In German, a concessive adverbial clause is standardly expressed with the help of the complementiser obwohl (‘although’), (1).

The complementiser obwohl is considered to be neutral regarding style and variety differences (Zifonun et. al 1997: 2311). In English, although is the prototypical representative of concessive complementisers. German trotzdem and English nevertheless can be taken as the representatives of concessive adverbs appearing in an independent clause and trotz and despite are representatives of concessive prepositions. The present paper deals in detail with verb-final adverbial clauses introduced by obwohl.1 Furthermore, there are special constructions which are associated with a concessive meaning. In our discussion the verb-first concessive construction (V1-concessive) in (2), which obligatorily contains the modal particle (MP) auch, will play a central role.

According to the standard assumption, the meaning of the concessive relation expressed for example by p, obwohl q involves two components. First it is expressed that p and q are true, and second that in the normal course of events non-p follows from q. Furthermore, the second meaning component is considered to be not-at-issue. Thus, the two components of the concessive meaning are often stated as in (3) (cf., e.g., König 1985; Kortmann 1997; Breindl 2004).2 The slightly different version of Antomo & Steinbach (2013) with the second meaning component expressed by a non-selective quantifier, a restrictor and a nucleus and with the assumption that it is conventional implicature instead of a presupposition is depicted in (4).3 The presupposition and the conventional implicature are associated with a lexical item like obwohl as illustrated in (3) and (4), respectively. Likewise, one would assume that these meaning components are associated with a special construction like the V1-concessive.

Based on d’Avis (2016), the present paper will argue that as a general condition the presupposition/conventional implicature depicted in (3)/(4) is too strong. In the standard use of a concessive, however, the non-at-issue condition in (3) or (4) is operative and if it is, it has far-reaching consequences. It is in the cases in which such a modalised not-at-issue condition is in operation that the concessive cannot have the status of a central adverbial clause. In these cases, the concessive relation obviously does not just claim that certain facts exist. Rather, a crucial part of its meaning is that the co-occurrence of two facts does not correspond to a rule which a thinking mind takes to hold usually. It is something rather special about concessives that in their standard use they crucially involve an estimation by a thinking mind, or, as we will say, by a judge. This is an important difference to, for example, the causal or the conditional relation. These are not necessarily understood as exhibiting an appraisal by a thinking mind as part of their meaning. Thus, in the standard case a concessive adverbial clause encodes a rather complex conceptual relation (e.g., König 1985; Di Meola 1997; Spooren & Sanders 2008).4

The article will proceed in the following way. For comparison, Section 2 presents some basic syntactic properties of temporal verb-final clauses (V_fin-clause), obwohl-V_fin-clauses with and without specific linguistic contexts and V1-concessives. Furthermore, it investigates the distribution of different kinds of root phenomena with these clause types and discusses whether these types carry illocutionary force. Then, the clause types are assigned to three classes of dependent non-complement clauses: central adverbial clauses (CACs), peripheral adverbial clauses (PACs) and non-integrated dependent clauses (NonICs). Section 3 introduces Krifka’s (2018, to appear) semantic layers of a speech act: a proposition, a judgement, a commitment and a speech act, as well as their corresponding syntactic categories: TP, JP, ComP and ActP. The categorical make-up of CACs, PACs and NonICs will be discussed. An obwohl-V_fin-clause in the use without a very special context necessarily involves a judgement and features a JP in syntax, and therefore cannot appear as a CAC but rather as a PAC. In contrast, a V1-concessive necessarily is a NonIC; it constitutes an independent speech act and is of the syntactic category ActP, containing ComP and JP. This section also compares Krifka’s classification with Sweetser’s (1990) three domains of interpretation (content, epistemic, speech act) and demonstrates that the two classifications are orthogonal to each other and additive. Section 4 is concerned with the syntactic licensing of concessives as PACs and as NonICs. Section 5 gives a summary.

In this section temporal V_fin-clauses, obwohl-V_fin-clauses and V1-concessives are compared regarding some basis syntactic properties, and it is determined how these clause types are to be classified under the classification into CACs, PACs and NonICs (e.g., Haegeman 2003; Frey 2016). Let us first consider whether binding is possible into these clauses when they appear without linguistic context:

The temporal in (5a) allows binding from the matrix clause more easily than the concessives (5b,c) do. The same grouping of these clause types can be found with other phenomena. For example, in a context which induces a wide focus reading a temporal may carry the nuclear accent of the complex [host + adverbial clause] more easily than an obwohl-concessive or a V1-concessive may. (Thus, the reader is asked to read the examples in (6) without a clausal accent in the matrix clause.)

Likewise, it can be shown that a temporal clause may be in the scope of a constituent negation appearing in the matrix clause, whereas an obwohl-concessive or a V1-concessive may not (cf., e.g., Brandt 1994).

It has been shown that regarding the properties tested in (5)–(7) a German causal weil-V_fin-clause (‘because’-V_fin-clause) and a German conditional wenn-V_fin-clause (‘if’-V_fin-clause) pattern with the temporal clause (cf., e.g., Frey 2016). According to an influential assumption, the generalisation in (8) underlies the data in (5)–(7) (cf., e.g., Haegeman 2006; Frey 2011).

Usually it is stated like this, and no explanation is offered. Yet, if one thinks about the meaning of concessives, an attempt at an explanation suggests itself. With the help of CACs arguably only relations in the outside world are encoded. However, in a context-free use, concessives do not just refer to a relationship between eventualities in the outside world.5 Thus, if the use of a context-free concessive necessarily goes beyond direct reference to the real world, it follows that such a concessive cannot be realised as a CAC.

Let us now come to examples in which an obwohl-concessive behaves as a CAC in contradiction to (8). Freywald (2018: 70) presents the authentic example in (9) (judgement by Freywald, her example 2.4: (23b)), which she takes to show that obwohl clauses can be CACs. Freywald (2018) asserts that sometimes concessives allow binding from the host, a sign of an adverbial clause being a CAC.

I think Freywald’s (2018) observation is important. However, in the following I will argue that (9) does not illustrate a general possibility for obwohl-clauses but depicts a special case.

D’Avis (2016) argues against (3)/(4) as correct explications of the general meaning of the concessive relation expressed for example by obwohl. D’Avis makes clear that there are usages of obwohl-clauses which do not involve the expression of a normal course of states of affairs. In these cases the conditional to which the concessive is in opposition is to be found in the context. D’Avis (2016) gives the following examples (translated into English):

In (10)–(12), we have the following situation. When S2 utters ‘although p, q’, a conditional of the form ‘if p, then not q’ is salient. Thus, in all these cases the conditional which is contradicted by the although-clause appears in the context either explicitly or as an appropriate deduction. Thus, according to d’Avis (2016), in such cases the assertion of p and q made by using the concessive is not in opposition to a modalised generalisation the speaker assumes but rather to a statement given in the context.

Interestingly, Freywald’s (2018) example in (9) is of the type in which the relevant conditional is introduced in the discourse. In her example this happens via a question. If one checks the internet source, one finds that (9) is part of the answer to the question in (13) (translation of the German source).

Thus, when (9) is expressed, among others, the following conditional is salient: ‘If a certain person has signed up for the marathon distance, this person has to drive the marathon distance’. (9) contradicts this conditional and states: ‘for all x, if x has signed up for the marathon distance, then x can drive 115 km, 165 km or the marathon distance’.6

D’Avis (2016) states the condition of use for concessives in (14). The set of context candidates contains propositions upon which it has not yet been decided by the discourse participants whether they will become members of the common ground.

In Section 3 it will be argued that the option of binding in (9) depends on the very fact that (9) does not come with a concept of the speaker of what normally is the case. According to d’Avis, such a conception of an expected course of events (CECE) is only associated with a stand-alone use of a concessive clause.

We can replicate the relationships observed with (9) with a constructed example. Without an explicit context which satisfies (14), (15) does not allow binding into the obwohl-clause.

Consider now (16). Here the conditional which is contradicted by the concessive is given in the context. Binding into the concessive conditional becomes possible.

Thus, (9) and (16) suggest that a concessive may appear as a CAC if no CECE is operative since the proposition to which the concessive is in opposition is given by the context.

Now, the question arises of how the general explications of the concessive meaning like the ones depicted in (3)/(4) can be reconciled with d’Avis’ (2016) condition for the use of concessives presented in (14). According to d’Avis the interrelationship is as follows. Suppose the following sentence occurs discourse initially.

Since (14) is in effect, the set of context candidates contains contradicting propositions.

It seems that by uttering (17) the speaker is contradicting her-/himself. One can assume that the speaker wants the at-issue content of (17), i.e., ‘p and q’, to become part of the common ground. Why does she/he use an expression which leads to a contradiction between the members of the set of context candidates which she/he her-/himself establishes? According to d’Avis, this is just the point of this kind of concessive. The hearer is requested to think about why the speaker is bringing a contradictory conditional into play. An obvious solution is that the conditional for the specific case can be derived from a conception of an expected course of events ascribed to the speaker. For (17) the appropriate CECE is given in (19).

D’Avis assumes that the maintenance of CECEs and the formulation of generalisations belong to the basic cognitive equipment of the human being. When expressing a concessive, the speaker assumes that the listener recognizes that the contradictory conditional, which she/he her-/himself has brought into play, is based on a CECE of hers/his. Thus, we have:

In contrast to the standard assumptions depicted in (3)/(4), d’Avis (2016) takes the CECE which appears in these uses of concessives as a generalised conversational implicature triggered by the speaker, the reason being that it is deletable by the speaker. D’Avis gives the following example.

As an intermediate summary we can state that in the context-free use concessives do not appear as CACs. It can be assumed that the reason is that in the context-free use the concessive comes with a CECE ascribed to the speaker. With a CECE a speaker goes beyond a pure reference to the external world. With a CAC only a reference to the external world is possible. Accordingly, if a concessive contradicts a statement about the external world which is salient in the discourse, the speaker does not evoke a CECE and the concessive can be realised as a CAC.

In the following let us concentrate on the context-free use of concessives. In this use, concessives do appear as PACs or, as we will see later, as non-integrated adverbial clauses (NonICs). Let us first consider their appearance as PACs. Under the assumption in (22), cf., e.g., Haegeman (2006) and Frey (2011), to be refined below, the differences to CACs illustrated in (5)–(7) make sense: binding by the host’s subject is possible into a CAC but not into a PAC,7 a CAC, but not a PAC, may carry the nuclear accent of the complex sentence, and a CAC, but not a PAC, may be in the scope of a constituent negation appearing in the TP-domain of the matrix clause.

Another important difference between CACs and PACs lies in the fact that CACs do not tolerate so-called root phenomena (RPs) whereas PACs do.8 This is how it is normally stated (cf., e.g., Haegeman 2012). However, in Frey (2011) and Frey & Meinunger (2019) it is argued that one should distinguish between weak and strong RPs; weak RPs have a wider distribution, and only they can occur in PACs. As weak RPs the following German phenomena will be discussed: modal particles (MPs) like ja,9 evaluative adverbials like leider (‘unfortunately’) and the marking of an aboutness topic with the help of the particle jedenfalls (‘at least’).

A temporal clause may not host the MP ja, (23a), whereas a concessive clause may, (23b). As a weak RP ja may also appear in the complement of a mental attitude verb, (23c), this holds for the other weak RPs as well.

(24) shows that a temporal clause may not host the evaluative sentence adverbial leider whereas a concessive clause may.

In German, the particle jedenfalls may serve as a marker for an aboutness topic (cf. Frey 2004). (25) demonstrates that a temporal clause may not host such a marked phrase whereas a concessive clause may (of course, (25a) would be fine without the appearance of the topic marking particle jedenfalls).

Let me stress the correlation between the different properties of being a CAC and the correlation between the different properties of being a PAC. Consider German weil-causal clauses. In principle, weil-clauses can appear as CACs and as PACs. This is demonstrated in (26). (26a) shows that binding into a weil-clause is possible, giving evidence that a weil-clause can be a CAC. (26b) shows that a weil-clause in the prefield may host a MP or a sentence adverbial, giving evidence that the weil-clause can appear as a PAC. (Its positioning in the prefield is to guarantee that it is a PAC and not a NonIC, see below.)

Crucially these two phenomena cannot co-occur. Binding into a weil-clause is no longer possible when the weil-clause hosts a MP or a sentence adverbial.

The correlation of these properties is also found with concessives. Above we have seen that a concessive which contradicts a conditional salient in the discourse can appear as a CAC. We established this by the possibility of binding into the concessive, cf. (16). What happens if in this case a MP is added? As illustrated in (28), the binding option is lost.10

(29) demonstrates that the construction is grammatical if no binding is involved.

Let us next address the subject of strong RPs and their differences to weak RPs. While weak RPs are legitimate in all root contexts, strong ones can only be found in a subset of these contexts. Strong RPs only occur in independent clauses and in a very small set of dependent clauses which show signs of being syntactically independent of their host clauses (Frey 2011). Following the terminology of Frey (2011), the latter will be called non-integrated dependent clauses (NonICs). German NonICs are, for example, concessive verb-first clauses (V1-concessives),11 causal verb-first clauses (V1-causals), consecutive so-dass-clauses (‘so-that’ clauses), continuative relative clauses and wobei-clauses (‘whereby’-clauses); wobei-clauses may have different readings, including the concessive one. In addition, as we will see below, some adverbial clauses which normally appear as PACs can be realised as NonICs if they are separated from their hosts by a pause and carry their own sentence accent.

Before we study strong RPs, let us consider some of the properties of NonICs. Like PACs NonICs do not allow binding into them from their host, they cannot carry the sole nuclear accent of the complex [host + NonIC] and they cannot be in the scope of a constituent negation appearing in the host clause; cf. Frey (2011, 2016). However, while a CAC or a PAC may occupy the prefield of its host, as demonstrated for instance in (23a,b), a NonIC may not. (30a,b) contain a V1-concessive, (30c,d) a V1-causal, (30e,f) a continuative relative clause, and (30g,h) a concessively interpreted wobei-clause. That NonICs cannot be placed in the prefield is of some importance since the prefield is a possible position for the constituents of a clause. Furthermore, from the fact that NonICs cannot occupy the prefield it can be concluded that in (30a,c,e,g) these clauses do not occupy the (integrated) position for extraposed elements.

Another property which differentiates NonICs from PACs (and from CACs) concerns the possibility of being embedded together with the host clause. This is illustrated in (31) with representatives for PACs and NonICs. Whereas a PAC can be embedded together with its host, (31a), a NonIC cannot be; cf. (31b) with a V1-concessive, and (31c) with a wobei-clause.

The data in (30) and (31) demonstrate that NonICs are less integrated than PACs (for more evidence, see Frey 2016). Altogether we find the integration hierarchy in (32).

Let us recapitulate the two main insights about the structural relationships between CACs, PACs and NonICs and their licensing clauses. First, CACs are in the c-command domain of the subjects of their licensing clauses; this does not hold for PACs and NonICs. Second, CACs and PACs are constituents of the structures of their main clauses, while NonICs are not. According to Frey (2011), the licensing of CACs, of PACs and of NonICs is therefore rather different. A CAC is licensed in the standard way inside its host’s (extended) V-domain. A PAC is syntactically licensed inside its host too, albeit in a very high position. According to Frey (2011), a NonIC is not part of the syntactic structure of its associated clause but is a true orphan in the sense of Haegeman (1991/2009). In Section 4 the licensing of CACs, PACs and NonICs will be considered again and a refined proposal will be presented.

Let us now look at strong RPs.12 An example is tag questions like German habe ich recht? (‘am I right’). Such tags add an inquisitive illocutionary act to an assertion. They invite the hearer to give a yes-no answer, with a clear bias toward a support of the assertion. Now, (33) demonstrates that question tags are not tolerable in an object clause of a verbum dicendi, (33a), or in a PAC, (33b), but they are possible, for example, in a concessive V1-clause, (33c), and in a wobei-clause, (33d).13 Crucially (33c) and (33d) are fine with tags that only relate to the contents of the V1-concessive and the wobei-clause, respectively.

The same contrast is shown by integrated causal verb-final clauses on the one hand and V1-causals on the other. Again, only the V1-causal yields a reading in which the tag is restricted to the semantically dependent clause.

Thus, in their distribution tags are much more restricted than weak RPs.

Another strong root phenomenon I would like to mention is the German interjection Mann. This interjection cannot appear with a PAC, (35a), but it can appear with a NonIC; cf. (35b) with a V1-concessive, and (35c) with a wobei-clause.

Crucially again, the interjection in (35b,c) can be understood as being related only to the content of the V1-concessive and the wobei-clause, respectively.14

Let us next observe that obwohl-Vf_in-clauses, which standardly appear as PACs, can include a strong root phenomenon if they are separated by a pause from their host clauses and carry their own sentence contour, cf. (36). It follows that obwohl-Vf_in-clauses may occur as NonICs.

Let us now note an important feature of NonICs. NonICs have their own illocutionary force. This can be most clearly seen by the fact that the illocutionary force of a NonIC can be different from the illocutionary force of its host. In (37a,b) the main clause is an imperative, in (37c) it is a (rhetorical) question, and in all examples the dependent clause is assertive. ((37c) is constructed based on an example in Pittner 2011 with a V1-causal.)

Another piece of evidence for the claim that NonICs have independent illocutionary force consists in the very fact that they can host strong RPs. Arguably, strong RPs presuppose that the clauses they are added to have illocutionary force. A tag, for example, demands the confirmation of an assertion. An interjection like Mann leads to a (rather rude) strengthening of the illocutionary force of the host, be it assertive or imperative. Note furthermore that the fact that a NonIC cannot be part of the syntactic structure of its host can be seen as a consequence of the fact that it has its own illocutionary force. There is a tradition in philosophy of language and in linguistics that assumes that an expression which has illocutionary force cannot be a genuine part of the structure of another expression (e.g., Green 2000). I follow this line of thinking. Given these considerations we can immediately draw an important conclusion, which is that PACs do not have independent illocutionary force. As clauses lacking illocutionary force, PACs can be embedded, but they cannot host strong RPs. Note that with the claim that PACs do not have independent illocutionary force we object to the assumption that concessives (and adversatives for that matter) necessarily have illocutionary force (e.g., König 1991), since, as we have seen, obwohl-clauses normally appear as PACs (the same is true for adversative während-clauses (‘while’-clauses); cf., e.g., Frey 2016). Furthermore we take issue with the often made claim that MPs serve as modifiers of a speech act or an illocutionary operator (e.g., Thurmair 1989; Jacobs 1991; Karagjosova 2004; Bayer & Obenauer 2011; Coniglio 2011). According to our findings, this assumption is too strong; the occurrence of a MP does not presuppose that its host has illocutionary force.

Let us restate the main observations of the present chapter: in the standard use concessives may not occur as fully integrated CACs but only as less integrated PACs or as even less integrated NonICs (this is in opposition to conjectures to the contrary in Breindl 2014: 922 and Freywald 2018); in contrast, other adverbial relations like the temporal or the causal one can be realised by a CAC.

Above we already hinted at an intuition as to why it might be the case that a concessive normally cannot be a CAC. In its standard use a concessive necessarily goes beyond a reference to the actual world; it contains a modalised rule, for which a thinking mind assumes responsibility. In the following, a suggestion is made which outlines how this circumstance is reflected in syntax. Furthermore, proposals for the syntactic structures will be made that allow an explanation of some of the differences between concessives realised as PACs and concessives realised as NonICs.

We have seen that evaluative sentence adverbials belong to the weak RPs. It can be shown that evaluative adverbials are not-at-issue expressions. This is demonstrated in (38) (cf., e.g., Krifka 2018 for such a test).

The evaluative sentence adverbial is not part of the target of B’s denial, i.e., B’s denial in (38) is not understood as ‘it is not true that it is unfortunate that it rains’. It rather is understood as ‘it is not true that it rains’; the evaluative adverbial is not in the scope of B’s denial. It can be deduced that the evaluative adverb in (39) has the status of a non-at-issue expression.

(38) should be compared with (39), in which an evaluative predicative occurs.

In (39) B’s denial can be understood as ‘it is not true that it is unfortunate that it rains’. Thus, the evaluative predicative is in the scope of B’s denial, and it can be deduced that the evaluative adjective in (39) is at-issue.

With Krifka (2018) we can draw the important conclusion from these observations that evaluative adverbials, being not-at-issue, are external to the core proposition which is communicated by the clause in which they occur, whereas evaluative predicatives, being at-issue, belong to the core proposition communicated.

We can extend these considerations to MPs, which also belong to the weak RPs. They are also not-at-issue. Let us consider the MP ja.

B’s denial in (40) is not understood as something like ‘it is not true that I know that Maria came to the lecture’, (cf. fn. 9), it rather is understood as ‘it is not true that Maria came to the lecture’. B’s denial is not concerned with any additional nuance or expressive feature that comes along with the MP.

The marking of an aboutness topic is equally not-at-issue.

B’s denial does not encompass A’s marking of an aboutness topic. Building on one popular explication of the concept of an aboutness topic (cf. Reinhart 1981), we can say that B’s statement does not deny that the ingredient of A’s statement which is that the proposition that Hans will assist should be added to the address ‘Hans’. B just denies that Hans will assist. Thus B’s utterance is not concerned with the additional meaning that comes along with topic marking.

The strong RP Mann is also not-at-issue, cf. (42).

The interjection Mann expresses a strong emotional involvement of the speaker regarding her/his statement. B’s contradiction does not concern A’s expression of emotional involvement. Thus, the meaning component of the interjection is not in the scope of B’s dissension. In the same way, if applying the denial test, the result would be that the other examples of strong RPs are not-at-issue.

It seems a reasonable assumption that by means of the denial test it could be shown that weak and strong RPs are generally not-at-issue. To check this is beyond the scope of the present article though. However, we can be sure that our example RPs are not-at-issue and we can, for the time being, assume that this holds for all RPs. In any case, all RPs considered here are external to the core proposition communicated. With Krifka (2018) we assume that the not-at-issueness of RPs is mirrored by the fact that in syntax, RPs are located outside TP. This is a natural assumption for strong and for weak RPs. Regarding weak RPs, for example, according to Cinque (1999), sentence adverbials are positioned outside of TP, and the marking of an aboutness topic always seems to occur outside TP; cf., e.g., Frey (2004), who shows this for German. For German MPs it is also standardly assumed that they are generated above TP; cf. Coniglio (2011). Regarding strong RPs, we can safely assert that they do not occur structurally below weak RPs.

Next, we can observe with Krifka (2018, to appear) that evaluative adverbials are not related to the speech act performed. Their semantic contribution does not lie in a modification or qualification of the speech act. Furthermore, evaluative adverbials may occur in the complement of a propositional attitude predicate (cf. John thinks that it will unfortunately rain), where they become part of the embedded propositional attitude. This too makes clear that they do not qualify a performed speech act.

Thus, the data show that evaluative adverbials are not part of the core proposition communicated, but that they also do not scope over speech acts. Therefore, Krifka concludes that there must be a distinct semantic layer in-between the layer of the core proposition and the layer of the speech act.

In Section 1, on independent grounds we already arrived at the conclusion that weak RPs in general are not related to speech acts. Furthermore, we have just observed that weak RPs do not belong to the communicated proposition denoted by the clause in which they occur. Thus, we can assume that weak RPs in general relate to a distinct layer between the speech act and the communicated proposition (cf. also Frey & Meinunger 2019).

Krifka points to writings of Frege (1918) and Peirce (cf. Peirce 1994; Tuzet 2006) in which such a distinct layer is introduced. Frege (1918) explicitly differentiates between the following aspects involved in an assertion:

Peirce developed similar ideas about the distinction between a thought, a private judgement on the truthfulness of the thought and a public assertion of this judgement.

Krifka (2018) generalises Frege’s and Peirce’s partition to all speech acts. Moreover, Krifka (2018, to appear) subscribes to the commitment view of speech acts. Regarding assertions, the commitment view holds that in asserting a proposition φ a speaker S expresses public responsibility for the truth of φ, backed by social sanctions if φ is false and S has no excuse. To the common ground (CG) shared by S and the hearers it is added that S is publicly committed to the truth of φ. According to the commitment view, the proposition φ itself is just added to the CG as a conversational implicature.

Thus, according to Krifka one has to assume that four distinct layers are involved in a speech act.

Given (44) we can formulate the following theses:

Now, for our concerns, the next step is crucial. We follow Krifka (2018, to appear) in assuming that the layers depicted in (44) are represented in syntax by different functional projections. Thus, in addition to the common assumption that the core proposition is syntactically represented by TP, we assume that the layers of the judgement, the commitment and the speech act are syntactically represented too.

In the syntactic structure, (46) leads to the hierarchy in (47):

We can reasonably assume that the occurrence of the projections in (47) is implicational from left to right, i.e., if a clause structure contains the projection α of (47), it also contains all projections below α. Thus every independent clause which represents a speech act and consequently contains an ActP also exhibits a ComP, a JP and a TP. What about dependent clauses? Let us investigate for some of them which is the highest node of (47) which they contain.

The claims in (48) (to be extended later) are natural consequences of the theses in (45).

Since weak and strong RPs can only occur in some types of dependent clauses, the possible or non-possible occurrence of a given RP in a clause together with (48a) and (48b) can tell us something about the syntactic structure of the clause. All clauses which allow weak RPs exhibit a JP and all clauses disallowing weak RPs don’t. Thus with (47) we can immediately deduce that clauses of the latter type are just TPs. Furthermore, it holds that all clauses which allow strong RPs exhibit an ActP and all clauses disallowing strong RPs don’t.

Does the projection ComP distinguish certain dependent clauses? The answer is yes. There is evidence that the complement clauses of speech predicates have the status of a ComP (cf. Krifka 2018, to appear, Frey to appear), whereas the complement clauses of mental attitude predicates are just JPs. The following German data provide evidence for this distinction (cf. Krifka to appear). There are high modifiers in German which arguably specify the commitment of a speaker to the truth of a proposition, cf. (49). These modifiers will be called ‘commitment modifiers’ here. It seems reasonable to assume that commitment modifiers are licensed by ComP.

Now, it is interesting to observe that commitment modifiers can very well occur in the complements of speech act predicates, (50a), but lead to ungrammatical results if they occur in the complements of mental attitude verbs, cf. (50b).

If we assume that the complement of a speech predicate is a ComP and the complement of a mental attitude verb does not contain a ComP but is just a JP, this pattern follows.

Commitment modifiers are not good in a PAC (cf. Frey to appear); in (51) this is illustrated with an adverbial obwohl-clause.

Thus, PACs do not contain a ComP. However, we know that they contain a JP, since they may host weak RPs.

We have seen that the concessive relation can also be realised via a NonIC. We expect that in this case the concessive may host a commitment adverbial since as a NonIC it is an ActP and an ActP contains a ComP. This expectation is confirmed.

The commitment modifiers lead to the announced extension of the licensing of RPs as stated above in (48). Commitment modifiers belong to the class of the semi-strong RPs (cf. Frey to appear), for which condition (53) holds.

Let us now come to the categorical endowments of CACs, PACs and NonICs. Our observations suggest the assignments in (54i–iii). (54iv–vi) are added in the interest of completeness. An independent clause of course has the status of an ActP.

[Spec,ActP] and Act⁰ constitute the left periphery of independent clauses and NonICs. According to Frey (to appear) all integrated dependent clauses are covered by a CP-shell.15 The CP-shell is transparent for the categorical features delivered by TP, JP and ComP. [Spec,CP] and C⁰ constitute the left periphery of integrated dependent clauses. ActP and CP do not occur together. Thus, ActP and CP are head-first. [Spec,ActP] and, in the case of embedded verb-second clauses (V2-clauses), [Spec,CP] constitute the prefield (according to Frey to appear, in an embedded V2-clause CP covers ComP). They can be the target of long-distance movement. In independent V2-clauses the finite verb targets Act⁰ and in dependent V2-clauses the finite verb is in C⁰.

ComP, JP and TP (and vP and VP of course) are head-final. Their heads host the finite verb in the different verb-final clauses or are the intermediate positions of the finite verb on its way to the left in V2-clauses. ComP, JP and TP are not projections of the left periphery. Thus, proposals which postulate a fine structure of the left periphery with different functional projections, as pioneered by Rizzi (1997), are largely orthogonal to the proposed series of functional projections in (47).

CACs are TPs with a CP-shell; they express a core proposition and do not allow any RPs. PACs express an extended proposition, i.e., they allow weak RPs; they do not allow strong RPs and semi-strong RPs. Thus, they are JPs with a CP-shell. NonICs additionally allow strong RPs and semi-strong RPs. According to our considerations, they have full illocutionary force, i.e., they are ActPs and as such contain a ComP, a JP and a TP.

Obviously by assuming the categorical assignments in (54) we stand in the tradition which postulates different structural sizes of different dependent clauses, especially of different adverbial clauses (e.g., Haegeman 2006; Coniglio 2011; Frey 2016). Note however that (54) is accompanied by the claim that the different syntactic sizes of dependent sentences are anchored directly in conceptual differences. Greater syntactic complexity corresponds to greater semantic complexity.

At this point it seems appropriate to add some notes about the comparison of Krifka’s layers of interpretation proposition, judgement and speech act and Sweetser’s (1990) famous three domains of interpretation content, epistemic and speech act.16 As we will see, these two classifications are not alternatives but stand in an additive relationship.

Since Sweetser (1990) it has been standard practice to distinguish between content related, epistemic and speech act related interpretations of different types of adverbial clauses. For example, a relation like the causal one can in principle be interpreted on three different cognitive levels – in Sweetser’s (1990) terminology these are the content domain, cf. (55a), the epistemic domain, cf. (55b), and the speech act domain, cf. (55c).

A causal relation interpreted in the content domain states that one fact of the outside world causes another fact of the outside world. A causal relation interpreted in the epistemic domain gives the reason why the speaker takes a certain proposition to be true. Finally, a causal relation interpreted in the speech act domain tells the motivation for why the speaker is performing a certain speech act.

Concessive although- or obwohl-clauses may express the three readings as well (cf. Sweetser 1990).

(56a) illustrates the reading in the content domain. Among other things, it expresses that in the speaker’s view it normally holds that being badly prepared has the consequence that one does not pass the exam brilliantly. (56b) illustrates the epistemic reading. Normally the speaker would deduce from the truth of the proposition that someone brilliantly passed the exam the truth of the proposition that she/he was not badly prepared. (56b) expresses that this normal reasoning does not hold for the case at hand. Finally (56c) has an interpretation in the speech act domain. Normally, given the content of the concessive clause, the speaker would not assert the matrix clause.

Sweetser (1990) uses the concept of three domains to describe whether an adverbial clause is used to express a relation to an eventuality of the outside world, to an assumption of the speaker or to a speech act by the speaker. Crucially the three domains do not refer to different objects in a semantic model which are denoted by the adverbials, like Krifka’s propositions, judgements, commitments and speech acts.

Since CACs are of category TP, PACs are of category JP and NonICs are of category ActP, it follows in Krifka’s semantic model that they have as denotations propositions, judgements and speech acts, respectively. Crucially it does not follow that NonICs only have a reading in the speech act domain, and PACs only have a reading in the epistemic domain. In fact in Frey (2016) it was shown on the basis of German adverbial causal clauses that the following correlations regarding Sweetser’s classification hold:

Sweetser’s distinctions refer to the communicative intention of the speaker and not to the semantic type of a relation. For example, Sweetser only interprets a causal relation as epistemic if the speaker wants to substantiate an assumption and not to state the cause of a fact. Note, however, that a causal relation between facts p and q can lead in a natural way to a causal relation between attitudes; a relation in the outside world is presented by the relation that the assumption that p is true leads to the assumption that q is true (see Volodina 2011). This happens when a causal relation in the content domain is expressed by a PAC. Consider in this context German causal clauses with the complementiser da. These cannot appear as CACs (cf., e.g., Frey 2016), but a statement can be made with a da-clause in Sweetser’s content domain (e.g. Maria ist sehr bleich, da sie krank ist. (‘Maria is very pale since she is ill.’)). Here it is the case that the causal relationship realised with the da-clause does not speak directly of a relation between facts but of a relation between assumptions that certain facts are true.

The corresponding considerations apply to Sweetser’s speech-act domain. Here in the causal area Sweetser refers exclusively to the justification of the execution of a speech act. Logically, a causal relationship between speech acts can have facts as content. Then, a speaker asserts a proposition q because she/he asserts the proposition p.

Now to confirm (57), here we first consider causals before we consider concessives. To see that (57i) holds in regard to weil-clauses, it is enough to realise that if a weil-clause is not interpreted in the content domain, then no binding into the clause is possible, (58). This in turn implies that the clause is not a CAC.

To see that (57ii) holds for weil-clauses, we first realise that a weil-clause can be a PAC which is interpreted in the content domain in Sweetser’s sense, cf. (26b), and it can be a PAC with an epistemic reading, cf. (59). That the weil-clauses in (26b) and (59) are PACs is evident because first MPs or sentence adverbials appear, and second the clauses occur in the prefield of their host.

Second, we observe that if a weil-clause has an interpretation in the speech act domain, then it cannot occupy the prefield of a V2-clause, (60a), and it cannot be embedded together with its host, (60b). Both facts imply that a weil-clause with this reading is not a PAC.

To see that (57iii) holds we first notice that a V1-causal may be interpreted in the content domain, (61a), and in the epistemic domain, (61b). As Pittner (2011) observes a V1-causal cannot operate in the speech act domain though.17 The reasons are not clear yet. However, that a causal NonIC may refer in the speech act domain can easily be shown with the help of a non-integrated weil-V_fin-clause preceding a yes-no question, (61c).

Let us next consider concessives. We have already seen that concessives too may have readings in the content domain, the epistemic domain and the speech act domain, and we have seen that in the standard case concessives cannot be CACs. That a concessive which is a PAC can express the first two readings is illustrated in (56a,b). (56c) shows that an obwohl-clause which is a NonIC can express the third reading. (62a) shows that a concessive which is a PAC cannot express this third reading.18 (62b,c) demonstrate that an obwohl-clause appearing as a NonIC preceding a V2-clause may also express the first and the second reading; this could also be shown with the help of a V1-concessive.

In sum, we see that concessive PACs are necessarily interpreted in the content or in the epistemic domain, and that concessive NonICs can be interpreted in the content, in the epistemic or in the speech act domain.

We can now relate the claims in (57) to Krifka’s system. A CAC denotes a proposition and propositions can be used to refer to and connect eventualities of the external world. A PAC denotes a judgement of a thinking subject and judgements can be used to refer to and connect eventualities of the external world, eventualities of possible worlds or inferences of a thinking subject. A NonIC denotes a speech act by a speaker and speech acts can be used to refer to and connect eventualities of the external world, eventualities of possible worlds, inferences of the speaker or speech acts performed by the speaker.

With these considerations in mind we can make sense of the fact that an adverbial clause with an epistemic reading cannot be a CAC; cf. (58a). Any epistemic reading necessarily involves a judgement and a judge. The sentence pair in (63) illustrates this once more.

(63a) can be understood as a simple description of a causal relation existing in the external world. An external cause is given for the external fact that Paul has a certain belief. Thus, the weil-clause in (63a) can be a CAC. This is confirmed by the possibility of binding into such a weil-clause, cf. (64a). The situation is different with (63b). The mind of a thinking subject, i.e., the speaker, concludes that Max is at home on the basis of the light in the study. For syntax this means that the adverbial clause is a modifier of the JP of its host. Furthermore the adverbial clause itself must be a JP. It is a reasonable assumption that the modifying element of a JP has to encode a judgement. For syntax it follows that the adverbial clause in (63b) is a PAC. A PAC does not allow binding into it from its host, cf. (64b).

(48a), repeated here, formulated the licensing condition for weak RPs.

(48a) A weak RP needs syntactic licensing inside its minimal clause by J⁰.

How are the presence of JP and J⁰ syntactically licensed? (65) expresses the obvious assumption. We have seen either that a JP has Com⁰ as the next higher functional projection above it or that a JP is a complement (with a transparent CP-shell) of a mental attitude verb or of a verb of perception.

A locally licenses B iff A minimally c-commands B modulo any adjunctions to B.

For a PAC, (65) has the consequence stated in (66).

Because of the locality requirement, (66) implies (67).

A PAC may also occur in the prefield of an independent V2-clause, which according to our assumptions stated in Section 3 equals [Spec,ActP]. This is for example the case in (1b). The PAC is moved to [Spec,ActP] from its JP-adjoined base position. Furthermore, a PAC may occur in the prefield of an embedded V2-clause.19 In this case the PAC is moved to [Spec,CP] of the CP-shell of the embedded V2-clause from its JP-adjoined base position (cf. Frey to appear).

With (67) we can immediately explain data like (5b). There is no c-command from the base position of the host’s subject into the PAC, cf. also (68a).20 However, given (67), the possibility of binding from a higher clause into a PAC is expected. This expectation is fulfilled, cf. (68b).

(67) also accounts for (6b) and (7b). Regarding (6b), we note that in order for a phrase to be able to provide the nuclear accent for a wide focus reading of a complex construction, the phrase has to be the most deeply embedded. This cannot happen with a constituent of a PAC. Regarding (7b), we observe that a PAC cannot be in the scope of a negation appearing in the matrix clause since the PAC is just too highly base-generated.

Let us next turn to the licensing of concessive ActPs. Above it was already stated that the present paper adopts the assumption that ActPs cannot be syntactically embedded.

We have seen that like independent clauses, NonICs are also ActPs, cf. (54iii). (69) accounts for the fact that NonICs cannot occupy the prefield of a German V2-clause, cf. (30b,d,f,h), and that NonICs cannot occur embedded together with their hosts, cf. (31b,c).

(69) allows us to explain another peculiarity of NonICs. NonICs cannot be coordinated, cf. (70a). Note that, as to be expected, PACs can be coordinated, cf. (70b).

For the syntax of coordination it is often assumed that the coordinating element is the head of the construction, with the right conjunct as sister and the left conjunct as specifier (cf., e.g., Progovac 2003). Thus, the conjuncts are syntactically embedded. But even with alternative approaches to the analysis of coordination, at least the second conjunct is treated as embedded. With (69) it follows that NonICs, being of the category ActP, cannot be coordinated. In contrast, PACs, being of the category JP, can be coordinated.21

How are NonICs licensed? I would like to adopt a concept from Pittner (2011) and consider NonICs as constituting subsidiary speech acts relative to the speech acts performed by their hosts (Freywald 2018 also adopts this concept for non-integrated concessives). Presumably, in syntax this ancillary function of NonICs is mirrored by their being adjoined to their hosts. Since a NonIC constitutes a subsidiary speech act for another speech act, the host of a NonIC has to be ActP. We arrive at the assumption in (71).

Note that concessive main clauses containing the German adverb trotzdem or the English nevertheless constitute not subsidiary speech acts but independent speech acts.

Let us now recall that strong RPs can only occur with independent clauses and NonICs. This led to the formulation of the licensing condition for strong RPs in (48b), repeated here.

(48b) A strong RP gets licensed by being adjoined to an ActP.

It follows that strong RPs are base-generated at the very edges of their host clauses and cannot occur as integral parts of their hosts, for example they cannot occur in a prefield. In (73) this is illustrated with an interactional expression, with an interjection and with a tag.

It is tempting to assume that strong RPs constitute their own speech acts, albeit again subsidiary speech acts relative to the speech acts with which they are associated. This assumption seems to be justified on the grounds of the semantics/pragmatics of strong RPs and on the grounds of their prosodic realisation. Alston (2000: 158) speaks of ‘subsentential acts’. Thus we slightly extend (48b) to (74).

D’Avis (2016) states that the standard account of concessive obwohl-clauses does not capture all uses. He points out that there are usages of the concessive obwohl-clauses which do not involve the presupposition/conventional implicature of a normal course of states of affairs. In these cases, the conditional to which the concessive is in opposition is to be found in the context. In Section 2 it is shown that in this particular use, and only in this one, obwohl-clauses occur as CACs. Standardly used concessives cannot appear as CACs but only as PACs and NonICs, as can be demonstrated by syntactic and prosodic data. The paper argues that the reason why they cannot be realised as CACs is that in their standard use concessives come with a modalised conception of an expected course of events. According to d’Avis (2016) this conception appears as a generalised conversational implicature triggered by the speaker. The present paper argues that if the semantics of a clause α involves such a conception of normality, α does not just refer to the external world as CACs do, but α includes the evaluation and judgement of a thinking mind.

The paper follows Krifka (2018, to appear), who builds on Frege and Peirce, in assuming that distinctions have to be made between a proposition, a judgement about a proposition, a commitment to the truth of a judgment and the public manifestation of a commitment. These distinctions are reflected in syntax by the functional projections TP, JP, ComP and ActP, respectively. It is the JP-projection which represents the judgement of a thinking mind. CACs only contain TP, PACs contain TP and JP, and NonICs contain TP, JP, ComP and ActP. Thus, all PACs encode a judgement, and all NonICs encode a speech act. It can be shown that this has consequences for the internal and external syntax of the different types of concessive clauses.

Standardly used obwohl-concessives, being PACs, are JPs, which are constituents of their matrix clauses, generated in a high position. They are illocutionarily dependent on their hosts. Non-integrated concessives like verb-first concessives, being NonICs, are ActPs, which are not constituents of their matrix clauses. They are illocutionarily independent of their hosts and encode a separate speech act.

Among other things this approach allows us to explain why integrated obwohl-concessive clauses, and PACs in general, may contain so-called weak root phenomena like modal particles and sentence adverbials. These are not-at-issue expressions dependent on judgements. Concessives realised as PACs may not contain commitment modifiers and so-called strong root phenomena like tags and interjections, though. These are not-at-issue expressions dependent on a commitment and the performance of a speech act, respectively. Importantly, strong root phenomena require the presence of an ActP. Therefore, only if the concessive comes as a NonIC like a verb-first concessive or as an independent clause may it host strong root phenomena.