1 Introduction

Most northern Italian varieties,¹ unlike languages such as English or Italian in (1a), allow the co-occurrence of WH elements and complementizers, especially in embedded questions such as (1b).

The co-occurrence of complementizers and WH elements (for short, WHs) within the same functional projection, as shown in (2), was dubbed Doubly-Filled COMP (DFC) by Chomsky & Lasnik (1977), who proposed a language-specific Filter to explain why DFC is barred in languages such as (1a).

Evidence from sluicing in (3) suggests that WHs and Cs are not merged in the same functional projection (Lobeck 1995; Baltin 2010) as Cs must be deleted along with the rest of the clause, as illustrated in (4).²

The analysis in (4)—or any variant thereof—is confirmed by evidence from DFC-dialects that marginally allow the occurrence of functional elements, such as question particles or subject pronouns, between WH and C (Poletto & Vanelli 1995: 153; Dagnac 2018: 10–11; Manzini & Savoia 2005, vol. I: 520).

This evidence challenges the syntactic representation of DFC in (2), but it does not necessarily contradict the DFC-Filter hypothesis, as neither the QP nor the CP in (4) is doubly filled. However, by general consensus, the DFC-Filter hypothesis—understood as a generalized condition on syntactic structures—should be abandoned in favor of a model where DFC arises from the featural endowment of functional items.

This paper will suggest an analysis of embedded DFC that relies on Bayer’s (2015) hypothesis that WHs can carry a {C} feature. Departing from Bayer & Brandner’s (2008a; b) and Bayer’s (2015) analysis of Bavarian varieties, I will account for the Italo-Romance data in light of a slightly revised version of Cecchetto & Donati’s (2010) labeling algorithm.

To lay the groundwork for a principled account, we must first clarify the following aspects, as misinterpretations could lead to erroneous conclusions:

• - Is DFC one and the same phenomenon in every clause type and in every variety? More specifically, what is the relationship between embedded and root DFC? Are they independent phenomena? Does root DFC originate from a reanalysis of embedded DFC?

• - Is there any significant association between the incidence of DFC and the type of WH? Should we analyse the complementiser in DFC as part of the WH item?

• - What is the relationship between DFC and clefts? Can DFC result from the reanalysis of clefts? Are clefting and DFC independent or mutually-exclusive phenomena?

• - Subjunctive clauses often display distinctive phenomena involving complementizers, such as complementizer deletion or recomplementation. Does mood influence the incidence of DFC? Can DFC be analyzed in parallel with recomplementation phenomena?

The above research questions will be answered through a quali-quantitative analysis of data extracted from the Syntactic Atlas of Italy (ASIt, see http://svrims2.dei.unipd.it:8080/asit-maldura/pages/search.jsp [last viewed: 30.11.24]).

The article has a bipartite structure. The first part presents some quantitative data on the distribution of DFC across Northern Italian regions³ and elaborates on the interplay between DFC and other syntactic phenomena: Section 2 contains some preliminary remarks on the nature of microvariation data and on the dataset; Section 3 focuses on the distribution of DFC in main and embedded interrogatives; Section 4 examines the incidence of DFC with respect to various types of WHs; Section 5 compares the distribution of DFC and clefts; Section 6 shows that there is no association between DFC and mood; Section 7 examines the distribution of DFC in adjunct and relative clauses; Section 8 resumes the findings of the previous sections. The second part of the article deals with the analysis of DFC. Section 9 elaborates on the role of Light-Headed Relatives (Citko 2004) in the emergence of DFC (Munaro 2000[2001]; Garzonio 2007; Poletto & Sanfelici 2019; 2021); Section 10 entertains the hypothesis that DFC results from the feature endowment of WH elements. Section 11 concludes.

2 Microvariation data: handle with care

According to Kayne (1996: xiii): “the technique of examining a large number of very closely related languages promises to provide a broad understanding of parameters at their finest-grained (microparameters).” In this respect, big collections of existing data, such as atlases and databases, are a valuable source of information. The material contained in such primary sources, however, should be handled with care. Ideally, linguists should rely on a significant sample of data that are elicited with a mixed methodology from a group of speakers that is representative of a specific datapoint. In practice, microvariation analysis is based on limited evidence, often drawn from a small sample of speakers or, in some cases, a single speaker. Additionally, most linguistic atlases and databases contain data that were collected mainly through questionnaire-based interviews, by eliciting out-of-context examples. Lacking any insight into the pragmatic and discourse conditions that the subject-speaker had in mind while translating the questionnaire, we are not entirely sure whether the elicited material is a faithful translation or not. Interrogatives are a case in point: given the same item of the questionnaire, some speakers may have produced an information-seeking question, while others may have produced a special question with a flavour of surprise, disbelief or reproach. It is well known that special questions may differ from information-seeking questions in syntactic respects (Obenauer 2004; 2006), but pragmatic factors were seldom controlled for.

Furthermore, the data may be biased by the language of the questionnaire. This holds particularly true for phenomena, such as DFC, that are attested in the dialects, but not in the language of the questionnaire. Translations may be primed by the language of the questionnaire, resulting in the underrepresentation of DFC in the dialect translations.

Task-related biases are not the only source of unsystematic variability in the sample. Consider that Italian, i.e. the official language, and Italian varieties are not clearly distinct linguistic systems in the competence of speakers. Speakers always mix or switch from one language to the other and the reliability of elicited data varies depending on the metalinguistic awareness of subjects. This leads to a certain degree of inter-subjective variation, which is particularly evident in major cities, where the bilectal competence of speakers is more endangered than in the rural areas. Take for instance the data in (5) and (6), elicited in the city of Vicenza, where two informants differ with respect to DFC. Patterns of this kind are rather common, in particular across generations of speakers.

The main issue when dealing with microvariation data is precisely how to distinguish structural/systematic variation from sociolinguistic/unsystematic variation, given the limited empirical evidence at our disposal. In my view, a combined qualitative and quantitative analysis is the best approach to examining data from existing atlases and databases, as it helps filter out idiosyncratic variation. As previously mentioned, certain examples may contain code mixing/switching, certain speakers are untrustworthy (namely, their degree of linguistic self-awareness is not sufficient to undertake a translation task), certain data are collected with more attention than others, certain interviewers have a better knowledge of the surveyed variety than others, certain questionnaires are better designed than others, etc.

Notice that quantitative analyses are not supposed to replace more traditional qualitative research, which remains the best way to examine single varieties or specific phenomena, if the documentation is thorough and fine-grained. However, I believe that some quantitative analysis is needed to (re)exploit the data that have been collected so far.

This work focuses on the data contained in the database of the Syntactic Atlas of Italy (ASIt, formerly ASIS, Benincà & Poletto 2007). A corpus of direct and indirect interrogatives was extracted from the ASIt database in November 2023. The corpus contains 2546 embedded WH interrogatives and 9583 main interrogatives elicited in 169 datapoints from seven regions, see Table 1. The corpus is formed by dialect translations of Italian questions (16 indirect questions and 59 direct questions) contained in the ASIt questionnaires. The distribution of the datapoints across regions is not homogeneous, as shown in the following table.

Comparisons are made with the data contained in Manzini & Savoia (2005, Vol. I: 388ff) on approximately 200 datapoints.

The examples stored in the ASIt database were collected since the early 80’s by a team of linguists and their research associates. Some ASIt questionnaires were distributed by mail and completed in written by trained linguists. The ASIt data were transcribed by adapting the grammatical conventions of Italian. Despite its variability, the transcription system does not undermine the value of these data for syntactic analysis.

For the purposes of the present study, the data contained in the ASIt were first mapped into numerical variables, which were organized into two spreadsheets, one containing main interrogatives, the other containing indirect questions (see Annex 1 and 2). Spreadsheets contain three kinds of metadata: 1) Source: sentence and datapoint identifiers, geographical coordinates, region/province; 2) Dependent variable: presence/absence of C in interrogatives; 3) Factors (independent variables): WH type, mood, presence/absence of clefting.

The main patterns that emerge from an exploratory analysis of this corpus are discussed in Sections 3–8.

3 DFC in main vs embedded clauses

The main goal of this section is to show that DFC is a feature of embedded WH clauses. The occurrence of the phenomenon in main clauses can be considered as an extension of the original pattern, probably triggered by the loss of subject-clitics inversion in main interrogatives (Section 3.1), and/or as a discourse-related phenomenon. This claim is crucial for the analysis proposed in Sections 9–10, which focuses on the conditions leading to the emerge of DFC in embedded WH interrogatives and free relatives.

First of all, it is worth noting that northern Italo-Romance DFC is a feature of finite WH clauses. DFC occurs neither in non-finite nor in in yes/no questions; the latter pattern is found, in embedded WH clauses, in Ibero-Romance and Picard varieties (Gallo-Romance; Dagnac 2012) exemplified in (7a) and (7b).⁴

The geographical distribution of DFC across northern Italy is shown in Figures 1 and 2. Different shades of gray represent the incidence of DFC in, respectively, direct and indirect questions (the darker the dot on the map, the higher the percentage of examples featuring DFC).

Figure 1: Incidence of DFC in direct interrogatives in 358 datapoints. Source: ASIt [November 2023] and Manzini & Savoia 2005, vol I.

Figure 2: Incidence of DFC in indirect interrogatives in 358 datapoints. Source: ASIt [November 2023] and Manzini & Savoia 2005, vol I.

The comparison between main and embedded clauses in specific linguistic sub-areas is illustrated in Figure 3. Ligurian varieties, in which DFC is absent or very marginal, will not be examined in the remainder of the article. DFC in main clauses is found predominantly in three discontinuous areas: Piedmont and Alpine Lombard, western Friulian, and southern Romagnolo. Embedded DFC is found in a wider area, including northeastern dialects such as Veneto (but excluding western varieties), Trentino, and eastern Friulian. The languages spoken in the central area of the Po plain, such as Emilian and southern Lombard, exhibit a lower incidence of DFC, especially in indirect questions.

The ASIt data confirm Poletto & Vanelli’s (1995) generalization that, in Italo-Romance, the dialects exhibiting DFC in main clauses are a subset of those exhibiting DFC in subordinates. Additionally, in every sub-area, the incidence of DFC is lower in direct than in indirect questions, with a few noticeable exceptions such as the dialect of Forlì in (8)–(10), where—according to the data from the ASIt—DFC is optional in indirect questions such as (8) and (9), but (almost) mandatory in direct questions such as (10).

Figure 3: Incidence of DFC in direct vs indirect questions per linguistic area. Source: ASIt [November 2023].

In light of the asymmetry between main and embedded clauses illustrated in Figure 3, Poletto & Vanelli (1995), Poletto & Sanfelici (2019; 2021) argue that DFC originated in embedded interrogatives and was later extended to main clauses. Therefore, languages with main DFC are expected to exhibit embedded DFC, but not vice versa.

Parry (2003) challenged this view showing that in Piedmontese DFC emerged in main interrogatives first and was later extended to embedded interrogatives. Parry’s hypothesis is that DFC was initially a feature of special questions, possibly borrowed from exclamatives such as (11), where DFC is found since the earliest attestations.

The pragmatic import of DFC in early Piedmontese might have been similar to the value of DFC in modern colloquial French,⁵ in which DFC occurs in contexts where speaker and addressee are in disagreement about the Question Under Discussion (Matuschat-Petersen & Faure 2024). According to Parry’s hypothesis, the symmetric pattern of DFC that is found in Piedmontese and the asymmetric pattern of DFC that is found in northeastern dialects such as Veneto, Trentino and Friulano did not necessarily originate under the same conditions.

However, the morphology of the Piedmontese what-WHs suggests that DFC emerged uniformly in embedded clauses, contra Parry (2003) (Munaro 2000[2001]). The WH for what in several Piedmontese and northern Lombard dialects is formed by a demonstrative head (e.g. Piedm. lo, lon, que [kwe, kye], Lomb. chel [kel] ‘that’) followed by a complementizer, as shown in (12).

This type of WH originated from the reanalysis of light-headed relatives such as (13a) (Citko 2004; more on this in Sections 9–10). Light-headed relatives, featuring a determiner-like element followed by a bona fide relative CP, can be embedded under interrogative predicates as shown in (13b): in Italian, French, and in the other northern Italo-Romance varieties (see Section 10), light-headed relatives alternate freely with what WHs. In the Piedmontese and Lombard varieties exemplified in (12), the sequence D+C was probably reanalysed as a WH form and eventually extended to direct questions.

The morphology of the WHs in (12) confirms the hypothesis that complementizers leaked from certain kinds of embedded clauses (light-headed relatives) into interrogative clauses in northwestern as well as in northeastern dialects (Poletto 2000; Poletto & Sanfelici 2019). In the latter, the spreading of DFC was confined to indirect questions (Poletto & Vanelli 1995), whereas in northwestern varieties orthogonal factors (see Section 3.1) led to a more symmetric diffusion. In this respect, we cannot exclude that an independent pattern of DFC existed ab origine (as suggested by Parry’s data), conveying a specific pragmatic meaning as in colloquial French (Matuschat-Petersen & Faure 2024).

4 DFC per Type of WH

DFC is subject to a certain degree of variation depending on the type of the co-occurring WH. The data show that the incidence of DFC is significantly lower when the clause is introduced by why. The lower incidence of DFC with why is relevant in light of the hypothesis, explored in Section 9, that DFC emerged from the reanalysis of free relatives (more specifically, from the reanalysis of light-headed relatives). The WH why, in fact, cannot introduce free relatives, whereas the other WH elements can introduce free relatives and adjunct clauses that can be analysed as free relatives (such as temporal and locative clauses). The lower incidence of DFC with why ultimately supports the hypothesis that free relatives were the cornerstone for the emergence of DFC in WH clauses (more on this in Section 10).

No other significant differences are found. The higher incidence of DFC in subject WHs found in Lombard dialects will be explained as a side-effect due to the incompatibility of DFC and in-situ WHs.

Let us review the data on indirect questions first. In embedded clauses (see Table 3, plotted in Figure 4), DFC is disfavored in combination with the WH element for why, while the ASIt data do not confirm that DFC is found more readily with subject WHs or with the WH for who.

Figure 4: Incidence of DFC in embedded WH interrogatives per Region and Type of WH element. Source: ASIt dataset [November 2023].

The ungrammaticality of DFC with why-WHs is illustrated in (19a), vs the pattern of DFC in (19b). Recall, however, that in certain dialects DFC is found in all types of indirect interrogatives, as shown in (20a) vs (20b).

Several explanations can be advanced to account for the lower incidence of DFC in combination with reason adverbs. First, perché ‘why’ + che ‘that’ sequences might be barred because the WH element has a transparent bimorphemic structure [per]+[ke], triggering the haplology of the following complementizer che [ke]. However, despite occasional evidence supporting this hypothesis, as the minimal pair in (21), there is no solid association between the morphology of why-WHs and DFC. In (21a), che can co-occur with the WH parcossa (lit. ‘for what’), which does not contain the formative [ke], while it cannot occur with the WH parché in (21b). Alternations of this kind, however, are not systematic.

Alternatively, we may suggest that the lower incidence of DFC with reason adverbial WHs depends on syntactic factors. In Section 10, I will argue that it results from the historical conditions that led to DFC. Why WHs are barred in free relatives (Cecchetto & Donati 2012), probably because reason adverbial WHs are externally merged to a non-focal position in the left periphery (Rizzi 2001), which is higher than the position hosting the other WH elements. Since free relatives are, in all probability, the context where DFC emerged (Poletto & Sanfelici 2019; 2021 a.o.), the lower incidence of why in DFC clauses is therefore expected.

In main clauses (see Table 4 and Figure 5), DFC with why is marginally attested in Piedmontese. It is worth recalling that embedded DFC with why is virtually unattested in all northern regions (see Table 3).

Figure 5: Incidence of DFC in main clauses per WH type and region. Source: ASIt [November 2023].

According to the data in Figure 5, DFC is particularly frequent with subject WHs in Lombard varieties. The data in Figure 6, however, suggest that this effect results from a competing syntactic phenomenon, namely WH-in-situ, exemplified in (22). WH-in-situ bleeds DFC, i.e. Cs are merged if and only if a WH element is merged in the left periphery.⁹ Since WH movement to the left periphery is relatively marginal with object and adjunct WHs in these varieties, an increased incidence of DFC results in subject WH questions.

In conclusion, the data introduced so far confirm a lower incidence of DFC in embedded clauses (and, to a lesser extent, in main clauses) featuring why. A morphological explanation of this restriction is quite unlikely. The hypothesis that DFC originated from the reanalysis of light-headed relatives, conversely, may explain the restriction.

Figure 6: Main questions exhibiting DFC or WH in situ in Lombard varieties (counts). Source: ASIt [November 2023].

No direct association was found between the incidence of DFC and other WHs in main and embedded clauses, all other conditions being equal. No complementizer is found if WHs remain in situ.

5 DFC vs Clefts

The goal of this section is to show that DFC and clefting are independent phenomena. The quantitative data suggest that they are almost in complementary distribution. We can therefore rule out the hypothesis that clefting played any role in the emergence of DFC.

The incidence of clefts in the ASIt dataset is shown in Figures 7 and 8, focusing on direct and indirect interrogatives, respectively. Subject WH interrogatives, which often exhibit a higher percentage of DFC, are plotted in separate black bars. In several languages, mostly north-eastern varieties, subject direct interrogatives are obligatorily clefted, as in the dialect exemplified (23).

The results in Figures 7 and 8 show that, in many region, the incidence of clefting is higher in direct questions than in indirect ones. On average, the data in Figure 9 show that clefting is found predominantly in main clauses, while DFC is a feature of embedded interrogatives. Notice that DFC seldom occurs in clefts.

Figure 7: Incidence of clefts in main WH questions per Region: subject interrogatives vs other WHs. Source: ASIt [November 2023].

Figure 8: Incidence of clefts in embedded WH questions per Region: subject interrogatives vs other WHs. Source: ASIt [November 2023].

Figure 9: Direct and indirect questions exhibiting clefting, DFC, or both (counts). Source: ASIt [November 2023].

The above data suggest that clefting and DFC are almost in complementary distribution. In the ASIt dataset, DFC is found in 20 clefted main interrogatives, exemplified in (24a–c), and in 5 subordinate clauses like (24d).¹⁰

Otherwise, dialects exhibit either DFC or clefting in contexts where one would expect both. For instance, in Trissino, exemplified in (25), embedded subject interrogatives are clefted, as shown in (25a–c), or exhibit DFC, as in (25d), like other indirect interrogatives in (25e–g). However, in the ASIt sample, complementizers never occur in subject interrogatives that are clefted.

The data of the ASIt database confirm that clefting is rather common in main interrogatives throughout northern Italy, especially in Lombardy (Poletto 2000: 62, but see Pescarini & Donzelli 2018). In most Venetan varieties, clefting is mandatory in subject main interrogatives. In embedded clauses, where DFC is very frequent, clefting is marginal and almost never co-occurs with DFC.¹¹

The almost complementary distribution of clefting and DFC may be accidental, in the sense that these two phenomena may have originated in different clausal environment, triggered by independent factors. Nothing in the following analysis of DFC hinges on this. What is relevant for the purpose of the present study is that the data above show that clefting played no direct role in the diachronic emergence of DFC, i.e. the sequence WH + C found in DFC cannot result from the reanalysis of clefts.

6 DFC and Mood

Subjunctive mood is often associated with the syntax of “low” complementizers, i.e. the complementizers that lexicalize the lower heads of the split CP, in particular in irrealis/subjunctive clauses. Crucially, there is no significant asymmetry in the incidence of DFC between veridical/realis and nonveridical/irrealis embedded clauses, as shown in Figure 10. We can therefore exclude any possible link between the account of DFC and the analysis of other complementation-related phenomena. DFC is not a form of recomplementation and the absence of DFC is not a form of complementizer deletion. It follows that previous analyses of recomplementation and complementizer deletion cannot apply to DFC.

Figure 10: Incidence of DFC in embedded clauses per Region and Mood. Source: Source: ASIt [November 2023].

The data in Figure 10 show that northern Italo-Romance differs from colloquial French, where DFC is preferred in nonveridical contexts (Matuschat-Petersen & Faure 2024):

Moreover, Figure 10 shows that DFC must be orthogonal to and independent from other mood-related phenomena that affect the complementation system of Italo-Romance varieties.

Northern Italo-Romance exhibits no dedicated class of nonveridical/irrealis complementizers, unlike southern varieties (Ledgeway 2003; 2005; 2006). Certain northern varieties, however, exhibit syntactic evidence confirming the occurrence of non-veridical complementizers in the lower left periphery of the clause. The most revealing phenomenon in this respect is recomplementation, i.e. the co-occurrence of multiple complementizers. Recomplementation is found in western dialects such as Ligurian and Piedmontese (Paoli 2007), where a low complementizer may occur in subjunctive completive clause after topic and/or focus material, as shown in (27a) vs (27b). Recomplementation phenomena, however, are independent from DFC as languages with DFC do not necessarily exhibit recomplementation and vice versa. Dagnac (2018) reaches the same conclusion with respect to Gallo-Romance data.

Moreover, consider that recomplementation is often optional, whereas DFC is mandatory, and, as previously mentioned, recomplementation, unlike DFC (cf. Figure 10), is strongly associated with irrealis/subjunctive mood.

Another phenomenon that, in principle, may interact with DFC is complementizer deletion. In several Italo-Romance varieties including standard Italian, complementizers introducing subjunctive or conditional complement clauses may be dropped, as shown in (28a) vs (28b) (Poletto 1995, Giorgi & Pianesi 1997; 2004; Cocchi & Poletto 2002; 2007; Isolani 2023):

To explain DFC, one could build upon prior analyses of complementizer deletion. In fact, complementizers can be deleted in languages such as Italian, which do not display DFC, while they cannot be deleted in northern Italo-Romance, i.e. in the area where DFC is attested. However, complementizer deletion, like recomplementation, is an optional phenomenon, triggered by discourse conditions and confined mainly to subjunctive clauses. For these and other reasons, complementizer deletion cannot be a relevant factor in the modelling of DFC and, likewise, current analyses of complementizer deletion provide limited or no insight into DFC.

To conclude, mood is not a factor in the distribution of DFC in northern Italo-Romance. There is no association between the distribution of DFC in northern varieties and other complementation-related phenomena that are usually found in nonveridical contexts, such as recomplementation or complementizer deletion.

7 DFC in relative and adjunct clauses

The ASIt data show that DFC is not confined to interrogatives, but occurs in locative relatives¹² such as (29a) and adjunct clauses such as (29b) and (29c).

The data from the ASIt in Table 5 reveal a strong correlation between the occurrence of DFC in relatives and in indirect questions. With a single exception, the varieties exhibiting DFC in indirect interrogatives display DFC in relatives and adjunct clauses that are introduced by WH elements, such as temporal and locative subordinates.

A similar association cannot be verified for other kinds of adjunct clauses, which are missing in the ASIt database, but the data introduced so far lead to the conclusion that DFC is not exclusive to interrogatives. DFC is found in any embedded clause that is equivalent to a WH construction, such as relative clauses, free relatives, and other kinds of adjunct clauses that can be analyzed as free relatives “in disguise”, such as temporal clauses. Free relatives will play a prominent role in the analysis of DFC that will be discussed in Sections 9–10.

8 Intermediate conclusion: empirical findings and open questions

The relevant findings of the previous sections are summarized below:

• - The dialects allowing DFC in main clauses are a subset of those exhibiting DFC in embedded clauses, with a few exceptions. It is likely that DFC emerged first in embedded clauses, although we cannot exclude that a similar, yet independent discourse-related phenomenon emerged in main clauses.

• - The absence of Subject-Clitic Inversion is a necessary but not sufficient condition for DFC: DFC occurs in languages or contexts that do not feature SCI, but several varieties exhibit neither DFC nor SCI.

• - DFC is less frequent with why.

• - DFC is incompatible with in-situ WHs (unless the in-situ WH is doubled by an ex-situ copy).

• - Neither subject WHs nor who WHs trigger a higher incidence of DFC all other conditions being equal. In Lombard varieties, the higher incidence of DFC in subject/who WHs results from an orthogonal factor such as the preferred in-situ placement of other WHs.

• - DFC is almost incompatible with clefting.

• - DFC is not affected by mood. Variation with respect to DFC is therefore independent from other phenomena that are normally found in nonveridical contexts such as recomplementation or complementizer deletion; the former is not associated with DFC; the latter is not allowed in northern varieties.

• - If DFC occurs in embedded questions, it will occur in relative and adjunct clauses as well.

In light of these findings, the following Sections elaborate on a possible analysis of DFC. Although the account I propose can be extended to all northern varieties, the hypotheses discussed in the following Sections will be illustrated mostly with data from central Venetan varieties.

9 Previous accounts of DFC in Italo-Romance

In many languages, including northern Italo-Romance, embedded questions (Q) and Free Relatives (FR) are identical, see (30).

There are, however, certain gaps in the distribution of WHs in FRs. FRs cannot be introduced by complex WHs, as shown in (31b), and cannot be introduced by why.

Moreover, certain simple WHs cannot occur in FRs. In most¹³ Italo-Romance varieties, as well as in other Romance languages such as French or Spanish, what-FRs are ungrammatical, see (32b), while what regularly occurs in embedded questions such as (32a).

The Romance languages exhibit an alternative strategy to form FRs that involves a Light Head (LH). Light-Headed Relatives (LHRs; Citko 2004) obtain when a non-deictic demonstrative—i.e. a demonstrative devoid of deictic features—is followed by a bona fide relative CP, as shown in (33).

Following Citko (2004), let us assume that LHs are externally-merged in D, while the relative CP is merged in D’s c-command domain as shown in (34), corresponding to Citko’s (36b) (Citko 2004: 110):¹⁴

In a subset of the Romance languages, including northern Italo-Romance, LHRs may occur under certain interrogative predicates instead of what-interrogatives, cf. (35a) vs (35b). In these contexts, LHR and WH interrogatives are therefore in free distribution.

Previous scholars such as Munaro 2000[2001] and Garzonio 2007 argued that the emergence of DFC resulted from the reanalysis of LHRs such as (35a). In a nutshell, the sequence LH+C in (35a) was first reanalysed as a sequence WH+C and, eventually, C was merged in other embedded WH clauses.¹⁵ The reanalysis was favoured by the similarity between demonstrative LHs, WHs and complementizers. In many dialects, the demonstrative chel [kel] ‘that’ is identical to the sequence formed by the complementizer che [ke] and the third person subject clitic l. Diachronically, che was also the WH for what.

Unfortunately, the data in Section 5 do not show a higher incidence of DFC in what-clauses, which would confirm the hypothesis that what-clauses were the pivotal context of the reanalysis. Previous qualitative analyses of narrow samples of datapoints (such as Polonia 2014, cited in Poletto & Sanfelici 2019) seem to corroborate this intuition, but studies on bigger dataset like the present one do not confirm this conclusion.

Poletto & Sanfelici (2019) try to confirm the hypothesis just outlined by examining diachronic data. However, it seems to me that the analysis of medieval texts is not conclusive as DFC is found with all kinds of WHs since the earliest attestations, see (36). Therefore, the hypothesis that DFC emerged first in what-clauses cannot be confirmed (nor rejected) on the basis of data from medieval texts.

Moreover, Poletto & Sanfelici (2019) notice that indirect questions in medieval varieties often feature an “external” (light) head, such as the locative element là ‘there’ in (37a) or the generic noun chaison ‘reason’ in (37b). The LHs in (37) precede WH elements, which are in turn followed by the complementizer (with the noticeable exception of perché ‘why’ in (37b)).

The presence of an external (light) head preceding the WH in (37a) and (37b) seems an additional and orthogonal phenomenon that occurs along with DFC. It seems to me that the data in (37), albeit interesting in their own right, do not provide compelling evidence on the nature of DFC.

Let us leave aside the question of whether DFC emerged from LHRs and instead focus on the synchronic relationship between DFC-clauses and LHRs. On the one hand, the parallelism between DFC-clauses and FRs/LHRs can provide us with a better understanding of DFC:

• - First, it is worth recalling that DFC is a feature of finite clauses, a restriction that can be explained straightforwardly in light of the fact that FRs/LHRs, unlike questions, must be finite.

• - Analogously, the lower incidence of DFC with why (Section 4) reflects the fact that why cannot occur in FRs (arguably, why is externally merged in C; Cecchetto & Donati 2012)

• - By the same token, the occurrence of DFC in adjunct clauses introduced by where and when (Section 7) can be accounted by assuming that adjunct clauses introduced by WHs are FRs.

On the other hand, however, we cannot conclude that DFC-clauses are LHRs “in disguise”, as proposed by Poletto & Sanfelici 2019. DFC-interrogatives, for instance, differ from LHRs in several respects. First, questions in DFC-languages allow sluicing, cf. (38a), while light-headed FRs such as (38b) cannot be sluiced even if they occur under interrogative predicates.

Second, WHs in questions may be modified by the-hell expressions, which therefore occur between the WH and the following C as shown in (39a), cf. Matuschat-Petersen & Faure (2024) for French. LHRs, conversely, cannot be modified, as shown in (39b).

In conclusion, despite some superficial similarities, LHRs and DFC introduce different clause types. There is no evidence in favour of the hypothesis that WHs in DFC-languages are a special kind of LHR. Likewise, there is no evidence in favour of Poletto & Sanfelici’s (2019) claim that the complementizer in DFC is the spell-out of an internal projection of the WH-item. Evidence from sluicing in (3) and (38a) rules out this possibility.¹⁶

10 A possible analysis of DFC

To summarise the discussion in Section 9, embedded interrogatives and LHRs are string-identical, but instantiate distinct clause types. Evidence from microvariation suggests that the string-identity/similarity of LHRs and WH clauses was the trigger that led to DFC, but the underlying syntactic representation that triggered the emergence of DFC is still unclear. In this section, it is argued that the missing link responsible for the emergence of DFC was a change in the feature endowment of WHs.

In current minimalist analyses, the label of a phrase X must be a subset of the features of the items {a, b} that form X. Cecchetto & Donati (2010: 245) argue that a single algorithm, in (40), can predict the label of X, regardless of {a, b}’s being internally or externally merged (see also Donati & Cecchetto 2011).

Under (40), a minimal element X projects if it probes a phrase YP, which eventually may be (re)merged in SpecXP, as in (41a). Likewise, X projects if it selects a complement, as in (41b), assuming that selection is a kind of probe-goal relation.

The algorithm in (40)–(41) yields an ambiguous result when the element that is remerged in (41a) is a minimal element such as a simple WH element, as shown in (42). Simple WHs—i.e. WHs that are not lexically-restricted—can be read by the algorithm as lexical elements that select the rest of the clause as their complement, as in (41b).

According to Cecchetto & Donati (2010), questions obtain if the clause is labelled by the (silent) C head, which selects the IP you read in (42) and probes the WH what as in (41a). A FR (i.e. a DP) obtains from the same configuration if, after being remerged in C, the WH word is read as a D element selecting a CP, as in (41b).

Complex WHs (i.e. lexically restricted WHs) cannot label because they cannot be “mistaken” for a selecting head. Consequently, free relatives cannot be introduced by complex WHs, see (43).¹⁷

Notice that WHs in Cecchetto & Donati’s (2010) analysis of FRs do not occur outside of the CP, as in so-called head-analyses of FRs (Bresnan & Grimshaw 1978). Head-analyses explain why WHs are licenced by the matrix predicate (e.g. *I like [to whom you talked]), but cannot explain why the WH must be extraposed along with the CP, as shown in (44), in languages such as German and Dutch that allow the extraposition of bona fide relative CPs (Groos & van Riemsdijk 1981).

According to Cinque (2020: 97, fn. 129), Cecchetto & Donati’s (2010) account, like traditional head-accounts such as Bresnan & Grimshaw (1978), cannot explain the ungrammaticality of (44a). However, it seems to me that the pattern in (44a) results straightforwardly from the relabeling algorithm in (41)–(42) as the WH element, in order to relabel, must immediately c-command the relative CP. If the CP is extraposed, as in (44a), the condition allowing the projection of WH’s D feature is not met.

Another potential issue undermining Cecchetto & Donati’s analysis of FRs is that, in Italian, WHs do not target the same position in FRs and in interrogatives (Benincà 2012; Guglielmo Cinque, p.c.). Relative WHs precede topics, as shown in (45a), while interrogative WHs follow topics, as in (45b). The facts in (45), however, are not necessarily at odds with the labeling algorithm in (42). The condition in (42) determines a 50–50 chance to get either a FR or an embedded question. The data in (45) suggest that, when a topic occurs, the ambiguity disappears, but this is not per se an argument against (41)–(42).

The minimal pair in (45) suggests that the model in (41)–(42) needs some refinement. For instance, one may say that, in order to reproject, the WH must occupy a higher position in the left periphery, as in (45a). Therefore, we should abandon the representation in (42), where the WH and the complementizer are merged in the same position. The same conclusion is corroborated by evidence from sluicing in DFC languages, repeated in (46).

To account for (45) and (46), we can assume, in keeping with previous cartographic analyses, that WHs elements such as chi ‘who’ are not probed by a C head, but by a distinct head, say {Q} (Chomsky 2000). The position of {Q} in the functional spine of indirect interrogatives corresponds to Rizzi & Bocci’s (2017) Q_emb (Q in embedded contexts; the same position was called Wh in Rizzi 2004). Finite complementizers such as che ‘that’ are merged in a lower projection, corresponding to Rizzi’s (1997) FinP:

The tree marker in (47) is compatible with Cecchetto & Donati’s (2010) Probing Algorithm in (41), provided that the head labelling questions is not C, as proposed by Cecchetto & Donati, but Q. The rest of Cecchetto & Donati’s theory remains unchanged: questions obtain when the clause is labelled by {Q}; FRs obtain when the clause is labelled by the {D} feature carried by the WH.

This minor modification of Cecchetto & Donati’s theory allows us to account for DFC in keeping with Bayer’s (2015) hypothesis that WHs may also carry a {C} feature (see also Bayer & Brandner 2008a; 2008b). In German dialects, DFC virtually never occurs with simple WHs such as was (‘what’) and wer (‘who’) (see also Penner and Bader 1995 on Swiss German and Weiß 1998; 2004 on Bavarian), whereas DFC occurs frequently in combination with complex WHs. Bayer concludes that phrasal WHs can combine with a complementizer because, if the WH element belongs to a branching structure (e.g. what kind of people), the WH element cannot project {C}. The WH phrase will then move to the specifier of an externally-merged complementizer. Word-size WHs, conversely, can project their {C} feature, as in Cecchetto & Donati’s algorithm. Notice that Bayer’s (2015) analysis cannot be extended straightforwardly to northern Italo-Romance, where DFC is obligatory and is found across-the-board with simple and complex WHs. Above all, Bayer’s analysis is at odd with evidence from Italo-Romance, indicating free relatives as the pivotal structure from which complementizers began to leak into bona fide indirect interrogatives (Munaro 2000[2001]; Garzonio 2007; Polonia 2014; Poletto & Sanfelici 2019).

The remainder of this section presents an analysis that brings together two key insights: (a) Bayer’s hypothesis that WH elements can function as complementizers, as evidenced by both Germanic and Romance varieties, and (b) Cecchetto & Donati’s theory of labeling, which captures the ambiguity between free relatives and indirect questions. The latter explains the (reconstructed) diachronic evolution and the synchronic distribution of the Romance DFC pattern (not of the Germanic one).

The following explanation does not adopt an approach that assimilates complementizers to D-elements. Instead, it builds on the idea that both C and D features are essential to satisfy c-selection requirements: matrix predicates select CPs or DPs. The additional QP in (47) is inert with respect to c-selection, while it arguably relevant for s-selection.

The {C} feature that allows c-selection can be lexicalized by a WH element, as shown in (48a). Conversely, in languages in which WHs do not carry {C}, a CP is required in order to satisfy the c-selection features of the matrix clause, as shown in (48b).

In both cases, the embedded clause can be (re)labeled by the {D} feature carried by the WH, yielding a FR.

It is worth noting that the syntactic configuration in (48b) yields a word order that is string-identical to Light-Headed Relatives in (49). The similarity was even higher in the dialects where the LH, the WH meaning ‘what’ and the complementizer were lexicalized by the same formative [ke]. Notice, however, that in (48b) the WH is internally merged and the QP can be s-selected by interrogative predicates, whereas the LH quello in (49) is externally merged and does not carry {Q}. Consequently, LHR are not interrogatives (see Section 9) and are usually restricted to FR interpretation, with some exception that, as previously mentioned, might have played a role in the emergence of DFC.

The analysis just outlined accounts for DFC in embedded clauses, since the occurrence of a {C} feature is required to satisfy c-selection, while DFC in main clauses, as found in languages such as Piedmontese (Section 3), remains unaccounted for. However, recall that DFC in main clauses is either a discourse-related phenomenon (as shown by Parry’s longitudinal analysis of Piedmontese and by the comparison with spoken French) or it is a later development, which is arguably related to the loss of subject clitic-verb inversion (Section 3.1). Since the loss of inversion is a necessary, but not sufficient condition for the spreading of DFC in main interrogatives, we can entertain the hypothesis that the internal merge of subject clitics (yielding inversion) and the external merge of a complementizer head are alternative ways of typing main questions by checking the {Q} feature. To test this hypothesis, however, we need to undertake a thorough analysis of inversion across northern Italo-Romance, which is beyond the scope of the present work.

11 Conclusion

DFC, i.e. the co-occurrence of WH elements and finite complementizers, was examined in light of quantitative and qualitative data extracted from the Syntactic Atlas of Italy (ASIt).

Previous descriptive generalizations were tested and revised. The list of the empirical findings of this work is repeated below for the ease of exposition:

• - The dialects allowing DFC in main clauses are a subset of those exhibiting DFC in embedded clauses. It is likely that DFC emerged first in embedded clauses, although we cannot exclude that a similar, yet independent discourse-related phenomenon emerged in main clauses.

• - Subject-Clitic Inversion is a necessary but not sufficient condition for DFC.

• - DFC is less frequent with why.

• - All other conditions being equal, DFC occurs more frequently neither with subject WHs nor with who.

• - DFC is incompatible with in-situ WHs.

• - DFC seems incompatible with clefting, but one should consider that the two phenomena emerged in distinct clausal environments: DFC is a feature of embedded clauses; (obligatory) clefting is found mostly in (subject) main interrogatives.

• - DFC is not affected by Mood. Variation with respect to DFC is therefore independent from other phenomena that are normally found in nonveridical contexts such as recomplementation or complementizer deletion; the latter is not allowed in northern varieties.

• - If DFC occurs in embedded questions, it will occur in relative and adjunct clauses as well.

From a theoretical standpoint, this article endorsed Bayer’s 2015 hypothesis that DFC obtains when WHs do not carry a {C} feature. In languages where WHs carry {C}, WHs licence the embedding of IPs, while in languages in which WHs do not carry {C}, an overt complementizer is requires to fulfill c-selection requirements.

This analysis, coupled with Cecchetto & Donati’s (2010) labelling theory, explains why DFC is found in embedded interrogatives and free relatives, including adjunct clauses introduced by WHs (where, when), which are better analysed as free relatives.

Free relatives played a pivotal role in the emergence of DFC in embedded clauses, as suggested by the previous literature on the topic (Munaro 2000[2001]; Garzonio 2007; Poletto & Sanfelici 2019). In particular, the occurrence of Light-Headed Relatives under interrogative verbs prompted the emergence of DFC in embedded clauses. The extension of the phenomenon to all types of WHs—excluding why-WHs—is explained by supposing that the mechanism of reanalysis led to the loss of the {C} feature carried by WHs.

The spreading of DFC in main clauses, which is typical of Piedmontese dialects, is a later independent phenomenon, which may have been originated from discourse conditions (Parry 2003). In keeping with Poletto & Vanelli 1995; Poletto 2000, I suggested that the external merger of C and the internal merger of subject clitics (yielding inversion) are alternative mechanisms to type main interrogatives in northern Italo-Romance.

Abbreviations

DFC = Doubly-Filled COMP

FCI = Free Choice Item

FR = Free Relative

LHR = Light-Headed Relative

SCI = Subject Clitic Inversion

WH = wh element

Data availability/Supplementary files

The data were extracted from the ASIt database (http://svrims2.dei.unipd.it:8080/asit-maldura/pages/search.jsp last viewed: 30.11.24) and were subsequently annotated manually. The annotated files are freely available for download from the open-access repository Nakala:

• Annex A (main questions): https://nakala.fr/10.34847/nkl.6fbd4096

• Annex B (embedded questions): https://nakala.fr/10.34847/nkl.0e7c4aq7

Ethics and consent

The data presented in this study were collected in compliance with all relevant EU regulations in effect at the time of the fieldwork campaign. The textual data are anonymous and in written form. The research adhered fully to ethical standards, and no ethical concerns or issues were identified throughout the study.

Funding information

This work is an outcome of the project ViGramm – Visualising Grammars across space and time (SOSI 2023).

Acknowledgements

I would like to thank Fiammetta Di Pasquale for manually annotating the corpus. A preliminary version of this work was presented at CIDSM18 and IGG50 and in workshops held at the university of Udine, at the university of Corsica ‘Pasquale Paoli’, and at the Beijing Language and Culture University (online). I thank the audience of these events and three anonymous reviewers for their valuable feedback.

Competing interests

The author has no competing interests to declare.