Apparent wh-in-situ in Bellunese: Microparametric variation in the locus of subject clitic inversion

The paper investigates wh-in-situ in Lamonat and Sovramontino, two understudied varieties of the Bellunese super-dialect area (Munaro 1998; 1999; Munaro et al. 2001; Poletto & Pollock 2004; 2009; 2015), from a new perspective, whereby patterns of non-canonical wh-constituent order are dictated primarily by discourse-pragmatic needs and only secondarily by morpho-syntactic constraints. I propose that the canonical vs. non-canonical position of the wh-element is determined by three factors: (i) the discourse-pragmatics of the wh-item, (ii) its morpho-syntactic status as clitic or tonic, and (iii) the height of verb movement that targets the split C-domain (Rizzi 1997) and assures the well-formedness of root interrogatives across North-Eastern Italian Dialects (abbreviated NEIDs). By comparing Bellunese with Friulian (a neighbouring NEID) with respect to the formation of wh-questions, I argue that the special in-situ position of whelements in Bellunese is only apparent: it is the result of a deeper micro-parametric variation in the left-peripheral projection targeted by T-to-C movement in questions across NEIDs, which can be either ForceP or FinP. In Bellunese root interrogatives, the left-peripheral head targeted by T-to-C movement, Force°, is higher than the left-peripheral position occupied by the wh-element, SpecFocP, generating apparent wh-in-situ and the rigid constituent order: verb, wh-item, subject. Finally, I put forward the hypothesis that, in root interrogatives, such micro-parametric variation is a V2 reflex; in fact, Wolfe (2016) argues that, across medieval NIDs, V2 could be satisfied either in FinP or ForceP.


Introduction
The paper analyses the syntactic behaviour of wh-elements in Lamonat and Sovramontino. Lamonat and Sovramontino are two closely related North-Eastern Italian Dialects (henceforth abbreviated as NEIDs) spoken in the south-western part of the province of Belluno (Italy). They are hence part of the super-dialect area of Bellunese. Bellunese has caught the attention of several linguists due to a syntactic phenomenon that is found only in a handful of Northern Italian Dialects (abbreviated NIDs): wh-in-situ (Benincà 1994;Munaro 1998;Munaro et al. 2001;Poletto & Pollock 2004;Manzini & Savoia 2005;. Wh-in-situ is also documented in northern Venetan varieties and in the eastern and alpine varieties of Lombard (Munaro & Poletto 2002;Manzini & Savoia 2005;Bonan 2017;2019;Donzelli & Pescarini 2019). The investigation of wh-in-situ across NIDs has important cross-linguistics implications: (i) it deepens our understanding of the phenomenon, and (ii) points towards a typology of wh-items, whose discourse-pragmatic and syntactic properties determine their structural position within the clause. This paper wants to contribute towards the on-going debate on the nature of wh-in-situ across NIDs by providing novel data from two understudied Bellunese varieties. My investigation will lend support to Bonan's (2017;2019) claim against a one-fits-all analysis of wh-in-situ across NIDs: the different empirical evidence across NIDs with respect to the properties of wh-in-situ are in fact due to different underlying syntactic derivations of the phenomenon.
Lamonat and Sovramontino exhibit both canonical and (apparently) in-situ wh-words, as shown in (1), (2) and (3): (1) Sovramontino E-lo ndà aonde Toni? be.3sg-3sg.m.scl go.ptcp where Toni 'Where did Toni go?' (2) Ond-e-lo ndà Toni? where-be.3sg-3sg.m.scl go.ptcp Toni 'Where did Toni go?' (3) Co che machina e-lo ndà Toni? with which car be.3sg-3sg.m.scl go.ptcp Toni? 'With which car did Toni go?' (1) is an (apparently) in-situ wh-question, in which the wh-element is realised postverbally, whereas (2) and (3) resemble two "canonical" wh-questions. I will now briefly summarise my main claims with respect to the distribution of wh-items in Lamonat and Sovramontino. Type (3) is an instance of discourse-linked wh-question (in the sense of Pesetsky 1987;, whereby the answer to the question is drawn from a set of entities that were previously introduced in discourse. Note that throughout the paper the label d-linked will be used in this sense and not as a mere descriptive label to refer to lexically restricted wh-items. Along the lines of De Cia (2018), I will propose that, in Lamonat and Sovramontino, the answer to a d-linked wh-question is discourse-prominent (i.e. it expresses contrastiveness), and the d-linked wh-element thus occupies a discoursepragmatically salient high left-peripheral position. I will also argue that another type of discourse-pragmatically salient wh-XP can appear in (seemingly canonical) preverbal position, namely wh-items encoding a mirative interpretation. Mirativity is a discoursepragmatic strategy used by speakers when they want to convey surprise, disappointment, incredulity or anger (see Cruschina 2012). As for type (1) and type (2) wh-questions, they are instead equivalent in meaning. The key difference is that the wh-item in (1) is tonic and carries focus prominence, while the wh-element in (2) is atonic and hence unable to carry focus prominence.
In light of the basic paradigm (1)-(3), I will argue that the different distribution of wh-elements can be accounted for by considering the interaction between syntax and discourse-pragmatics. More specifically, my analysis pivots around two basic tenets: (i) wh-elements can serve different discourse-pragmatic functions, and (ii) wh-elements do not form a homogeneous class, but exhibit morpho-syntactic differences. In this respect, XPs bearing different discourse-pragmatic readings (i.e. d-linked, mirative, focal) tend to appear in different syntactic-pragmatic configurations, and XPs exhibiting different morpho-syntactic statuses (i.e. clitic vs. tonic) tend to occupy different structural positions (i.e. X° vs. SpecXP). My analysis and discussion of wh-in-situ in Bellunese hence departs from the existing accounts of the phenomenon (Munaro 1998;Poletto & Pollock 2001;2009;Munaro & Pollock 2005;Manzini & Savoia 2005;Bonan 2019), which mainly seek an explanation on purely syntactic grounds.
Eastern-Central Friulian (see Benincà & Vanelli 2016). Friulian has received relatively more scholarly attention than Lamonat and Sovramontino; nevertheless, no comprehensive research has been carried out on its syntactic mechanism of content question formation.
In light of the sociolinguistic situation that characterises Friulian, Lamonat and Sovramontino, data collection focused on speakers ranging between 60 and 84 years of age. The bulk of consultants were between 60 and 70 years old. In order to maximise the naturalness of the elicited discourse-pragmatic data, interviews were carried out in small groups (two groups per speech community) of three or four speakers. Approximately 27 hours of interviews were recorded (12 hours Lamonat, 9 hours Sovramontino, 6 hours Friulian). All consultants were bilingual in their dialect and in Italian, the influence of the dominant language over the speakers' native language was a variable that I hence had to control for in the interview setting. This was done through the use of visual stimuli or the simulation of semi-natural conversations where participants could freely interact among themselves (semi-staged acts of communications). I also relied on questionnaire-based elicitation and, most importantly, natural occurring speech (see Himmelmann 1998;. Data manipulation and subsequent acceptability judgements was a particularly effective tool of investigation (Chelliah & de Reuse 2011). Note that all patterns found through elicitation were also attested in natural occurring speech. Unfortunately, no corpus exists of these (exclusively) spoken varieties: this greatly limits the possibility to conduct quantitative research.

Wh-in-situ in Lamonat and Sovramontino: A root phenomenon
Wh-in-situ in Bellunese is only attested in root interrogatives (Munaro 1998;Poletto & Pollock 2004;. In Lamonat and Sovramontino embedded wh-in-situ is ungrammatical, as shown in (4) and (5) (5) shows that the wh-adverb aonde 'where' cannot be realised within the embedded clause: neither in its first external-merge position (5c) nor in a TP-internal position that resembles the constituent order of the root interrogative in (1), as in (5b). In this paper, I will hence limit the discussion of wh-in-situ to root contexts where the phenomenon is attested.
As far as Friulian is concerned, it does not allow wh-in-situ in both root interrogatives and in embedded interrogatives, as shown in (6) and (7) In Section 6, I will show that, given the apparent wh-in-situ nature of the phenomenon in Lamonat and Sovramontino, the derivation of the wh-question in (6) is comparable to that of the Sovramontino example in (1), with the crucial difference that, in the two NEIDs, the interrogative force of the question is satisfied in different left peripheral heads.
In Lamonat, Sovramontino and Friulian, root interrogatives are marked by a special type of morpho-syntactic construction: subject clitic inversion (SCLI). Note that, in Lamonat and Sovramontino, wh-in-situ is orthogonal to SCLI: in root interrogatives, SCLI mandatorily takes place independently of the position of the wh-element (see Manzini & Savoia 2011). Nevertheless, in order to provide a suitable definition of locus of SCLI, I will briefly discuss the phenomenon in this section. Subject clitics (SCLs) are a characteristic trait of NIDs and, across NEIDs, SCLI arises in an obligatory fashion in root interrogatives (see Renzi & Vanelli 1983;Brandi & Cordin 1989;Poletto 2000), as shown in (8) The yes/no question in (8b) shows SCLI. In purely descriptive terms, the SCL al inverts with the inflected verb a, forming a single unit a-lo. As the term suggests, for locus of SCLI I mean the left peripheral head in which SCLI surfaces. Across NEIDs, SCLI is a necessary condition for the well-formedness of root questions: the verb moves to the left peripheral space to satisfy the interrogative force of the clause, licensing SCLI (see Goria 2004). T-to-C movement in root interrogatives across NEIDs is considered a residual V2 property (Rizzi 1996;Chomsky 2005;Salvi 2012;, and hence intimately related to the left peripheral space of the clause (Rizzi 1997).
Empirically, across NEIDs, the intimate link between SCLI and the lexicalisation of a left peripheral head in the C-domain can be transparently observed in a subset of seemingly declarative sentences that encode irrealis mood, more specifically an optative/desiderative or counterfactual interpretation (see Munaro 2001). Let us consider the examples in (9) and (10) The data in (9) and (10) show that SCLI and the complementisers se/che are mutually exclusive, demonstrating that SCLI is intrinsically linked to T-to-C movement and the lexicalisation of a left peripheral head. Since genuine root interrogatives are never introduced by an overt complementiser, this is more difficult to appreciate in wh-and yes/no questions.
In this paper, I will not put forward a novel account of SCLI, but assume a derivation of SCLI along the lines of Goria's (2004) and Robert's (2010). In order to account for the different distribution of SCLs in declaratives and root interrogative sentences, namely proclisis (8a) vs. enclisis (8b), I assume the existence of two distinct sets of SCLs. 3 Declarative SCLs are adjoined to T°, whereas interrogative SCLs surface in the relevant C° targeted by T-to-C movement (see Goria 2004;Roberts 2010). In Table 1, I show the paradigm of declarative SCLs and the paradigm of interrogative SCLs in Lamonat and Sovramontino.
As shown in Table 1, in Lamonat and Sovramontino, the interrogative paradigm is heavier than the assertive one: SCLs in declarative clauses show gaps in their paradigm. Declarative SCLs also display different morpho-phonological forms with respect to their interrogative counterparts. The marked differences between the two paradigms in terms of (i) morpho-syntactic distribution (proclisis vs. enclisis), (i) paradigm gaps, and (iii) morpho-phonological realisation have been taken as evidence in favour of the existence of two discrete sets of SCLs with different underlying derivations (Goria 2004;Roberts 2010). There is general agreement in the literature not to treat declarative SCLs as weak pronominal elements (à la Cardinaletti and Starke 1999) on par with French atonic subject pronouns (Rizzi 1986;Brandi & Cordin 1989), but as clitic agreement markers that encode the feature specifications of the subject. In line with Goria (2004), I do not assume that interrogative SCLs are located in a specific agreement position within the left periphery (à la Poletto 2000), but in the left peripheral head responsible for the interrogative nature of the sentence. Interrogative SCLs should hence be regarded as interrogative affixes licenced by T-to-C movement, which surface on the inflected verb and spell out the feature specifications of the subject. Even if the derivation of SCLI per se is not central to the claims put forward in this paper on wh-in-situ in Bellunese (see Manzini & Savoia 2011), I want to suggest a derivation of SCLI along the lines of Bonan's (2019) in Trevigiano, namely via Agree (Chomsky 2001), whereby the C° targeted by T-to-C movement probes for the feature specifications of the subject.
As previously mentioned, T-to-C movement in root interrogatives across NEIDs is considered a residual V2 property (Rizzi 1996;Chomsky 2005;Salvi 2012;. By adopting a split-CP model (Rizzi 1997), Wolfe (2016) argues that Medieval Romance V2 (see Benincà 1984;Ledgeway 2008) was not satisfied in the same left peripheral head across Romance, but could either involve the lexicalisation of Force° or Fin°. Given the residual V2 nature of verb movement in root interrogatives, in Section 6, I will put forward the claim that the locus of SCLI is also subject to micro-parametric variation across NEIDs: ForceP in Lamonat and Sovramontino, whereas FinP in Friulian. Despite the different locus of SCLI, I assume that the mechanism of question formation (T-to-C movement SCL surfaces in questions as the inflected verb is moved out of T°. See also Calabrese and Pescarini (2014) and Manzini (2012) for a derivation of SCLI based on the one-paradigm hypothesis. and derivation of SCLI) is the same across the three NEIDs under investigation. Following Baker's (1970) and Bresnan's (1972) arguments on clausal typing and the Q-morpheme as well as the subsequent literature on the topic (Chomsky & Lasnik 1977;Huang 1982;Cheng 1997;Cable 2010;Bocci 2013 a. o.), I assume that, from a synchronic perspective, residual V2 movement in root interrogatives questions is justified by postulating that the locus of SCLI hosts a Q null-question particle that is primarily responsible for attracting the verb to the left peripheral space in root interrogatives. 4 Traditionally, Q is contained under the head accountable for the well-formedness of questions in the C-domain. In root wh-and yes/no questions, Q attracts the finite verb, which moves from T-to-C to satisfy the structural requirements on interrogatives. Chomsky (1995) argues that Q is affixal in nature and attaches to the overt head it attracts. I hence assume that, across NEIDs, Q is externally merged in the locus of SCLI. It is important to note that, in the NEIDs under investigation, in embedded interrogatives Q is satisfied by the external merge of an obligatory (default) complementiser che (cf. 11a), which de facto impedes T-to-C movement and hence the surfacing of SCLI, as shown in (11b) The discussion of embedded wh-questions is beyond the scope of the present paper; nonetheless, the obligatory lexicalisation of che in embedded interrogatives has been taken as evidence of a residual V2 feature in interrogatives across NEIDs (Poletto 2000;Benincà & Poletto 2004). I believe, however, that in a purely synchronic perspective, postulating the existence of a null affix Q as opposed to a strong V2 feature has its own theoretical and empirical advantages. In Section 6, I will show that assuming the existence of a null interrogative affix Q in the locus of SCLI helps us account for why SCLI is ungrammatical in root negative-interrogatives in Lamonat and Sovramontino. The postulation of Q also aids the analysis of those NEIDs that do not exhibit SCLI in root interrogatives: it can be assumed that the silent interrogative affix is absent at the morpho-syntactic level, but present at the prosodic level (see Reglero & Ticio 2013).
Having shown that wh-in-situ is uniquely a root phenomenon in the Bellunese varieties under investigation, and having outlined my main assumptions on the syntax of root 4 Note that, by drawing on Cable's (2010) work, Bonan (2019) puts forward an elegant model to capture Trevigiano's high degree of optionality between the fronting and the in-situ position of wh-elements that involves the Q particle. She claims that Trevigiano can resort to both QP-selection, in case of wh-fronting, and Q-adjunction, in case of wh-in-situ (see Cable 2010; Bonan 2019 for further discussion). In the model put forward in this paper for the NEIDs under investigation, Q must solely be interpreted as an affix that is always externally merged in the C° responsible for the well-formedness of all root interrogatives and primarily responsible for attracting the verb in the left peripheral space. I hence assume that Q is independent of the wh-fronting/wh-in-situ alternation. In Lamonat and Sovramontino, where no genuine optionality between the preverbal vs. postverbal wh-poition is attested, a more elaborate account would result as uneconomical: the position of wh-items can be fully constrained by (i) their discourse-pragmatic interpretation (focal vs. d-linked or mirative) and (ii) their morpho-syntactic status (tonic vs. clitic).
interrogatives across NEIDs, in the next section I will outline and discuss the existing analyses of wh-in-situ in Bellunese, stating how and why the proposal put forward in this paper differs from them.

Bellunese wh-in-situ: Existing analyses
Wh-in-situ is a trait of NIDs that has been long noted. Several scholars have investigated this phenomenon: the two most influential and conflicting accounts are those of Munaro, Poletto and Pollock (2001) on one side, and Manzini and Savoia (2005;, on the other side. Munaro et. al (2001) argue against an actual in-situ analysis of postverbal wh-items, whereas Manzini and Savoia (2005; argue that postverbal wh-items truly surface in their base-generated in-situ position. Bonan (2017;2019) provides a third theoretical account of wh-in-situ across NIDs based on Trevigiano by adopting Belletti's (2004; periphery of the vP. In this section, I will show that the empirical evidence from Lamonat and Sovramontino lend support to an analysis of Bellunese wh-in-situ whereby the postverbal wh-element is not in TP-internal position, but has undergone movement to the C-domain.
In this regard, Bellunese should not be treated as a wh-in-situ language, but as a wh-fronting language. The discussion put forward in this section also supports Bonan's (2017;2019) claim that wh-in-situ across NIDs should not be treated as a unitary phenomenon, but at least three different types of insituness can be identified: (i) Trevigiano-type insituness, in which the wh-item is realised in TP-internal position in the periphery of the vP (Bonan 2017;2019), (ii) Lombard-type insituness, in which the wh-item appears TP-internally in its first external-merge position (Manzini & Savoia 2005; and (iii) Bellunese-type fake insituness, in which the wh-element actually surfaces in the C-domain (Munaro 1998;Munaro, Poletto & Pollock 2001;Poletto & Pollock 2004;2009;. Lamonat and Sovramontino abide to this typology and display the behaviour of type (iii) in-situ languages. I will now outline the existing analyses of wh-in-situ across NIDs and state how and why they differ from the analysis of wh-in-situ in Bellunese proposed in this paper. I will start from Poletto and Pollock's (2004;2009; derivation of Bellunese fake insituness. Munaro (1998;, and Poletto and Pollock (2004; agree that the in-situ position of wh-words in Bellunese is only apparent: the wh-element is not in TP-internal position, but has moved to the C-domain in the same fashion as canonical wh-words in Italian and in other Romance languages. 5 The account of wh-in-situ in Bellunese put forward in this paper fully supports this claim, but differs in explaining the alternation between apparently in-situ and preverbal wh-elements with respect to a number of points.

Against remnant-TP movement: The complex T°U
nder Poletto and Pollock's (2004;2009; account, the apparent in-situ position of wh-items along with SCLI is derived through overt remnant-TP phrasal movement to the C-domain. Poletto and Pollock (2004) argue that after wh-movement takes place, the whole TP layer undergoes remnant movement to the left periphery, as shown in the derivation of the apparent wh-in-situ question in (12): Bellunese ( Among their arguments against Poletto and Pollock's (2004;2009) analysis, Manzini and Savoia (2011) point out that, in Bellunese, the only element actually vacating the TP layer is the past participle (if present) along with the inflected verb, as shown more transparently in (13) The fact that the subject Mario and the oblique prepositional phrase al can remain in their seemingly canonical position theoretically and empirically weakens the remnant movement analysis: we would, in fact, expect all the material contained in the TP-layer to be able to appear in the C° domain. Note that, in (13), no major prosodic break is present between the wh-item and the rest of the clause; hence no right dislocation is evident. I will further discuss this point in Section 4.1.3, where I will show that in Lamonat and Sovramontino there is no sentence finality requirement on the postverbal wh-item (à la Etxepare & Uribe-Etxebarria 2005), as opposed to what claimed by Munaro (1999), Poletto and Pollock (2004) for Pagotto (the variety of Bellunese studied by Munaro et al.).
I propose that, in Lamonat and Sovramontino, the constituent order in (13) can be instead derived by assuming T-to-C movement of a complex T° to a left peripheral position higher than that occupied by the wh-item. Assuming a T-to-C movement analysis of the phenomenon has also the advantage of a more theoretically sound derivation of SCLI, as opposed to a derivation of SCLI via overt phrasal movement (see Bonan 2019 for further discussion). In order to fully justify this analysis, we must however account for the behaviour of the past participle in analytic tenses, which, if present, always precedes the wh-item. In his discussion of Lamonat and Sovramontino's residual V2 properties in relation to information structure, De Cia (2018) claims that, in the two NEIDs, T forms a complex head that hosts (i) the finite verb/auxiliary, (ii) the past participle and (iii) any satellite clitics (satellite clitics in the sense of Benincà 1994). By adopting De Cia's (2018) analysis, we can straightforwardly account for the behaviour of the past participle without having to resort to Poletto and Pollock's (2004; remnant movement analysis. In the next paragraphs, I will explore De Cia's (2018) claim comparatively by considering Lamonat and Sovramontino, on one side, and Friulian, on the other side.
As discussed in Section 3, a common trait between the NEIDs under investigation is that T-to-C movement satisfies the interrogative force of the sentence, displaying SCLI, as shown in (14) and (15) In (14) and (15), assuming T-to-C movement for the fronting of the inflected verb proves the most economical and straightforward analysis (as opposed to a remnant movement analysis), whereby the finite verb vacates the TP-layer via T° to land in the left-peripheral C° that satisfies the interrogative force of the question. 6 However, a complication arises if we consider wh-questions that exhibit an analytic past, where the presence of the past participle seems to challenge the proposed head movement analysis for the derivation of apparent wh-in-situ in Lamonat and Sovramontino. Let us consider (16) and (17) In order to account for the constituent order in (17a) in Bellunese, Poletto and Pollock (2004; assume that the past participle is located in a topical position lower than the projection hosting the inflected verb: the past participle surfaces in that position due to remnant-TP movement, which takes place after wh-movement. By looking at examples (15) to (17) under closer scrutiny, however, it is possible to appreciate that the constituent order between (14) and (15), and (16a) and (17a) has not changed except for the presence of the past participle that follows the inflected auxiliary. If we try to separate the auxiliary and the past participle, as in (16b) and (17b) where the unmarked subject Peter intervenes, the result is ungrammatical in Lamonat and Sovramontino, whereas grammatical in Friulian. These facts can be unitarily accounted for by assuming that the derivation of the wh-questions in (16) and (17) involves T-to-C head movement in the same fashion as (14) and (15), with the only difference that (16a) and (17a) exhibit a complex T° that hosts both the auxiliary and the past participle. 7 6 Note that V-to-T movement systematically takes place in Lamonat, Sovramontino and Friulian, as well as in virtually all NIDs (see Roberts 2010). Evidence will be nonetheless provided later in Section 4.1.1. 7 Note that if we adopt an analysis of SCLI via Agree (à la Bonan 2019), the derivation of SCLI is not problematic in (16) and (17): the past participle is not morphologically fit to spell out the feature specifications of the subject (i.e. show subject-verb agreement), hence the interrogative SCL must surface on the auxiliary. On the contrary, by adopting a phrasal movement analysis of SCLI, we would expect the interrogative SCL to surface encliticised onto the past participle.
De Cia (2018) claims that, in Lamonat and Sovramontino, auxiliary and past participle form a single unit (in the sense of Frascarelli 2000) due to the clitic status of Lamonat and Sovramontino's auxiliaries. The past participle hence must undergo V-to-T movement to support the tensed clitic auxiliary and ensure the cliticisation of the SCL as well as any other satellite clitics. Before reviewing De Cia's (2018) evidence in support of this claim, let us consider the Friulian example in (16) above. The grammaticality (16b) clearly shows that, in Friulian, the auxiliary is able to undergo T-to-C movement on its own and does not need the past participle for its PF realisation; yet, the constituent order in (16a), comparable to that of Sovramontino and Lamonat, is frequently attested. Making reference to root interrogatives in Friulian, Salvi (2016) points out that the constituent order in (16b) is rather archaic and, across NEIDs, more closely resembles vestigial V2 T-to-C movement. In this respect, the constituent order in (16a) is more innovative, suggesting that the clitic status of auxiliaries across NEIDs is also an innovative trait of these languages. In Friulian, auxiliaries have not yet achieved clitic status. On the contrary, in Lamonat and Sovramontino, auxiliaries display cliticlike behaviour, as shown by the ungrammaticality of (17b). This is not surprising, as, across Romance, the development of the periphrastic perfect/analytic past from the Latin possessive-resultative periphrasis is by no means a unitary phenomenon (see Loporcaro 1998;. Ledgeway (2016) notes how, across Romance, auxiliaries are moving away from their original lexical predicates, becoming more and more morpho-phonologically specialised forms that have already achieved clitic status in some Romance varieties. 8 De Cia (2018) puts forward two main pieces of evidence in support of the clitic status of auxiliaries in Lamonat and Sovramontino: (i) the unavailability of stylistic inversion, which is linked to the residual V2 root phenomena shown by these varieties and, (ii) the inability for a low tonic adverb to appear between the auxiliary and the past participle. 9 I will briefly discuss the second point and add some independent evidence on the morphophonological status of Lamonat and Sovramontino's auxiliaries and past participial form. In (17b), we have seen that an unmarked subject cannot intervene between the auxiliary and the past participle, the same is true of low tonic adverbs (Cinque 1999), which appear post-verbally after the past participle, as shown in (20) and (21). Examples (18) and (19) instead illustrate that, in line with virtually all NIDs, V-to-T movement canonically takes place in Lamonat and Sovramontino; as a result, low adverbs are found postverbally (Cinque 1999;Manzini & Savoia 2005;Roberts 2010): (18) Lamonat No i copa miga gnesuni. neg 3pl.m.scl kill.3pl not-even nobody 'They won't kill anyone.' 8 Full-fledged clitic auxiliaries are found in Romanian and in some varieties of Abruzzese (Dobrovie-Sorin 1994; Stampone 2017). 9 Clitic auxiliaries must not be confused with auxiliary clitics across NIDs (Poletto 2000;Pescarini 2016), which accompany auxiliaries, coexist with subject clitics and generally have a fixed form. Lamonat and Sovramontino do not have auxiliary clitics. Nevertheless, the link between clitic auxiliaries and auxiliary clitics should be properly investigated, as the latter are fully compatible with the clitic status of auxiliaries: in those varieties, the auxiliary paradigm has developed fully independent forms from its original lexical verb paradigm, suggesting a higher degree of grammaticalisation (see Dobrovie-Sorin 1994 for a comparison with Romanian). The ungrammaticality of (20b) and (21b) shows that the past participle undergoes V-to-T movement in the same fashion as the inflected lexical verbs in (18) and (19). Nevertheless, De Cia (2018) notes that a limited set of adverbs can be found between the auxiliary and the past participle. These adverbs are atonic, only licensed between the auxiliary and the part participle, and never found in isolation. This is shown in examples (22) and (23) An adverb like despes or migo cannot appear between the clitic auxiliary and the past participle (cf. 20b, 21b) due to its "phonological weight", whereas their phonologically reduced counterparts spes and mia are felicitous only in that position. The difference in distribution suggests that those adverbs that can be placed between the clitic auxiliary and the past participle also have clitic status. They can be stacked under the complex T° with the clitic auxiliary, the subject clitic and any other satellite clitics. Clitic adverbs cannot be focalised, behaving like weak elements in the sense of Cardinaletti and Starke (1999). The clitic status of certain low adverbs is not unattested in the literature on Romance. Dobrovie-Sorin (1994) shows that in Romanian a restricted class of adverbs can be placed between the clitic auxiliary and the past participle. 10 In Romanian, the past participle must undergo V-to-T movement to allow for the cliticisation of the auxiliary clitic and any satellite clitics. In Lamonat and Sovramontino, I argue that the same mechanism is in place: given the clitic status of auxiliaries and their inability to support any satellite clitics, V-to-T movement of the lexical verb, be it inflected or in the past participial form, must take place.
In Lamonat and Sovramontino, the clitic status of auxiliaries can also be empirically appreciated by considering their morpho-phonological form in relation to that of the past participle. Let us consider examples (24), (25) and (26) (24) is uttered bearing stress as in (26), the result is an infelicitous utterance. Making reference to the paradigm of auxiliary have, there are no evident morphological differences with the paradigm of lexical-possessive have; nevertheless, the auxiliary paradigm is strikingly atonic. The monosyllabic paradigm of have has unquestionably contributed towards the development of clitic auxiliaries, but the auxiliary paradigm has not yet developed autonomous morphological forms (see Romanian clitic auxiliary paradigm Dobrovie-Sorin 1994). The verb paradigm of lexical verbs is considerably impoverished and displays idiosyncratic forms for the 2SG, 3SG and 3PL grammatical persons, which can only be set apart by the discrete forms of the preverbal SCLs. The same is true for lexical/auxiliary have whose paradigm is monosyllabic, as shown in (27) The presence of the subject clitic is hence crucial to disambiguate the syncretic forms of the paradigm; however, by virtue of being atonic, auxiliary have cannot phonologically support the subject clitic. The PF realisation of the SCL and the auxiliary have thus relies on the past participle. In turn, the contrast between the identical morphological forms of the inflected verb magna in (25) and the past participial form of the verb 'to eat' magnà in (24) shows that past participles are equipped with an extra layer of stress, which falls on the final syllable (this is particularly evident in the past participial forms of the first and second verb conjugations like magnà 'eaten' and beù 'drunk'). The prosodic prominence of the past participial form must be considered the result of the phonological weakening of auxiliary forms that, being unable to support themselves as well as any satellite clitics at PF, phonologically rely on the past participle for their realisation.
The empirical observations put forward in this section strongly suggest that, in Lamonat and Sovramontino, T is a complex head which hosts (i) the inflected verb (lexical or auxiliary), (ii) the past participle, if present, and (iii) any satellite clitics that prosodically rely on the lexical verb. In root interrogatives, in order to satisfy the interrogative force of the question, the complex T° undergoes T-to-C movement as a single unit, deriving so Bellunese apparent wh-in-situ. For instance, clitic adverbs can be found sandwiched between the auxiliary and the past participle in root interrogatives, but tonic adverbs cannot (they remain in TP-internal position), as shown in (28) and (29) As far as the internal syntactic structure of the complex T° is concerned, our model must keep into account the diachronic development of the Latin possessive-resultative construction have + past participle into the periphrastic perfect, which, across NEIDs, has fully replaced the synthetic past. Two accounts are possible: (i) the periphrasis auxiliary + past participle behaves like a synthetic form (replacing de facto the synthetic past); 11 or (ii) assume that the auxiliary is base-generated and adjoined to T, forming a clitic cluster with the subject clitic and the other satellite clitics. As for (i), treating the perfect periphrasis as a single syntactic unit would be desirable for the derivation of the complex T°; however, it leaves open the question of clitic adverbs. Clitic adverbs allegedly developed by "remaining trapped" between the auxiliary and the past participle, while the periphrasis was reanalysed as a single phonological unit. By adopting the account in (ii), the placement of adverb clitics is more straightforward, as they their cliticisation on the past participle supposedly takes place before V-to-T movement. Nonetheless, the derivation in (ii) forces us to assume that the cliticisation of satellite clitics takes place in T° and not in V° prior to V-to-T movement (see Roberts 2010). I will leave this point open for debate.

A more economical left-peripheral space
The analysis put forward in the present paper is more economical in terms of C-domain architecture than Poletto and Pollock's (2004;2009), and solely relies on Rizzi's (1997) seminal split-CP model. Manzini and Savoia (2011) criticize the remnant-TP movement analysis for the legitimacy of the numerous left-peripheral projections that are involved in Poletto and Pollock's (2004;2009) account: the existence of such projections is in fact not always independently justified. In their analysis of the alternation between apparently in-situ and canonical wh-elements in Bellunese, Poletto and Pollock (2004) (30) depicts the left periphery of root interrogatives across Romance. They argue that in Bellunese wh-elements appear in Operator2P when they exhibit canonical behaviour; wh-elements instead occupy Operator1P when they are apparently in-situ. In Poletto and Pollock's model, the finite verb is in SpecForceP, the interrogative form of the encliticised SCL in GroundP and, finally, the past participle in TopP. As previously mentioned, their analysis does not involve T-to-C movement, but the finite verb moves to SpecForceP through remnant movement of the TP layer after the whelement is attracted to Operator1P in the C-domain.
Besides the aforementioned left-peripheral projections, Poletto and Pollock's (2004;2009) account crucially involves a null restrictor in SpecOp2P that licences the apparently in-situ wh-element in Op1P. They argue that the null restrictor is base-generated in the TP layer with the wh-word, with which it forms a complex wh-phrase. In root interrogatives, the complex wh-phrase is moved to Op1P as a single unit and, as soon as Op2P is merged, the null restrictor is moved from Op1P to Op2P, leaving the wh-word in Op1P in apparently in-situ position. If the null restrictor is not licenced, the question cannot have the wh-word surfacing in Op1P: the wh-word itself must move to Op2P, yielding a canonical wh-question. Let us exemplify Poletto and Pollock's (2004) analysis by considering the Bellunese apparent wh-in-situ question in (31a) and its tree representation is (31b): Bellunese a. Se-tu ndat andè? be.2sg-2sg.scl go.ptcp where 'Where did you go?' As previously mentioned, Poletto and Pollock (2004) argue that some Bellunese whwords have the property of licensing a null restrictor, namely those corresponding to English who, what, where, when and how. The null restrictor is thus defined as a non-lexical DP that forms a complex wh-phrase with the wh-word. In (31), the null restrictor is base-generated in the T-domain with the wh-element ande as follows: [ TP tu se ndat [ PRN ande n-rest]]. The whole wh-phrase is then moved to SpecOp1P. Once Op2P is merged in the derivation, it attracts the null restrictor, leaving ande in Op1P in apparently in-situ position. Poletto and Pollock (2004;2009) claim that, across Romance, wh-questions obey this syntactic mechanism. In their account, the key difference between Bellunese and the other Romance languages is that the latter do not possess wh-words that can license a null restrictor: hence, as soon as Op2P is merged, the wh-phrase lacking the null restrictor must move to SpecOp2P to prevent the derivation from crashing. For instance, let us consider the Friulian counterpart of (31) in (32a) and its tree representation in (32b). Note that Friulian does not exhibit wh-insitu. The null restrictor hypothesis is based on the empirical observation that, in some NIDs exhibiting wh-in-situ, the apparently in-situ wh-element is doubled by a wh-clitic in canonical position (Poletto & Pollock 2004;2009;Manzini & Savoia 2011). Poletto and Pollock (2004;2009) thus claim that the null restrictor can have a lexicalised form consisting of a wh-clitic. Let us briefly have a look at some data from Lamonat and Sovramontino. Poletto and Pollock's (2004;2009) account captures the wh-question in (33), but does not fully explain (34) as well as the unacceptability of (35) Under their analysis, (35) should be grammatical, as the wh-clitic ond should act as null restrictor and license aonde in apparently in-situ position. Speakers, however, do not accept as grammatical the simultaneously lexicalisation of the preverbal wh-clitic and the postverbal tonic wh-element. The only exception is what-doubling, which I will briefly discuss in Section 5.2. Also, wh-questions like (34) are quite problematic under Poletto and Pollock's (2004;2009) account: if ond is a wh-clitic acting as null restrictor and is licensed by the wh-element aonde 'where', how come the licensor is not present in the clause? If aonde is absent in apparently in-situ position, what licenses the acting null restrictor ond? In addition, Friulian also exhibits a wh-clitic item for where that alternates with its tonic counterpart, as shown in (36) and (37)  The wh-clitic element in (36), namely la, has the same morpho-syntactic status as ond in Sovramontino (see Section 5.2); nevertheless, apparently wh-in-situ is not possible in Friulian, as shown in (37b). Across the two NEIDs, the different distribution between clitic and tonic wh-elements further weakens the null restrictor hypothesis. Another important point worth noting is that, in Lamonat and Sovramontino, prepositional whelements like with what, to whom etc. can also appear in apparently in-situ position: the licensing of a null restrictor would thus not be limited to only a restricted set of genuine wh-elements, as Poletto and Pollock (2004;2009) claim, but to all wh-phrases. These empirical observations make Poletto and Pollock's (2004;2009) account unsuitable for the analysis of Lamonat and Sovramontino. In order to account for the alternation between apparently in-situ and preverbal wh-words, I will propose a revision of Poletto and Pollock's (2004;2009) model for Lamonat and Sovramontino. I will hence not resort to the existence of two distinct ad hoc operator positions for wh-elements, but my analysis will consider (i) the discourse-pragmatic nature of wh-words and (ii) their morpho-syntactic status.

Absence of a clause finality requirement on the postverbal wh-item
The remnant-TP movement analysis has been partly justified by the observation that, in Bellunese, a clause-finality requirement is in place on the postverbal wh-element. Poletto and Pollock (2004; claim that the XPs following the postverbal wh-element are set off by the rest of the sentence by a "comma intonation". Such clause finality requirement, however, is not observed in Lamonat and Sovramontino. Let us consider the examples in (38) and (39) The sentence in (38b) is uttered in an out-of-the-blue context. All the constituents of the clause are hence in broad focus (Lambrecht 1994), except for the anchoring frame-setting element ier, which is topical in nature (Poletto 2000;Benincà & Poletto 2004;Greco & Haegeman 2016) and followed by a pause. 12 Sentence focus is generally used as a test to reveal the default constituent order of a language (see Cruschina 2012). Lamonat and Sovramontino exhibit a strict SVO constituent order. The direct object is strictly followed by the indirect object, which in turn is followed by any spatio-temporal adjuncts. The order of constituents cannot be modified without giving rise to an ungrammatical sentence. The same strict constituent order is observed in (39): an alteration in the basic constituent order leads to ungrammaticality, as shown in (39b). Contra to what observed by Munaro et. al in Pagotto,(39c) shows that, in Lamonat and Sovramontino, the postverbal wh-item cannot be followed by a comma intonation that sets it off from the rest of the clause. The sentence in (39c) can be rescued by a resumptive object clitic, as shown in (39d). In that case, the relative strict order subject > direct object becomes irrelevant with the direct object that is able to appear before the subject without affecting the grammaticality of the sentence (cf. 39e). The data in (39) suggests that, in Lamonat and Sovramontino, a comma intonation after the wh-item à la Poletto and Pollock (2004; can be only compatible with clitic right dislocation (Cinque 1990). In such case, the direct object must be resumed by an agreeing pronominal clitic, which obligatorily triggers object-past participle agreement (cf. 39d), and the right dislocated XPs can be stacked in any order in a topic field (à la Benincà & Poletto 2004 -cf. 39e). In Lamonat and Sovramontino wh-questions, a clause finality requirement à la Poletto and Pollock (2004; is therefore not attested. The rigid constituent orders in (38b) and (39) also rule out the possibility of marginalisation (see Antinucci & Cinque 1977;Cardinaletti 2001; Samek-Lodovici 2015 for an account in Italian). As a result, in Lamonat and Sovramontino, the order of constituents that follow the postverbal wh-item must not be seen as the result of an optional dislocation (i.e. clitic right dislocation or marginalisation), but as the result of wh-movement itself in tandem with T-to-C movement of the complex T°.
As far as the prosodic contour of (39a) is concerned, it is different from that of (38b) for the simple reason that (39a) contains an XP in narrow focus, namely the wh-item a chi. Nevertheless, this difference is not compatible with Poletto and Pollock's (2004; clause finality requirement. Let us consider the wh-question in (40) and its pitch contour in Figure 1.
give.ptcp what the doctors to Simon 'What did the doctors give to Simon?' Poletto and Pollock (2004; claim that, in a sentence like (40) in Bellunese, the dative complement is marginalised and separated from the wh-element by a "comma into-12 Note that frame-setters are particularly common in Lamonat and Sovramontino. They are located in a projection higher than ForceP, namely FrameP, and are generally non-recursive (see De Cia 2018;Greco & Haegeman 2016). They are hence compatible with root interrogatives, as self-contained speech acts (Krifka 2001). Frame setters have a scene-setting function, which anchors the speech act in terms of locative and temporal deixis, and speech participants. Their abundant presence in Lamonat and Sovramontino is linked to the residual Force-V2 phenomena attested in these languages (see De Cia 2018): being unable to resort to bona fide left peripheral topics due to the unfitness of Top° to host verb movement (Rizzi 1997), Lamonat and Sovramontino more abundantly resort to frame-setting elements (see Wolfe 2016). With respect to the unavailability of TopP in Lamonat and Sovramontino, background information must hence be realised insitu. See Bianchi and Frascarelli (2010), and Frascarelli and Ramaglia (2013) for the cross-linguistic parameterisation of G-topics, which can be either realised in-situ or as bona fide topics in the C-domain.
nation". As previously discussed, this is not the case in Lamonat and Sovramontino. As shown in Figure 1, the XPs following the wh-items are instead prosodically subordinated and characterised by a post-focal pitch contour. This is determined by the prosodic realisation of focus prominence on the wh-item (Zubizarreta 1998;D'Imperio 2002;Donati & Nespor 2003;Bocci & Avesani 2005;Frota et al. 2007). In Lamonat and Sovramontino, apparently in-situ wh-elements are in fact marked by a prosodic high-pitch plateaux along with the verb cluster, which signals the purely focal nature of this type of wh-elements (cf. Figure 1). Such change in prosodic pattern is not compatible with a "comma intonation", but is simply the result of prosodic subordination of the material following the focus-prominent wh-item. In the specific case of wh-in-situ in Lamonat and Sovramontino, it can be considered as marking the boundary between the C-domain (hosting the complex moved T° and the wh-item) and the T-domain (with the material that is left TP-internally after T-to-C movement and wh-movement): cross-linguistically, it is, in fact, common that prosodic phrasing be conditioned by major syntactic phrase boundaries (Kratzer & Selkirk 2007). Finally, I want to briefly discuss the position of the subject in (39) and (40). Given that Lamonat and Sovramontino are null-subject languages (see Roberts 2010 on the relation between the null-subject parameter and the declarative SCL paradigm across NIDs), the subject in (39) and (40) can be omitted without affecting the grammaticality of the sentence. Nevertheless, if the subject appears in root interrogatives, it is systematically placed after the postverbal wh-item or, in case of yes/no questions, after the past participle. I hence assume that the subject TP-internal position is either SpecTP or SpecvP. The placement of tonic low adverbs, however, seems to suggest that, in Lamonat and Sovramontino, the subject raises as high as SpecTP, as shown in (41): If we assume that the adverb despes is located in a functional projection higher than vP (à la Cinque 1999), the ungrammaticality of (41b) is accounted for by postulating that the lexical subject does not stay in SpecvP, but raises to SpecTP (hence the felicity of 41a). Note again that the sentence in (41b) can be rescued by adding a pause after despes through the right dislocation of the subject and the spatial adjunct.
Another argument in favour of the subject raising to SpecTP in root interrogatives comes from subject questions. In Lamonat and Sovramontino, subject what or who must be obligatorily clefted (see also Munaro 1998), 13 as shown in (42) and (43) The subject wh-element follows the third person dummy copular be + SCL expl and is in turn followed by the complementiser che. Subject chi and che must appear in a cleft structure in order to be correctly interpreted as the subject of the question. If not clefted and realised in postverbal position, 'who' or 'what' are by default interpreted as the object of the clause, as shown in (44) The sentence in (44) literally means 'who did the wine drink?'. Also note that, by virtue of its post-wh-item position (and lack of clause finality requirement), l vin is by default interpreted as the subject of the clause. However, if chi 'who' is the subject of a core unaccusative verb like die or arrive and, hence, structurally behaves like an object (see Belletti 1988;Levin & Rappaport Hovav 1995), the wh-element chi is not clefted, but realised in postverbal position, as shown in (45) and (46): 13 Cleft-structures are frequent in Lamonat and Sovramontino. They generally host an XP in narrow focus (Lambrecht 1994), including tonic wh-words. In Lamonat and Sovramontino, two types of clefts can be identified: (i) syntactically motivated and (ii) pragmatically motivated. Syntactically motivated cleftstructures arise as a last resort strategy to focalise XPs (i.e. transitive subjects or focal XPs in co-occurrence with a strong negation) whose realisation in the C-domain would otherwise be illicit. Pragmatically motivated cleft-structures, on the other hand, prompt an exhaustive reading on the clefted focal element (see Lambrecht 2001 The data in (44) to (46) suggest that true transitive subjects, if overt, are attracted from their base-generated position to the canonical subject position, SpecTP. Once they move to SpecTP, they are frozen in place due to criterial freezing (see Rizzi 2006;Rizzi & Shlonsky 2007). The subject cleft-questions in (42) and (43) thus arise as a strategy to overcome the problem of extracting an overt subject that is frozen in SpecTP (i.e. chi or che). This is further confirmed by the behaviour of unaccusative subjects, which, instead, do not require a cleft structure for their focalisation, as they do not syntactically behave like canonical subjects. In (45) and (46) the canonical subject position is instead filled by a pro expletive (whose feature specifications are shown by the surfacing of an interrogative subject clitic expletive): chi does not structurally behave like a subject, but is basegenerated in VP-internal position, making the extraction of the wh-item fully licit. The surfacing of obligatory cleft-questions with non-unaccusative wh-subjects is therefore also evidence that wh-in-situ in Bellunese is only apparent: postverbal wh-items are not in their TP-internal first external-merge position, but have undergone wh-movement out of their base-generated position.

Subject clitic inversion and dynamic agreement
Finally, I want to point out that I will adopt Poletto's (1993) proposal of dynamic agreement (see also Poletto & Pollock 2004). Dynamic agreement concerns the well-formedness of questions across NEIDs and is based on Rizzi's (1991;1996) wh-criterion: (47) Rizzi (1996: 64) wh-criterion: a) a wh-operator must be in spec-head configuration with an X° [+wh], b) an X° [+wh] must be in spec-head configuration with a wh-operator. Poletto (1993) proposes a process that is complementary to Rizzi's wh-criterion, whereby the [WH] feature can be transmitted from the C° position to the corresponding specifier, hence dynamic agreement. Across NEIDs, the interrogative nature of root clauses is morphologically marked by SCLI: SCLI signals verb movement to the C° that hosts the null-question particle Q (the locus of SCLI). Poletto (1993) argues that through SCLI, C° licenses in SpecCP a silent operator that inherits the feature [WH] from the verbal head: the wh-criterion is thus satisfied. The surfacing of SCLI alone hence satisfies the structural well-formedness of questions. This hypothesis crucially supports the independence of wh-movement and SCLI/verb movement across NEIDs (see Manzini & Savoia 2011).
In wh-questions, the wh-item does not need to land in the specifier position of the head hosting the moved verb, as the structural requirements on interrogatives are satisfied by SCLI alone. I will thus assume that, in root interrogatives, the specifier position of the C° hosting the moved verb is filled by a null wh-operator that inherits the feature [WH] from the verbal head. Manzini and Savoia (2011) argue against Poletto and Pollock's (2004;2009) remnant-TP movement account in favour of an actual in-situ analysis of postverbal wh-items across NIDs. The Lombard data with which they challenge Poletto and Pollock's (2004;2009) analysis, however, yield ungrammaticality when tested in Lamonat and Sovramontino. This is especially true with respect to three points. First, as previously mentioned, in Lamonat and Sovramontino there is a rigid root vs. embedded asymmetry with respect to wh-in-situ. The phenomenon is only attested in root interrogatives and, conversely to Lombard varieties, wh-in-situ is ungrammatical in embedded interrogatives. Second, Lamonat and Sovramontino wh-in-situ is sensitive to strong and weak island constraints, whereas Lombard proper wh-in-situ is not. Munaro (1999) in fact notes that, in the Bellunese variety of Pagotto, apparent wh-in-situ is sensitive to strong and weak island constraints (Munaro, Poletto & Pollock 2001;Poletto and Pollock 2004). If Bellunese were a true wh-in-situ language, wh-islands would not hold (see Huang 1982). The data from Lamonat and Sovramontino are in line with Munaro's (1999). Consider (48), (49) and (50) The ungrammaticality of (48), (49) and (50) is respectively due to a complex-NP island, an adjunct island and a relative island (see Munaro 1999 for further discussion). Islands constrain the unbounded nature of wh-movement: a wh-element can potentially be an unlimited number of clauses away from its in-situ null copy, with which it establishes a long distant dependency. Munaro (1999), Poletto and Pollock (2004) argue that sensitivity to islands in Bellunese is evidence in favour of the ex-situ nature of the postverbal wh-item. It is important to note, however, that Manzini and Savoia (2011) argue that the differences between Lombard-type and Bellunese-type wh-in-situ can be accounted for in the context of surprising micro-variation in LF conditions among closely related varieties. In their account, wh-in-situ vs. wh-movement across NIDs is explained in terms of the LF interpretative alternation between overt scope (wh-fronting) and scope construal (wh-in situ). They argue that sensitivity to islands and the inability of wh-movement in embedded clauses can be explained on the basis of grammar-specific conditions on LF interpretative construals, rather than on different syntactic conditions on wh-movement itself. 14 In light of Manzini and Savoia's (2011) critique, the most compelling evidence against an actual in-situ analysis of postverbal wh-items in Lamonat and Sovramontino comes from the strict relative order between wh-adverbs/wh-indirect objects with respect to the in-situ lexical direct object. Let us consider the wh-question in (51) Given our discussion in Section 4.1.3, the ungrammaticality of (51b) shows that the wh-adverb quand cannot appear in its base-generated position, but must undergo wh-movement (cf. 51a). The different empirical evidence with respect to wh-in-situ between Bellunese and Lombard strongly indicates that the syntactic mechanism at the basis of the phenomenon is not the same in the two NIDs. In line with Bonan (2017;2019), I believe different accounts should be sought out for wh-in-situ across NIDs. Manzini and Savoia's (2011) analysis fully captures wh-in-situ in Lombard, but an actual in-situ analysis of postverbal wh-items is incompatible with the Lamonat and Sovramontino data.

TP-internal in-situ analysis
Finally, I want to point out that, in Lamonat and Sovramontino, wh-items can appear in their TP-internal base-generated position in echo questions. Let us consider the echo question in (52) and its non-echo question counterpart in (53) The echo question in (52) translates into English 'The dog ate WHAT?'. In Sovramontino and Lamonat, echo questions are marked prosodically by an echo intonation on the interrogative pronoun, which is uttered duplicating its final vowel. In (52), the wh-word chee truly occurs in-situ in its TP-internal base-generated position. A crucial difference between (52) and (53) is that the echo question does not exhibit SCLI, whereas, in the apparent wh-in-situ question in (53), SCLI is pivotal for the grammaticality of the question. Thus, the echo question in (52) is not characterised by an interrogative structure. The in-situ position of chee in (52) can be tested by replacing the bare wh-item with an aggressively non-d-linked wh-item (Pesetsky 1987): the result is an ungrammatical sentence, as shown in (54) Aggressively non-d-linked wh-elements must move out of their TP-internal in-situ position, hence the ungrammaticality of (54). Aggressively non-d-linked wh-elements can instead surface in postverbal position in wh-questions with an interrogative structure (i.e. exhibiting SCLI), as shown in (55) (55) is further evidence that, in root wh-questions, postverbal wh-items are not in their TPinternal base-generated position. In Section 5.3, I will further discuss the case of d-linked and aggressively non-d-linked wh-elements.

Wh-movement to the periphery of the vP
By adopting Belletti's (2004; periphery of the vP, Bonan (2017; 2019) proposes a third theoretical account of wh-in-situ across NIDs based on Trevigiano. 15 Bonan (2017;2019) does not impose her analysis on all NIDs that exhibit wh-in-situ and crucially advocates a non-unitary analysis of wh-in-situ across NIDs. She claims that her analysis applies to Trevigiano and related varieties. As convincingly shown in her work, she claims that wh-in-situ in Trevigiano is the result of short wh-movement to Foc in Belletti's (2004; left periphery of the vP. There are two main reasons not to adopt Bonan's (2017;2019) analysis in this paper. First, empirically, there is a clear difference between the Trevigiano data and the Lamonat/Sovramontino data with respect to wh-in-situ. Second, one of the aims of this paper is to show that the derivation of wh-questions in Sovramontino/Lamonat is comparable to that of Friulian (crucially, a NID that does not show wh-in-situ): by virtue of being a wh-fronting language, the Friulian data would be extremely difficult to account for by adopting an analysis that involves wh-movement to Belletti's (2004; Foc in the vP periphery. Let us briefly discuss the empirical differences between Trevigiano and Bellunese with respect to wh-in-situ. The first difference lies in the availability of wh-in-situ in embedded clauses. Trevigiano allows wh-in-situ in embedded contexts, whereas, as previously mentioned, Lamonat and Sovramontino do not. If, in Lamonat and Sovramontino, the postverbal wh-element targeted a TP-internal position (Foc in the vP periphery), the unavailability of embedded wh-in-situ would be difficult to justify. 16 The second difference concerns the distribution between preverbal and postverbal wh-items. Bonan (2017;2019) shows that, in Trevigiano root interrogatives, wh-items can indistinctively surface in either preverbal or TP-internal position (with the exception of cossa 'what' that is banned TP-internally). Crucially, the same is observed for d-linked wh-elements, which can be found in either wh-positions. In Trevigiano, wh-in-situ hence seems to be an optional phenomenon. 17 In Lamonat and Sovramontino, on the other hand, there is a clear distinction between the wh-items that are licensed in preverbal position and those that are licensed in postverbal position: d-linked wh-elements, wh-items encoding mirativity and clitic wh-elements must surface preverbally, whereas tonic non-d-linked/non-mirative wh-items must surface in postverbal position. This fact suggests that a different set of functional projections is involved in the realisation of Bellunese wh-items, which is tightly linked to (i) the information structure status of the wh-items themselves (d-linked, mirative or focal), and (ii) their morpho-syntactic status (tonic vs. clitic). An analysis whereby d-linked and nond-linked wh-items landed in the same TP-internal position would not be able to capture these differences. Third, the empirical evidence put forward in the previous sections of the paper on Lamonat and Sovramontino's (i) complex T° and clitic auxiliaries, (ii) placement of tonic and clitic adverbs, 18 and (iii) subject cleft-questions, would be difficult to 17 As pointed out in ftn. 4. Bonan (2019) puts forward a "competing-grammar" model to capture Trevigiano's high degree of optionality between the fronted and the in-situ position of wh-elements that involves the Q particle (in the sense of Cable 2010). She claims that Trevigiano can resort to both QP-selection, in case of wh-fronting, and Q-adjunction, in case of wh-in-situ (see Cable 2010; Bonan 2019 for further discussion). 18 There are some interesting differences with respect to tonic adverb placement and the position of the postverbal wh-element: some configurations that are grammatical in Trevigiano are ungrammatical in Lamonat and Sovramontino. This is evidence that in Lamonat and Sovramontino a different derivation of (apparent) wh-in situ is in place. This is particularly noticeable with respect to the position of the past participle and the postverbal wh-item in relation to tonic adverbs. Bonan (2019) shows that in Trevigiano low adverbs are placed within the TP-internal space above the vP; in her account, the past participle also targets the same bundle of projections and lands in a functional projection lower than the functional projections occupied by virtually all low adverbs. The only exception is tuto 'all', which instead appears in a functional projection lower than that occupied by the moved past participle (Cinque 1999). In Trevigiano, the ungrammatical Sovramontino wh-question in (i) would hence be grammatical: (i) *A-lo sistemà tut quand? have.3sg-3sg.m.scl tidy.up.ptcp all when (When did he tidy up everything?) In fact, as opposed to Sovramontino, in Trevigiano, the tonic adverb tut 'all' is able to appear between the past participle and the wh-item in the left periphery of the vP. Given the unsuitability of Top° to host the T-to-C movement (see Rizzi 1997) and thus the absence of bona fide topical elements in the left periphery of Lamonat and Sovramontino, the past participle/inflected lexical verb and the postverbal wh-item are always adjacent in the two Bellunese varieties (also, linear adjacency with the verb cluster contributes towards the assignment of focal prominence on the postverbal wh-item, see Section 4.1.3). As a result, the grammatical linear order of the Sovramontino wh-question in (i) features tut after the wh-element, as shown in (ii): (ii) A-lo sistemà quand tut? have.3sg-3sg.m.scl tidy.up.ptcp when all 'When did he tidy up everything?' This difference shows that, as opposed to Trevigiano, in Lamonat and Sovramontino, the wh-item has moved higher than the left periphery of the vP, more specifically in the C-domain. Let us consider another example with the tonic low adverb sempre 'always': the following liner order is grammatical in Lamonat, but ungrammatical in Trevigiano (in line with Bonan's 2019 account): (iii) Va-lo aonde to fradel sempre a-l mar? go.3sg-3sg.m.scl where your brother always to-the seaside 'Where does your brother always go to the seaside?' In Trevigiano, Bonan (2019) claims that, if not dislocated, the subject surfaces in SpecvP and that, as previously mentioned, the adverb sempre is located above the vP periphery. In this light, the grammatical counterpart of (iii) in Trevigiano would be the ungrammatical Lamonat sentence in (iv): (iv) *Va-lo sempre aonde to fradel a-l mar? go.3sg-3sg.m.scl always where your brother to-the seaside (Where does your brother always go to the seaside?) The empirical difference in adverb placement again suggests that postverbal wh-items in Lamonat and Sovramontino do not target the left periphery of the vP. reconcile with an analysis whereby postverbal wh-elements surface TP-internally in the periphery of the vP.
Having discussed the main analyses of wh-in-situ across NIDs, in the remainder of the paper, I will: (i) summarise the proposed analysis for apparent wh-in-situ questions in Lamonat and Sovramontino, (ii) account for Lamonat and Sovramontino's different distribution of wh-items between the apparently in-situ and preverbal position, and (iii) show the resemblance between Lamonat/Sovramontino and Friulian with respect to the derivation of wh-questions.

The behaviour of wh-words in Lamonat and Sovramontino
Lamonat and Sovramontino exhibit both apparently in-situ and preverbal wh-elements. Letting aside for now preverbal d-linked and mirative wh-elements, which will be discussed separately in Section 5.3, let us consider the following wh-questions in (56) and (57) Ond-e-la ndà to sor? where-be.3sg-3sg.f.scl go.ptcp your sister 'Where did your sister go?' (56) exemplifies an apparent wh-in-situ question, whereas (57) shows a wh-question with a non-d-linked/non-mirative preverbal wh-item in seemingly canonical position. So far, we have amply discussed apparently in-situ wh-questions (cf. 56), but we have not proposed a derivation for (57). In the next two sections, I will (i) summarise the proposed analysis for apparent wh-in-situ in Lamonat and Sovramontino, and (ii) propose a derivation for wh-questions featuring a non-d-linked/non-mirative preverbal wh-item.

Apparently in-situ wh-words: SpecFocusP as the targeted left-peripheral projection
In the previous sections, we have established that, in root interrogatives, the wh-element occurring in postverbal position undergoes wh-movement. We now need to determine which position it targets in the C-domain. In order to do so, I will adopt Rizzi's (1997) seminal split-CP model. Rizzi (1997) claims that the CP encodes at least two types of information relating to both the 'outside' and the 'inside' of the clause. In his model, ForceP encodes information that looks at the higher structure of the clause and specifies its clausal type (i.e. interrogative, declarative or exclamative force); on the other hand, FinitenessP looks at the inside of the clause, namely at the content of the TP, and is responsible for marking finiteness. Rizzi's (1997) split-CP model does not only encode force and finiteness, but also discourse-pragmatic information that relates to information structure. He claims that the topic-comment and the focus-presupposition articulations, namely TopicP and FocusP, are sandwiched in that hierarchical order between ForceP and FinitenessP. It is important to note that I will not adopt Rizzi's model in its original form, but abide by Benincà and Poletto's (2004) claim against the existence of a lower TopicP between FocusP and FinitenessP.
Let us go back to the interrogative pronoun che in (56). I repeat (56) in (58)  We cannot establish the landing site of che within Rizzi's (1997) split-CP without first considering the behaviour of the surrounding constituents. Following Poletto's (1993; claim that in Bellunese the wh-criterion is satisfied by SCLI alone, there is no need for the wh-element to land in the specifier position of the moved verb cluster (a-lo magnà) to fulfil the structural needs of the interrogative clause. Assuming that the locus of SCLI in Lamonat and Sovramontino is Force°, the wh-element must land in a position lower than ForceP. The next lower position is TopP. This position is recursive in nature (Rizzi 1997) and can give rise to a bundle of projections that characterises a Topic Field (Benincà & Poletto 2004). Given the focal nature of wh-elements and the lack of recursiveness of wh-items, I exclude that che in (27) lands in TopicP. In addition, bona fide topical elements (i.e. non-contrastive) are optional as they are the presupposed-known portion of the sentence (Lambrecht 1994); by contrast, wh-words always refer to some novel piece of information that the speaker wants to elicit from the hearer. There is also another reason to exclude TopicP: Top° is an unsuitable head for the transit of head movement (Rizzi 1997). Given that T-to-Force movement is necessary for the well-formedness of questions in Lamonat and Sovramontino, bona fide topical elements can never surface in the left periphery of the clause in wh-and yes/no questions (only frame-setters are in fact able to appear preverbally in root interrogatives, see Sections 4.1.3 and 6). The focal properties of wh-elements strongly suggest that the most obvious landing site for che in (27) is FocusP: a position that is above the TP layer, lower than ForceP and not topical in nature.
Focal XPs and wh-items share in fact the following properties: uniqueness, lack of a resumptive clitic, prosodic sentential stress and are incompatible with one another (Rizzi 1997;2018;Cruschina 2012). As for the last point, apparently in-situ wh-items are incompatible with both contrastive/corrective and informational focal XPs, as shown in (59) and (60)  Note that the relevant constituent in narrow focus is in CAPITALS. In (59), the wh-element che cannot coexist with the contrastive/corrective focal element al gat, which in Lamonat and Sovramontino always surfaces preverbally (see De Cia 2018). The two NEIDs seem only to allow one focal element per sentence (see Rizzi 2018 on uniqueness of focus). This possibly translates into the inability of focal prominence to be recursive in the grammar of Lamonat and Sovramontino: focal prominence can only be assigned to one constituent per sentence. In (60), on the other hand, the wh-element and the informational narrow focal XP al gat potentially compete for the same syntactic position. This is true regardless of the order of focal element and wh-word (cf. 60b). 19 Having identified the landing position of the apparently in-situ wh-word che in (58), let us briefly outline how the wh-element targets that position. In addition to an intrinsic [WH] feature, I assume that the wh-element also carries a [FOCUS] feature (à la Bianchi 1999). The [FOCUS] feature is primarily responsible for the surfacing of the whelement in SpecFocusP. As for the [WH] feature, it does not need to be satisfied through spec-head agreement: the structural well-formedness of the question is ensured by SCLI alone via dynamic agreement (Poletto 1993). The wh-element is thus free to surface in SpecFocusP, fulfilling its discourse-pragmatic nature. This suggests that, in Lamonat and Sovramontino, the discourse-pragmatic nature of wh-elements is more prominent than their syntactic status as question operators. The simplified tree in (61) shows the proposed derivation of (58)

The clitic status of non-d-linked/non-mirative preverbal wh-words
We now need to account for the behaviour of non-d-linked/non-mirative preverbal whelements, like the interrogative pronoun ond in (62): Ond-e-la ndà to sor? where-be.3sg-3sg.f.scl go.ptcp your sister 'Where did your sister go?' In a nutshell, I propose that, in Lamonat and Sovramontino, non-d-linked/non-mirative preverbal wh-elements exhibit a different morpho-syntactic behaviour from canonical preverbal wh-elements in other Romance languages. Due to their clitic status, they do not undergo wh-movement on their own, but rely on the inflected verb for their realisation, mirroring very much the behaviour of preverbal clitics across NEIDs (see Benincà 1994). Non-d-linked/non-mirative preverbal wh-elements hence form a special class of wh-items that have clitic status.
In Sovramontino and Lamonat, non-d-linked/non-mirative preverbal wh-items and apparently in-situ (bare) wh-elements exhibit clear morpho-phonological and morphosyntactic differences. Non-d-linked/non-mirative preverbal wh-elements are generally phonologically reduced and, unlike apparently in-situ wh-elements, cannot receive focal stress: they are ultimately atonic and prosodically rely on the verb cluster. The inability of wh-clitics to receive focal prominence can be appreciated by the fact that wh-clitics can co-occur with an XP in narrow focus within the same sentence. Given that the semantics of wh-interrogatives, it is difficult to devise a suitable and clear context in which an element in narrow focus coexists with a clitic wh-element; however, if the focal XP is the subject of the clause, a cleft-question obligatorily arises: the subject in narrow focus and the wh-clitic can coexist, as shown in (63): Quan-e-lo MARIA che la a catà n om? when-be.3sg-scl expl Maria comp 3sg.f.scl have.3sg find.ptcp a husband 'When did Maria find a husband?' Conversely to Lombard varieties (see Manzini & Savoia 2011), Lamonat and Sovrmontino wh-clitics cannot co-occur with apparently in-situ wh-items of equivalent meaning, as shown by the ungrammaticality of (64): (64) Lamonat *Onde-son-e aonde? where-be.1pl-1pl.scl where 'Where are we?' The only exception is the wh-clitic sa 'what', which can optionally co-occur with the equivalent tonic postverbal wh-item che, as shown in (65) (65b) is comparable to Italian che cosa. The lexicalisation of the two wh-items is not an instance of a multiple wh-question, but the wh-clitic and the tonic wh-item are semantically interpreted as a single unit. Wh-doubling configurations are common across NIDs (see Manzini & Savoia 2005;Bonan 2019), but only optionally attested with sa in Lamonat and Sovramontino. Again, note that the co-existence of the clitic wh-item and the postverbal tonic wh-item is only possible because the former does not carry focal promi-nence. All other non-d-linked/non-mirative preverbal wh-elements (i.e. wh-clitics) show a strict syntactic distribution that is complementary to that of tonic postverbal wh-elements.
Let us consider examples (66), (67), (68) and (69) (69) show that onde and aonde are not interchangeable. Onde can only appear pre-verbally, while purely focal aonde (i.e. non-d-linked/non-mirative) can only appear in apparently in-situ position. Onde is the clitic form of its tonic counterpart aonde. Besides its atonic status and the impossibility of carrying focal prominence, another straightforward piece of evidence can be put forward in support of the clitic status of onde: onde cannot be used in isolation, while aonde can (e.g. 'where?' to ask for a precise location). 20 Across NIDs, clitic wh-elements are robustly attested (Manzini & Savoia 2005;Poletto & Pollock 2009;. Poletto and Pollock (2009; put forward a three-way 20 Note, however, that under Merchant's approach to ellipsis (2001; whereby wh-fragments/wh-phrases in sluicing are derived through the elision of the head located under FocusP in the C-domain, we would not expect focal tonic wh-items to be able to appear in isolation. This prediction is partially borne out. In a context like Speaker A: 'I saw Toni', Speaker B is very likely to reply the Lamonat wh-question in (i): (i) L-a-tu vedù aonde? 3sg.m.ocl-have.2sg-2sg.scl see.ptcp where 'Where did you see him?' In (i), the complex T° precedes the focal wh-item. In this respect, Merchant's prediction is borne out. Nonetheless, if Speaker B still does not get the place where Speaker A saw Toni, Speaker B would utter aonde? in isolation. It is important to note that aonde used in isolation very often resembles the intonation of an echo-question: a different structure, for instance, may be in place. I will not further discuss the issue in this paper. Nevertheless, I want to point out that clitic wh-items like onde can never be used in isolation. If it is used as a fragment, it must be cliticised onto the complex T°. For instance, Speaker A: 'I saw Toni', Speaker B must utter the wh-question in (ii) with the wh-clitic onde cliticised onto the complex T°: (ii) Onde-l-a-tu vedù? where-3sg.m.ocl-have.2sg-2sg.scl see.ptcp 'Where did you see him?' If Speaker B still does not get the location, Speaker B cannot utter the clitic wh-item onde with an echointonation, but must resort to its tonic counterpart aonde. distinction between tonic, clitic and weak (à la Cardinaletti & Starke 1999) wh-elements. In this paper, I will only distinguish between clitic and tonic wh-words: a two-way distinction is, in fact, sufficient to account for the distribution and morpho-syntactic behaviour of non-d-linked/non-mirative preverbal wh-elements in Lamonat and Sovramontino (see also Manzini & Savoia's 2011 criticism on weak wh-elements). Munaro and Poletto (2014) acknowledge the clitic status of some wh-elements across NIDs, exhaustively discussing the diachronic development of the wh-word where across NIDs. In order to account for the tonic-clitic divide across NIDs, they reconstruct the internal layering of the wh-word where based on its morphological composition. They claim that the wh-item where is made up of three sub-elements: i) a prepositional formative, which corresponds to either one of the following prepositions in 'in', di 'of', and da 'from', or a combination of them; ii) a vocalic formative o/u namely the wh-operator feature that developed from Indo-European wh-formative qw; and iii) a deictic locative element derived from various sources -nd, -v, or -la, expressing locative deixis. Following Munaro and Poletto's (2014) analysis, in Table 2, I show the internal make-up of the tonic wh-word aonde in Lamonat and Sovramontino.
In light of Table 2, in its diachronic development, the clitic wh-element onde lacks the prepositional layer, resulting in a reduced morpho-phonological form. Also, the epenthetic final vowel -e is absent if the inflected verb to which it attaches starts with a vowel. Based on Poletto and Munaro's (2014) discussion of the diachronic development of clitic and tonic where across NIDs, we can safely claim that a clitic wh-form of where is also present in Friulian. This will be particularly important in Section 6 where I will compare the derivation of wh-questions in Friulian and Lamonat/Sovramontino. Much like onde in Lamonat and Sovramontino, Friulian clitic la 'where' lacks Poletto and Munaro's (2014) prepositional layer (compared to its tonic counterpart dulà 'where'). The wh-questions in (70) and (71) are respectively examples of Friulian clitic form of 'where' la and its tonic counterpart dulà: Friulian La-is-al Jacum? where-be.3sg-3sg.m.scl James 'Where is James?' Dulà is-al Jacum? where be.3sg-3sg.m.scl James 'Where is James?' The Friulian clitic wh-form la exhibits the same behaviour as onde in Lamonat and Sovramontino: it cannot appear in isolation and, phonologically, relies on the inflected verb to which it attaches. In addition, both la and dulà cannot co-occur in the same sentence, as shown by the ungrammaticality of (72): (72) *Dulà la-is-al Jacum? where where-be.3sg-3sg.m.scl James Besides where and what, Lamonat and Sovramontino exhibit a full set of clitic wh-items, as shown in Table 3 along with their tonic counterparts. Clitic wh-forms systematically appear preverbally. Tonic wh-forms can only appear preverbally if they carry a salient discourse-pragmatic reading (i.e. d-linked or mirative). It is important to note, however, that how, how much, who and, in Sovramontino only, what have the same form in the tonic and clitic paradigm. The clitic wh-forms are nonetheless unstressed and tend to be phonologically reduced. For example, the clitic form of 'how much' loses its final -t when the following inflected verb starts with a consonant, but retains it if the following verb starts with a vowel, as shown in (73) and (74)  These phenomena are not attested with their tonic counterparts, which do not show any sensitivity to the following phonological environment and always appear in their full form in apparently in-situ position (or preverbal position if they bear a d-linked or a mirative interpretation, see Section 5.3).
The clitic nature of non-d-linked/non-mirative preverbal wh-elements is primarily responsible for their preverbal position. In the syntactic treatment of wh-clitics, I adopt Kayne's (1991) generalisation that clitics attach to the heads of functional categories. In this regard, I propose that wh-clitics attach to the verb cluster and hence surface in the locus of SCLI. The behaviour of wh-clitics is hence comparable to that of satellite clitics across NEIDs (see Benincà 1994). In Lamonat and Sovramontino, wh-clitics cliticise onto verb cluster, which undergoes T-to-Force movement, surfacing thus in preverbal position, as shown in (76a) and (76b) below: As argued in Section 4.1.4, the well-formedness of the question is assured by SCLI alone, which licenses a silent wh-operator in SpecForceP through dynamic agreement (Poletto 1993;. It is reasonable to assume that clitic wh-elements are also equipped with a [WH] feature and hence are able to lexicalise the wh-operator in SpecForceP. However, their clitic status is evidence against such analysis. I instead propose that wh-clitics cliticise onto the verb cluster under the complex T°, which undergoes T-to-C movement to the locus of SCLI, yielding their preverbal position. Having accounted for the different distribution between apparently in-situ purely-focal wh-elements and non-d-linked/non-mirative preverbal wh-elements (i.e. wh-clitics), in the next section, I will discuss the last type of wh-items found in Lamonat and Sovramontino: preverbal tonic wh-elements, which carry either a d-linked or a mirative interpretation.

The discourse-pragmatically salient nature of tonic preverbal wh-items
Let us start our discussion by accounting for the behaviour of d-linked wh-items (Pesetsky 1987). I have in fact not yet proposed an analysis for the type of wh-questions in (77)  In (77) and (78) a tonic wh-element is fronted. (77) and (78) seem to challenge the analysis of the behaviour of wh-elements in Lamonat and Sovramontino that I have developed so far: the data, in fact, show that tonic wh-elements can appear in preverbal position (as well as in apparently in-situ position). However, in (77) and (78), the preverbal position of the tonic wh-elements is systematically justified by their d-linked interpretation (in the sense of Pesetsky 1987;. It follows that the preverbal position of tonic wh-elements is not identical to the preverbal position of clitic wh-elements, but it is a discourse-pragmatically salient structural position. In (77), which jumper refers to a limited set of jumpers whose existence and characteristics are known by the interlocutors: it is part of the knowledge shared by speaker and hearer. Similarly, in (78), what refers to a limited set of items that the speakers can choose from: implicitly, the speakers know what these items are. De Cia (2018) provides an analysis of d-linked wh-elements in Lamonat and Sovramontino. He claims that, by virtue of being equipped with a [WH] feature and a [CONTRASTIVE] feature, d-linked wh-elements can occupy SpecForceP, which is a privileged structural discourse-pragmatic position in Lamonat and Sovramontino where contrastiveness is satisfied. It is important to note that following Molnár's (2002; arguments on contrastive focus, De Cia (2018) does not conceive contrastiveness as a categorical discourse-pragmatic notion, but as a continuum. The degree of contrastiveness is determined by the properties of the set containing the contrastive element with respect to its size and the accessibility of alternatives. In the contrastiveness continuum, De Cia (2018) argues that d-linked wh-elements are placed in the middle, serving the membership in a finite set of entities discourse-function. This is particularly true of lexically restricted wh-items of the type 'which X', which, not surprisingly, virtually always trigger such discourse-pragmatic reading. The d-linked interpretation of a wh-element is tightly linked to common ground management (Krifka 2007;Krifka & Musan 2012): the landing position of a d-linked wh-element is hence a discourse-pragmatically salient one.
De Cia (2018) notes a key difference between non-d-linked wh-elements and their d-linked counterpart: d-linked wh-elements cannot be clefted, as shown in (79b) The postverbal position of che in (79b) makes it impossible for the tonic wh-item to encode a d-linked interpretation: che in (79b) can only have a focal exhaustive interpretation as regular clefted tonic wh-items. In order to be interpreted as a d-linked wh-item, the whelement must appear out of the cleft in preverbal position, as shown in (79a): it cannot be realised in apparently in-situ position, but must move to the left peripheral position, in which it can check its [CONTRASTIVE] feature, namely in SpecForceP. The same tonic wh-item (being it bare or prepositional) can hence appear either preverbally or postverbally, depending on its discourse-pragmatic interpretation as either purely focal (postverbal position) or d-linked (preverbal position), 21 as shown in (80): Lamonat a. Laore-lo con chi to pare? Non-d-linked interpretation work.3sg-3sg.m.scl with who your father 'With whom does your father work?' b. Con chi laore-lo to pare? D-linked interpretation with who work.3sg-3sg.m.scl your father 'With whom (of these people) does your father work?' Crucially, the preverbal position of the prepositional wh-item in (80b) triggers a d-linked interpretation, which can be paraphrased as 'with whom of a previously mentioned/presupposed set of people does your father work?'.
The discourse-pragmatic specialisation of the preverbal position, which, according to De Cia (2018) is also activated in declarative sentences hosting a contrastive XP, is a residual property of Lamonat and Sovramontino's Force-V2 system (in the sense of Wolfe 2016): the V2 prefield has been reanalysed as a discourse-pragmatically salient position that hosts contrastive XPs (De Cia 2018). As far as root interrogatives are concerned, we have claimed that SpecForceP is occupied by a wh-operator, which satisfies the wh-criterion and derives from dynamic agreement (Poletto 1993;. This position, however, is available to d-linked wh-elements by virtue of carrying both a [WH] and a [CONTRASTIVE] feature. The [WH] feature allows the d-linked wh-element to lexicalise the wh-operator. Purely contrastive XP (equipped with a [CONTRASTIVE] feature, but lacking a [WH] feature) are indeed banned by this position in root interrogatives. 22 Note that, given the structural position of the preverbal wh-item and the verb cluster, it follows that nothing can intervene between the d-linked wh-item and the verb cluster. This prediction is indeed borne out in Lamonat and Sovramontino. As previously mentioned, the only non-interrogative XPs that can appear preverbally in root interrogatives are scene-setting elements that are base-generated in a projection higher than ForceP (Poletto 2000;Benincà & Poletto 2004;Greco & Haegeman 2016). As a consequence, it is possible to have a frame-setter preceding a preverbal wh-item, but nothing can intervene between the wh-item and the verb cluster. The special behaviour of this type of wh-elements in Bellunese is known in the literature, but has not been fully addressed in terms of discourse-pragmatics. Poletto and Pollock (2004) claim that d-linked wh-elements are not licensed in focal postverbal position, but higher in the left-peripheral space. Munaro (1998) notes that, in Bellunese, wh-elements that are structurally complex, like (77), are not licensed in apparently insitu position, but must appear preverbally. In terms of frequency, d-linked wh-elements are more often part of a complex DP ('which X'), than bare wh-words. As a consequence, d-linked wh-elements are often more structurally complex than non-d-linked wh-words; however, this is not always the case, as shown in (78). I claim that the internal structural complexity of the tonic wh-word does not predict its syntactic position within the clause: what determines the preverbal position of a tonic wh-element is its salient discoursepragmatic interpretation.
Two predictions follow from our discussion so far: (i) lexically restricted wh-items that do not carry an immediate d-linked interpretation should appear postverbally in apparently in-situ position, (ii) aggressively non-d-linked wh-elements (see Pesetsky 1987) should appear postverbally, as, de facto, they lack a d-linked interpretation. While the first 22 Consider the Lamonat sentence in (i): (i) No se che majon che l a comprà neg know.1sg which jumper that 3sg.m.scl have.3sg buy.ptcp 'I don't know which jumper he bought'.
In this case, 'which jumper', che majon, is followed by an obligatory complementiser, che, that introduces the embedded interrogative. I assume that che majon fills the specifier position of the head hosting che. It is important to note that a d-linked/non-d-linked asymmetry is not attested in embedded contexts, but all tonic wh-items precede the overt complementiser che. I take this asymmetry as evidence in favour of the V2-related nature of SpecForceP as a discourse-pragmatically salient structural position.
prediction is fully borne out, the second needs some further discussion and the introduction of the concept of mirativity. As for (i), lexically restricted wh-items without a d-linked interpretation must appear postverbally, as shown in (81): (81) Sovramontino A-tu ledhest quanti libri l an pasà? have.2sg-2sg.scl read.ptcp how.many books the year past 'How many books did you read last year?' In (81) no d-linked interpretation is available, hence the postverbal position of the lexically restricted wh-item. Note, however, that the same lexically restricted wh-element must instead surface preverbally if it bears a d-linked interpretation, as shown in (82): Mi o ledhest diese libri, e ti? Quanti libri I have.1sg read.ptcp ten book and you how.many books a-tu ledhest? have.2sg-2sg.scl read.ptcp 'I read ten books, and you? How many books did you read?' In (82), there holds a contrastive relationship between the number of books read by the speaker and the number of books read by the hearer. The latter is encoded in the lexically restricted wh-item, which must thus appear preverbally.
As for (ii) aggressively non-d-linked wh-items, they can unexpectedly appear in both preverbal and postverbal position. It is possible to identify two categories of aggressively non-d-linked wh-items: (i) those roughly translating into 'what the hell' (che dal ostia/ che ostia, che dal demonio/che demonio), and (ii) those translating roughly into 'what the fuck' (che mona, che cazzo). The first group can surface in either preverbal or postverbal position, whereas the second group seems to be only felicitous in preverbal position. In Lamonat and Sovramontino, these wh-items are mainly uttered when the speaker wants to convey surprise, disappointment, incredulity or anger. This is especially true of the second group of aggressively non-d-linked wh-items. This aspect brings us to the last type of discourse-pragmatically salient tonic wh-items that can appear in preverbal position: wh-XPs encoding mirativity (in the sense of Cruschina 2012). The possibility of aggressively nond-linked wh-items to surface preverbally is hence not justified by the fact that they carry a [CONTRASTIVE] feature (in fact, they cannot as they are by definition non-d-linked), but because of their mirative interpretation (see Cruschina 2012). I here therefore propose two possible analyses for aggressively non-d-linked wh-items (as well as more generally tonic wh-items) that carry a [MIRATIVE] feature and hence must surface in preverbal position. Either (i) we expand on De Cia's (2018) proposal and argue that the residual V2 prefield is a generalised discourse-pragmatically salient position that can host both contrastive (i.e. d-linked) and mirative XPs endowed with a [WH] feature, or (ii) we assume that mirative XPs land in a dedicated functional projection (à la Bianchi et al 2015;, which, according to Bianchi et al. (2015;, is located higher than ForceP. Both analyses correctly predict that no XP can intervene between the wh-item and the verb cluster. Given that d-linked wh-items cannot co-occur with a mirative XP, I believe that adopting an analysis whereby the V2 prefield has been reanalysed as a salient discourse-pragmatic position that can host contrastiveness and mirativity is more empirically and theoretically sound. I will nonetheless leave this point open for debate. Going back to aggressively non-d-linked wh-items, it is important to note that if a mirative interpretation is not in place, they must appear postverbally. Mirativity is not only a discourse-pragmatic reading available to aggressively non-d-linked wh-items, tonic bare and prepositional wh-items can also have a mirative interpretation. In such case tonic wh-items must surface preverbally to satisfy their [MIRATIVE] feature. Let us consider the wh-question in (83) The wh-question in (83) features a tonic wh-item (aonde) and was uttered by an angry mother, whose son was late for lunch. In this particular context, (83) can be paraphrased as 'where the hell have you been?!' employing an aggressively non-d-linked wh-item. Given the discourse-pragmatic context, the wh-element in (83) does not land preverbally because of a d-linked interpretation: its preverbal position is instead due to its mirative interpretation. Figure 2 summarises the overall distribution of wh-items in Lamonat and Sovramontino. 23 Having analysed and discussed the behaviour of clitic and tonic wh-elements, and the effect of information structure on tonic wh-items, it is time to take the investigation a step forward with the following question: why are apparently in-situ wh-elements only present in certain NEIDs and not in others? By comparing Lamonat and Sovramontino with Friulian, I will claim that wh-in-situ is the result of a deeper microparametric variation across NEIDs that involves the locus of SCLI.

Microparametric variation in the locus of subject clitic inversion
Let us start our discussion by considering the Friulian yes/no question in (84) and its Lamonat equivalent in (85) The preverbal position is only licensed by a mirative interpretation. Given the special properties of 'why', its postverbal position, however, may not be the same as that of purely focal wh-items, but it may be externally merged into the C-domain in a different functional projection (see Rizzi 2001). 24 As argued in Section 4.1.1, note that the word order in (i) below is also grammatical in Friulian: In contrast to Lamonat and Sovramontino, the auxiliary can be morpho-phonologically independent of the past participle and undergo T-to-C movement on its own. I assume that, in Friulian, if the past participle does not cross the subject position, it stays within TP: V-to-T movement of the past participle does not take place. In a nutshell, I will argue that, in Lamonat and in Sovramontino SCLI fails (cf. 85a) because of the intervening strong negation that is able to act as host of Q under Force°, namely under the locus of SCLI. In Friulian, on the other hand, the locus of SCLI involves a left peripheral projection lower than that in which the strong negation checks its [NEG] feature; as a result, in Friulian, SCLI is not affected by the intervening strong negation. As shown by the ungrammaticality of (85b), in Lamonat and Sovramontino, SCLI is impossible in root negative-interrogatives. In Friulian root negative-interrogatives, SCLI instead exhibits the exact opposite behaviour: in (84a), SCLI takes place regardless of the presence of the preverbal strong negation, and, if SCLI does not take place, the sentence is ungrammatical (cf. 84b). Let us now briefly explain the assumptions around the ungrammaticality of SCLI in root negative-interrogatives in Lamonat and Sovramontino. Consider the following negative sentences in (86) and (87) The negation in the root declarative in (86) is not prosodically marked and allegedly occupies a syntactic position within the TP layer that is below the subject Mario. On the other hand, the negation in (87) is prosodically marked and has scope over the whole sentence. In this respect, the negation in (87) is an instance of sentential negation whose realisation is rooted in the C-domain above the TP-layer (see Haegeman 1995;Zanuttini 1997;Horn 2001). In order to distinguish the negation in (87) from the negation in (86), I will henceforth refer to the former as strong negation, which must be solely interpreted as a descriptive label that identifies the negation employed in a negative yes/no question. In (87), the constituent order is altered, yielding T-to-C movement. Given the negative-interrogative nature of the sentence in (87), the strong negation must check its [NEG] feature in the C-domain (Zanuttini 1997;Poletto & Zanuttini 2013). I assume that this is done in Polari-tyP (see Laka 1990), which is located lower than ForceP and higher than FinP in the left peripheral space. Once the negation has checked its feature, it moves to Force° where it adjoins to Q. I claim that Q is satisfied by the adjacency with negative element: the intervening negative element hence blocks SCLI, as the strong negation itself is a suitable host for Q. To some extent, the present analysis resembles Zanuttini's (1997:44) account of the behaviour of the negation in root negative-interrogatives in Paduan. In Paduan, much like Lamonat and Sovramontino, SCLI is blocked by the intervening negation: I will argue, instead, that the negative marker itself (or, perhaps, its features) moves to C° in negative yes/no questions […] for concreteness, I will assume that the yes/no operator is in the specifier of CP and that no can check the features of C° either by adjoining to it in the syntax or by LF-raising of its features (Zanuttini 1997: 44).
The crucial difference with Zanuttini's (1997) account is that she does not assume T-to-C movement in root negative-interrogatives in Paduan: she claims that only the negation moves to the left peripheral space and the negation alone satisfies the wh-criterion (Rizzi 1991). Lamonat and Sovramontino's constituent order in negative-interrogatives and the data from negative-interrogatives in Friulian (84) suggest that, across NEIDs, T-to-C movement does take place in negative-interrogatives allegedly to the head of PolP. I assume that, in such case, T-to-C movement ensures (i) the satisfaction of the yes/no operator and (ii) verb cluster adjacency with the strong negative element, which, in light of its focallike properties (see Villa-García 2016), receives discourse-prominence prosodically. 25 The derivation of (87) is schematically represented in (88) (84) and (85)  Note that SCLI inversion does not take place; nonetheless the subject clitic appears proclitically. This is important because supports Goria's (2004) view on the licensing of interrogative subject clitics. SCLI does not indistinctively take place when T-to-C movement occurs, but when such movement satisfies the interrogative nature of the sentence (T° adjoins to Q under the locus of SCLI). This is also why, for example, SCLI does not take place when T-to-C movement satisfies focus adjacency requirements in root declaratives in the NEIDs under investigation (see De Cia 2018). In such cases, I assume that the SCL is adjoined to T°.
The presence or absence of SCLI in negative-interrogatives can hence be explained in terms of the microparametric variation in the locus of SCLI: Force° in Lamonat and Sovramontino, and Fin° in Friulian. In sum, assuming that the negation in a yes/no question is licensed in PolarityP (Laka 1990), which is located lower than ForceP, but higher than FinP, in Friulian T° is attracted to Fin° by Q: SCLI takes place before the negation checks its [NEG] feature in PolarityP. In Lamonat and Sovramontino, on the other hand, the negation is licensed in PolarityP before T° reaches Force°: the negation itself is then able to support Q and SCLI is not triggered (via verb movement to Force°). The position of PolarityP, and the difference in the locus of SCLI between Friulian and Lamonat/Sovramontino are schematically shown in (90) If we instead assume an account whereby the strong negation in negative-interrogatives has purely focal properties and hence must check a [FOCUS] features in FocusP (à la Villa-García 2016), the intervening effect of the negation on SCLI can still be appreciated, as FocusP, the same as PolarityP, is lower than ForceP, but higher than FinitenessP.
Let us now consider the Friulian wh-questions in (91) and (92) and their Lamonat counterparts in (93) and (94) In Section 5.2, I claimed that the Friulian wh-word dulà (cf. 92) is a tonic wh-element that bears focus prominence: it has the same morpho-syntactic status as aonde in (94). The corresponding clitic form of dulà is the wh-clitic la in (91): la behaves like onde (cf. 93) in Lamonat and Sovramontino. By assuming a one to one correspondence between Friulian and Lamonat/Sovramontino clitic and tonic wh-elements, dulà surfaces in Spec-FocP, whereas its clitic counterpart la is cliticised onto the verb cluster in the locus of SCLI. If the Friulian wh-elements in (91) and (92) morpho-syntactically behave like their Lamonat/Sovramontino counterparts in (93) and (94), why do they exhibit a different linear order? The difference in linear order can be accounted for by assuming the aforementioned microparametric variation in the locus of SCLI: Force° in Lamonat and Fin° in Friulian. In Friulian, since SCLI takes place under Fin°, the tonic wh-element dulà and the clitic wh-element la both appear preverbally. In reality, the wh-clitic la appears cliticised onto the verb cluster in Fin°, while the tonic wh-word dulà surfaces in SpecFocP. The identical derivation of Friulian dulà and Lamonat/Sovramontino aonde as purely focal tonic wh-elements in (92) and (94) becomes more transparent when the two sentences are shown under the same arboreal representation (cf. 95). Note that Friulian also exhibits SCLI in root interrogatives, hence, through dynamic agreement, the wh-element has no need to land in the specifier position of the head hosting the moved verb (i.e. SpecFinP). The tonic wh-element can thus freely satisfy its discourse-pragmatic function, namely narrow focus, landing in the specifier position of FocusP. This is shown in (95)  As a result, Lamonat and Sovramontino's apparent wh-in-situ is solely the manifestation of a parametric choice that involves the locus of SCLI, namely the left peripheral head that hosts Q in root interrogatives and where SCLI surfaces. In this respect, Lamonat and Sovramontino should not be treated as wh-in-situ languages, but as proper wh-fronting languages on par with Friulian.
Another piece of evidence in support of the proposed microparametric variation in the locus of SCLI comes from information structure, more specifically, from the limitation on the number of preverbal topics in root interrogatives. For the present investigation, we are interested in the number, rather than type of topics that can appear preverbally. Let us consider the Friulian sentence in (97): (97) Friulian Iar, t-al bosc a-tu viodût to barbe? yesterday in-the woods have.2sg-2sg.scl see.ptcp your uncle? 'Did you see your uncle in the woods yesterday?' Given the recursive nature of topics (Rizzi 1997;Benincà & Poletto 2004), Friulian's C-domain can theoretically host an unlimited number of topics in a yes/no question like (97). On the other hand, in the exact same yes/no question, Lamonat and Sovramontino can only host a single topic preverbally that functions as a frame-setter (Poletto 2000;Benincà & Poletto 2004;Greco & Haegeman 2016), as shown in (98) Bona fide topical elements are not licenced in the C-domain of root interrogatives in Sovramontino and Lamonat. In Lamonat, the same linear order as (97) would yield an ungrammatical sentence, as in (98a), which can be rescued by realising the presupposed XPs in-situ within the TP layer, as in (98b). Again, this difference can be accounted for by assuming a micro-parametric difference in the locus of SCLI across the NEIDs under investigation: since Friulian SCLI takes place in Fin°, there is room for a more articulated left periphery. By contrast, in Lamonat, the presence of one or more topics would impede the verb cluster in its movement to Force°, as TopP is an unsuitable host for the transit of T-to-C movement (see Rizzi 1997). 26 Finally, the satisfaction of Q under either Force° or Fin° is not random or arbitrary, but is rooted in the diachronic development of the NEIDs under investigation. Starting from the assumption that T-to-C movement in root interrogatives across NEIDs is a vestigial V2 trait (Rizzi 1991;1996;Salvi 2016 a.o), verb movement in root interrogatives directly reflects where, within the C-domain, the V2 constraint was satisfied during their V2 stage (see Benincà 2006;Ledgeway 2008;Salvi 2004;. The present microparametric variation in the locus of SCLI suggests that V2 was satisfied in different left-peripheral projections, namely ForceP in Lamonat and Sovramontino, and FinP in Friulian. Crucially, Wolfe (2016) independently argues that Medieval Romance varieties could either satisfy V2 in Fin° or Force°. The residual V2 properties of the NEIDs under investigation supports Wolfe's (2016) V2 parameterisation.

Conclusion
In conclusion, I have shown that wh-in-situ in the Bellunese varieties of Lamon and Sovramonte is not a genuine phenomenon, but the result of a deeper microparametric variation that concerns the locus of SCLI across NEIDs. In line with Munaro (1998;, Poletto and Pollock's (2004;2009) account of Bellunese, I argued that the postverbal wh-item does not truly appear in-situ in TP-internal position, but has undergone wh-movement to the left periphery of the clause. Lamonat and Sovramontino cannot hence be considered wh-in-situ languages, but wh-fronting languages on par with Friulian. Differently from NIDs that exhibit genuine TP-internal wh-in-situ, Lamonat and Sovramontino do not show true optionality between the (apparently) in-situ and the fronted wh-position. The distribution of wh-items between the preverbal and postverbal position can be fully constrained on the basis of (i) the morpho-syntactic status of the wh-item as clitic or tonic, and (ii) its discourse-pragmatic interpretation as focal, d-linked or mirative. Wh-items with a purely focal interpretation must appear postverbally, whereas wh-items with a mirative or a d-linked interpretation must surface preverbally. As for wh-clitics, they are not affected by information structure and surface preverbally cliticised onto the verb cluster. The case of Lamonat and Sovramontino has important cross-linguistic implications for the study of wh-syntax: it shades light on the pervasiveness of discourse-pragmatics in content question formation and points towards a typology of wh-items that is characterised by different degrees of sensitivity to information structure. Assuming that the degree of pervasiveness of discourse-pragmatics in wh-syntax is language specific, content question formation should be considered a discourse-pragmatic phenomenon as well as a syntactic one.