1.1 External syntax of adverbial clauses

1.1.1 Coordination

One piece of suggestive evidence for the distinction in external syntax between CAC and PAC is coordination. This is already demonstrated on the basis of example (1a): while, informally speaking, both while clauses in this example are in effect related to the same main clause, they cannot be coordinated, regardless of whether the resulting coordination is initial (1b) or final (1c).

(1)	b.	*While2 this ongoing lawsuit probably won’t stop the use of lethal injection and while1 the Supreme Court decides what to do, it will certainly delay its use.
	c.	*This ongoing lawsuit will certainly delay the use of lethal injection while1 the supreme court decides what to do and while2 it probably won’t stop its use.

Haegeman (2012: 164–165) proposes that central adverbial clauses only coordinate with central adverbial clauses and that peripheral adverbial clauses only coordinate with peripheral adverbial clauses (cf. the work cited for more data). A full discussion of the syntax of coordination is not appropriate here but ultimately the ungrammaticality of (1b) and (1c) can be related to the external syntax. Coordination is subject to Williams’ (1978) “Law of Coordination of Likes”. Following a suggestion in Huddleston & Pullum (2006) “likeness” can be interpreted in structural terms as meaning that constituents merged in hierarchically distinct positions in the tree do not qualify as “like” and hence cannot coordinate. If CACs are merged within the TP layer, as shown in (5a),¹ and if PACs are merged TP-externally, as informally represented in (5b) and (5c), their inability to coordinate would follow.

(5)	a.
	b.
	c.

Observe that in terms of representations (5b) and (5c), the matrix clause CP1 and the associated PAC, CP2, which modifies it are hierarchically on near-equal footing, meaning that their structural relation is close to that found in coordination. Indeed, that PACs share properties of coordination has been observed independently by Hornstein (1993: 206, note 19), who writes:

There is a secondary conjunctive interpretation that all these connectives (as, while, when) shade into. They get an interpretation similar to and in these contexts. And is not a temporal connective, and these conjunctive interpretations do not tell against the theory [of temporal subordination and complex tense structures].

It should be pointed out, though, that PACs cannot simply be equated with coordinated and clauses (Haegeman 2012: 168). Typically, for instance, gapping is available in coordinated patterns but not in PACs (6). While coordination with and allows for subject ellipsis in the second conjunct (7a), this is not available in the majority of PACs (7b), though, for completeness’ sake, it should be added here that though clauses allow subject ellipsis (7c). (7d) is attested and such examples are not infrequent in journalistic prose.

(6)	a.	John reads the Guardian and Mary the Times.
	b.	??John reads the Guardian while Mary the Times.

(7)	a.	John is doing a Ph.D. in Oxford but did his first degree in Cambridge.
	b.	*John is doing a Ph.D. in Oxford while did his first degree in Cambridge.
	c.	John is doing a Ph.D. in Oxford though did his first degree in Cambridge.
	d.	Observer (02.09.2007, page 6, col. 2)
		Other holiday makers enjoying a late summer break in Kalamata, in the southern Peloponnese, and near Athens, reported seeing smoke and dashes in the air, though were not close enough to the fires to see the flames.

1.2 Internal syntax: Are central adverbial clauses “structurally impoverished”?

As reported in the literature, CACs and PACs differ in terms of their internal syntax: typically English CACs resist argument fronting, PACs allow it:

Central conditional clauses have been reported to resist speaker-anchored modal expressions (Stowell 2004; Nilsen 2004; Zagona 2007; Ernst 2007; 2009; Haegeman 2005; 2006c; 2011). In particular, they are not easily compatible with the four topmost expressions of modality in Cinque’s functional sequence (1999; 2004: 133) – speech act, evaluative, evidential and epistemic modality. PACs, on the other hand, allow such modal expressions. Consider the examples in (11) based on Nilsen (2004: 811). (11a) contains an event conditional, on which the actual realization of the main clause event is dependent; (11b) is a conditional assertion: it echoes a contextually accessible assumption, meaning something like “if, as you say, …”. The CAC in (11a) is not compatible with an epistemic modal: (11c) is unacceptable. In contrast, Nilsen’s (11d) shows that the conditional assertion, i.e. a PAC, is compatible with an epistemic modal.

The incompatibility of CACs with the encoding of speaker-related modal markers is related to the fact that CACs are taken to be presupposed (Hooper & Thompson 1973). As such, CACs lack independent illocutionary force, and are part of the speech act expressed in the superordinate clause. In contrast, PACs have illocutionary potential (cf. Declerck & Reed 2001: 131), as illustrated, for instance, by the fact that they can contain (rhetorical) questions or imperatives, as shown in the attested (12). On the assumption that speaker-anchored modal markers have to be anchored to speaker and/or to the speech time (cf. Ernst 2009), the availability of such markers in PACs (11d) correlates with their illocutionary potential. Relevant proposals can be found in the literature. Zagona (2007: 230) states: “epistemic modals should appear only in finite clauses that make statements relative to the speaker’s world of known propositions at the moment of speaking.” In a similar way, in their seminal paper, Hooper & Thompson (1973) relate the distribution of argument fronting in English to assertive illocutionary force.

1.3 Relating internal and external syntax

The generalization emerging from the discussion above seems to be then that the more structure is manifested within the adverbial clause (1.2), the higher it is merged (1.1). This leads to two (related) questions:

1. How can the observed impoverishment in the internal syntax of CACs be captured?

2. How can the observed correlation between the internal impoverishment (1.2) and the external syntax (1.1) be captured?

We will deal with these two questions in turn. Section 2 develops a proposal to account for the “defectiveness” in CACs.

2.1 CACs are not incompatible with encoding information structure

As suggested above, one might relate the fact that CACs are incompatible with argument fronting and with expressions of speaker-related modality to their presupposed status (see Hooper & Thompson 1973). Referring to Güldemann (1996: 178), van der Wal (2014: 61), for instance, proposes that the backgrounded status of such clause types leads to a compact, largely unstructured presentation of the state of affairs. This “compact” format of presentation would then be incompatible with the availability of internal (information) structuring of the relevant clause, for instance, by focusing a constituent. From this reasoning, one might be tempted to deduce that CACs cannot encode information structural (IS) properties at all, but this conclusion would be too rash. CACs can encode IS effect, as also pointed out by Van der Wal. We provide some illustrative examples here.

In English, CACs are compatible with in situ focus (13a), with clefting (13b), and with heavy NP shift (13c), a pattern typically used for focusing objects.

In French, Stylistic Inversion, a syntactic configuration typically used for focusing a post-verbal subject, is available in CACs (Lahousse 2003a; b; 2010). Consider (14): (14a) represents the default word order; in (14b), an example of Stylistic Inversion, the subject nominal les enfants ‘the children’ occupies a sentence-final position, following the non-finite participle arrivés ‘arrived’.

These empirical data suffice to show that CACs are not categorically incompatible with the encoding of IS related concepts such as, for instance, focus marking. In view of the ungrammaticality of (10a), one might reformulate the restriction on the encoding of IS effects in CACs by proposing that IS effects cannot be encoded in the left periphery, possibly because CACs lack a left periphery. In the next section, we show that this revised hypothesis too would be inadequate.

2.2 Central adverbial clauses do not lack a left periphery

While argument fronting is indeed unavailable in English CACs (10a), clitic left dislocation (CLLD), which also deploys the left peripheral space is available in French CACs. (15) illustrates the pattern: the nominal cette chanson ‘this song’ is in a left peripheral position of the temporal CAC; it is resumed by the pronoun le ‘it’ in the matrix clause.

This difference between English, which disallows argument fronting in CACs, and French, which allows CLLD, cannot be attributed to a parametric difference in the availability of the left periphery in CACs. The left periphery of French CACs displays restrictions on fronting operations reminiscent of those in English. For instance, in French root clauses, when a PP is fronted to the left periphery, the resumptive clitic, where available, is optional (as noted in Rizzi 1997), but when a PP is fronted in a CAC, the resumptive clitic becomes obligatory. (16) provides examples of the latter restriction: ²

Similar effects obtain with fronting of an infinitival complement (Authier & Haegeman 2015). While an infinitival complement can be fronted in a root clause without the need for clitic resumption (17a), fronting of an infinitival clause in a CAC without a resumptive clitic is not acceptable (17b); infinitival fronting with a resumptive clitic (17c) is acceptable.³

In addition, while arguments cannot precede the subject in English CACs, adjuncts can be found to the left of the subject. In Section 3.2, we provide evidence that the adjunct is in a left-peripheral position.

These data suggest that the left periphery of CACs is available for encoding IS effects, but that it is subject to some constraints.

2.3 Constraining the left periphery of adverbial clauses

3.1 Head movement to the left periphery and locality

In the discussion above, it is tacitly assumed that CACs are derived by the leftward movement of a maximal projection (e.g. when in 24a, b) and that this is compatible with the presence of adjuncts and CLLD constituents because, by hypothesis, these constituents do not give rise to intervention effects.

While CLLD constituents remain compatible with XP movement to the left periphery, it should be noted that they are not compatible with head movement to the left periphery. Cardinaletti (1997; 2004: 141) discusses the restrictions on the availability of the left periphery in the Italian Aux-to-Comp pattern, in which a non-finite verb moves to the C domain, inverting with the subject. The pattern is illustrated in (25), from Cardinaletti (2004: 141, her 96). Of interest for the current paper is the observation that while pre-verbal subjects are available in Aux-to-Comp configurations, as shown by acceptability of the pre-verbal DP Gianni and the weak pronominal subject egli in (25a), CLLD is not available in the same context (25b, c).

Similarly, while adjuncts preceding the subject remain compatible with XP movement to the left periphery, they are not available in a configuration created by head movement. Rizzi (1997: 303, 57) shows that in Italian the adjunct improvisamente ‘suddenly’ can be sentence-medial (26a) or sentence-initial (26b). In contrast, (26c) and (26d) illustrate the Aux-to-Comp environment: the non-finite auxiliary essendo ‘being’ has moved to C. (26c) shows that Aux-to-Comp, i.e. T to C movement, is compatible with the presence of an adjunct in a medial position; (26d) shows that Aux-to-Comp is not available when the adjunct precedes the subject.

These patterns are replicated for English. In the embedded interrogative context (27a), the adverbial adjunct at that time may occupy a position between the fronted wh-phrase which models and the subject they. (27a) confirms that the adjunct does not constitute an intervener for leftward movement of the wh-phrase. In root contexts, the same adjunct may appear to the left of the fronted wh-phrase as shown in (27b). But crucially, the adjunct cannot follow the inverted auxiliary (27c). Thus, while the adjunct is compatible with XP movement to the left periphery, (27c) shows that it is incompatible with head movement to the left periphery. The data in (28) replicate the effect for yes-no questions:

Rizzi (1997: 303) interprets the incompatibility of a left peripheral adjunct with T to C movement as a violation of the Head Movement Constraint (HMC) (Travis 1984). For him, the fronted adjunct occupies the specifier of a left peripheral TopP, whose head, Top, blocks head movement of the auxiliary. (28d) would represent the derivation of the illicit (28c):

Observe in passing that the incompatibility of adjuncts with T to C movement and their compatibility with XP movement militates against a simple adjunction analysis for adjuncts. If initial adjuncts were simply TP adjoined it is not clear why they should interact with head movement.

3.2 Head movement and the left periphery of English adverbial clauses

We have seen that T to C movement imposes more important restrictions on the left periphery than does XP movement. This leads to the prediction that in cases in which adverbial clauses are derived by head movement, their left periphery will appear to be more impoverished, in that, for instance, in such contexts adjuncts will also be banned. This is confirmed for English, the relevant data for which were already discussed by Rizzi (1997: 303), and which we will discuss in this section, and it also applies to Japanese, as shown in Endo (2011b; 2014). We discuss the Japanese data in Section 3.3.

3.3 A gradient typology of adverbial clauses

So far, we have been operating with a binary opposition between CACs and PACs. However, while this opposition can be used as a rough typology of adverbial clauses, it turns out that a more subtle approach is needed, as revealed by Endo’s (2011a; b; 2012; 2014) work on Japanese adverbial clauses. In particular, exploring a body of descriptive work on Japanese grammar, Endo has shown that in terms of the availability of internal functional structure, i.e. the internal syntax, the typology of adverbial clauses must ultimately be recast in terms of a gradient system.

Endo’s proposal is based on the distribution of post-verbal functional elements in adverbial clauses. Japanese (37a) illustrates the sequencing of functional elements and (37b) represents the corresponding functional hierarchy for the clause which will be adopted here. The labels S-mood and A-mood refer to two speech act related moods, with, as suggested by the abbreviation, S-mood being speaker-related and A-mood addressee related. The label A-mood refers to the locus for the encoding of the speech act. A-mood is also sometimes referred to as interpersonal mood. (See Endo 2007 for A-mood sentence-final particles and S-mood elements in Japanese.) In (37a): the element rare represents passive voice, the element tei represents progressive aspect, the element na represents negation and the element i represents present tense.

Minami (1974) and Noda (1989; 2001) demonstrate that there is a “gradience” in the availability of post-verbal functional elements in adverbial clauses. Table 3 inventorizes the relevant gradience: at one extreme, aspectual nagara ‘while’ clauses only allow for a voice head to be present in their functional field, while, at the other extreme, ga ‘though’ clauses allow almost the full array of functional elements, and ba and nara clauses are somewhere in between.⁵

In Table 3, Noda uses the term “Pol” for “Polarity” to cover both the negation morpheme na and – by hypothesis – its non-overt counterpart, the affirmation morpheme φ.⁶

(39) illustrates the patterns. In the nagara clause in (39a), the aspectual element tei is not available. Note that the same element tei remains available in the associated main clause, a point to which we return in Section 4.2.⁷ In contrast in the ga clause in (39b), not only the aspectual element tei but also the polarity element na, the tense element ta and the S-mood element daroo are available. Ba and node clauses in (39c) and (39d) are somewhere in the middle. In a ba adverbial clause in (39c), the aspectual element tei and the negation element na are available, but the tense element i is not. The ba adverbial clause in (39c) is minimally different from the node adverbial clause in (39d), in which the tense element i is available, while the S-mood element daroo is not.⁸

Superficially, then, one might say that nagara ‘while’ clauses are highly defective in that only the voice element can be present, that ga ‘though’ clauses have an extended functional structure and that node/ba ‘when’/‘if’ clauses are somewhat less reduced than nagara clauses. This entails that we no longer have a binary opposition between CACs and PACs and that instead there is a more fine-grained cline of available functional structure with the different clauses displaying various degrees of reduction. The gradience observed for Japanese cuts across Haegeman’s binary opposition between CAC and PAC, with nagara, zuni, ba and toki plausibly seen as introducing CACs and node and ga introducing PACs.

For Japanese, the appearance of gradient structural reduction as reflected in the availability of functional elements can be made to follow if we assume with Endo (2011b; 2014) that adverbial clauses are derived by head movement. This is so because, as we have seen, in line with what has come to be known as the Head Movement Constraint (Travis 1984), head movement has more drastic effects on the available structure than XP movement (see Section 3.2). Assuming, for instance, with Endo (2011b; 2014) that the aspectual nagara clause is derived by the movement of the functional head Asp, and that intervening heads on the movement path constitute interveners for the movement of the head, it follows by the HMC that none of the hierarchically higher heads in the functional sequence (38b) will be available, mimicking a deep truncation effect. Also, nagara clauses will not host the aspectual element tei because nagara itself and the aspecual element compete for the single position of the Asp head. In Section 3.4, we develop the head movement analysis in more detail.

3.4 The head movement derivation of Japanese adverbial clauses

To capture the gradient clause-internal variation in the presence of functional material in Japanese adverbial clause, Endo (2011b; 2014) develops a head movement derivation. In his analysis, the observed gradience in the availability of the functional heads reflects the derivation of the adverbial clause.

Concretely, we have seen that adverbial clauses may show defectivity in the availability of the functional heads. Endo’s proposal is based on the hypothesis that the lowest functional head that is missing in the adverbial clause corresponds to the head that has been moved to the left periphery. Due to the HMC, as a direct consequence of the movement of this head, all higher functional heads also become unavailable, i.e. must be missing. We illustrate the proposal with two concrete examples. Endo proposes that nagara ‘while’ clauses are derived by the movement of the aspectual head to the C domain. On the one hand, in line with the discussion of the English data in (32), and due to the HMC, head movement of Asp entails that the functional projections dominating Asp become unavailable, while the projections dominated by AspP (e.g. VoiceP) can be present. On the other hand, the proposal also entails that the features associated with the moved Asp head will end up on the C-head that is responsible for clause typing. Similarly, if zuni clauses are derived by movement of the Pol head, then this means that, while AspP can be projected, projections dominating PolP are unavailable. We refer to Endo’s own work (2012b; 2014) for more details.

There remain a number of questions concerning the implementation of this derivation, which, though important, are somewhat tangential to our initial focus, namely that of the relation between internal syntax and external syntax of adverbial clauses.

First, the implementation of the head movement analysis as outlined here will ultimately be heavily dependent on the specific assumptions about the architecture of the clause and in particular about the granularity of the functional hierarchy. So, any decisions taken at that level will determine the labels of the functional hierarchy at play. A further question is whether, in a head movement derivation of adverbial clauses, the relevant head which undergoes movement to the left periphery, is associated with an operator and if so, what the fate of that operator is. A question of a different nature is whether the head movement analysis could in fact be reformulated in terms of XP movement in an account in which the final position of the post-verbal elements in Japanese results from leftward XP movement (in terms of Kayne’s 1994 Antisymmetry approach). Under such a view, it is conceivable that the adverbial clauses be derived by a movement of the clause which smuggles (in the sense of Collins 2005) the relevant head/operator.⁹ Endo (2012) proposes a truncation analysis of adverbial clauses in Japanese. As discussed here, the effect of truncation follows from head movement.

3.5 Head movement and XP movement

The head movement analysis of adverbial clauses raises more general questions of comparative syntax. One question concerns cross-linguistic variation: we need to examine to what extent languages vary parametrically in the way that adverbial clauses are derived by head movement or by phrasal movement.

At first sight, it might have seemed natural and tempting to try to relate the fact that Japanese systematically implements head movement rather than operator movement to derive adverbial clauses to the fact that Japanese is a wh in situ language. However, a rigid classification according to which languages either deploy head movement or XP movement, depending on other properties of their syntax, is empirically incorrect. The variation (head movement vs. XP movement) is not absolute: we have already shown, for instance, that in English, head movement is available for a subset of conditional clauses, as shown in (29)–(32), see also Biberauer & Roberts (2014) on the parametrization of head movement.¹⁰ Similarly, West Flemish by and large can be argued to derive adverbial clauses by XP movement (see Reuland 1979 for arguments for the XP movement derivation in Dutch and Flemish), but a subset of conditional clauses can be derived either by XP movement or by head movement. These are illustrated in (40): in (40a), the conditional clause is introduced by the conjunction oa ‘als’, and the assumption is that this is derived by the movement of a null operator to the left periphery (see Haegeman 2012: 220 for arguments); in (40b) the adverbial clause is derived by movement of the finite irrealis auxiliary moest ‘should’ to the left periphery (see Haegeman 2010a for discussion and motivation).

As discussed, the further implication of Endo’s analysis is that we have to abandon the binary opposition between CACs and PACs. Concretely, if the functional hierarchy determines the internal structure of adverbial clauses in the case of head movement, as suggested above, and if, as also suggested, head movement and XP movement are alternative ways of deriving adverbial clauses, it follows that also in the case of XP movement, a more fine-grained approach is needed, in which the relevant operator that is extracted will vary depending on the type of adverbial clause. Though the complete gradience has yet to be determined and will depend on the granularity of the functional hierarchy, the idea that the binary opposition must be replaced by a more fine-grained approach taking into account the functional hierarchy of the clause is foreshadowed in the literature. We illustrate this point briefly here.

In Haegeman’s original proposal (1991, etc) adopted here both temporal clauses and event conditionals are CACs. Adopting a minimalist functional hierarchy, Demirdache & Uribe-Etxebarria (2004; 2012) propose that temporal adverbial clauses introduced by when are derived by movement of temporal operator. In their proposal, the launch site of the operator that derives temporal clauses is SpecAspP, which hosts the reference time. (41) summarizes the analysis. We refer to their work for detailed motivation.

(41)	a.	Demirdache & Uribe Etxebarria (2012)
		when Zooey left
	b.

In terms of a more fine-grained functional hierarchy such as Cinque (1999: 87–88), the relevant projection will have to be reinterpreted in terms of the specialized temporal projections.

Bhatt & Pancheva (2006) propose that conditional clauses are derived by the movement of the World operator, which, in terms of the Cinque hierarchy of functional projections, Haegeman (2010a; 2012) reinterprets as the Irrealis (mood) operator, which is associated with a projection that is dominated by the temporal projections (see Haegeman 2010a).

Thus, while temporal adverbial clauses and conditional clauses are CACs and while both are derived by operator movement, the launch sites of the operators differ with the temporal operator originating in a higher position than that associated with conditional clauses.

If such a more fine-grained typology is called for, the question arises whether there is any need to maintain the binary opposition between CACs and PACs at all.¹¹ Though clearly of interest, we will not explore this point here and we return to the second research question which we have raised: how can one account for the correlation between the internal and the external syntax of adverbial clauses?

4.1 Introduction

In a nutshell, the correlation uncovered in relation to adverbial clauses is that PACs are merged in a higher position than CACs and that the higher the adverbial clause is merged (Section 1.1), the more structure is available within the adverbial clause (Section 1.2).

Building on Hooper & Thompson (1973) and Haegeman (2003: 6), Frey (2011) postulates that, being root like, PAC’s host a Force projection and that, by virtue of this, PACs must be licensed through the matrix Force. We refer to his paper for details. Interestingly, Frey’s analysis can be taken to entail that there is a “matching” condition on the merger of PACs in that a property of their internal structure, in particular, the PAC-internal presence of the projection ForceP, imposes restrictions on their licensing and as a result determines the level of merger with the matrix clause. Put informally, an adverbial clause which contains the Force head must merge at the level of Force in the matrix clause.

Similarly, in English, temporal clauses introduced by since impose restrictions on the temporal and aspectual coordinates of the matrix clause (Quirk et al. 1985: 1015–1018), as shown in (43): in British English, modification by a temporal since clause requires a main clause perfective aspect:

The intuition that there is a matching condition between the adverbial clause and the matrix clause which it modifies is fully exploited in Endo’s work on Japanese adverbial clauses, which elaborates earlier work by Noda (1989), where a general algorithm is developed for the merger of adverbial clauses. Put informally, Endo’s core proposal is that the functional make-up of the adverbial clause, which is itself a function of its internal syntax, determines the locus of merger in the matrix clause. As the functional make-up of the adverbial clause itself is a function of its derivation, as shown in Section 3.4, this ultimately means that the derivation of the adverbial clause determines the merger with the main clause it modifies. In this section, we present Endo’s proposals in some detail.

4.2 Adverbial concord

Noda (1989) was the first to discuss what he called the “concord” relation between the presence of a specific adverbial clause and the availability of functional heads in the clause it modifies. For instance, the presence of an aspectual nagara ‘while’ clause restricts the choice of the matrix aspectual element. This is illustrated in (44). Examples (44a) and (44b) illustrate two aspectual elements in root clauses: progressive tei and inceptive hazime. As shown by (44c) and (44d), a matrix clause modified by a nagara clause may contain the progressive aspectual element tei, and cannot contain the inceptive aspectual element hazime.

One might claim that Noda’s proposal simply expresses certain semantic compatibility conditions between the matrix clause and the adverbial clause. However, Noda’s proposal is also syntactic in nature, because he claims that the point of attachment of an adverbial is determined by concord with a functional element in the matrix. This raises the question of what underlies this concept concord. Noda’s (1989) idea is that each adverbial clause is associated with a functional element such as Aspect, Tense, etc. and that it restricts the choice of the matrix functional element that it modifies rather than selecting a specific lexical item in the main clause. Noda labels this matching restriction “koou” in Japanese, which is usually translated as “concord” into English. Another way of expressing this relation might be to use the term “correlation” as in Iori (2017: 377), in which the contrast between (44c) and (44d) is characterized as follows: “Since, [reference omitted, lh & ye] a nagara clause showing ancillary state fits well with a stative aspect but not so well with a non-stative aspect, it can be said to correlate with stative aspect”.¹³

The syntactic nature of Noda’s concord relation can be seen in the case where two adverbial clauses appear at the same time. If, as suggested above, the matrix functional head signals the timing of the merger of the adverbial clause, this would lead to the prediction that the higher an adverbial clause is situated in the clause (i.e. the lower its position in Table 3), the later it will be merged with the associated clause and the further from V on its left it will appear. This prediction is largely borne out, as illustrated below. In (45) the matrix clause combines with a conditional ba clause and with a zuni ‘without’ clause. As can be seen, the highly-preferred order is that according to which the zuni clause is to the right of the ba clause. This corresponds to their merge order, the zuni clause being merged before the ba clause. Similarly, in (46) the node clause is merged later, hence it is structurally higher, than the nagara clause and precedes it.

Noda (2013) discusses such co-occurring adverbial clauses in detail: he examines the combinations of two adverbial clauses of each type in Noda (1989) to show that the structurally higher adverbial clause precedes the structurally lower adverbial clause, while the reversed word order of the two adverbial clauses is unnatural or ungrammatical. (See also Endo 2012 on this point).

Returning to other types of adverbials, the concord relation between a zuni ‘without’ adverbial clause and the matrix clause is reflected in the availability of polarity elements: a matrix clause modified by a zuni clause is incompatible with the negative polarity element na. We assume that it must instantiate the non-overt positive element.

For node ‘because’ clauses, the selection of mood markers in the matrix is constrained:

Finally, PACs introduced by ga ‘although’ only impose restrictions at the level of the topmost layer, which encodes A-mood. When modified by a ga clause, the matrix proposition must be associated with the (non overt) neutral A-mood (=φ) and it is incompatible with overt A-mood element such as the question marker ka. As illustrated in (49), all other functional heads are instantiated: (49a) and (49b) illustrate root speech act elements; (49c) shows that the neutral mood is available in a root clause modified by a ga clause, while (49d) shows that the question marker ka is not available.

The concord effect, i.e. the correspondence between the adverbial clause and the functional material of the clause which it modifies, is interpreted then as the outcome of a matching relation between the internal syntax of the adverbial clause, the launch site of the head movement that derives the clause, and its external syntax, i.e. the point in functional sequence of the matrix clause where the clause merges. In the next section, we tentatively formalise this matching relation.

4.3 Adverbial concord as a reflex of merge

Let us assume that adverbial clauses are merged as modifiers of the matrix clause by the intermediary of a functional head Mod (cf. Bowers 1993; Rubin 1994; 2003; Rizzi 2004): specifically, we propose that the adverbial clause is merged as a specifier of the functional head Mod. Alternatively, assuming a less articulated structure, the adverbial clause could also be taken to adjoin to the relevant matrix functional projection. We will not examine the difference between these two implementations. The observed “concord” effect in adverbial clauses is a reflex of the matching condition on the external merge of the adverbial clause: to be more precise, there is a featural matching condition on the specifier of Mod and the complement of Mod. Concretely, our analysis rests on two hypotheses.

1. An adverbial clause is derived by operator movement or by head movement, and the features of the moved operator or the moved head are instantiated as a clause typing feature at the C layer of the adverbial clause. For instance, if nagara clauses are derived by head movement of an aspectual head, then the feature [ASP] or [+stative aspect] will be instantiated as a clause typing feature of the adverbial clause.

2. The matching condition on the merger of adverbial clauses requires that an adverbial clause with the clause typing feature [x] merge with a head Mod that selects the functional layer instantiating the (“concording”) feature [x].¹⁵

This is schematically represented in (50). We assume that the matched feature [x] is also instantiated on Mod.

Concretely: a nagara clause will be merged as the specifier of a Mod head, which selects the matrix aspectual projection whose head is spelt out as tei (or has the feature [+stative aspect]).

Depending on the relative position that the adverbial clause takes in the gradience scale in Table 3, then, the launch site of the head that moves to derive the clause will vary in height and its launch site will in turn determine both the internal functional structure of the adverbial clause and its external merge in the matrix domain. Observe that ga ‘though’ clauses retain almost all functional structure (see 49c). On the assumption that they are derived by head movement, the moved heads must then be topmost in the functional hierarchy. This hypothesis would tie in with the hypothesis formulated in Section 2.3.3, inspired by Arsenijević’s (2009b) work on finite complement clauses, according to which the operator that moves to derive PACs is launched in the speech act projection.

The schematic representation (52) is in line with existing proposals in the literature. (52c) is very close to Haegeman’s representation (5b), and for (52) we can also take the matching of the [Mood A] feature on the PAC with that on the associated clause to correspond to the licencing condition on Force that was discussed in Frey (2011).

4.4 Polish: By concord in counterfactuals

Tomaszewicz (2009; 2012) develops a head movement derivation of Polish counterfactual conditional clauses. Effects of the matching condition are manifested there too. In the matrix counterfactual clause (53a), the modal particle by appears to the right of the subject. In the counterfactual conditional antecedent of (53b), the particle by is adjacent to the conjunction gdy ‘when’, which serves to introduce the conditional clause, and it precedes the subject. (53c) shows how the derivation meets the matching condition:

For completeness’ sake, we add that in PACs, by follows the subject, like it does in main clauses. This is expected since if PACs are derived by movement, the moved constituent will originate in a higher position.