2.1 Selection of the data & geographical distribution

The empirical basis of this article consists of ten dialect phenomena attested in Dutch dialects spoken in Belgium, the Netherlands, and the north of France. All data originate from the Syntactic Atlas of Dutch Dialects (SAND) (the SAND-project, Barbiers et al. 2005; 2008). The data from this project are freely available through the project’s online interface, see Barbiers et al. 2006.

The SAND-project collected data in three phases: an initial written questionnaire, followed by in-person oral interviews, and finally telephone interviews. This paper draws exclusively from the second and third phases. Fieldwork was carried out across 267 dialect locations, with each interview conducted in the presence of at least two native speakers of the dialect. The interviews featured various task types, including translation exercises, sentence completion tasks, and grammaticality judgments. In total, approximately 75 (morpho)syntactic variables were examined. For a comprehensive discussion of the project’s methodology and design, see Cornips & Jongenburger (2001).

For this paper, we selected ten syntactic variables, which we introduce in detail in subsection 2.2. The selection of this particular set of phenomena is based on two main considerations. First, some of these constructions have already been analyzed in the formal linguistic literature on Dutch dialects. Existing analyses provide a starting point for identifying potential underlying parameters that may connect these phenomena. Second, their geographical distribution played a crucial role in the selection process. As illustrated in Figure 1, all ten phenomena are predominantly found in the southwestern part of the language area. This raises the possibility of a shared underlying grammatical principle. At the same time, though, their distributions are not identical, making it interesting and challenging to investigate how exactly the phenomena are related and to what extent they can be traced back to common underlying syntactic principles. These ten linguistic phenomena are introduced and exemplified in the following subsection.

Figure 1: Geographical distribution of the ten syntactic phenomena.

2.2 Description of the ten dialect phenomena

In this subsection we discuss and illustrate the ten phenomena introduced above. We have grouped them thematically, i.e. based on the part of grammar or part of speech they pertain to. We start with phenomena that revolve around the complementizer domain. The first involves the choice of the complementizer in conditional and comparative clauses. In the majority of the Dutch dialect area as well as in colloquial Standard Dutch, the comparative and conditional complementizers are identical and surface as (a variant of) als ‘as’. This is illustrated for colloquial Standard Dutch in (1) for comparative clauses and in (2) for conditional clauses.

However, in certain Dutch dialects, all located in the south west of the language area, the two complementizers are not identical. While conditional clauses use a variant of Standard Dutch als ‘as’, the comparative complementizer is of ‘if’. This is illustrated in (3) and (4) for the dialect of Oostkerke. We will henceforth refer to this phenomenon as comparative if, or CMPR-IF for short.

A second complementizer-related phenomenon is complementizer agreement, abbreviated as CA throughout this paper (see Van Koppen (2017) and the references cited there for an in-depth discussion of complementizer agreement). In dialects with CA, a complementizer introducing a finite embedded clause agrees with the subject of this clause. This is illustrated for the dialect of Gistel in (5), where the complementizer da ‘that’, agrees in third person plural with the subject of the embedded clause ze ‘they’ by means of an agreement ending -n which appears at the end of the complementizer.

CA agreement affixes are identical to the agreement endings found in the verbal paradigm (see De Vogelaer (2005) for detailed discussion). This is clear in the example in (5), where the complementizer agreement ending, -n, is identical the agreement ending found on the verb e-n ‘have-pl’.

A phenomenon that looks related at first sight, but is quite different on closer inspection is subject clitic doubling (henceforth CD, see e.g. Haegeman 1992; Van Craenenbroeck & Van Koppen 2002; De Vogelaer 2005; Van Craenenbroeck & Van Koppen 2023b for discussion). In clitic doubling constructions a strong pronoun is doubled by a clitic pronoun. The Aalst Dutch example in (6) contains two instances of clitic doubling: the third person plural pronoun zeir ‘they’ is doubled by the clitic se ‘they’ and the second person plural pronoun geir ‘you’ is doubled by the clitic ge ‘you’.

Given that clitic doubling frequently occurs in embedded clauses and that the clitic cliticizes onto the complementizer in that case (as in (6)), it might be tempting to try and unify this phenomenon with complementizer agreement. On closer inspection, though, there are clear differences between the two phenomenona that argue against such a unification. First, CD is restricted to pronominal subjects, while CA can show up in combination with lexical DPs as well. Secondly, CA-endings are identical to verbal agreement endings, while subject clitics are clearly distinct from them. Thirdly, CA is restricted to embedded clauses introduced by a complementizer, while CD can also occur in inverted main clauses and in some dialects even in subject-initial main clauses (Van Craenenbroeck & Van Koppen 2019a; 2023a). Fourthly, CA and CD can co-occur in the same example. These final two points are illustrated by the Nieuwpoort Dutch example in (7).

This example features both clause-initial clitic doubling in the form je..gieder ‘you_clitic..you_strong’ and the co-occurrence of CA and CD on the finite complementizer da: da-n-ze zieder ‘that-pl-they_clitic they_strong’. It is on the basis of these kinds of data that we decide to keep complementizer agreement distinct from clitic doubling.¹

The final construction under discussion in this paper that is related to the clausal left periphery involves imperatives. This clause type is normally characterized, both in Standard Dutch and in its dialects, by the occurrence of a single verb in clause-initial position. Positioning more than one verb in this peripheral position leads to an ungrammatical result. This is illustrated in (8).

In certain dialects of Dutch, however, a close equivalent of (8b) is well-formed. This is illustrated in (9) for the dialect of Ghent.

This imperative clause features two verbs in clause-initial position: the imperative form of the main verb halen ‘to get’ (i.e. haalt) is immediately preceded by the infinitival form of the aspectual light verb gon ‘to go’. This phenomenon, which we will refer to as GO-GET in the remainder of this paper, is reminiscent of so-called quirky Verb Second in Afrikaans (see de Vos 2006 for discussion), but it is more restricted in its scope and its syntactic distribution: the dialect Dutch phenomenon illustrated in (9) is only attested in imperatives—whereas Afrikaans quirky V2 can also be found in declaratives—and apart from ‘go’, the only other light verb that can partake in this pattern is komen ‘to come’, as shown in (10) with an example from the dialect of Moerzeke.

The following set of phenomena we want to introduce are all related to polarity. The first we will refer to as short do replies (SDRs for short), following Van Craenenbroeck (2010). Consider an example from the East-Flemish dialect of Berlare in (11).

In this example A’s negative statement, IJ zal niet komen ‘He won’t come’, is contradicted B’s short answer IJ doet ‘Yes, he will’. That short answer is what we will refer to as an SDR. Intuitively, this phenomenon seems related to English-style VP-ellipsis, but there are also clear differences between the two. For one, whereas a VPE-reply to A’s statement in (11) would include the auxiliary (Yes, he will.), SDRs invariably contain (a present tense form of) the dummy verb doen ‘to do’. Secondly, SDRs are more restricted in their distribution: unlike VPE, they can only be used in short contradictory replies to declarative statements. For a more detailed discussion of the empirical properties of SDRs as well as a formal analysis, we refer to reader to Van Craenenbroeck (2010) (and see also §4.3.1 below).

The second polarity-related phenomenon under discussion in this paper are clitics on yes and no (henceforth CYN). We are referring here to the construction whereby the polarity markers ja ‘yes’ and nee ‘no’ combine with a subject clitic in short elliptical replies to a preceding yes/no-question or statement (see Van Craenenbroeck (2010) and references cited there). The subject clitic invariably refers back to the subject of the preceding question or statement. CYN is exemplified in (12) where the polarity marker ja is combined with the subject clitic -k ‘I’ to form an elliptical reply meaning ‘Yes, I would like some more coffee’.

The third and final polarity-related phenomenon related is the occurrence of two-part negation, in particular the use of the negative clitic en in addition to the regular clausal negator nie(t) (see Haegeman (2005) and Haegeman & Breitbarth (2014) for discussion). We will henceforth refer to the occurrence of the negative clitic as NEG. An example from the dialect of Tielt is given in (13) in which the negative clitic en is combined with the negative adverb nie ‘not’.

The next group of phenomena involve expletive pronouns. In Standard Dutch as well as in many of its dialects, the expletive pronoun is (a variant of) er ‘there’, i.e. a pronoun that is morphologically and etymologically related to a locative adverb. A basic example is given in (14).

In certain dialects, however, the expletive is not locative-related, but surfaces as t. An example from the dialect of Brugge is given in (15).

At first glance, this t-element looks like a reduced form of the third person neuter personal pronoun het ‘it’ (though see Van Craenenbroeck (2022) and §4.3.2 below for arguments against this idea). It is only attested in sentence-initial position: in inverted main clauses and embedded clauses all Dutch dialects use a locative-based expletive and the t-element is completely missing in these contexts. We will refer to this phenomenon as EXPL-T in this paper.

A second vector of variation related to expletives concerns the optional versus obligatory use of (locative-based, see above) expletives in embedded clauses and inverted main clauses. Consider in this respect the contrast in (16)–(17) (data from Van Craenenbroeck 2022: 390–391).

The example in (16) shows that in the dialect of Torhout the expletive pronoun is obligatory in inverted main clauses—and the same applies to embedded clauses. By contrast, in the dialect of Wambeek illustrated in (17), the expletive is optional in these contexts. The obligatory nature of locative-based expletives in inverted main clauses and embedded clauses is one of the empirical phenomena under investigation in this paper. We will abbreviate it as ER-OBL.

The tenth and final dialect phenomenon under discussion in this paper involves demonstrative pronouns. In Dutch and its dialects—as in English and many other languages—demonstrative pronouns can be used attributively, and when they are, they stand in complementary distribution with definite determiners (see the Standard Dutch example in (18)). This complementary distribution persists in contexts where the noun is elided, as in (19).

The co-occurrence restriction between determiners and demonstratives illustrated in (18) holds for all dialects of Dutch, but the one in (19) does not: there are dialects where, when the noun is elided, a definite determiner and a demonstrative pronoun can co-occur (henceforth THE-THAT, see Corver & Van Koppen 2018 for discussion). An example from the dialect of Merelbeke is given in (20).

This concludes our introduction of the empirical basis of the present paper. Table 1 presents a schematic overview of the ten phenomona, the abbreviation we use for them, and the key example we have provided.

In the remainder of this paper, we analyze these data in two steps. First, we use quantitative-statistical methods to identify the main patterns and tendencies in the data. This is the topic of section 3. Secondly, we interpret and analyze the quantitative findings from a formal-theoretical point of view in section 4.

4.1 Parameter #1: ϕ-features in CP

The first dimension of the Correspondence Analysis sets complementizer agreement (CA) apart from the remaining nine phenomena. Let us therefore zoom in on the linguistic analysis of this phenomenon. Reconsider our basic CA-example in (5), repeated below as (22).

Following Van Koppen (2017) we analyze CA as the overt reflex of unvalued ϕ-features on C undergoing Agree with the DP in specTP. These C-features probe their c-command domain, find the DP in specTP, and as a result get valued. A schematic representation of this analysis is given in (23).³

A noteworthy corollary of this analysis is that the ϕ-feature specification on the complementizer is independent of the relationship between T and the subject. In other words, complementizer agreement is not the result of T-to-C-movement (Zwart 1993; 1997), nor of C-to-T feature inheritance (Chomsky 2008). An indication that this line of thinking is on the right track is provided by the example in (24), where it is not the whole subject-DP that occupies specTP, but only a subpart of it (from Haegeman & Koppen 2012).

The subject in this example is the possessed DP die venten underen computer ‘those guys’s computer’. Interestingly, the complementizer omdan ‘because’ shows (plural) agreement with the possessor, while the finite verb was ‘was’ agrees with the (singular) head noun of the subject, i.e. computer ‘computer’. Examples like these clearly show that the ϕ-feature specification of C is independent of that of T, which in turn is compatible with the analysis proposed in (23) (see Van Koppen (2005); Haegeman & Van Koppen (2012) for further discussion, including arguments against non-syntactic accounts of complementizer agreement). In light of that proposal, and in an attempt to formalize the distinction between dialects that do and those that do not display complementizer agreement, we propose the parameter in (25).

Dialects with complementizer agreement have a positive setting for this parameter, whereas dialects without complementizer agreement have a negative one. Note that the formulation in (25) remains agnostic about the exact formal implementation of this parameter: it might boil down to the feature specification of an individual functional head (Borer 1984), but it could also be that there is a specified functional projection that is missing or not projecting in the CA-less dialects (cf. Shlonsky’s (1994) AgrCP), or that what we are dealing with in CA-dialects is a Fin/Force-split (Rizzi 1997).

4.2 Parameter #2: a split left periphery in DP

Let us now turn our attention to clitic doubling (CD) and determiner-demonstrative doubling (THE-THAT). Our basic examples of these constructions are repeated below as (26) and (27) respectively:

These two phenomena clearly pattern together in the plot in Figure 5 and in this section we want to propose a theoretical reason for why this might be the case.⁴ More specifically, both for CD and for THE-THAT we will draw on existing accounts of these phenomena, from which we distill a common core, and that common core will form the basis for the parameter we propose. Looking ahead already, that parameter will be formulated as in (28).⁵ Languages with a positive setting for this parameter are ones that feature CD and THE-THAT, whereas varieties with a negative setting lack these phenomena.

Let us first turn to clitic doubling, i.e. the example in (26). In the dialects under consideration here, this phenomenon always involves the combination of a clitic pronoun (se and ge in (26)) and a strong pronoun (zeir and geir in (26)) (see de Vogelaer & Devos (2008) for an overview of the different types of subject doubling in Dutch dialects and ways to differentiate them). In what follows we introduce and adopt the analysis by Van Craenenbroeck & Van Koppen (2008; 2019b) of this phenomenon (henceforth VCVK). It is a version of the so-called big DP-analysis of pronominal doubling (see Uriagereka (1995); Laenzlinger (1998); Grohmann (2000); Belletti (2005); Kayne (2005); Poletto (2008) for accounts that are similar in spirit, though different in their analytical details), which has as a general starting point the idea that the two components of a doubled subject form a single constituent upon first merger which gets broken up in the course of the derivation. VCVK’s starting point is Déchaine & Wiltschko’s (2002) classification of pronominal systems into pro-DPs, pro-ϕPs, and pro-NPs. As shown in (29), those three types are ordered in a containment relation:

After applying Déchaine & Wiltschko’s tests to Dutch dialect data, VCVK conclude that while strong subject pronouns behave as DPs, subject clitics correspond to ϕPs. In terms of the representation in (29), this implies that structurally speaking, clitics are contained in strong pronouns. This is the starting point of VCVK’s analysis. They propose that clitic doubling is a case of double spell-out, whereby ϕP moves to the edge of the DP and gets spelled out as a clitic, while the remaining (DP-)portion of the structure gets spelled out as a strong pronoun. The structure in (30) is a schematic representation of this analysis.

The projection that the ϕP moves to is agnostically labeled FP here, but Van Alem (2023) further develops this analysis and argues that it should be identified as a focus projection. Both Van Alem (2023) and Van Craenenbroeck & Van Koppen (2019b) argue that the movement of ϕP cannot target specDP as that would result in an anti-locality violation in the sense of Abels (2003).⁶ This means that in order for clitic doubling to occur, the nominal left periphery has to be rich enough to host the landing site of this movement operation. This is how we tie the presence of clitic doubling in a dialect to the parameter in (28): only an extended nominal left periphery can host the clitic doubling operation shown in (30).

Interestingly, Barbiers et al. (2016) argue for a highly similar and compatible analysis for determiner-demonstrative doubling. They start from the observation introduced in section 2 above that while demonstratives are normally compatible with a overt noun, this is not the case in determiner-demonstrative doubling (data from Barbiers et al. 2016: 13,15):

Moreover, THE-THAT is also incompatible with adjectives and numerals (Barbiers et al. 2016: 21):

Barbiers et al. (2016) and Corver & Van Koppen (2018) take this to mean that the element de ‘the’ found in THE-THAT is not a regular determiner merged in D, but rather an element that spells out or pronominalizes a portion of the nominal structure, particularly ϕP. Parallel to the analysis of clitic doubling in (30), this ϕP moves into the left periphery of the nominal constituent where it gets spelled out as de, while the demonstrative is spelled out in specDP. This analysis is illustrated in (33).

Needless to say, the high degree of similarity between the analyses in (30) and (33) jibes well with clitic doubling and determiner-demonstrative doubling patterning together in the output of the Correspondence Analysis in Figure 5. Accordingly, we want to link the presence or absence of these two constructions across the Dutch dialects to this analysis as well, which is why we propose the parameter in (28), repeated below as (34).

Both clitic doubling and determiner-demonstrative doubling involve movement into the extended left periphery of the DP. In dialects that lack such an extended periphery, this movement cannot take place. As a result, both clitic doubling and determiner-demonstrative doubling are missing in such dialects.

Evidence in support of this line of thinking comes from determiner-possessor doubling. This is a construction parallel to determiner-demonstrative doubling, but with a possessive pronoun instead of a demonstrative. An example from Rotterdam Dutch is given in (35) (M. van Koppen, p.c.).

Unlike determiner-demonstrative doubling, determiner-possessor doubling is not restricted in its geographical distribution: as shown by Corver & Van Koppen (2010) and Corver & Van Koppen (2018), most Dutch dialects allow for the construction in (35). Given the fact that the two constructions share a number of central properties, though, like their incompatibility with overt nouns, numerals, and adjectives, Corver & Van Koppen propose an analysis for determiner-possessor doubling that is highly similar to the analyses in (30) and (33) (see also Grohmann & Haegeman 2003; Haegeman 2004 for related proposals). It is represented in (36).

Just like in (33), the de-element in determiner-possessor doubling spells out not a D-head, but rather the entire ϕP, which has moved into the left periphery of the DP. The main difference with the analysis of (33), however, is that now a PossP is projected in between DP and ϕP (see Schoorlemmer 1998 and references cited there). This means that movement to specDP no longer violates anti-locality, and so a split nominal left periphery—i.e. a positive setting for the parameter in (34)—is not required to license this construction. At the same time, however, dialects that do have a positive setting for (34) should have additional structural room available to the left of the determiner. That this is indeed the case is suggested by contrasts such as the one in (37)–(38) (data from Corver & Van Koppen 2010: 131 and M. van Koppen p.c.).

The example in (37) illustrates another, more extreme case of possessor doubling, whereby the possessive pronoun zinnen ‘his’ is accompanied not only by the determiner de ‘the’, but also by an overt possessor DP, in this case the proper name Teun. This example is taken from the dialect of Asten, a dialect with a positive setting for the D-parameter. As its counterpart in (38) shows, however, this type of doubling is not allowed in all varieties of Dutch: while the dialect of Rotterdam allows for the basic case of determiner-possessor doubling—as do most if not all Dutch dialects—this construction cannot be combined with another overt possessor. The fact that Rotterdam Dutch has a negative setting for the D-parameter provides us with a clue as to what underlies the contrast between (37) and (38). In a dialect with a positive setting for the D-parameter, like that of Asten in (37), there is an additional specifier available in the left periphery that can host the additional possessor DP, while dialects with a negative setting, like that of Rotterdam in (38), lack such a position and as a result lack this type of construction. The structure in (39) represents our analysis of this additional type of possessor doubling:

To summarize, in this subsection we have provided an analysis for the clustering of clitic doubling and determiner-demonstrative doubling in the first two dimensions of the Correspondence Analysis (see the plot in Figure 5). We have argued that the presence or absence of these two phenomena is regulated by the parameter in (40), which essentially determines how much structural space there is in the nominal left periphery. Given that both clitic doubling and determiner-demonstrative doubling—as well as the additional type of double possessor doubling we discussed—make crucial use of this extra space, we correctly predict these phenomena to only occur in dialects that have a positive setting for this parameter.

In the next subsection we turn to the third cluster we identified in the quantitative analysis of our data, the one that grouped together the remaining seven phenomena.

4.3 Parameter #3: a split left periphery in CP

This subsection contains the third part of our qualitative analysis of the results that came out of the Correspondence Analysis and the Cluster Analysis discussed in section 3. Recall that the starting point of our formal analyses has been the visual representation of the first two dimensions of the Correspondence Analysis in Figure 5, repeated below as Figure 6.

Figure 6: The first two dimensions of the Correspondence Analysis.

So far, we have provided a parametric analysis of two aspects of this plot. We have associated the outlier position of complementizer agreement (CA) with the C-parameter, which regulates the presence or absence of ϕ-features in the CP-domain, and we have attributed the grouping of clitic doubling (CD) and determiner-demonstrative doubling (THE-THAT) to the D-parameter, which determines how much structural space there is in the nominal left periphery. Now we turn our attention to the lower left-hand quadrant of the plot, and argue that the grouping we find there can also be plausibly reduced to a single parameter, in particular one that regulates how much structural space there is in the clausal left periphery. In so doing, we mainly focus on the highest-level grouping, i.e. the one that takes all seven phenomena together in a single group. The further subgrouping of the negative clitic (NEG) and clitics on yes and no (CYN) is something we briefly comment on, but will generally leave as a topic for further research.

As already hinted at, the line of approach we will be adopting towards these phenomena is similar to the one advocated in the previous subsection: we argue that dialects featuring the seven phenomena under consideration have more structural space available than dialects lacking these phenomena. That structural space is not situated at the nominal left periphery this time, but at the clausal one. The parameter we will end up proposing is given in (41).

For some of the seven phenomena there are existing formal analyses available which suggest that they do indeed require the presence of specific left-peripheral functional projections. For others, there is to the best of our knowledge no existing literature, which also means that our discussion of them will inevitably be more speculative. We have structured our discussion of the seven phenomena into three parts: (1) the polarity-related phenomena, (2) the expletive-related phenomena, and (3) the rest.

4.4 Revisiting clitic doubling vs. determiner-demonstrative doubling

The third dimension of the Correspondence Analysis is represented in the plot in Figure 7.¹³

Figure 7: The first and third dimension of the Correspondence Analysis.

Against the backdrop of the discussion in §4.2, the data pattern revealed in this plot is highly unexpected: in that section we argued that clitic doubling (CD) and determiner-demonstrative doubling (THE-THAT) are regulated by one and the same parameter—the D-parameter in (28)—such that when this parameter is set to ‘yes’, both phenomena occur, whereas when it is set to ‘no’ they are both absent. In Figure 7, however, these two phenomena are at polar opposite sides of the plot, suggesting a strong divergence in their geographical distribution. In order to understand what is going on, let us zoom in on the geographical distribution of these two phenomena, by overlaying the two relevant maps from Figure 1. This resulting map is shown in Figure 8.

Figure 8: The geographical distribution of clitic doubling vs. determiner-demonstrative doubling.

At an intuitive level, this map makes clear where the seemingly conflicting Correspondence Analysis dimensions come from: on the one hand, there is a large degree of overlap between the two phenomena. They overlap in 84 locations, which amounts to 74% of the clitic doubling locations and 87% of the determiner-demonstrative doubling locations. This high degree of co-occurrence is what the second dimension of the Correspondence Analysis picked up on. On the other hand, the overlap is not complete, and there are a number of regions or dialect locations that feature one of the two phenomena, but not the other: (1) the north-eastern tip of France—a region known as French Flanders—features clitic doubling but not determiner-demonstrative doubling, (2) the Dutch province of Zeeland shows the opposite pattern, (3) in the region Flemish Brabant-Antwerp-North Brabant (the central area on the map) there is a lot of overlap between the two phenomena, but at the same time there are a couple of locations, spread throughout this region, that feature clitic doubling, but not determiner-demonstrative doubling, and (4) there are two locations in the province of Belgian Limburg (near the eastern border) that feature determiner-demonstrative doubling but not clitic doubling. These discrepancies were picked up by the third dimension in the Correspondence Analysis and they constitute the main focus of the current section. We split up the discussion into the two logical options that threaten the parametric account we presented in §4.2: (a) dialects with determiner-demonstrative doubling, but without clitic doubling, and (b) dialects with clitic doubling but without determiner-demonstrative doubling.

4.4.2 Dialects with clitic doubling but without determiner-demonstrative doubling

As is clear from the map in Figure 8, there are two types of dialects that fit this description: on the one hand, the French region known as French Flanders (the south west corner of the map) and on the other, scattered throughout the core overlap region between the two phenomena (an area covering the provinces of East and West Flanders, Flemish Brabant, Antwerp, and North Brabant). In this subsection we argue that both these patterns are epiphenomenal, i.e. they are artefacts caused by the specific methodology used in the SAND-project. With respect to the second pattern, it is worth starting form Corver & Van Koppen’s (2018) in-depth discussion of determiner-demonstrative doubling. What they point out is that this phenomenon is only attested in contrastive contexts. Consider in this respect the examples in (87) and (88).

1. (87)

1. Ik
2. I
1. ging
2. went
1. vaker
2. more.often
1. bij
2. with
1. deze
2. this
1. tante
2. aunt
1. logeren
2. stay
1. dan
2. than
1. bij
2. with
1. ??(de)
2.      the
1. die.
2. that
1. ‘I used to stay with this aunt more often than with that one.’

1. (88)

1. [Speaking of an aunt:]
2.  
1. (*De)
2.    the
1. die
2. that
1. is
2. is
1. altijd
2. always
1. heel
2. very
1. aardig.
2. nice
1. [Speaking of an aunt:] ‘She is always very nice.’ (Southern Dutch)

The comparative construction in (87) creates a clear contrast between two female referents, and as shown by the judgments, in such a context determiner-demonstrative doubling is highly preferred. Cases like (88), however, where the demonstrative simply picks up a non-contrastive, continuing topic, are incompatible with determiner-demonstrative doubling. Against this background, let us consider the way in which this phenomenon was probed in the SAND-questionnaires. The informants were asked to provide a grammaticality judgement about (a dialectal version of) the sentence in (89).

1. (89)

1. De
2. the
1. die
2. those
1. zou
2. would
1. ik
2. I
1. niet
2. not
1. durven
2. dare
1. opeten.
2. eat
1. ‘I wouldn’t dare to eat those.’

Given that this sentence was presented to the informants without any further context, there is no guarantee that they interpreted the object of opeten ‘to eat’ contrastively. If they didn’t, then we predict the informants to reject the sentence, regardless of their setting of the D-parameter. We conjecture that this is what underlies the judgments under consideration here: while these informants have a positive setting for the D-parameter—as is evidenced by the presence of clitic doubling—the lack of a clear contrastive reading in (89) led them to reject determiner-demonstrative doubling. There is some indirect historical evidence which seems to back up this conjecture. In 1938, the linguist and dialectologist P. Peters published a series of papers on DP-internal modifiers in Dutch dialects, and the one on possessive pronouns and demonstratives (Peters 1938) contains the map in Figure 9.

Figure 9: Determiner-demonstrative doubling in 1938 (Peters 1938: 232).

This map provides an overview of the non-attributive use of demonstrative pronouns, and in so doing, offers insight into the geographical distribution of determiner-demonstrative doubling in the early part of the twentieth century. Note how the area in which this phenomenon is attested shows great similarity to the modern SAND-map in Figure 8 and crucially, how it also includes the locations with clitic doubling but without determiner-demonstrative doubling in the SAND-data. This would make it unlikely that this phenomenon is completely absent in the modern versions of these dialects.¹⁴

The other area in the map in Figure 8 that displays clitic doubling but not determiner-demonstrative doubling is French Flanders, a region in the northern tip of France, where a dialect of Dutch is spoken among the older members of the local population. Here, too, we believe the absence of determiner-demonstrative doubling is epiphenomenal: as it turns out, the question probing for this phenomenon was not included in the SAND-questionnaires that were used in this region.¹⁵ This means that from the map in Figure 8 we cannot conclude that determiner-demonstrative doubling is absent in this region; we simply don’t know. Given that French Flanders is clearly included in the relevant area in the historical map in Figure 9, however, we conjecture that the French Flemish dialects have a positive setting for the D-parameter and that they display both clitic doubling and determiner-demonstrative doubling.

4.5 Synthesis: parameter interactions & geographical distribution

In the preceding subsections we have shown how the results of the quantitative, statistical analysis presented in section 3 can be translated into a formal, parametric account of the underlying variation. In so doing, we have proposed the following three parameters:

In this final subsection we want to explore the interactions between these parameters as well as the geographical patterns that underlie them. One way of visualizing this is through the map in Figure 10.

Figure 10: Interactions between the AgrC-, D-, and C-parameters.

This map lists for each of the SAND-dialects its setting for the parameters in (90)–(92). Let us first make explicit how exactly we determined this for each dialect and each parameter: (1) the setting for the AgrC-parameter is based on the presence/absence of complementizer agreement in the dialect; (2) the setting of the D-parameter is based on the presence/absence of clitic doubling, and in addition the two Limburg dialects that show determiner-demonstrative doubling but no clitic doubling have also received a positive setting for this parameter (see §4.4 for discussion); (3) the setting of the C-parameter is based on the three polarity-related phenomena discussed in §4.3.1: a dialect that features at least one of these phenomena gets a positive setting for the C-parameter. We have chosen this approach because there is an implicational relation among the set of seven phenomena regulated by the C-parameter (i.e. the ones discussed in §4.3): whenever a dialect features one of the four non-polarity-related phenomena, it also invariably features at least one of the three polarity-related ones. We take this to mean that polarity serves as the trigger for the language acquirer that the C-parameter should be set to ‘yes’.

With this as background, we can now turn to the patterns displayed in the map in Figure 10. There are two key takeaways that we want to draw attention to. First of all, there is a clear sense of geographical patterning in these data: particularly the white, red, and green areas—and to a slightly lesser extent the brown and black ones—show a fair amount of geographic homogeneity. This means that the parametric account we have proposed identifies fairly well-defined dialect regions. A second point to notice is that three of the eight logically possible parameter setting combinations are dramatically underrepresented in the data. These are patterns that at most three dialects belong to. In order to get a clearer sense of which combinations are common and which ones are not, let us present the data underlying the map in Figure 10 in table form. This is what is shown in Table 6.

Table 6: Interactions between the AgrC-, D-, and C-parameters

We have used the same color coding in this table as in the map in Figure 10. What becomes clear is that the underrepresented patterns—blue, gray, and yellow—are all situated on the same diagonal in the table, i.e. the one where the setting for the D-parameter diverges from that of the C-parameter. In fact, the four least frequent patterns in the table are the ones where the values for the D- and C-parameters do not match. In 89.5% of the SAND-dialects—239 out of 267 dialects—the two parameters show the same setting. The AgrC-parameter on the other hand seems completely orthogonal to the other two. More or less half of the dialects have a positive setting for this parameter (155 out of 267 dialects, or 58%), and the other half a negative one (112 dialects, 42%), and those settings do not correlate with the values for the D- or C-parameters. What this seems to suggest is that even though they are logically independent, the C- and D-parameter might be connected at some level, with a rich nominal left periphery preferably being combined with a rich clausal left periphery and vice versa.¹⁶ One way of modeling such correlated behavior is via so-called parameter hierarchies, as proposed by Biberauer et al. (2014) and Biberauer & Roberts (2015). In a nutshell, what these authors suggest is that parametric variation can be relativized according to the scope that a parameter has. Take for instance the null-argument parameter hierarchy in (93) (Biberauer et al. 2014: 112).

This hierarchy includes one type of parametric variation, namely the presence or absence of uϕ-features on probes. The higher we are on the hierarchy, the wider the scope of the parameter: in the two topmost branches of the representation in (93), the parameter applies to no or all probes respectively. The cross-linguistic differences triggered by these choices are situated at the macrolevel: radical pro-drop languages versus languages with pronominal arguments. The more we move downwards in the hierarchy, the more selective the parameter becomes and the more restricted the set of probes it applies to. The variation triggered at this level is more of a meso-, micro-, or even nanoparametric type. The difference between macro- and microparametric variation, then, is not qualitative, but quantitative in nature. They are at opposite ends of the same scale and reflect a difference in the range of application of a parameter.

Inspired by this approach, we want to tentatively explore to what extent the C- and D-parameter might be part of the same parameter hierarchy, as this could account for their coordinated behavior in Table 6. What we would like to suggest is that both parameters are ultimately related to a parametrization of phase edges and the role A^′-features play in these domains (Chomsky 2007, cf. also Van Craenenbroeck & Van Koppen 2007; Ouali 2008; Miyagawa 2010; Jiménez-Fernández & Miyagawa 2014). Assuming both DP and CP to be phasal, the presence of an extended left periphery can be taken to be an indication of syntactically active—i.e. movement-triggering—A′-features on phase heads. One way of implementing this into a hierarchy is shown in (94).¹⁷

Assume that at the highest, macrolevel stage of this hierarchy we find languages that have no grammatically active A′-features on phase heads at all. This might lead to such heads being altogether absent from the language (see e.g. Bošković (2012) on D-less languages). When A^′-features are consistently present, the languages have phase heads, but they can differ in whether or not these features trigger movement into the left periphery (cf. Miyagawa 2010; Jiménez-Fernández & Miyagawa 2014). The lower portions of the hierarchy could then regulate the distribution of grammatically active A′-features across the syntactic structure, including the extent to which the various phase heads show similar behavior in this respect. In languages where they do, we might expect the C- and D-parameter to operate in tandem, in the way shown in Table 6. Needless to say, at this point all of this constitutes mere speculation, but we do want to highlight the way in which parameter hierarchies can not only provide connections between seemingly independent parameters, but how they can also situate the type of low-level microparametric variation that is the topic of this paper into a much broader macroparametric narrative. It is a direction of research we hope to explore further in future work.