2.1 Comparison particles as prepositions or complementizers?

As far as the internal syntax of comparison constructions is concerned, one central issue is the syntactic position of the comparison particles such as English as/than or German als/wie. In the literature, it is widely assumed that they constitute either prepositions or complementizers. Some authors such as Napoli (1983), Hoeksema (1983), Hendriks (1995), van Gelderen (2004: 124f.), SAND (2005: 13), Quirk et al. (2008: 1132), and Hubers/de Hoop (2013: 93) propose a prepositional analysis only for comparisons in which the standard of comparison on the surface consists of a mere DP.⁴ Others hold that comparison particles are generally prepositions, cf. Chomsky (1977), Kennedy (1999) for English than, Zimmermann (1987) for German als.⁵ The question of the syntactic position of the comparison particle is thus also related to the question of the syntactic status of phrasal vs. clausal comparisons that will be discussed in Section 3.

Besides the fact that comparison particles may, like prepositions, combine with mere DPs, the main argument for a supposedly prepositional status of comparison particles is that, in these phrasal comparisons, a pronominal DP often appears in a morphological shape corresponding to an oblique case form rather than to the form it would take in a full comparison clause. This is illustrated for English, Dutch and French in (2)–(4), cf. Hoeksema (1983), Hendriks (1995), Hubers & de Hoop (2013: 93), SAND (2005: 13) on Dutch, Quirk et al. (2008: 1132) on English, see also Haspelmath & Buchholz (1998: 308f.) on Swedish and Bulgarian.

However, I would like to argue that the special morphological form of the pronoun is not due to case government by the comparison particle. It rather constitutes the strong form of the respective pronoun in the sense of Cardinaletti & Starke (1999), which, in the same languages, is optionally or obligatorily used in typical contexts such as coordination, cf. (5) (other such contexts include cleft-constructions, contrastive focus etc.). While data such as (2)–(4) do therefore not constitute arguments for a prepositional analysis of comparison particles, they are significant with respect to the syntax of phrasal vs. clausal comparisons as well as the link to the rest of the clause, cf. Sections 3 and 4.

In German, there is no case government by the comparison particle either (whether past or present). Als and wie do not govern case but are followed by DPs in all cases available in the language, including nominative case, cf. (6), which again strongly speaks against a prepositional analysis because prepositions do not combine with nominative DPs. Duden (2016: 636) also rejects the prepositional analysis because of the obvious lack of case government by the comparison particles.

Besides these morphosyntactic observations, further syntactic evidence against the prepositional analysis comes from the fact that not only a DP but also for instance a mere AP may follow the comparison particle cf. (7a), which is ungrammatical with prepositions, cf. (7b).⁶ Furthermore, the comparison particle may be followed by a PP or even a PP embedded within a PP, cf. (8a). While stacking of two PPs is attested (but, as one anonymous reviewer points out, restricted to von ‘from/about’ as in (8a), and bis ‘until’), stacking of three PPs does not seem to be possible in German, cf. (8b). These data therefore corroborate the fact that comparison particles are not prepositions.

The second common syntactic analysis of comparison particles treats them as complementizers, cf. for English the classic analysis by Bresnan (1973) and much subsequent research. For German, this has been proposed by Bierwisch (1987), Bergerová (1997), Eggs (2006) and Helbig & Buscha (2013).⁷ A central argument for the complementizer analysis, besides the fact that comparison particles may also introduce full clauses, comes from word order: In German clausal comparisons, the finite verb obligatorily occurs in final position just as in regular subordinate clauses introduced by a complementizer (or wh-item). The classic explanation for the word order asymmetry of verb-final (Ve) order in complementizer clauses versus verb-second (V2) order in main clauses goes back to den Besten (1989): The complementizer occupies the landing site of the finite verb (C⁰) and thus prevents verb movement to the left periphery of the clause. Accordingly, the comparison particle could be taken to occupy C⁰ so that the structure would look as in (9) illustrated for sentences (1a–c), cf. Bresnan (1973) for English and Bierwisch (1987) for German. However, data from the present-day standard language as well as from dialectal and diachronic variation constitute important arguments against this analysis and in favour of an analysis according to which the comparison particle resides in a functional head above the standard-of-comparison CP, as will be demonstrated in Sections 2.2 and 2.3.

2.2 The syntactic position of als and its diachrony

Regarding the German comparison particle als, one crucial set of data against the analysis as a complementizer comes from hypothetical comparisons, cf. (1d) above, repeated and analysed in (10). Within the left clausal periphery in these constructions, als does not only co-occur with the complementizers ob (‘if’) or wenn (‘if’), which have occasionally been considered to form one syntactic unit with als in these cases (Pasch et al. 2003; Eggs 2006), but also with the finite verb in its left-peripheral position (cf. als liefe … in (1d)). The position C⁰ is thus clearly available to be filled by other elements (complementizer or finite verb).⁸ Als must therefore be in a syntactic position above C⁰.⁹

As als is not phrasal and does not pass typical constituency tests (fronting, substitution, coordination, question test etc.) it cannot occupy SpecCP. It must accordingly be placed in a syntactic position above the standard-of-comparison CP as also argued in Jäger (2010).¹⁰ Because this position cannot obviously be equated for instance with one of the split-CP sublayers in terms of Rizzi’s (1997) account, it will simply be labelled Conj⁰ for conjunction, the relevant part of speech, signaling that als is neither a preposition nor a complementizer. The suggested structure of comparisons with als in present-day German thus looks as in (10), illustrating hypothetical comparisons with ob, wenn and verb-initial (V1) order, cf. (1d), as well as comparatives, cf. (1a), in which the C⁰ position is phonetically empty, as will be motivated below.¹¹ (Note that this analysis is in itself neutral as to the external syntax of comparisons, i.e. the way the standard of comparison is linked to the rest of the clause. It only concerns the internal syntax of the standard of comparison. However, it obviously sits very naturally with a coordination-like link to the rest of the clause discussed in Section 4.)

While the C⁰ position remains phonetically empty in comparatives with als in Modern Standard German as illustrated in the last line in (10) for the sentence in (1a), this position may be filled even in this type of comparison by a complementizer such as dass, was etc. (Doubly Filled Comp/DFC) in some dialects such as certain Swiss German varieties, see (11). These dialectal data thus constitute further evidence for a syntactic position of the comparison particle als above C⁰, as illustrated in (12).¹²

The assumption that als is placed in a functional head above the standard-of-comparison CP is further corroborated by diachronic data. The contrast with historical stages of the language in which als did in fact occur as a complementizer underlines the fact that it stands in a hierarchically higher position (Conj⁰) today.¹³

Based on non-degree equatives in which the OHG equative particle so ‘as’ occurred immediately following an adverbial al ‘fully’ in the matrix clause, also (>alse/als) was beginning to be grammaticalized into a new equative particle in late OHG.¹⁴ As indicated in Section 1, it already occurred as the main particle used in non-degree equatives and occasionally in degree equatives in MHG. In late OHG, MHG and at the beginning of the ENHG period, the new equative particle also took up the C⁰ position, unlike in Modern German. One piece of evidence for this comes from phonology: With also, so-called synaloepha occurs, that is, in order to avoid a hiatus, a following word starting in a vowel clitizices to also under vowel elision. This is illustrated in the MHG example in (13) for also + pronoun er ‘he’ forming alsor.

The phenomenon of synaloepha is known to occur with elements in C⁰, viz. the finite verb in left-peripheral position or complementizers such as wanta ‘because’ or oba ‘if’. Notably, it does not occur with coordinating conjunctions such as inti ‘and’ or odo ‘or’, cf. de Boor (1928), see also Weiß (2018). Thus, also must be positioned in C⁰, not yet in Conj⁰.

Another set of data that constitutes evidence for als(o) filling the complementizer position in historical German is hypothetical comparisons, cf. Jäger (2010). The syntactic interaction with verbal placement in these contexts shows that als(o) is placed in C⁰ originally, while the change in verbal placement during ENHG forms a crucial diagnostic for a change in the position of als. Until the 15th century, hypothetical comparisons introduced by simple als(o) typically showed Ve word order as illustrated in (14a) for MHG and in (15a) for ENHG, cf. Jäger (2018: 144; 215f.). By contrast, from the late 15th and especially 16th century on, hypothetical comparisons introduced by mere als generally show verb-initial order as is still the case today while verb-final placement is ungrammatical, cf. the Modern German translations in (14b/c) and (15b/c).¹⁵

The tree marker in (16) illustrates the syntactic structure of the ENHG example (15a) in contrast with the corresponding Modern German sentence (15c) in the line below. In the historical example, als fills C⁰ and thus blocks verb movement to this position. By contrast, als occurs in the higher functional head Conj⁰ in Modern German so that the finite verb can move to C⁰. The observed change in verbal placement indicates that the change of als from C⁰ to Conj⁰ took place approximately during the 16th century. A number of further constructions that started to occur around the same time further support this: During the second half of the 16th century, als is attested in combination with complementizers such as dass and, in hypothetical comparisons, wenn and ob, cf. (17)–(19), indicating that als is positioned higher than C⁰.¹⁶ Furthermore, at the same time, als is first attested in relative clauses before the relative pronoun and possibly a pied-piped preposition, cf. (20), showing that als is also hierarchically higher than SpecCP.

The diachronic movement of als from C⁰ to Conj⁰ represents a common type of syntactic change, viz. upwards reanalysis typically involved in grammaticalization (cf. Roberts & Roussou 2003). In terms of van Gelderen (2004: 28) it can be explained on the basis of the syntactic economy principle Late Merge, which states that it is economical to merge an item maximally late in the derivation, and therefore leads to diachronic upwards movement. The history of comparisons in German thus constitutes another area in which this principle, which has been formulated independently for a range of other phenomena, can be shown to apply. Besides, as indicated above, the historical data and the syntactic changes observed represent additional valuable evidence in favour of an analysis of the Modern German comparison particle als as a functional head (Conj⁰) above the actual standard-of-comparison CP.

2.3 The syntactic position of wie and its diachrony

For the Modern German comparison particle wie, I would like to argue that – like als in (10) – it is placed in the head position Conj⁰ above the actual standard of comparison. As such, it is syntactically distinct both from the homophonous modal/degree wh-item wie ‘how’, which occupies SpecCP, as well as from the homophonous complementizer wie ‘that/when’ in C⁰, supporting the view that the comparison particle wie occupies neither position but is placed higher in the structure.

At first sight, it might seem straightforward to analyse the comparison particle wie in the same way as the modal/degree wh-adverb wie ‘how’, illustrated in (21), i.e. as a phrasal element (AdvP) occupying SpecCP. This approach is taken by Duden (2016: 1052: in sentential comparisons, wie as relative adverb, but partly also as relative particle) and Bücking (2015; 2017: wie in SpecCP in hypothetical comparisons, but as PP rather than an adverb).¹⁷ As we will see below, the wh-adverb wie in SpecCP in modal free relatives indeed formed the original historical source for the grammaticalization of this comparison particle. However, in the present-day language, the syntactic behaviour of wie ‘as’ in degree and in non-degree equatives, cf. (24), is distinctly different from that of the homophonous modal/degree wh-adverb wie ‘how’ as well as generally of wh-items in free or headed relatives, cf. (22a/b) and (23a/b).¹⁸ This is evident from – depending on the analysis (see Section 3) – the possibility of ellipsis or (according to the direct analysis argued for in Section 3) of connecting mere DPs, PPs etc., cf. (24). This is entirely ungrammatical with the wh-adverb wie, cf. (21), as well as with other wh-items whether argumental as in (22) or non-argumental as in (23). (For differences with respect to the possibility of Gapping see (45a vs. b) below.) The comparison particle wie is thus syntactically distinct from the wh-adverb wie and should not be analysed as a SpecCP element. Further support for this view comes from the fact that, in many languages, the equative comparison particle is also phonologically clearly distinct from the modal/degree wh-item, cf. English as/like vs. how or French comme vs. comment (see also Eggs 2006).

Besides being used as a wh-item meaning ‘how’ as in (21), wie also occurs as a complementizer in Modern German, viz. instead of dass ‘that’ especially with verbs of perception, cf. (25), and instead of als ‘when’ as a temporal complementizer, cf. (26).¹⁹ Again, dialectal data provide additional evidence for this use: As illustrated in (27), Bavarian, wie (wia) may be used as a complementizer in relative clauses (i.e. relative particle) following the relative pronoun which is positioned in SpecCP, and thus clearly occurring in the C⁰ position itself (DFC). (For differences between comparison particle wie and complementizer wie with respect to the possibility of Gapping see (45a vs. c) below.)

The question is thus, whether the comparison particle wie could instead be assimilated to this use and be analysed as a C⁰ element (cf. Bierwisch 1987; Jäger 2010; Duden 2016: wie partly relative particle in sentential comparisons see fn. 7). However, the same kind of syntactic contrasts as illustrated above between wie as a comparison particle and wie as a SpecCP item obtain with wie as a complementizer: Again, the occurrence of a mere DP, which is possible with the comparison particle wie cf. (28), is ungrammatical with the complementizer wie (just as with dass), cf. (29), since it needs to occur in a full CP. Note that this holds even if the elided material is phonetically identical to material earlier in the clause. Accordingly, the comparison particle wie cannot be analysed as a complementizer occupying C⁰, either.

Depending on the analysis of hypothetical comparisons including wie, the position of wie in a syntactic head (Conj⁰) above C⁰ is further corroborated by these constructions. In hypothetical comparisons, wie may co-occur with wenn ‘if’, which is clearly a C⁰ element, cf. (30), see also (1d) above. One option would be to analyse this as an instance of elliptical CP stacking: an elliptical wie-clause embedding a wenn-clause (cf. Jäger 2010). However, Bücking (2015: 268) has argued that the supposed ellipsis cannot always be reconstructed under identity of the overall semantics, cf. also Pasch et al. (2003: 356ff.; 619).²⁰

If accordingly, wie wenn is not analysed as an instance of elliptical CP stacking, this is additional evidence that wie fills a position above C⁰.²¹ As mentioned above, I assume that, like als, the comparison particle wie is positioned in the functional head Conj⁰ above CP. The structure of comparisons with wie that is suggested here thus looks as given in (31). In hypothetical comparisons including wie wenn, the CP contains wenn in its C⁰ position, as illustrated in the second line. In ordinary equatives, the C⁰ position is empty (with a degree operator in SpecCP in degree equatives see fn. 11).

Note that the comparison particle als wie, which occurs both in historical and dialectal variants of German, does not constitute an argument against the suggested analysis. In contrast to what is assumed in the previous literature (DWB 1: 249; 29: 1475; Dückert 1961: 208f.; Ebert et al. 1993: 478; and on this basis also Jäger 2010), this particle does not constitute a combination of the two comparison particles als and wie that occurred diachronically as an intermediate step when als was replaced by wie, or in dialects in areas where an als-area and a wie-area partly overlap. Diachronically, it occurs first in equatives, but only after wie had firmly supplanted als as an equative particle, cf. Jäger (2018). The observation that it is first used in non-degree equatives indicates that als wie in fact constitutes a comparison particle that was, like Latin sicut, Gothic svasve, Dutch zoals etc. grammaticalized on the basis of the reanalysis of a matrix-final correlate ‘as’ – in historical German, specifically in MHG and ENHG (cf. Jäger 2018: 140f.; 196), mostly als(o) rather than simple so – and the adjacent equative particle (see also fn. 14 and (72)). Also in Modern German dialects, als wie is not areally restricted to overlaps between als and wie areas, but mostly occurs within larger wie areas, supporting this diachronic scenario. Synchronically, als wie is therefore (despite the common separate spelling) to be considered as one syntactic unit which, like als and wie, occurs in Conj⁰ in the syntactic structure.

With wie occupying Conj⁰ just as als, one might expect that it should be possible to combine wie with V1 clauses in hypothetical comparisons, too, as the C⁰ position would be available for the finite verb. Yet, this is ungrammatical in Modern Standard German. However, this ungrammaticality is probably due to independent reasons (e.g. garden-path effects due to ambiguity with embedded wie-wh-questions):²² In fact, there are Modern German dialects in which V1 in hypothetical comparisons introduced by wie is indeed attested, cf. (32) and (33),²³ showing that the C⁰ position is available for the finite verb as predicted by the analysis.²⁴

Furthermore, there is another set of dialectal data supporting the suggested analysis. Just as with als cf. (11), the comparison particle wie may co-occur with a complementizer such as dass ‘that’ (DFC) in certain, especially Upper German dialects, demonstrating that wie does not occupy C⁰ but a higher position.²⁵ This occurs for instance in degree as well as non-degree equatives, cf. the Bavarian examples in (34a–c), and also in comparatives, in which wie is also overhwelmingly used in most High German dialects (see Table 1), cf. the Swabian and Swiss German examples in (34d/e).²⁶ As illustrated in (35), both hypothetical comparisons with wie + V1 clause as well as comparisons with wie followed by a complementizer are adequately captured by assuming that the comparison particle wie is positioned in a functional head above CP. Dialectal data thus constitute valuable additional evidence for the syntactic analysis suggested here.

Further support for this analysis comes from historical data. As mentioned above, the use of wie as a comparison particle is based originally on its use as a wh-item (interrogative/relative ‘how’). Accordingly, its syntactic behaviour at earlier stages of the language contrasts with the behaviour of the comparison particle wie today, providing additional evidence for the above analysis. In OHG, the predecessor wio was used only as a wh-item. In classical OHG, it is indeed attested exclusively as a modal/degree interrogative adverb ‘how’ in SpecCP position. Over the course of the OHG period, it is beginning to occur also as a modal relative adverb in free relatives, notably in combination with correlative so … so, which is typical of free relatives in OHG (so wio so ‘how(ever)’, cf. also so wer so ‘who(ever)’, so waz so ‘what(ever)’ etc.). As illustrated in (36), the first so arguably originally constituted a correlate within the matrix CP (CP1) while wio filled the specifier position of the free relative clause CP (CP2),²⁸ and the second so appeared as a relative complementizer (relative particle) in the C⁰ position of this latter CP. However, already in the 9th century, so wio so occurs as a complex inside CP2, as evidenced by a caesura in poetic texts such as Otfrid’s gospel book, cf. (37), i.e. the first so has been reanalysed as part of the relative clause. In fact in MHG, so and the wh-item are fused into one word, viz. swie. Also single wio occurs in modal free relatives in OHG which resemble non-degree equatives, cf. (38), constituting another source for the later comparison particle wie. At any rate, its predecessors always occur in SpecCP and only in full clauses, cf. DWB (29: 1474–1480), Lerch (1942: 345), Dückert (1961: 205f.). First rare instances of non-degree equatives including (s)wie are attested in MHG, cf. (39).

A first syntactic change of wio/wie itself occurred when it was reanalysed from SpecCP to C⁰. As this reanalysis does not affect general word order such as verb placement, it is challenging to find conclusive diagnostica for when this change actually occurred. A number of arguments suggest, however, that this change can be dated to MHG: Firstly, (s)wie also starts to occur as a complementizer in other functions at that time, notably as a temporal or concessive complementizer, cf. (40), see also DWB (29: 1450–1452), Paul (2007: 417; 419). In these functions, it is clearly positioned in C⁰ (note also the occurrence of synaloephe in (40): swie + er > swier).

Secondly, as mentioned above, the second so of the original so wio so is dropped in MHG (while the first one fuses with the wh-item). This can be explained if swie indeed changed from SpecCP into C⁰ so that this position was no longer available for the relative complementizer so and accordingly, it disappeared, cf. (41).²⁹

The change from a specifier to the head of the same projection is again a common kind of syntactic change, notably since both generally agree with respect to crucial grammatical features (Spec-head agreement, cf. Chomsky 1995). It also occurred for instance in relative clauses with English that, French que and German dass as part of the relative cycle, cf. van Gelderen (2004: 81–88), Axel-Tober (2012).³⁰ Again, this change is caused by syntactic economy as formulated in van Gelderen’s (2004) Head Preference Principle (or Spec to head principle) which demands for a syntactic item to be a head rather than a phrase, if possible. A head comes with less syntactic structure than a full phrase. Accordingly, the learner will assume head status in ambiguous cases, which leads to the common diachronic reanalysis from Spec to head.

A second syntactic change of wie occurred during the ENHG period – the time when wie started to be increasingly used in equatives, notably in non-degree equatives, cf. Table 1. As illustrated in (42), wie changed from C⁰ of the second CP to Conj⁰, the position that it still occupies in the present-day language. This change from C⁰ to Conj⁰ can, as in the case of als, be explained on the basis of the syntactic economy principle of Late Merge (van Gelderen 2004) as an instance of upward reanalysis (Roberts & Roussou 2003).

As discussed above, before the ENHG period, i.e. during OHG and MHG, wio/wie had been restricted entirely to full clausal contexts. Only these contain a CP layer in which it could occur – first as a wh-item in SpecCP, then as a complementizer in C⁰. If one assumes that phrasal comparisons, on the other hand, may constitute single phrases (DPs, PPs etc.) connected to the rest of the clause by the comparison particle (so-called direct analysis, cf. Section 3), phrasal comparisons do not include a CP layer. Accordingly, wie could only appear in phrasal comparisons once the reanalysis from C⁰ to the higher functional head Conj⁰ had taken place, providing a diagnostics for when this change occurred. Phrasal comparisons including wie are first attested in the 15th/16th century as illustrated in (43). For some time after this, however, wie continues to be preferably used in clausal comparisons (see also Table 2, Section 3).

Three further sets of data also constitute evidence for the reanalysis of the comparison particle wie from C⁰ to Conj⁰ during the ENHG period. At this time, wie is first used in comparisons including Gapping (deletion of the finite verb with several constituents remaining, e.g. subject and object), cf. (44). This type of ellipsis is typical of coordination-like structures, but is ungrammatical in subordinate clauses introduced by a wh-item or complementizer, as discussed in more detail in Section 4. Until today, the possibility of Gapping after the comparison particle wie, cf. (45a), is another distinctive feature in contrast to the ungrammaticality of Gapping after the interrogative wh-phrase wie, cf. (45b), and the complementizer wie, cf. (45c). In the investigated diachronic corpus, Gapping is first attested with wie in the 16th century. Furthermore, during the same time, coordinating use of wie is first attested cf. (46), see also DWB (29: 1450–1452). This fact again constitutes evidence that wie is positioned in a higher – possibly coordination-like position (cf. Section 4). Around the same time, viz. since the 15th century, wie co-occurs with the complementizer dass ‘that’, as illustrated in (47), so that wie must clearly be in a higher syntactic position than C⁰. The diachronic data thus support the analysis that, since ENHG times, wie is placed in Conj⁰.

4.1 Relative-clause like characteristics of comparisons

Traditionally, comparisons are considered as subordinate clauses and are treated in the context of adverbial clauses, cf. Schrodt (2004: 167–172), Paul (2007: 425–428), or analysed as a kind of relative clauses, cf. already Erdmann (1874–1876, I: 111), more recently Reich & Reis (2013), Bücking (2015; 2017 for hypothetical comparisons), Duden (2016: 1052f. for clausal non-degree equatives and hypothetical comparisons).³⁹ Indeed, syntactically, comparison constructions resemble relative clause constructions. Thus, the standard of comparison in the equatives in (63) just like the relative clauses in (62) may occur with a head noun or (demonstrative) proform (correlate) as in the examples in a, free (headless) as in b, or continuatively as in c.⁴⁰

Generative syntactic theory has revealed further similarities between comparisons and relatives: As illustrated in (64) and (65), extraction from the standard of comparison produces wh-island effects (both in equatives and comparatives) as it does with relative clauses and other wh-clauses. While extraction from a that-clause embedded inside the standard of comparison is possible, cf. (64b)/(65b), extraction from a wh-clause embedded in the standard of comparison, cf. (64a)/(65a), results in ungrammaticality as it does with embedded wh-clauses as opposed to embedded that-clauses inside relative clauses. According to Chomsky (1977), this is because the wh-phrase – unlike the syntactic head that – blocks the intermediary landing position thus preventing successive-cyclical movement and causing a subjacency violation. He concludes that, as in relative clauses, wh-movement takes place within the standard of comparison and comparison clauses should accordingly be analysed as free relatives.⁴¹

These syntactic similarities correspond to semantic parallels between comparison and relative constructions. According to the standard semantics of comparisons (von Stechow 1984; Heim 1985; 2000), just as in relative clauses, in comparisons semantic abstraction takes place due to the movement of a (possibly empty) operator leaving behind a bound variable (an individual variable in relatives, a degree or property variable in comparisons).⁴²

Diachronic data also highlight similarities of comparison and relative constructions. In historical German, asyndetic comparison clauses, i.e. standards of comparison not introduced by a comparative particle as in (66)⁴³ are attested. These are reminiscent of asyndetic relative clauses that are also found in historical German, cf. (67), see also Behaghel (1923–32, III: 742–745), Schrodt (2004: 174f.), Axel-Tober (2012: 230–234). Both kinds of constructions are ungrammatical in Modern German.

The link between relatives and comparisons is furthermore evident in a diachronic perspective from the fact that equative comparison particles are commonly grammaticalised into relative particles (relative complementizers), cf. Erdmann (1874–1876, I: 56f.), DWB (16: 1341–1385), Paul (2007: 426f.), Ferraresi & Weiß (2011).⁴⁴ The equative particle so occasionally occurs in (modal/temporal) relative clauses already in the course of OHG, develops into a general relative particle in the 12th century and is attested as such well into the early NHG period, cf. (68), see also Paul (1920: 238), Behaghel (1923–32, III: 285–288; 729f.), DWB (16: 1341–1385), Schrodt (2004: 168f.). Since MHG, also is predominantly used as a comparison particle in equatives (for an overview of the development of comparison particles in German see Table 1). It is secondarily also attested in relative clauses into the ENHG period, cf. Paul (2007: 426), MWB (1: 168–173), Ebert et al. (1993: 447) – partly preceding a relative pronoun as in (20) above, see also Ebert et al. (1993: 446), partly on its own as in (69). In MHG, und(e) is also very occasionally used as an equative particle, cf. (87) below. It has developed into a relative particle on this basis, too, as illustrated in (70), see also Paul (2007: 427), Ferraresi & Weiß (2011). Wie, which is becoming widespread in equatives in ENHG, also developed into a relative particle during this period as illustrated in (71), see also Ebert et al. (1993: 448; 479). It is still evidenced as such in Modern German dialects such as Bavarian, cf. (27) above.

In many European languages⁴⁵ and in historical German, too, we can also witness a grammaticalisation path from wh-items (relative adverbs) used in free modal relatives into equative particles, again due to the closeness of comparisons and relatives. As discussed in Section 2.3 above, OHG (so) wio (so)/MHG (s)wie ‘how(ever)’ turned into an equative particle on the basis of its use as a wh-item in free modal relatives (and on this basis was later reanalysed into a relative particle, see above).

Another parallel between relative and comparison constructions is finally evident from the development of new equative particles on the basis of originally matrix-internal elements and the original equative particle. Due to the lack of a possibly intervening Tertium Comparationis in non-degree equatives, recurring matrix-internal elements such as the equative correlate or a matrix-internal noun referring to manner, extent etc. regularly occur in a position adjacent to the equative particle. They are then reanalysed as part of the introduction of the standard of comparison, forming a new equative particle by univerbation with or substitution of the former equative particle. This process is repeatedly attested in historical German, cf. (72a) for the OHG equative particle soso, early NHG als wie and inmassen ‘like/as’, as well as in many other languages (cf. Latin tamquam, sicut, Gothic svasve, Dutch zoals etc.). Similarly, on the basis of degree equatives, the equative correlate and Tertium-Comparations AP may be reanalysed into a complementizer, cf. (72b).

These developments are reminiscent of the common grammaticalisation of new complementizers on the basis of a relative clause and its immediately preceding matrix-internal head noun. This can for instance be observed in the development of the noun wîle ‘time (span)’ with adjacent relative clause into the complementizer weil introducing a temporal, later causal adverbial clause, as illustrated in (73), see also Paul (2007: 406; 417; 430), Szczepaniak (2011: 175–178), Weiß (2012: 147). The combination of an equative comparison clause and its equative correlate (and possibly Tertium Comparationis) thus behave in the same way as the combination of a relative clause and its head noun. This syntactic change, too, can be argued to be driven by economy: Due to the univerbation with or substitution of the element originally introducing the subordinate clause, one item less needs to be merged in the derivation. As the original relative clause is simply merged in the place of its former head element, the syntactic structure is reduced by one layer in line with the Subset Principle (Berwick 1985) or Least Effort Principle (Roberts 1993).⁴⁶

Accordingly, the diachronic data suggest that the standard of comparison is linked to the equative correlate and, if present (i.e. in degree equatives), the Tertium-Comparationis AP/AdvP in a relative-clause like fashion.⁴⁷

This syntactic analysis is also supported by topicalisation data as illustrated for Modern German in (74)–(76). In degree equatives, the standard of comparison forms one constituent with the equative correlate so and the Tertium-Comparationis AP, as it may be topicalised together with them, i.e. moved to SpecCP, which can be filled by only one constituent in German, cf. (75). By the same argument, the standard of comparison can be demonstrated to form a constituent with the (optional) correlate in non-degree equatives, cf. (76), and with the Tertium-Comparationis AP in comparatives, cf. (74). Notably, this holds in exactly the same way for the head noun of a relative clause or a relative clause correlate (demonstrative proform) together with its relative clause, cf. (77).

Relative-clause like analyses in which the standard of comparison forms one constituent with the Tertium-Comparationis AP and/or the equative correlate have been suggested various times in the literature, as illustrated for German in (78)–(80). It is generally assumed that the equative correlate (in degree equatives) and the comparative morpheme –er, respectively, form the head of a degree phrase. The analyses differ in detail as to the relative position of this DegP, the Tertium Comparationis AP and the standard of comparison. Under one approach, the degree phrase acts as a modifier within the AP as in (78),⁴⁸ cf. Bresnan (1973), Heim (2000) and Bhatt & Pancheva (2004) for English and Zimmermann (1987) and Bierwisch (1987) for German. Alternatively, the DegP dominates the Tertium-Comparationis AP, taking it as an argument, too – either as its complement with the standard of comparison occurring in the specifier of DegP as in (79),⁴⁹ cf. Abney (1987) and Kennedy (1999), or as its specifier while the standard of comparison forms the complement of Deg⁰ as in (80),⁵⁰ cf. Lechner (2004). (In the tree markers in (78)–(80), I have kept the label ConjP introduced in Section 2 above for the standard of comparison while, depending on their analysis of the syntactic category of comparison particles, the respective authors refer to it as CP/S, PP, or – somewhat agnostically – as than-phrase.)

In all of these structures, some additional movement operation is necessary to derive certain types of comparison constructions: In (78) rightward-movement of the complete DegP in comparatives and of the standard of comparison (here called ConjP) in equatives has to take place. In (79) rightward-movement of Deg⁰ in (synthetic) comparatives has to be involved, whereas equatives and English analytic comparatives with Deg⁰ more are straightforwardly derived. In (80) leftward-movement of the Deg⁰ correlate in equatives is necessary, whereas comparatives are straightforwardly derived (note that the comparative morpheme on the adjective is in Spec-Head-Agreement with a comparative feature in Deg⁰, here).⁵¹ They all share the assumption, however, that the standard of comparison is an argument of a Deg⁰ that forms one constituent together with the Tertium-Comparationis AP. Depending on its syntactic function (predicative: … ist so schnell wie … ‘is as fast as’/adverbial: … läuft so schnell wie … ‘walks as fast as’/attributive: eine so schnelle Läuferin wie … ‘as fast a runner as’), this entire constituent (AP or DegP) will in turn be linked to the rest of the clause just as other predicative phrases (e.g. within a Small Clause or AgrP), adverbials (e.g. adjoined to VP/V’),⁵² or attributive phrases/nominal modifiers (e.g. adjoined to N or NP), respectively.⁵³

Applying these syntactic approaches to non-degree equatives, which have not been considered in the literature so far, the minimal structure to assume would be the one in (81). The standard of comparison is again an argument of the equative correlate so. Since we are dealing with non-degree equatives, the latter is not taken to constitute a Deg⁰ but is simply labelled Adv⁰, here.

Since the equative correlate is optional in non-degree equatives (similarly to the optional correlate of a free relative clause e.g. … (das), was …), one could assume a phonetically empty correlate (pro) in these cases, cf. Axel-Tober (2012) for historical predecessor constructions of dass-clauses. Alternatively, the standard of comparison may be linked to the rest of the clause not only via the AdvP shell assumed in (81) but also directly, like a free relative clause. It is conceivable that linking via pro vs. direct merger of ConjP constitute two consecutive diachronic stages as in the case of dass clauses. Depending on its syntactic function, the AdvP (or possibly mere ConjP) will be linked to the rest of the clause again as an adverbial (… läuft (so) wie … ‘walks like’),⁵⁴ predicative phrase (Das ist (so) wie … ‘That’s like …’) or attributive phrase, i.e. nominal modifier ((so) ein Läufer wie …‘(such) a runner as/like’). Clausal non-degree equatives that are used as attributes (nominal modifiers) are thus particularly close to ordinary relative clauses, structurally, as they constitute subordinate clauses that modify a noun directly (rather than via an overt equative correlate or Tertium-Comparationis AP). This is the basis for the frequent development of equative particles into relative particles discussed above, cf. (68)–(71).⁵⁵ On the other hand, adverbial clausal non-degree equatives without an overt correlate correspond to adverbial free relative clauses – thus the frequent development of modal free relative wh-items into comparison particles.

4.2 Coordination-like characteristics of comparisons

While the widely-held relative-clause like analysis of comparisons can thus explain a number of crucial syntactic and diachronic observations, there are also several characteristics of comparisons that clearly set them apart from (free and headed) relatives as well as other subordinate clauses and that they share with coordinate structures.⁵⁶ The most notable of the coordination-like features of comparisons is the fact that, as mentioned above, the standard of comparison may consist of the comparison particle and a mere phrase such as DP, PP or AP (phrasal comparisons, see Section 3). This is ungrammatical in relatives, see examples (22)/(23) vs. (24) above, and in subordinate clauses introduced by a complementizer, cf. (28) vs. (29). However, it is perfectly grammatical in coordination where also infinitival constructions, mere syntactic heads or even elements below the word level (e.g. verbal particles) may be conjoined, as is also possible in comparisons, cf. (82) and (83). In these examples (a) and (b) exemplify comparisons, with (a) illustrating a comparative and (b) a degree and a non-degree equative, (c) exemplifies coordination, (d) a (free) relative clause, and (e) a subordinate clause introduced by a complementizer, respectively. Thus the External Homogeneity Condition (syntactic identity of conjuncts) typical of coordination also appears to play a role in comparisons.

Another feature that comparisons share with coordination is the possibility of Gapping already mentioned above, see examples (45) and (52), repeated here as (84a) with an additional example of degree and non-degree equatives in (84b). While this is also typical of coordination, cf. (84c), it is ungrammatical in relatives (including free relatives) and subordinate clauses introduced by a complementizer, cf. (84d/e).

Concerning deletion and movement of syntactic material, one may furthermore observe so-called Coordinate Structure Effects in comparisons. For instance, the Tertium-Comparationis AP is typically deleted in the standard of comparison (Comparative Deletion) but not in the preceding part of the clause. Similarly, identical material is deleted in the second, not the first conjunct in coordination. A fourth characteristic that comparisons share with coordination is the possibility of Across-The-Board movement (ATB). Here, several traces of the same type are linked to a single left-peripheral wh-item. As (85) illustrates, this is again possible in comparisons but not in relatives or other subordinate clauses.

Furthermore, Right Node Raising (RNR, “Linkstilgung”) may also apply in comparisons as in coordination, but again not in relatives or other subordinate clauses, as demonstrated in (86).

Finally, as discussed in Section 2.1, certain languages and varieties optionally or obligatorily use a strong pronoun in phrasal comparisons, cf. examples (2)–(5) and (57) above. Crucially, the same form of the pronoun also typically occurs in coordination (e.g. taller than me/you and me).

The diachronic data also constitute evidence for the intimate link between comparison constructions and coordination. As mentioned above in Section 4.1, und(e) which is primarily used as a coordinating conjunction in MHG and ENHG also appeared as an equative particle as illustrated in (87) (and subsequently as a relative particle, cf. (70) above), see also Paul (2007: 427), Ferraresi & Weiß (2011). The use of coordinating conjunctions as equative or comparative particles can also be observed in other languages, consider for example Latin atque and et, Ancient Greek kaí, Lithuanian kaĩ, Old Norse og, enn, Old English and and Javanese karo, cf. Small (1924: 35; 55), Ziemer (1884: 199f.; 217f.) and Stassen (1985: 60; 189).

The link between comparison and coordination is also exploited diachronically in developments from comparison particles into coordinating conjunctions. In German for instance, the comparison particles als and wie secondarily developed coordinating uses on the basis of non-degree equatives during the ENHG period, cf. (88) for als and (46) above for wie. During the same time, the bipartite coordinating conjunction sowohl … als was beginning to be grammaticalised on the basis of degree equatives including the Tertium-Comparationis AP wohl ‘well’, cf. (89) (compare also English as well as). On the basis of coordinating so … als as in (90) and – after the substitution of equative als by wie cf. Table 1 – so … wie as in (91), the coordinating conjunction sowie ‘and’ was formed in the 18th century.⁵⁷

On the basis of the similarities between comparisons and coordination, it has occasionally been suggested that comparison constructions represent instances of coordination syntactically and that comparison particles are in fact coordinating conjunctions. This has been proposed for German by Lang (1977) and Eisenberg (2013), and for English than by Napoli (1983).⁵⁸ For coordination, a functional coordinating head is generally assumed, whose arguments – complement and specifier, respectively – are the two conjuncts as in (92) after Johannessen (1998), see also Reich & Reis (2013).

The same basic structure has been suggested for comparatives by Matos & Brito (2008). More specifically, they assume correlative coordination as in the case of bipartite coordinating conjunctions such as either … or. The second part of the conjunction corresponds to the comparative particle and the first part, that is moved out of the first conjunct by quantifier raising at LF, corresponds to the comparative morpheme.

Applied to German, the structure looks as given in (93). Here, the comparison particle fills the functional head position of the coordinating conjunction, joining the standard of comparison as a second conjunct (equivalent to YP in (92)) to the rest of the clause as the first conjunct (equivalent to XP in (92)). Adapting this analysis to equatives, one could assume a syntactic structure as in (94). (The analysis of comparison particles as Conj⁰ elements argued for in Section 2 sits very naturally with this analysis, although Conj⁰ status of the comparison particles is in principle also compatible with a relative-clause like analysis.)

Assimilating comparisons to coordination can easily explain why typical characteristics of coordination such as Gapping, ATB movement, RNR etc. may occur in comparisons. However, an obvious problem for this analysis is the fact that, as discussed above, the Tertium-Comparationis AP and/or the correlate may be topicalised together, i.e. moved to SpecCP in German, cf. (74)–(76), suggesting that they form one constituent. In (93) and (94), the standard of comparison is situated too high within the syntactic structure to derive this constituency in a straightforward way. The relevant constituent could only be formed by complex and hardly motivated syntactic operations such as emptying of the entire ConjP except for the Tertium-Comparationis AP and the standard of comparison (Conj’) and subsequent remnant movement. Furthermore, as Reich & Reis (2013: 554) state, comparisons generally show characterstics of integration/embedding of the standard into the rest of the clause (note, however, that while Reich & Reis assume that comparisons are always integrated, non-degree equatives as well as hypothetical comparisons also allow unintegrated, e.g. continuative use, see example (63 c) and fn. 54). While it can obviously derive the coordination-like features of comparison, the coordination analysis of comparisons described above is therefore not fully unproblematic, either.

4.3 Comparisons between coordination and subordination

The challenge for a syntactic analysis of the external syntax of comparisons thus consists in capturing their subordinate-clause like, especially relative-like characteristics (including the option of forming a constituent with the correlate and/or Tertium-Comparationis AP) as well as their coordination-like characteristics.⁵⁹ As demonstrated above, both of these do not only show up in syntactic phenomena in the present-day language but are also evident from the diachronic data. The Janus-headed nature of comparisons is reminiscent of a number of other constructions that have come into the focus of syntactic research in recent years, that share this peculiar status of being somehow in between subordinate and coordinate structures, viz. instances of non-canonical sub-/coordination. Thus, one should consider the option of modelling comparisons on these.

One such kind of structure in between coordination and subordination is so-called asymmetric coordination. It consists of two clauses joined by the coordinating conjunction und ‘and’ which, however, do not conform to the External Homogeneity Condition typical of coordination or allow ATB movement etc., cf. Höhle (1990), Reich (2009). Büring & Hartmann (1998) therefore suggest a structure in which und ‘and’ does not take the conjuncts as its arguments but is adjoined to the second conjunct, which in turn is adjoined to I’ of the first conjunct and thus integrated into it.⁶⁰ However, as asymmetric coordination exactly lacks the typical characteristics of coordination that we do find in comparisons, the latter should not be modelled on these.

Another possible model structure to be considered is V2 adverbial clauses, in particular denn/weil-V2 clauses. Denn/weil ‘because’ in these constructions behave similar to coordinating conjunctions because, unlike subordinating conjunctions, they do not trigger verb-final placement in the respective clause but leave the main-clause typical V2 order intact. Antomo & Steinbach (2010) therefore suggest a structure of two coordinated CPs with weil residing in the coordination head joining the remainder of the weil-clause to the main clause. Reich & Reis (2013: 549), however, argue that, since again typical characteristics of coordination are lacking, weil-V2 clauses as well as denn clauses can be characterised as unintegrated but in fact subordinate clauses and should therefore be analysed as adjoined to the matrix-CP (like e.g. continuative relatives).⁶¹ Comparisons thus differ from denn/weil-V2 clauses not only in showing typical characteristics of coordination (a trait that also sets them apart from asymmetric coordination), but also in generally showing characteristics of clause intergration (apart from continuative use etc.). Accordingly, comparisons should not be modelled on these structures, either.

A third kind of structure at the intersection of coordination and subordination that appears particularly promising is V2 relatives, which – like comparisons – combine relative-clause like and coordination-like characteristics according to recent analyses.⁶² While canonical relative clauses are generally taken to be adjoined to their head NP, cf. Stockwell et al. (1973), Haegeman (1991) etc., non-canonical relative clauses in German, which differ from canonical ones in a number of syntactic and semantic features, notably in displaying V2 order, have been argued to possess the paratactic, coordination-like structure in (95), cf. Gärtner (2001), for historical German Axel-Tober (2012).⁶³ This structure is exactly identical with the structure of coordination in (92) with the relative clause and the rest of the clause constituting the two conjuncts YP and XP joined by a functional coordinating head (which here additionally possesses a relative feature). While canonical relative clauses are subordinate (dependent) as well as integrated (embeddded) clauses, V2 relatives can be characterised according to Reich & Reis (2013: 554f.) as neither subordinate nor syntactically integrated clauses but as an asyndetic paratactic/coordination structure with merely prosodic integration of the second conjunct into the first one.

Applying the structure in (95) to comparisons, in fact basically yields the analysis suggested by Matos & Brito (2008) with the comparison particle filling the position of the conjoining head. However, again a number of syntactic differences between comparisons and V2-relatives speak against assigning the same structure to both: While comparisons show typical characteristics of coordination, e.g. the possibility of a simple phrase (DP, PP etc.) after the comparison particle as well as Gapping etc. as demonstrated in Section 4.2, this is again impossible in V2 relatives, cf. (96).

However, rather than speaking against a coordination-based analysis of comparisons, this shows that comparisons are even more coordination-like than V2-relatives and thus in fact supports the coordination analysis for comparisons whilst casting some doubt on whether the coordination analysis in (95) is indeed correct for V2-relatives as it would predict these operations characteristic of coordination to be possible.⁶⁴ Put differently, while comparisons and V2-relatives should not receive the same analysis in view of these differences, a coordination-like structure seems more plausible for comparisons. As in the case of V2-adverbial clauses, another difference between comparisons and V2-relatives is that, whilst not constituting subordinate clauses, V2 relatives are also syntactically unintegrated, whereas, as mentioned above, comparisons typically (but not always) constitute integrated structures.

In sum, none of the non-canonical sub-/coordination constructions that are also somehow ‘in between’ subordination and coordination, viz. asymmetric coordination, denn/weil-V2 clauses or V2-relatives (the structures of which are partly also still under dispute), share with comparisons the typical features of coordination such as Gapping, RNR, ATB movement etc. so that comparisons cannot simply be modelled on either of these structures. Note, however, that due to their clearly coordination-like features in contrast to all other constructions discussed above, a coordination analysis seems quite feasible for comparisons.⁶⁵ The problem that remains is the fact that the constituency of equative correlate and/or Tertium-Comparationis AP with the standard of comparison – as well as their typically integrated behaviour – is not straightforwardly derived under this analysis.

While a final analysis of the external syntax of comparisons that captures all relevant features will have to be left to future research, a tentative solution that combines a relative-like and a coordination-like analysis will be outlined here, based on the reanalysis of an original correlative construction as embedding, i.e. integration into the main clause. In most European languages, equatives are based on correlative constructions, cf. Haspelmath & Buchholz (1998). In so far as comparatives are diachronically often secondarily derived from equatives, as is also the case in German cf. Table 1, this kind of structure will also be relevant in comparatives. For OHG equative so … so, Desportes (2008) explicitly assumes that this pattern is based on the Indoeuropean correlative construction. This kind of construction, also referred to as the correlative diptych, is still found in some Indoeuropean languages today. It typically consists of a free relative clause and a clause containing a demonstrative (correlate, optionally realised as pro) that is anaphorically linked to the relative clause (coindexing of relative clause and demonstrative). Because of the quasi-paratactic characteristics of the construction, it is generally assumed that the relative clause is not embedded, i.e. integrated, in the matrix clause as an argument or modifier but instead left- or right-adjoined to the matrix CP, cf. Keenan (1985), Kiparsky (1995), Dayal (1996). For a number of subordinate clauses in German, it has been argued recently that they originate historically from a reanalysis of this kind of correlative construction, cf. Axel-Tober (2012) for temporal adverbial clauses and dass-clauses.

In order to integrate the analysis of the internal syntax of comparisons including a functional head Conj⁰ above the standard-of-comparison CP, as argued in Section 2, and to capture the coordination-like charactersitics of comparisons, an underlying correlative structure including a coordination head will be assumed basically following Matos & Brito (2008).⁶⁶ Based on this construction, cf. (97 a) (illustrated here for OHG non-degree equatives with so … so ‘so … as/like’), syntactic reanalysis as extraposition would have resulted in an embedding structure with the relative clause integrated into the matrix clause, however still including the functional coordination projection.⁶⁷ A tentative structure for this is given in (97 b).⁶⁸ The comparison clause thus partly resembles the V2 relatives discussed above (however, it is integrated into the matrix clause) and is also reminiscent for instance of de Vries’ (2006) analysis of non-restrictive relatives as being joined to their head DP via a functional coordination projection. The reanalysed construction accordingly shows relative-clause like features as well as characteristics of intergration, in particular the standard of comparison forms one constituent with the correlate (and/or Tertium-Comparationis AP in degree equatives/comparatives), whilst at the same time constituting a basically coordinate structure.

(97)

a.

b.

As indicated by the dotted lines in (97a/b), the reanalysis discussed in Section 2 of the comparison particle⁶⁹ into Conj⁰ may in principle occur before or after the reanalysis of the correlative structure into an embedding structure. (Secondarily, the sequence of correlate and equative particle may then be reanalysed into a new equative particle as discussed in Section 4.1.) The structure in (97 b) largely corresponds to the ones in (78)–(81), however, the standard of comparison is not linked to so (or -er in comparatives, respectively) as its argument but via coordination.