditional clause (= main clause for the relative clause).
Can we order the two alternatives diachronically? Possible evidence that the type with double mān before the subordinate and main clause arose earlier than single mān prosodically within the relative clause is that the former shows more variety: it is attested much more frequently with various kinds of subordinate clauses, not only relative, but also temporal and manner clauses. However, a more convincing chronological argument is lacking, as both types are attested in MH/MS texts as well as in later NH/NS texts (no attestations in OH/OS texts are known to me).
Thus the diachrony does not give us a clear answer as to which strategy was the original one-the one with spell-out of only of the higher copy or that with spell-out of both copies. Both are attested at the same time, starting from the Middle Hittite period.
In view of the structural analysis of mismatch sentences with conditional subordinators above, I follow Lyutikova & Sideltsev (2021) in taking both types (41a) and (41b) to have come about as (different) results of the reanalysis of the original structure where the relative clause is in Spec,FinP. Because the relative clause is located low enough in the structure of the conditional clause and conditional clauses contain an overt clause-initial subordinator C in , whereas the relative clause contains overt constituents in the position of C prev , there is an opportunity to reanalyze the adjoining relative clause as the complement of the initial subordinator, as represented schematically in (42)  If this happens, the rest of the conditional clause may be treated as a separate full-scale clause (CP) (Lyutikova & Sideltsev 2021). This produces type (42a) without repetition of the subordinator. But if ZP is analyzed as an incomplete clause (FinP), then it is identical to the structure of two coordinated FinPs, dominated by ForceP; subordinator doubling in this case results from restoring the head in the second conjunct (Lyutikova & Sideltsev 2021). Thus we get (41b).
Indo-European Linguistics 11 (2023) 1-69 It is also significant that clausal scope conjunctions (-(m)a and -( y)a) are always doubled together with the subordinator save one case where -( y)a is present only on the higher copy of the subordinator (ex. 35b). It is never only on the lower copy of the subordinator. All the cases are attested in texts where -(m)a, -( y)a cliticize to mān 'if' . As argued in Lyutikova & Sideltsev (2021), the most common variant (with doubling of -(m)a, -( y)a) results from (a) -(m)a, -( y)a in Disc lowering to the nearest head, the initial subordinator of the main clause, and (b) restoration of the material in the second adjunct: [

Conclusion
It has been demonstrated that Hittite attests a taxonomically separate subtype of complex sentences, called here "mismatch sentences", in which (mostly) relative clauses occur within other subordinate (conditional, more rarely temporal) clauses or main clauses. The relative clause may be preceded by a very limited array of constituents from the matrix clause: mān namma LD -(m)a -wa(r) relative clause conditional clause (= main clause for the relative clause)/ main clause.
There are also sentences of the same type which attest the conditional subordinator and the irrealis particle in two positions simultaneously, at the left edge of the sentence and in the conditional clause (= main clause for the relative clause). It was shown than the type is not reducible to other constructions standardly recognized for Hittite such as bare indefinites, embedded relative clauses, or parenthetical clauses. Structurally, mismatch sentences are heterogenous and of three types. The subordinate (mostly relative) clauses in these structures are themselves common Hittite adjoining relatives, although they adjoin at different levels (rele-Indo-European Linguistics 11 (2023) 1-69 vant for the paper are Spec,FinP and Spec,ForceP). The three types differ by the material occurring before the relative clause as well as structurally: a. namma (and other temporal adverbials) to the left of the relative clause; relative clauses following namma adjoin to the CP of the main clause, namma adjoins to CP higher than the relative clause; b. conditional or temporal subordinator to the left of the relative clause; this type came about due to a complex reanalysis of the original structure with the relative clause in Spec  (Lyutikova & Sideltsev 2021); c. left dislocations: the direct speech particle -wa(r) and discourse markers -(m)a, -( y)a to the left of relative clauses; these elements are supraCP, relative clauses following them adjoin to the CP of the main clause. Since all three structures share a number of surface properties, I believe that a common label is justified. The most conspicuous property is an intriguing mismatch between prosodic domains and syntax/semantics: the constituent to the left of the subordinate clause that is part of the main clause from a semantic and syntactic perspective is prosodically part of the subordinate clause. Thus syntactically, there are two CPs, whereas for clitics and for nu there are two prosodic domains, labelled below (a)  The clause boundary (#) is here marked not between the two CP s, but after one of the CPs. In contrast, the syntactic boundary between the two CP s is in another place, before (a1) and after (a2) Prosodically, there is a clause boundary delimiting the end of the subordinate clause within the main clause; thus the clause boundary is syntactically within the main clause as well.
It is important that all of these subyptes of mismatch sentences are attested only if there is a one-word constituent in the main clause to the left of the subordinate clause. If there are two constituents in the first part of the main clause, the enclitics of the subordinate clause remain within the subordinate clause and do not surface in the first part of the main clause. Clitics of the second subordinate clause also always remain within the clause.
The new evidence allows us to obtain a fuller understanding of the Hittite left periphery than was previously possible, including the structure of the left periphery in CP and supraCP. It thus offers an important window into the