2.1 Slavic

Cross-linguistically, the set of complementizers—items introducing complement clauses—is very restricted. Slavic languages do not constitute an exception to this general pattern as they usually have only two complementizers:² one for interrogative contexts which we dub [+Q] and another one for non-interrogative contexts, [–Q]³ complement clauses; cf. Table 1.⁴,⁵

Polish, for instance, has the complementizers że ‘that’ and czy ‘if,’ illustrated in (1a) and (1b), respectively.

On the other hand, all Slavic languages have a multitude of subjunctors to introduce diverse types of adverbial clauses. This is illustrated in Tables 2 and 3. A number of comments are in order:

The majority of adverbial subjunctors in Tables 2 and 3 are composed of a preposition, a demonstrative and a complementizer. Thus, for instance, Belarusian paslja taho jak ‘after’ in Table 2 comprises the preposition paslja ‘after,’ which governs the genitive case on the demonstrative pronoun taho ‘this.GEN,’ as well as the C-element jak ‘how.’ Essentially the same component parts can be found in Polish dlatego że ‘because’ in Table 3: preposition dla ‘for,’ demonstrative tego ‘this.GEN,’ and complementizer że ‘that.’

But there are examples in the tables that lack one or another of the components mentioned: Thus, for instance, Belarusian tamu što ‘because’ in Table 2 lacks a preposition, while Bulgarian văpreki če ‘despite’ in the same Table lacks a demonstrative. On the other hand, Polish podczas gdy ‘while’ features the (abstract) noun czas ‘time’ in place of a demonstrative pronoun.

However, we argue that these are only apparent exceptions:

First, examples like Belarusian tamu što ‘because’ are instances of so-called semantic cases where a(n oblique) case-marker “does more than just indicate the surface grammatical function of the DP. The DPs in question are not arguments of the main verbal predicate, so semantic case must connect them with the verb […]” (McFadden 2004: 54). In the current example, it is the dative which seems to connect, not a DP, but the CP introduced by što ‘that’ with the matrix clause, at the same time establishing a causal relation between the two. One can both account for the presence of this relation and maintain a uniform analysis for adverbial subjunctors by the assumption that semantic cases are not merely case-marked DPs, but PPs with a null head (cf., a.o., Bresnan & Grimshaw 1978; Emonds 1985; 1987; Nikanne 1993; McFadden 2004; Zimmermann 2013). Under this view, Belarusian tamu što is structurally parallel to e.g. Russian potomu čto ‘because’ from Table 3 with the only exception that tamu što contains a null preposition which is responsible for both the causal relation and the dative case on the demonstrative, whereas the preposition is overt in potomu čto; cf. (2):

Second, the existence of examples like Bulgarian văpreki če ‘despite’ where there is no overt demonstrative is not quite surprising, given that the “missing” element can easily be reconstructed. Still, the specific criteria allowing for the demonstrative to be absent/null need further scrutiny. Possible factors that come in mind are (i) properties of the preposition at hand and (ii) the case governed by that preposition. Additionally, the system/extent of case inflections of the relevant language might play a role – note that Bulgarian and BCMS have a reduced set of case markers as compared to other Slavic languages, and it is these two languages that exhibit the relevant examples in Tables 2 and 3.

Finally, examples with a noun in place of a demonstrative pronoun are well attested in many languages. In these cases, an abstract noun specifies the semantic domain to be modified by the embedded clause. For instance, czas ‘time’ in Polish pod-czas gdy ‘while’ defines that the relevant domain is the temporal one.

After this brief discussion of the data presented in Tables 2 and 3, we would like to go into more detail, using Polish data as particular examples. The subjunctors in (3)–(5), for instance, are combinations of a preposition or prepositional phrase with either the complementizer że ‘that’ or the temporal subjunctor gdy ‘when’ (see Section 4.3 concerning the latter). The meaning of these complex subjunctors derives straightforwardly from their individual component parts, which is why it is fair to speak of syntactic complexity as well as of semantic transparency.

Like (3) and (4), (6) and (7) contain the complementizer że ‘that.’ This time, however, it follows on non-prepositional material, more precisely on the comparative relator jako ‘as’ and the adverb chyba ‘probably,’ respectively. Irrespective of this difference, (6) and (7) pattern with (3) and (4) in that the subjunctors are syntactically complex. On the other hand, they are semantically opaque: Their meaning cannot be calculated from the meaning of their component parts in a compositional way:

Furthermore, there are instances where the complexity of a subjunctor is more intricate, but nonetheless recognizable due to the fact that the relevant items exhibit internal morphological structuring; cf. (8)–(10):

We argue that examples like these are morphologically complex.⁶ Like most morphologically complex subjunctors, the above examples are semantically opaque.

Finally, some Polish adverbial subjunctors seem to be simplex; cf. (11) and (12):⁷

To summarize, while Slavic complementizers are simplex throughout,⁸ the vast majority of adverbial clauses is introduced by complex subjunctors, hence items that (i) involve more than a single syntactic object or (ii) display internal morphological structuring.

Adverbial subjunctors that are, or at least seem to be, simplex will be dealt with in more detail in Section 4.2.

2.2 German

This section extends observations from the previous one to German in support of the CSA. Documentation on subjunctors to verb-final clauses is quite extensive, cf. Fabricius-Hansen (2007) for a detailed descriptive survey and functional classification. We cannot do full justice to the rich repertoire of these elements but will here merely point to what we consider representative cases, highlighting properties relevant to the current analysis.

As in Polish, there are two lexicalized complementizers for finite German complement clauses: one for [–Q] and another one for [+Q] complement clauses, dass⁹ ‘that’ and ob ‘if’, respectively:

In the context of indirect WH-questions, standard German is commonly taken to have a third complementizer, the [+Q] null-C. In doubly-filled Comp variants of German, like Bavarian, this complementizer is dass-C or null-C, a variation that is subject to complex conditions beyond the scope of the current paper (cf. Bayer 1984; Bayer & Brandner 2008).

We follow virtually all work on the issue and assume that the null-complementizer is syntactically simplex, i.e. an instance of C.

As in Slavic, there is a plethora of German subjunctors to introduce adverbial clauses of different types. Both the internal structure and the lexical or derived meanings of the subjunctors are complex in many ways, so that a clear-cut classification proves difficult. However, these complexities do not undermine the purpose of this section: an illustration of the CSA. As a preliminary stab, we would like to submit a threefold descriptive classification, summarized in Table 4.

The first class comprises simplex prepositional elements plus the obligatory occurrence of dass, shown in (15)–(17).

The second class comprises what we take to be an anaphoric or cataphoric demonstrative pronoun in addition to the prepositional element. We take such demonstratives to be part of a DP in the containing PP. (20) shows that in case this demonstrative precedes the preposition, it is an R-pronoun familiar from such PPs, cf. Noonan (2017). A subset of these elements can feature the complementizer dass in addition to the subjunctor, subject to dialectal and sociolectal variation.

Notice that some of the above cases can be modified by degree or adverbial elements, e.g. gleich/sofort nachdem ‘right after,’ clearly indicating the phrasal (i.e. PP) nature of these units.

The third class, finally, involves either C- or P-elements like ob ‘if’,¹⁰ als ‘as’/’than’ or zu ‘to,’ accompanied by a nominal or adverbial element or a sentence particle.¹¹ Orthographically, these elements are one word and we indicate their contained material in the gloss for perspicuity.

As concerns the distinction between morphological and syntactic complexity, the examples show that – much like their Slavic counterparts – the German complementizers are simplex, while subjunctors are either morphologically (e.g. obwohl ‘despite,’ zumal ‘the more so as’) or syntactically (e.g. bis auf ‘except that,’ als dass ‘rather than’) complex.

This concludes the survey on complementizers and subjunctors in Polish (and Slavic more generally) and German. An important question arises: What is the underlying reason for the observed asymmetry between clause types functioning as arguments to embedding verbs on the one hand, and adverbial clauses on the other? Given that the CSA is a strong cross-linguistic tendency, we believe that an account is needed which addresses the following two questions which are, even though related, logically independent: Why are complementizers syntactically and morphologically simplex throughout? Why do subjunctors exhibit a strong tendency for syntactic and morphological complexity? In the following section, we briefly subject to scrutiny previous approaches to the syntax of adverbial clauses in Slavic and German and show that they fall short of even addressing these questions, let alone provide clues of an answer to them.

4.1 Theoretical background

This section provides theoretical background on the analysis that we subsequently develop. Though it might appear circumlocutory and not immediately relevant to the phenomenon at hand, we ask the reader to bear with us: The exposition is crucial for the account that we give.

There is a recent and ongoing effort to formulate grammatical properties in terms of the interaction of (a) the free application of the set-forming operation Merge, (b) properties of the Conceptual-Intentional and the Sensorimotor systems and (c) Third Factor principles, i.e. efficient computation (cf. Chomsky 2013). The framework within which this effort takes place imposes stringent conditions and standards on the postulation of grammatical operations. Next to common scientific standards of parsimony, part of the motivation for this Spartan handling of technical grammatical means comes from considerations regarding the evolution of the Language Faculty:

“not only asking what mechanisms suffice to determine I-language from data available, but why these mechanisms should exist, and whether they are real or just dispensable descriptive technology. The task of accounting for the evolution of language would also be correspondingly eased, for the same reasons that hold for inquiry into evolution generally: the less attributed to genetic information (in our case, the topic of UG) for determining the development of an organism, the more feasible the study of its evolution.” (Chomsky 2004: 4)

There is a containment relation between the different explanatory goals of generative grammar. A theory which goes “beyond explanatory adequacy” in aiming at a plausible scenario for the evolution of the Language Faculty must derive properties of a theory that meets the goal of explanatory adequacy, i.e. a theory that describes UG. One such property is the X̄-module (cf. Chomsky 1970; Jackendoff 1977; Chomsky 1986). Even in its early Minimalist incarnation (cf. Chomsky 1995a), it comprises numerous properties which we consider “dispensable descriptive technology” at least in principle; showing that equal descriptive coverage can be obtained when pursuing a more parsimonious analytical alternative is, of course, a different matter. According to recent efforts (cf. Collins 2002; Seely 2006; Chomsky 2013; 2015), the X̄-theoretic notion of endocentricity derives from and dissolves into two more fundamental principles:

The former is one hypothesis regarding the question why sets formed by Merge need to have a syntactic category in the first place; as the principle is stated, it doubtlessly needs sharpening and operationalizing. The latter is one hypothesis regarding the question how the grammar detects the label and what the most principled conception of the mechanism is, if one does not resort to notions like projection, feature percolation or the like. A leading idea is that an overarching principle of efficient computation, Minimal Search, guides the way in which a label is detected. More specifically, for a given structure {X, YP} = α, the LA picks the simplex element as a label, i.e. the lexical item X. The reason for preferring X over YP within α is that YP is itself a set and the detection of the head within YP requires deeper search – which Minimal Search precludes. The following scheme illustrates the procedure:

That is, upon applying the LA to α, α=XP. Notice that in the representation on the right in (36), the label is not structurally represented. It must be assumed that the effect of applying LA at the phase level stores the information of treating this (structurally invariably unlabeled) set as XP.

There are two more logical combinations Merge can create: {Y, X} and {YP, XP}, and we will for the time being consider the latter configuration only.¹⁷ Chomsky (2013) suggests two strategies to repair the labeling problem of {XP, YP}-structures: symmetry-breaking movement (based on ideas by Moro 2000) and sharing of a prominent feature. In the first case, the leading idea is that only the head of a movement chain is visible to syntactic operations,¹⁸ while lower chain members, defined as “occurrences” within a given syntactic context, are invisible. Consequently, LA fails to see a lower occurrence and its sister wins through. In (37) XP is the ensuing label after evacuation of {XP, YP} by YP.

In the second case, there is an underlying assumption that the heads of the two terms must share a prominent feature and agreement must hold between them (for reasons pending clarification; cf. Chomsky 2013: 45). LA scans both these terms, finds one interpretable and one uninterpretable feature on the heads of both, and forms a complex label from them.

We leave it to future research to decide on the question if the procedure of searching within two phrases in parallel is a viable option and compliant with Minimal Search. Chomsky (2013) appears to assume so and only structures involving AGREE between the heads yield a convergent output for the Conceptual Intentional systems.

By now these strategies have been employed in analyzing diverse phenomena, e.g. copular clauses with a contained small clause, successive-cyclic Ā- and A-movement, EPP-raising in English, split topics in German, coordination, etc. (cf., a. o., Blümel 2012; Chomsky 2013; 2015; Seely et al. 2014; Ott 2011). Notice that the adoption of this framework opens a host of questions, and a myriad of phenomena formerly captured under the rubric “specifier” needs to be rethought and recast, an enterprise that is obviously beyond the scope of this paper. Here we would like to make one modest proposal within the approach regarding adverbial clauses in Slavic and German.

Blümel (2017) proposes a third strategy. He proposes that under narrow conditions, sets formed by Merge can remain unlabeled.¹⁹ Thereby the general validity of (34) is called into question. He argues that V2-languages feature a structure {XP, CP_V2} in which the “prefield” is defined as the category in the sister-position of CP_V2. In the absence of the notion specifier, units formerly analyzed as contained in a V2-CP must receive a different treatment. He argues that in the labeling-analysis, placing an XP in the prefield fulfills the following interface condition:

What is the rationale behind (39)? There is a tight connection between root clauses and speech acts, so the latter have to be part of the picture. One way to think of (39) is to say that root clauses are typed in different ways, say, by their featural content in addition to a particular prosody. There might be conventionalized rules that connect the clause type with a given speech act. Pragmatically then, speakers make use of particular structural formats to convey certain meanings (proposition, question, etc.). Every such meaning is syntactically typed in a particular way. Root declaratives, on the other hand, might be an elsewhere case, i.e. the unmarked option speakers resort to when making assertions. These do not require any particular marking and, in fact, must not have any particular marking if (39) is true. That is, (39) is a strategy to unmark the clause – and this unmarked clause type is conventionally associated with assertions.

Now syntactically, placing an XP in the prefield emerges as one strategy to meet (39). The principle (34) needs refining in that a label is needed for interpretation, the ongoing derivation and more specifically: selection. This qualification was hinted at in Chomsky (2008: 141):²⁰

“Each S[yntactic]O[bject] generated enters into further computations. Some information about the SO is relevant to these computations. In the best case, a single designated element should contain all the relevant information: the label. […] the label selects and is selected in E[xternal]M[erge …].” (our supplements and emphasis)²¹

Since V2-clauses are root phenomena, the derivation terminates, which renders labels superfluous, Blümel (2017) suggests. If root declaratives must not be labeled, prefield occupation is but a means to ensure this, given (35).²² Core properties of V2 follow, such as obligatory XP in the prefield (40),²³ XP’s heterogeneity (41) and the expletive prefield-es ‘it’ (42).

Following Reis (1997) who endorses an adjunction view of embedded V2 in German, Blümel (2017) proposes that phenomena like (43), which, roughly put, are licensed in the presence of so-called bridge verbs, involve the introduction of the subordinate V2-clause by Pair Merge (Chomsky 2004) to yield a structure <α, TP>:

Crucially, unlabeled syntactic objects are restricted to specific contexts: root contexts, of which adjunction – here conceived as Pair Merge – can be a subcase.²⁴ In addition, contexts could qualify which exempt syntactic objects from being interpreted for some independent reason, i.e. objects which do not violate Full Interpretation (cf. Obata 2016 for this reasoning).

With this much in mind, consider the question posed at the outset of this paper: Why do adverbial clauses exhibit a strong tendency to feature complex subjunctors, in contrast to complement clauses? That is, what is the deeper reason of the CSA? In light of the labeling theory just presented, we would like to conjecture the following generalizations:^25,26

One deeper reason behind the CSA is that the PP-part of the complex subjunctor effectively clogs the application of LA because it can, and no derivational or interface principles prevents it. The same is not true for complement clauses, which must receive a label as they enter into selectional relations. This way, we recast the analysis of adverbial clauses while providing an explanatory rationale behind CSA (cf. Hornstein & Nunes 2008 for an approach to adjuncts more generally, which is similar in spirit).²⁷

4.2 Complex subjunctors

So far we have shown that there are reasons to believe that the syntax of natural languages employs more exocentric structures than hiterto assumed.²⁸ We believe that a case can be made that adverbial clauses have this property and that the hypothesis can shed light on the CSA. Taking the transparent complexity of the subjunctor elements at face value, we propose that a subset of subjunctors are phrasal, full-fledged XPs (mostly PPs), not merely heads,

Each of these complex subjunctors comprises a prepositional element plus additional elements. The following list exemplifies the variation in the complexity:

• A nominal element statt ‘stead’ in (46a) ≈ czas ‘time’ in (47a);

• A demonstrative dem in (46b), (46c) and (46d) ≈ tego/tym/to in (47b), (47c) and (47d);

• An R-pronoun da- in (44e) ≈ -nim in (47e).

Add-ons to the prepositional element either directly or indirectly have a correlating function. Particularly, they correlate with the proposition that they co-occur with. It should be noted that, in German, adverbial subjunctors show three patterns of co-occurrence restrictions with complementizers:

• C = dass is optional, i.e. dass and null-C are in free variation with some XPs; cf. (48) (cf. Müller 1995):

• C = dass is obligatory with some XPs; cf. (49):

• C = dass is impossible with some XPs; cf. (50):

Note that clauses introduced by the subjunctors in (48) can only be finite; cf. (51).

On the other hand, clauses introduced by the subjunctors in (49) become necessarily non-finite in the absence of dass; cf. (52).

This difference with respect to the availability of non-/finiteness is indicative that the presence of dass is required in case a subjunctor allows for both finite and non-finite clauses/TPs, hence is a means of desambiguation; cf. (49) vs. (52). If there is no choice anyway, dass is optional but standardly omitted; cf. (48). To explain the ungrammaticality of dass in cases like (50), we will propose to analyze them as (hidden) relative clauses with weil, da and als located in C; see Section 4.3.1.

For subjunctors as in (46) and (47), the structure of adjunct clauses is something like (53) and (54) (compare with this the traditional analyses (25) and (27)):

(53)

More specifically, in most cases the PP embeds a demonstrative DP (e.g. German dem ‘this.DAT,’ Polish tego ‘this GEN’) or a nominal (e.g. German statt ‘stead,’ Polish czas ‘time’). This DP or nominal entertains a relationship to the CP that we might implement in terms of either coindexation (in case of cataphors in D) or semantic restriction (in case of abstract NPs/DPs). The demonstrative or the nominal itself is a (semantic) placeholder which needs to be explicated, and it is the CP which fulfills the function of explicating it.

It should be added that some items seem to be lexically cataphoric, i.e. they cannot freely function as either an anaphor or cataphor but must obligatorily precede the “antecedent.” While this is not the case in German and Polish (where the d- and t-pronoun, respectively, can be both cataphoric and anaphoric), Russian exhibits a systematic distinction between the strictly cataphoric pronoun tot (as in po-tomu čto ‘because’) and the strictly anaphoric pronoun ėtot (as in po-ėtomu ‘because of this’).

Returning to (53), the PP and the CP are hierarchically on a par. While traditional analyses highlight the prepositional nature of adverbial clauses, our analysis does not attribute a privileged or dominant status to either term in {PP, CP}, thus expresses both the prepositional and the sentential character of these clause types. Due to X̄-theoretic constraints, previous analyses were forced to structurally subordinate the CP. But the intuition that the sentential component partakes in the nature of the whole adverbial clause has long been around. As Müller (1995: 85) points out:

“Actually, there is diachronic evidence that all finite adjunct clauses involve a ‘doubly-headed’ structure in German. Lenerz (1984: 110 ff.) has pointed out that in Old High German, adjunct clauses were clearly decomposable into a preposition (-like element), belonging to the matrix clause, and an embedded genuine daß-clause, the two elements merging only later in the development of German. Further evidence for splitting the left periphery of adjunct clauses into a CP with a semantically empty complementizer as its head, and a higher XP the head of which contains the relevant semantic information (temporal, consecutive, causal, etc.) can be gained from French or Russian. …”

In the current context, the notion of a double head of adjunct clauses can be reinterpreted in a way current theorizing allows (if arguably not intended and/or envisaged in this way by Lenerz/Müller).

Notice that (53) raises a labeling problem, given the LA: Neither P nor C labels α. The problem is resolved neither by XP-movement nor by sharing of a prominent feature. We claim that the labeling problem remains unresolved, i.e. α is exocentric, cf. also Hornstein & Nunes (2008) on adjuncts more generally. Absence of a label is arguably the null hypothesis for unselected clause types like adjunct clauses (or root clauses; cf. Emonds 2004; 2012):

Notice that in languages like German, there is a potential corollary of (55): Can uncategorized syntactic objects occupy the German prefield?

There are two possible responses to this observation. One holds that (56) is what it looks like: an adverbial clause in the prefield, occupying SPEC-CP (in traditional parlance). In that case, there appear to be syntactic objects which simultaneously remain unlabeled throughout and which are introduced into the derivation by Set Merge. This might not be a problem after all insofar as prefield occupation is special in that it does not represent an intermediate step of the derivation but in fact its termination. If it is true that elements require a label which are introduced in intermediate stages and which partake in selectional relations, the element to halt the derivation might not be contingent on the syntactic category. That is, we have an exception which however, might be a principled one.³⁰ Adopting the approach to V2 by Blümel (2017), we can say that such structures involve {α, CP_V2} = β, where both α and β are unlabeled. An anonymous reviewer asks why it is not CP_V2 which wins through in this structure, because, impressionistically, Minimal Search detects it before the unlabeled sister α. Arguably, the notion Minimal Search needs to be sharpened. Here we would like to contend that the labeling procedure stops immediately at the point of failure, i.e. there is no “comparison” between α and CP_V2 in the search procedure: β is inspected, no label is found, search ends.

The second approach to (56) treats it as an instance of left-dislocation. In that case, (56) is analytically on a par with (57) modulo the overtness of the proform da ‘there.’

For now we leave open the issue of the precise analytical details.

Merger of PP – a complex complementizer – is but one strategy to suppress labeling of an adjunct.

4.3 Simplex subjunctors

All languages investigated exhibit a number of (apparently) simplex adverbial subjunctors. Table 5 lists some examples.³¹

There are at least two theoretical options to explain the “simplicity” of the elements in question:

1. Despite structural simplicity at the surface, a subset of the relevant items are underlyingly complex, presumably involving an overt XP to the left of a zero-headed CP. If this is on the right track, the examples in question do not differ structurally from the transparently complex examples analyzed in Section 4.2. A special subcase are subjunctors that introduce “hidden” relative clauses involving a covert relative operator within an XP to the left of CP. See Section 4.3.1.

2. Another subset of the relevant items turn out not to be (hypotactic) subjunctors altogether. Rather, they are paratactic relators, hence conjunctions with a specific adverbial (relational) meaning. If so, the respective items and clauses are irrelevant to the present discussion. See Section 4.3.2.