Je-Desto, Je-Umso: An Analysis of the German Comparative Correlative Construction

Abstract

This paper argues that German je-desto-sentences are regular verb-second structures. Unlike left-dislocation(-like) structures (such as wenn-dann-clauses or free relative clause constructions with case mis-match and resumption), je-desto-strings are normal, regular prefield structures. A proposal from the literature is developed further, accord-ing to which je-clauses, like relative clauses, belong structurally to the clausal constituent into which they are integrated. The head of these constituents remains in its canonical middle field position, whereas relative as well as je-clauses usually extrapose to the right. If the host constituent is moved to the prefield, the attributive clause is carried along with it, with the je-clause realized initially and the relative clause usually finally. However, relatives may also precede their head noun. Such an analysis highlights the common features of clause types that are otherwise treated as fundamentally distinct from one another, without denying the differences and postulating construction-specific verb-third realizations. In addition, various other realization options are discussed and dismissed as grammatical illusions.*

1. Introduction

Je-desto-clauses have been a highly debated structure in German linguistics—see, for example, Bech 1964, Beck 1997, Reis 2009, Speyer 2011, Fortmann 2016, Meinunger 2018. Often, but certainly not always, these contributions have been inspired by the work on the so-called comparative correlative construction, which has mainly been proposed for English (Thiersch 1982, McCawley 1988, Culicover & Jackendoff 1999, Borsley 2004, and particularly den Dikken 2005).

Although nearly all of the approaches cited above have been developed in the tradition of generative grammar, most of them agree that the structure has something “construction-like” about it; that is, these approaches assume specific characteristics of the je-desto-structure, so that it cannot be generated in a purely compositional way without preconditions. Typical sentences are given in 1.

1. (1)

   

One of the main reasons why the proponents of this view claim that the structure is construction-like and cannot be derived straightforwardly by regular rules is the (apparent) fixed order ‘dependent clause > matrix clause’. Reversing the order results in an extreme lessening of accept-ability, as shown in 2.

1. (2)

   

Other syntactic realizations, although only slightly divergent, are fine: In English, the structure becomes acceptable without the in the initial matrix clause, and in German things become acceptable if the main clause contains the element umso instead of desto:

1. (3)

   

   ¹

This interesting pattern suggests a line of argumentation which appears reasonable at first glance. However, such an approach creates a problem that seems to be more difficult than the issues it solves. Although this problem does not become evident and perhaps does not even have any effect in most languages (such as English), it arises in verb-second languages, especially in German. Thus, the analysis proposed here may be adopted for German (only) and disregarded for English and other languages, including Dutch, which is a verb-second language but is different from German (see den Dikken 2005). However, perhaps—and hopefully—it can be adopted more generally and finally capture the structural conditions more elegantly, convincingly, and adequately.

The structure of this article is as follows: In section 2, I introduce two left dislocation structures that have been repeatedly discussed as parallel constructions, that is, related to the je-desto/umso-clauses. However, I show that there is a fundamental difference. In section 3, I discuss previous research on the status of the je-protasis, adopting the analysis of the je-clause as a clearly subordinated embedded clause, but not subscribing to any of the previous existing structural proposals, which assume a verb-third structure. In section 4, I show that none of these structural descriptions can account for realizations where the je-protasis surfaces in slots other than the absolute initial position. In section 5, I highlight the common features of je-clauses on the one hand and restrictive relative clauses on the other. First, I highlight the features that both types (of subordinated clauses) have; then I discuss actual and apparent differences. Finally, I analyze both types as attributive clauses that contain a subordinated constituent. Section 6 discusses other positional options for the je-clause, some of them previously undiscovered. In particular, I argue that certain linearizations usually considered ungrammatical are, in fact, well-formed; at the same time, certain attested realizations that are relatively unobtrusive are, in fact, grammatical illusions and should be analyzed as ill-formed. Section 6 provides a first rough formal structural proposal. In section 7, I argue two more detailed internally differentiated variants of this proposal. Section 8 gives a summary of the results. The appendix presents various cases in which the relevant morphemes or words (je, umso, desto) are interchangeable to varying degrees.

2. The Comparative Correlative and Its Relatives

For both semantic and syntactic considerations, parallels have been drawn between je-desto-clauses on the one hand, and if-then conditionals and certain left dislocation structures on the other. Authors such as Beck (1997) and McCawley (1988) point out that the so-called protasis—the je-clause—expresses a condition similar to an if-clause:

1. (4)

   

In addition, den Dikken (2005 for Dutch and in general) and Reis (2009) and Meinunger (2018) (for German) also observe the similarity of the je-desto-construction to left dislocation structures:

1. (5)

   

However, je-desto-clauses differ from the other two constructions with respect to one important feature: In conditional sentences, such as 4a, as well as in canonical NP left dislocation structures, such as 5, the two preverbal expressions are synonymous, or coreferential. That is, the left-peripheral (subordinate) clause and the resumptive element have the same referent. This, in turn, means that in the given cases, one of the two constituents can be omitted without creating an ungrammatical structure—and without changing the meaning of the complex assertion. Examples 6a,b demonstrate that the semantically empty resumptive pronoun is optional. In contrast, the possibility of omission does not exist for the semantically substantial desto/umso-phrase: It must be realized, as shown in 6c. Moreover, it must be realized in right adjacency to the je-protasis. A simple je-phrase topicalization with the correlative umso-phrase occupying the middle field position leads to ungrammaticality as well, as shown in 6d.

1. (6)

   

Under the traditional analysis sketched in more detail in the next section, the je-desto-construction would be the only construction where the initial constituent must be realized; this makes it the only obligatory verb-third structure. In any other case, where the finite verb is in C° and where the pre-prefield (initial) constituents have been argued to be structurally (syntactically) affiliated with their respective clauses, these initial constituents can be omitted without any consequences for the grammat-icality of the sentence. This important difference should lead to a recon-sideration of the structure.

3. Main or Subordinate Clause Status?

There is no doubt that the je-clause (or the respective protasis) is a syntactically subordinated clause (much like conditional wenn-clauses or free relative clauses). The previously influential approach of Culicover & Jackendoff (1999), which argued for a paratactic structure, has been successfully refuted by many researchers, for instance, by den Dikken (2005) for English (or generally) and by Reis (2009) for German. There are several arguments in favor of a structure in which the je-protasis is integrated into the complex sentence. ² The most convincing one to be mentioned here is (semantic) binding, as shown in 7 (Beck 1997:236; for more details concerning binding, see section 5).

1. (7)

   

This pattern is exactly the same as with donkey anaphora inside conditionals, such as 8, where an indefinite in the protasis delivers a variable bound by the (same) operator that also binds the pronoun in the matrix clause, that is, in the consequence.

1. (8)

   

Examples 7 and 8 show that unselective donkey sentence binding in je-desto/umso-clauses clearly parallels binding in conditional wenn-clauses.

The other arguments are based on the illocutionary independence of the apodosis and the inherent nonspeech act status of the je-clause. Related to this is the incompatibility of a (question) tag with the je-part and its exclusive attachment to the root (that is, the matrix desto/umso-) clause. ³

Taking these observations as evidence for the embedding constellation, Beck (1997)—elaborating on and referring to von Fintel 1994—proposes the structure in 9a. Reis (2009) proposes essentially the same structure shown in 9b (as does den Dikken 2005). Both structures contain a left-dislocated CP.

1. (9)

   

However, these trees do not account for one feature, which remains perplexing. Curiously, an indispensable part of the construction—whether a clausal constituent (of the first degree) or an attribute—is or must be realized outside the core clause: The obligatory je-clause, if it appears sentence-initially, appears detached from and (right) before the prefield, which only contains the desto/umso-adjectival phrase (DegP in Beck’s representation). In Beck’s as well as in Reis’s structure, the je-protasis is adjoined to the CP, which, in principle, is already a complete clause. Such an analysis does not seem satisfactory.

4. Further Weaknesses of the Left Dislocation Approaches

In light of the root versus nonroot distinction, the left dislocation approaches have other problems. Two further interrelated reasons to reject the CP-adjunction analysis are as follows. The analyses in 9 treat je-desto/umso-structures on a par with i) wenn-dann (or wenn-so, both: if-then) structures (see 4a, 6b, 8) and/or ii) free relatives in combination with resumption (see 5 and 6a). Both of these allegedly parallel structures are instances of left dislocation. Left dislocation is a phenomenon of the left periphery (alternatively called the C-domain; Altmann 1981, Grewendorf 2002, Frey 2004). In fact, the preverbal string—with its two parts—cannot surface in any lower position, which means that it cannot be placed in the middle field. In contrast, the complex je-umso-constituent can appear in a lower position. Reis (2009:233) cites 10a and Fortmann (2016:132, 133) gives the examples in 10b,c, with the latter even showing that the je-clause provides a binding domain for the subject of the matrix clause.

1. (10)

   

Such a distribution is completely excluded for the two other true left-dislocation structures:

1. (11)

   

   ⁴

What is more, left dislocation is a main clause phenomenon (Frey & Meinunger 2019). Main, or root clause phenomena are argued to be possible in root clauses only. Such phenomena include, for example, the so-called modal particles (in German), speaker-oriented adverbials, certain expressive items, specific types of topicalization, tags, etc. In most cases, clauses containing root (clause) phenomena are syntactically inde-pendent and have illocutionary force. However, it has been known for some time that these phenomena are also licit in a very restricted set of dependent clauses. Such dependent clauses are arguments of assertive predicates, such as verbs of saying, thinking, evidential predicates, and the like. Factive verbs, in contrast, (and response stance verbs) are assumed to not license root phenomena. Frey & Meinunger (2019) report the contrast in 12.

1. (12)

   

   ⁵

For the sake of completeness, it should be mentioned that the same degree of unacceptability is observed in other versions of 12a. For example, in 13a, the complementizer wenn ‘if’ is used, which would be another very natural option in this context. Compare it with the acceptable example in 13b, where the factive verb bedauern ‘regret’ is followed by verb second (or verb third); in 13b, the name Otto is not topicalized but appears inside its regular clause. Example 13c is already excluded due to the ban on verb fronting (verb second) under factive predicates.

1. (13)

   

Note that the ban on the occurrence of verb fronting under factive and response stance verbs does not seem to apply in je-desto/umso-structures, which embed quite naturally:

1. (14)

   

As one can see, however, the embedding in 14 is not a left dislocation structure. The je-clause surfaces either in the middle field, as in 14a, or in the extraposition site (postfield), as in 14b. An attempt to mimic 12b, where the je-clause as a left-peripheral constituent precedes the core CP introduced by dass and would hence count as an instance of left dislocation, results in ungrammaticality: ⁶

1. (15)

   

An example from Fortmann (2016:133) in 16, where a je-desto/umso-structure appears inside a regular central adverbial clause, points to the same conclusion.

1. (16)

   

Thus, in all of the pertinent cases, the je-desto/umso-structure occurs in the middle field. This placement is incompatible with the left dislocation analysis.

Another piece of evidence that the two allegedly parallel left dislocation structures are actually different comes from the following observation. If the initial string is a je-clause, it may also surface in a short(ended) form. Fortmann (2016:128) discusses this variant (and related ones) in more detail. If the clause is or corresponds to a copular construction, the relevant verbal part of sein ‘be’ can remain silent:

1. (17)

   

Looking at the other two types, one sees clearly that this type of ellipsis is not possible, as shown in 18.

1. (18)

   

This uneven pattern indicates that the structures are different, that is, je-desto/umso-clauses are not (completely) syntactically parallel to the left dislocation structures—conditional wenn-dann-clauses or free relatives with resumption. In section 6 I give an explanation for this observation.

5. The Alternative

5.1. Je-Clauses and Relative Clauses

There is a relatively unnoticed side path in German linguistics that advocates for another structural representation of the construction. The present paper builds on this approach and ends up with a structural proposal that analyzes je-desto/umso-sentences as regular verb-second clauses. The main idea to which this analysis can be traced back is laid down in Speyer 2011. Speyer argues that je-clauses should be considered analogous, or parallel to canonical restrictive relative clauses. Such an idea does not seem unreasonable in view of the structures in 19.

1. (19)

   

In both cases, the anchor expression is in the middle field and the associated subordinate clause—the je-clause in 19a and the relative clause in 19b—is in the so-called postfield, which is the target area of extraposition. ⁷

Note once more that all the proponents of the construction’s left dislocation status (Beck 1997, den Dikken 2005, Reis 2009, and others) explicitly deny the relatedness of this structure—with the linearization ‘main clause > je-clause’—to the “real” correlative structure displaying the order ‘je-clause > main clause’. However, interestingly enough, they all refer to the order in 19a whenever they want to argue for a specific point that cannot (easily) be made clear with the ‘je-clause > main clause’ linearization. Reis (2009), for example, refers to it as a plausibility argument (“Plausibilitätsargument”) when she discusses the degree of integration of the je-clause. Despite all the good and valid arguments that she presents for the dependence of a je-clause on the main clause, she argues for a very loose connection between the two clauses (disintegration, unembeddedness; Reis 2009:230). There is some literature on the serialization of subordinate clauses in the postfield (for example, Reis 1997 or Haider 1997, 2010). ⁸ In order to demonstrate the relatively high independence of the je-clause (in general), Reis shows that a je-clause in the postfield must be “outermost” in the periphery, as opposed to a relative clause, which tends to appear first.

1. (20)

   

In section 5.3, I elaborate on the ordering options in the right periphery. The main goal here is to confront Reis with the following question: If there is no real kinship between left- and right-placed je-clauses, why put forward such an argument? Also, Beck (1997:236), while denying the basic relatedness of left- and right-peripheral je-clause occurrences, lists left- and right-peripheral je-clauses side by side in a set of examples. What is more, Beck dedicates a great deal of space to an impressive formal derivation showing that 21a,b (her 15 and 48) (can) have the same meaning (a point, which Reis makes as well).

1. (21)

   

In contrast, in Speyer’s analysis, je-phrases, like regular relative clauses, are originally and actually subconstituents of a larger phrase: (Restrictive) relative clauses are embedded in a nominal phrase whose head is a noun; je-clauses, for their part, are embedded within a complex constituent that contains the umso-phrase as a (sort of) head. In this respect, relative clauses are attributive clauses just like je-clauses, that is, they are not first-degree clausal constituents in their matrix clause.

1. (22)

   

To make the structural relationship between the two types of sentences plausible, Speyer compares them—for the most part inspired by the remarks in Reis’s article—and shows how and to what extent they behave in the same way. In particular, je-clauses and regular restrictive relative clauses share the following essential properties: Both are incompatible with root clause phenomena and are infelicitous as elliptical answers. Furthermore, they provide a domain for variable and operator binding, and they cannot occur unaccompanied in the prefield. Each of these properties is discussed in the next section.

5.2. Similarities Between Je-Clauses and Relative Clauses

The first similarity between the two types of clauses concerns their incompatibility with so-called root, or main clause phenomena, such as speaker-oriented or other high sentence adverbials, modal particles, epistemic adverbial expressions, etc. (see discussion in section 4). As shown in section 3, je-clauses are incompatible with root phenomena, which suggests that they are subordinate structures. Speyer (2011) clearly and convincingly demonstrates that restrictive relatives pattern with je-clauses in that respect. In contrast, appositive relative clauses, which are unintegrated and can often be considered quasi-independent, syntactically as well as semantically, allow root clause phenomena. Tag questions are another example of a structure that is only compatible with root clauses: Tag questions do not attach to je- or restrictive relative clauses, but are fine with appositives (see, among others, Frey & Meinunger 2019).

The second important feature is that je-clauses and relative clauses are both problematic if used as a stand-alone answer to a question (as a so-called term answer). Examples 23 and 24 show that they can only figure as true subconstituents. In this respect, a kind of pied-piping effect is observed.

1. (23)

   

1. (24)

   

Another clear similarity between the two sentence types is (semantic) binding. Certainly, binding into unambiguously integrated, that is, subordi-nated, sentences is straightforward. It is argued and generally accepted that quantifier binding is an unambiguous sign of subordination (see section 3): A pronoun in a subordinate clause can be interpreted as bound only if it is c-commanded by the associated quantifier. This, in turn, is only possible if the quantifier in the host clause is syntactically higher than the host clause of the pronoun. If binding is not possible, this suggests a higher position—perhaps even structural independency—of the pronoun’s host clause. Again, je-clauses and restrictive relative clauses behave the same way in that they allow binding, as shown in 25a and 25b, respectively. In contrast, appositive clauses do not, as shown in 25c.

1. (25)

   

Things are trickier with the order in which the binder appears linearly after the bindee. As shown in section 3, unselective “donkey sentence” binding in je-desto/umso-clauses clearly parallels binding in conditional clauses. The work of Haegeman (2004, 2006) and Frey (2011)—as well as von Wietersheim (2016) on the experimental side—shows that subordinate clauses expressing a condition originate as deep in the matrix clause as an adverbial clause can possibly originate, that is, below the canonical subject position. In this sense, they are “central” adverbial clauses (Haegeman’s term). ⁹ Example 26a illustrates a conditional construction parallel to the je-construction in 25a: The subordinate je/wenn-clause appears on the right side, and the binding works in the parallel fashion, with the ‘binder > bindee’ order. However, canonical variable binding as in 26a becomes marginally more difficult when the if-clause is sentence initial, as in 26b. ¹⁰

1. (26)

   

The same is true for je-clauses. For Reis, 27a is fully acceptable, and 27b gets a single question mark, which means it is slightly marked but quite grammatical. For Frey (and many more, pers.commun.), 27d,e are perfectly grammatical, whereas 27c is not.

1. (27)

   

These data show clearly that je-clauses are subordinated and (originally) embedded deeply enough.

The third clear parallel between je-clauses and restrictive relatives involves variable binding. It is commonly accepted that in order to allow for variable binding, relative clauses can—but do not have to—reconstruct, as in 28b. The same is true for je-clauses, as shown in 27a,b,d. Thus, no principle C effects arise with preposed relative clauses: A name (an R-expression) and a coreferential pronoun can appear in either the matrix or the subordinated clause. The same holds for je-desto-clauses.

1. (28)

   

   ¹¹

These data thus once more point to commonalities between restrictive relative clauses and je-clauses.

Finally, neither je-clauses nor restrictive relative clauses can appear alone in the prefield. This commonality is very clear and straightforward, yet it is not mentioned anywhere. As a true attribute, a relative clause cannot be moved to the sentence-initial position while leaving the head noun behind, as shown in 29a; nor can a je-clause be separated in this way from the umso- or desto-constituent, as shown in 29b.

1. (29)

   

To summarize the discussion so far, all these parallels suggest that both clause types are of a very similar structural make-up. Restrictive relative clauses as well as je-clauses are attributes, that is, they originate inside a larger sentential constituent, where they are c-commanded by an essential element of the host: i) the determiner in relative clause structures or ii) the desto/umso-item in connection with the comparative mor-pheme(s).

5.3. Apparent (and Real) Differences

Despite the similarities outlined above, there seem to be some ways in which restrictive relative clauses and je-clauses behave differently. I argue, however, that the differences are only apparent. It was pointed out by Speyer (2011) that relative clauses and je-clauses contrast with respect to their linearization options. This observation can be traced back to an (apparent) discovery by Haider (1997, 2010) that in the postfield, relative clauses necessarily precede argument clauses, as in 30a,c. If the order is reversed, as in 30b,d, the sentence is ungrammatical (in the examples below the relative clauses appear in square brackets and the argument clauses are underlined). Speyer’s evidence for this claim is given in 30a,b; the original examples from Haider are in 30c,d.

1. (30)

   

Speyer (2011) argues that in contrast to relative clauses, je-clauses in the postfield necessarily follow the argument clause. The alleged difference is illustrated in 31. The minimal pair is meant to show that je-clauses, unlike relative clauses, cannot be first in the postfield, but can only appear after argument clauses. This accounts for the ungrammat-icality of 31b, where the je-clause precedes the argument clause of glaubt ‘believe’ …dass der Angeklagte unschuldig ist ‘…that the accused is innocent’.

1. (31)

   

This observation, if correct, would mean the end of parallelism between restrictive relatives and je-clauses. However, Schrinner & Opitz (2016) and Richter & Opitz (2018) are able to show that, contrary to Haider 1997, 2010, under certain circumstances, the linearization ‘argument clause before relative clause’ is not excluded. In other words, relative clauses may also sometimes follow the argument clause, just like je-clauses in 31a. Among the factors that play a role, finiteness is a decisive feature. It is important whether the argument clause is finite or nonfinite: Infinitival clauses are very likely to come first, as shown in 32a. However, even finite relative clauses are quite acceptable if the argument clause realizes the object and the relative clause modifies the subject, as in 32b (quite opposite to 30d, where the relative clause is linked to the object and the argument clause realizes the subject).

1. (32)

   

Richter & Opitz (2018) argue that there is no grammatical rule or principle that excludes argument clauses after relative clauses, as shown by the grammaticality of 32. In experiments they did indeed find that in some contexts, there was a gradient preference for relative clauses before argument clauses; however, there are no constraints that would rule out the other order categorically. With grammaticality judgment elicitations, they showed that both serializations are accepted. None of their examples in 33 were judged bad.

1. (33)

   

   ¹²

What is more, Speyer’s (2011) judgment in 31b, in which the je-clause follows the argument clause, is much too strong. ¹³ This sentence is no worse than the examples in 32, in which the relative clause follows the argument clause. In this respect, yet another commonality between the two attribute clause types can be established: Not only can they both precede the argument clause in the postfield, they can also both follow it.

In sum, then, it can be said in general terms that all the common features cited argue for a quasi-equal treatment of je-clauses and restrictive relative clauses. However, this is not to say that restrictive relatives and je-clauses are semantically and/or syntactically identical, with the one—presumably, je-clauses—forming a subtype of the other (restrictive relatives). ¹⁴ For example, one difference is brought up in Meinunger (2011): je-clauses seem to potentially be able to contain multiple operators. This is also true for (embedded) questions and free relatives, but never simple regular relative clauses. Furthermore, researchers, especially Bech (1964:56), have claimed that je-clauses are not (or hardly ever) compatible with negation—a situation which is completely different from the one with relatives.

What is more, although there is no grammatical difference with respect to the potential linearization options for argument clauses, regular relative clauses, and je-clauses, there are some differences in terms of frequency and intuitive evaluation of available options. Je-clauses are less likely to occupy a linearly preceding position, leaving the first postfield slots for canonical restrictive and/or object clauses (not to mention their propensity to appear initially in the fronted version, which is only exceptional for relatives, see below).

A similar argument comes from another tendency. Relative clauses are typically optional elements: In traditional terms, they are considered modifying adjuncts that are not required to be realized. In contrast, je-clauses seem to be an obligatory part of the construction. These absolute claims, however, turn out to be too strong. There is a type of restrictive relative clause that is obligatory; namely, the type that appears with the determiner der∼/die∼/dasjenige (Blümel 2011:1; see next paragraph):

1. (34)

   

At the same time, Reis (2009:226, for example) mentions that there are umso-constituents that do not require a correlative je-part, as shown in 35.

1. (35)

   

Conversely, je-clauses can also be licensed by elements other than umso—for example, by comparative forms of adverbs, as in 36a, or by verbs of gradual change, as in 36b (the examples are from Reis 2009:226, who credits Bech 1964).

1. (36)

   

These data show that—despite the observable tendency to cluster differently—, relative clauses and je-clauses show no difference, in principle. However, what is important—and this is the essential point of this paragraph—is that both types of subordinate clauses, relative clauses and je-clauses, are attributive clauses: They are subconstituents of a single compact, complex constituent in the prefield. Thus, in the base structure, they are properly embedded within this larger phrase.

6. The Proposal

6.1. Inversion of the Attributive Part and Its Consequences

An analysis that comes closest to the structural proposal in this paper is provided by Fortmann (2016). His analysis is similar—or, from a certain perspective, even identical—to the one proposed in Roehrs et al. 2002 or Huber 2007. Fortmann bases his approach on Speyer’s (2011) view that relative clauses and je-clauses should receive a similar treatment. However, unlike Speyer, Fortmann analyzes the prefinite verb string as a complex but single constituent that properly includes the initial je-clause. Thus, in contrast to den Dikken (2005), Beck (1997), Reis (2009), and even Speyer (2011), who all adopt the CP-adjunction structure in 37a, Fortmann adopts the structure in 37b.

1. (37)

   

This is exactly the structure that I want to advocate here as well (for more on the derivation, see below).

Fortmann (2016) points out one feature that sets je-clauses and relative clauses apart, namely, the position of the clause relative to the construction head. Je-clauses virtually always precede the head of the AP constituent (that is, umso/desto), whereas relative clauses obligatorily follow the head they belong to: “With relative clauses the unmarked order is N before relative clause, while with comparative conditionals it is the je-protasis before desto” (p. 137). I argue, however, that this difference is neither decisive nor crucial, nor is it always present. Upon closer inspection, it seems that NP-initial relative clauses do exist. Note that it is not uncommon for a nominal syntagma to consist of an article-like element with no overt head noun, followed by a relative clause. Blümel (2011), for example, discusses cases with the complex article-like element derjenige, which appears alone in 38.

1. (38)

   

Blümel (2011) discusses the status of derjenige as a determiner or a pronoun. This status is not important for the argument here. What is more crucial is that the same pattern can be observed with simple forms, typical for definite articles, as in 39.

1. (39)

   

In 39, the relative clause follows the relative pronoun. However, in this type of relative clause the reverse order is possible, as in 40.

1. (40)

   

Admittedly, it is not far-fetched to analyze these examples as instances of left dislocation. Fuß & Grewendorf (2014) have looked at these and related constructions, and their analysis is different from the one proposed here. ¹⁵ The latter authors are also interested in those cases where a bare (headless) relative clause appears, as in 41.

1. (41)

   

To account for 41, Fuß & Grewendorf (2014) argue for an input, or base structure, where the head determiner is present and is followed by the relative pronoun opening the relative clause. Then deletion under element repetition takes place. They further argue that this operation, which they refer to as syntactic haplology, reduces formally identical adjacent syllables—in this case, identical pronouns (homophonous; see Perlmutter 1971). Example 42a shows Fuß & Grewendorf’s analysis: The element that gets elided is the second occurrence, that is, the relative pronoun. ¹⁶ Under this approach, one has to assume that an example such as 40b contains a left-dislocated structure; the “big” relativized constituent is outside the core clause and itself starts with a resumptive pronoun in the prefield, as shown in 42b. The alternative analysis of 40b makes use of the unorthodox stance that a relative clause can be initial and hence predeterminer positioned, as shown in 42c. ¹⁷

1. (42)

   

Both analyses have pros and cons. A lack of clarity in the Fuß & Grewendorf approach emerges with respect to structures with case mismatch, such as 43. Example 43a is provided by Fuß & Grewendorf themselves.

1. (43)

   

With such data, the Fuß & Grewendorf analysis runs into problems. For their haplology deletion device to work, the authors are forced to assume base structures such as 44a,b. Note that it is the second occurrence of the pronominal in Fuß & Grewendorf’s approach that gets deleted under formal identity. For reasons of online processing, if spoken out loud, 44a,b sound awkward and fuzzy, but not necessarily unacceptable. However, after “thinking away” the relative clause attribute—which is a reliable test—it becomes clear that the haplology approach is not straightforward, see 44c,d.

1. (44)

   

In contrast, a “reversal-by-raising” approach, as shown in 45, would work easily and elegantly.

1. (45)

   

Additionally, it is even possible to have DP-internally preposed relative clauses where a nominal head is present, as in 46. Speyer (2011) quotes this example (his 24) to justify the common assumption that relative clauses cannot appear initially.

1. (46)

   

However, I believe that 46 is actually acceptable, albeit somewhat “sloppy”, or too colloquial. The conviction that it is ungrammatical may be due to a linguistic understanding that is very normative and oriented toward written language use. Under specific circumstances and in the right context, structures similar to 46 can be found and argued to be possible. Rarely in the literature have linguists argued that the structure [_DP RelCl [_D’ D° [_NP noun]]] is an option indeed (Müller 2003, Meinunger 2018). Thus, a comparable DP-internal realization has been proposed for some cases where an attributive PP precedes its head noun. In particular, Müller (2003), in order to argue for a regular verb-second structure, analyzes the cases in 47 as NP topicalization. Under his analysis, the prefield constituent is a single NP, and the PP is an NP-internal attribute that happens to precede its head.

1. (47)

   

Haider (1992) already proposed a DP-internal preposing for these cases: [[auf dem Tisch]_i die Bücher t_i]. Müller (2003) defends the same structure in which a relative clause surfaces. Referring to Abb (1994), Müller (2003:35) argues that “Bei (72) [48a] sieht man besonders deutlich, dass es sich nicht um eine Mehrfachbesetzung des Vorfelds handeln kann, da der Relativsatz ja allein nicht vorfeldfähig ist” [one sees particularly clearly that [48a] cannot be a case of multiple occupation of the prefield because a mere relative clause cannot fill the prefield on its own]. ¹⁸ In Meinunger 2018:13, I adduce the authentic find shown in 48b.

1. (48)

   

These examples are taken here as evidence that DP-internal relative clause inversion is possible. However, things are not so simple. It has been shown that this type of inversion is quite restricted. There is research showing that DP-internal PP fronting is not possible when the DP itself occurs inside a PP (Gallmann & Lindauer 1994 and Popp & Tebay 2019). This explains the ungrammaticality of 49, in which the DP die Bücher auf dem Tisch is an object of the preposition für ‘for’.

1. (49)

   

Yet in spite of any concerns about the validity of the data and difficulties with the analysis, some interesting conclusions can be made on the inverse relative clause structure. The first observation is that the inversion option seems to only be possible if the relative clause is not straightforwardly intersective in the sense of canonically restrictive. The classic idea of Partee (1975) that a restrictive relative clause must be in the c-command domain of the determiner holds for surface structures of this kind, which means that in terms of sequence, the relative must follow its restrictor. Hence it does not come as a surprise that truly quantified expressions, which are inherently restrictive and thus not possible as appositives, are not allowed in this pattern: ¹⁹

1. (50)

   

Another observation concerning this structure is that the inversion is only licit in the sentence-initial position.

1. (51)

   

However, this very pattern can be analyzed in such a way as to again unify relative clause structures and the je-desto-construction. For some linguistic structures it can be stated that the observable shape of a constituent is possible only in a derived position, that is, the relevant format is not licit in the original or base position. A very similar case is represented by a construction originally discussed by Bayer (2001) and known as emphatic topicalization in Bavarian:

1. (52)

   

In these cases, an object clause has been moved to the prefield. The interesting feature is that within the preposed subordinate clause, the topicalized constituents—an Mantl and da Hans—have been fronted to the embedded prefield, so to speak. Bavarian is known to tolerate a constituent in Spec, CP while also spelling out the subordinating element, thus there is no doubly filled COMP filter. Crucially now, this type of doubly filled COMP is licit only if the object clause has been fronted. If it appears (or remains) in a right-peripheral (that is, presumably in the base or canonical) position, this order is not possible, as shown in 53.

1. (53)

   

One must conclude that the object clause internal topicalization triggers an obligatory preposing of the complete object clause. In parallel fashion, those complex NPs where the relative clause has been inverted are likewise good only if they appear initially. NP-internal relative clause preposing thus triggers the obligatory leftward movement of the complex NP to the initial position, as in 48; leaving it behind causes ungrammat-icality, as in 51.

6.2. Explaining the Adjacency (Differently)

As laid down by Reis (2009), who bases her analysis on Bech 1964, many combinations and versions of je-desto/umso(-je)-realizations are attested (see the Appendix). The standard and normatively correct and recom-mended use is the following: The je-clause precedes the other part—whether it is introduced by umso or desto. Roehrs et al. (2002) provide the example in 54a. If the order is reversed (which for linguists like den Dikken, Reis, and Beck means that a different construction must be assumed), only the umso-variant is unobjectionably fine; the desto-variant is not, as shown in 54b,c. The same applies to embedded cases, such as 54d.

1. (54)

   

The interpretation of this interesting pattern in the present paper is the following: The desto version is not licit in the middle-field internal base position. It is only acceptable with i) the je-clause to its left and ii) the whole unit in the prefield. This explains the observed adjacency requirement. In 54c,d, the head of the AP remains middle-field internal, and the je-clause may extrapose—hence no obligatory adjacency for the je-clause with its host. However, if the je-clause has moved AP-internally to the AP-initial position and the correlative item is desto—as in the cases with NP-initial relative clauses or Bavarian emphatic topicalization—the inversion structure must move to the matrix prefield as a compact, coherent chunk. This obligatory move can be attributed to the wh-character of the AP. In Meinunger 2011, it has been argued that je- can be considered a bound morpheme, much like the w- or wh-part in question words or relative pronouns. If this element is part of an AP-affiliated A-bar Spec position, the whole expression—containing the item desto—is an operator that must be moved to its designated preverbal position. This explains i) the obligatory order (je before desto) and ii) the mandatory adjacency.

Given all the considerations above, je-desto-clauses behave as expected. The structure in 55 reflects the fact that the je-clause originates as a relative-clause-like attribute inside the AP. In a first step, it raises to a left-peripheral scope position, turning the whole constituent into a newly created operator, which in turn must be moved to the highest scope position, that is, the matrix prefield. All these dependencies and interconnections explain the adjacency claimed in Roehrs et al. 2002 in a different way.

1. (55)

   

Note that this analysis is valid for the standard register. Here it is offered as the genuine structural description of it, assuming that a large majority of native speakers agree that the order in 55 is grammatical in the standard language. However, as mentioned above and in the Appendix, the linguistic reality is much more diverse. Contrary to the quite rigorous evaluation of Roehrs et al. 2002 (in accordance with the standard and normative common practice also adopted here), the largest German grammar (IDS-Grammatik, Zifonun et al. 1997) cites the following examples:

1. (56)

   

In the rest of the paper, I would like to discuss some further examples and argue that the analysis remains valid, although there are structures attested that seem to resist it.

6.3. Further Cases of Nonadjacency

Occasionally, one encounters a different type of example in which the adjacency is not present (see the examples in 57). Here, the je-clause appears separately from the umso- or desto-constituent. The je-clause seems to have been shifted (scrambled, topicalized) to the left and is separated from its host by intervening material. In the remainder of this paper, I argue that these and other related examples fall most likely under the phenomenon grammatical illusion (prefixed with ⸕ to suggest a magic wand).

1. (57)

   

   ²⁰

It becomes apparent that if confronted with the discontinuous version, speakers regularly opt for the adjacent version as the better one (although the distance realizations are not clearly unacceptable; see also den Dikken 2005:505, 506.) Thus, native speakers always prefer 58a to 58b, and 58c to 58d.

1. (58)

   

So, what speaks for the ungrammaticality of examples 58b,d despite their apparent acceptability? It can be observed that people often—almost systematically—produce or readily accept structures such as 59.

1. (59)

   

However, in this type of example, the subject is clearly missing in the relative clause, which is the matrix to the complex je-umso-configuration. This is not untypical for center-embedding structures. The same happens with intraposed adjunct clauses, as in 60.

1. (60)

   

Such sentences are very inconspicuous but nevertheless ungrammatical.

Consider now wenn-dann- and je-desto-constructions, which exhibit a similar pattern when it comes to embedding and main clause verb order (for example, Meinunger 2014 arguing against Freywald 2008).

1. (61)

   

In both examples, dass is followed by a clause that should be verb final but is not. It comes along as a sort of verb third: In 61a, the order is [dass +] je-clause + desto-AP + finite verb …, and in 61b, it is [dass+] wenn-clause + dann + finite verb… Such a realization is ungrammatical, but the accumulation of true, potential, or illicit clause beginnings blurs the regular processing. Only a more careful analysis reveals that these structures are ill-formed. What is more—and this is crucial and clear in terms of native speakers’ judgments—the je-clause cannot be fronted alone to the prefield leaving its associate umso (desto) in the middle field; 29b is repeated here as 62.

1. (62)

   

Only the compact fronting of the complex constituent [je weniger Gepäck wir haben, umso leichter] can be fronted. The wenn-clause and the proform or correlate dann must appear adjacent, if they appear early, in an initial position, as in 63a. Only if the wenn-clause appears in the extraposed postfield position is discontinuity possible, as in 63d.

1. (63)

   

Example 63e is as bad as 62. It shows that in the case of leftward movement, dann and the associate wenn-clause cannot be separated. However, as with the center-embedding examples in 60, intraposition delivers quite an acceptable result, as shown in 64.

1. (64)

   

The reason for this behavior seems to lie in the parser-unfriendly structure of center embedding. A long tradition in psycholinguistics has shown that center embedding—especially, double center embedding—renders un-grammatical structures acceptable (see Huang & Phillips 2021 for very recent work and the references cited therein). Thus, after closer scrutiny, it can be stated that je-clauses cannot appear in isolation (that is, remote from the umso-part) apart from being extraposed into the postfield, just like relative clauses.

6.4. Cases of Apparent Deletion: “Short” Je-Forms

In this section, I present another argument against the left dislocation analysis of je-clauses. It has been observed in connection with the comparative correlative construction that in some cases, the involved parts can be reduced or realized with less linguistic material than in the form of fully fledged finite clauses (Culicover & Jackendoff 1999:554 for English and Fortmann 2016 for German; see section 4 above). The most salient approach would probably be to assume ellipsis, perhaps as some form of copula deletion or omission (Borsley 2004:89, den Dikken 2005:497), as shown in 65. Example 17 is repeated below as 65b.

1. (65)

   

Fortmann (2016) suggests an alternative analysis based on various considerations. The main reason is that the usual circumstances for verb deletion—such as gapping, for example—are not present. Furthermore, there seems to exist the option to realize the je-constituent as part of a prenominal adjectival attribute:

1. (66)

   

Hence, the conclusion can be drawn that the je-phrase does not need to be clausal in nature but can be of a smaller size. Fortmann (2016:129) suggests a degree or adjectival phrase: [je fetter der Braten]_DegP/AP. The following new observations support this analysis. ²¹ First, copula-free options for the je-constituent are restricted. A copula-free je-clause is natural and unproblematic in the canonical initial, that is, left-peripheral position—see 65b or the authentic example found by a reviewer in 67.

1. (67)

   

It turns out that short(ened) copula-free versions of je-clauses show a clear preference for the initial position, as in 68a. Example 68b, in which the je-clause appears in the middle-field internal position, is marked and must be somewhat artificial, whereas the extraposed variant in 68c is impossible: It appears to be incomplete and hence ungrammatical.

1. (68)

   

The explanation can be as simple as this: Clausal constituents can extrapose and do so preferably; smaller ones (nominal and adjectival) can never extrapose and hence must not undergo extraposition. As soon as the je-constituent is obviously clausal, exhibiting a (finite) verb, such a minimally different version is completely fine:

1. (69)

   

The grammatically well-formed prefield and middle-field versions are unproblematic. Both je-constituents appear inside their host as the fronted but not removed attributive subpart projection. None of the classic left dislocation approaches have a straightforward explanation for this pattern. Under these approaches, one might expect a difference between sentences with initial versus noninitial je-constituents, but not a similar behavior of initial and medial je-constituents in contrast to right-peripheral realizations.

7. A Structural Proposal

In this section, a formal structural proposal is suggested. The main claim of the present paper is that je-desto/umso-constructions are regular verb-second structures. The apparent verb-third shape is an illusion. It emerges because a complex constituent, which necessarily consists of two subcon-stituents, is moved to the sentence-initial position. This complex unit is the extended je-desto/umso-AP, which is base generated inside the lower part of the clause, that is, inside the domain of the predicate. For this adjectival constituent several structural analyses are possible. The one given in 70 is along the lines of the analysis offered by proponents of restricted structure building (à la Kayne’s 1994 anti-symmetry and Linear Correspondence Axiom, or Haider’s 1992, 1997, 2010 rigorous right branching). In the tree in 70, the uppermost head is umso or desto—here assigned the label d. This head takes as its complement a DegP, which is supposed to capture the fact that the complement AP must surface in its comparative form. The head of DegP is the relevant comparative morpheme: For example, if, as in most cases, the comparative is synthetic, the adjective moves to the head position and fuses with the -er morpheme. What is more, the comparative head supplies a specifier position that hosts the je-clause (or a smaller je-constituent). Under the specifier-head agreement, the je-constituent aligns with the comparative head in that it, too, must contain comparative morphology.

1. (70)

   

Another option is given in 71, where CorrP stands for Correlative Phrase.

1. (71)

   

This at first glance perhaps less attractive option has a few advantages compared to 70. The first one is that it can capture very well the similarities between je-clauses and relative clauses worked out in the previous sections. One proposal that has had occasional impact comes from Sternefeld 2006:376. Sternefeld agrees with the new and very influential approaches that assume that the relative clause is an argument of the determiner. He proposes a structure where the determiner is transitive and may (first) combine with the NP and (then) the CP, as shown in 72 (Sternefeld’s example 42c).

1. (72)

   

There are many advantages to this analysis, despite one (apparent) disadvantage, that is, the unpopular right-hand specifier (of the dP). The structure in 71 captures the parallelism between je-clauses and relative clauses perfectly. The desto-element, which, formally, also very much reminds one of a determiner-like element (see den Dikken 2005:501), takes an AP as the first argument and the je-clause as the second one. Additionally, just like the head Deg in 70 (the -er morpheme), it may impose the requirement of comparative morphology on its specifier (or on its two arguments).

Another reason that the structure in 71 is possibly superior to the one in 70 is the position of the je-clause: The structure in 71 has the right position for it, under the syntax-and-semantics approach by Sternefeld. Except for the “comparative agreement”, there is not much evidence for its position in 70, that is, as the specifier of DegP. The je-clause can never surface in this hypothetical position overtly. It is impossible for it to appear between desto and its associated adjectival expression. This means that the relevant surface orders would have to be achieved by a cascade of movement operations, which are not motivated straightforwardly. In contrast, the rare intraposed order, where the je-clause might be assumed to occur in its base position, follows directly from the structure in 71:

1. (73)

   

Be that as it may, whichever analysis is adopted as the final internal microstructure, the proposal of the present paper is that the macrostructure of the je-desto/umso-correlative is the verb-second analysis in 37b, repeated below as 74.

1. (74)

   

The adjectival phrase, which embeds the internally proposed je-clause, occupies the prefield as a single but internally complex constituent.

8. Conclusion

It has been shown that German je-desto/umso-sentences are regular verb-second structures. The hitherto influential analyses assume a left dislocation structure, which can be considered a verb-third structure. It has been shown that such an approach cannot capture many features of je-desto-(/umso)-structures, such as i) middle-field internal occurrence, ii) clause-initial obligatory adjacency, iii) short (verb-less) variants in nonextraposed positions and the impossibility thereof in the postfield position, and iv) the semantically equivalent contribution of initial, medial, and postposed je-clauses to the compositional meaning of the whole complex structure. In the present article, the proposal has been put forward and developed, according to which je-clauses are similar to relative clauses: In their base structure, both types of clauses are attributes that belong to a clausal constituent in which they are embedded.

The specific structure of je-initial complex sentences invites a new perspective on inverted NPs, that is, on nominal groups where an attribute precedes the determiner (and the head noun). This leads to some speculative, but promising proposals such as the existence of a class of prenominal relative clauses in German.

Furthermore, the paper discusses middle-field realizations of je-clauses where the host, that is, the matrix clause, seems to license an empty subject. It has been argued that this type of structure should be considered a case of grammatical illusion, much like cases of embedded verb fronting je-structures. The paper concludes with the proposal of two (noncom-mittal) potential formal options for how the given analysis of je-clauses as attributes can be worked out in a syntactically more elaborated way using very specific types of NP- or AP-internal functional projections.