An exploratory, corpus-based study of concessive markers in English, German and Spanish: The distribution of although, obwohl and aunque in the Europarl corpus Volker Gast∗ Abstract This contribution presents a comparative, corpus-based study of three concessive subordinators from English, German and Spanish, i.e., although, obwohl and aunque. It investigates differences in the distribution of these operators on the basis of richly annotated data from the Europarl corpus, taking into consideration structural properties of the relevant clauses (position and length), the type of semantic relation holding between the main clause and the concessive clause, the level of linking (propositional, illocutionary, textual), and information structural parameters (status of the concessive clause, topic-comment structure). The study shows that there are significant differences between German obwohl on the one hand, and English although and Spanish aunque, on the other. German obwohl is more restricted in its distribution and mostly occurs in ‘canonical’ concessives, while English although and Spanish aunque are used in a broader range of contexts, beyond the canonical distributions (e.g. in ‘restrictive’ uses, and at a textual level). This observation is related to the fact that German, unlike English and Spanish, uses a specific type of word order in subordinate clauses, which seems to block distributional extensions, at least in the formal register investigated in the present study (political speech). Keywords: concessivity, adversative, restrictive, rich annotation, information structure, modality ∗ Earlier versions of this paper were present at the Conference on Discourse Relational Devices (Valencia, 24–26 January, 2016) and at the Olinco 2016 conference (Palacký University Olomouc, 9–11 June, 2016). I would like to thank the audiences for valuable feedback. Furthermore, I have greatly benefited from comments made by V. Atayan, D. Hole. E. König and P. Siemund. Any inaccuracies are of course my own responsibility. 1 1 Introduction This study deals with concessive clauses introduced by the three arguably most common concessive subordinators of English, German and Spanish, i.e., although, obwohl and aunque as illustrated in (1a) and its translations into German and Spanish in (1b) and (1c) (all sentence have been taken from the Europarl corpus/version 7, cf. Koehn 2005; Cartoni and Meyer 2012; the identifier at the top indicates the number of the document; see Sect. 3 for remarks on the data). (1) EP-10-05-05-014 (original English, emphasis mine) a. I voted in favour of the Van Dalen report on maritime transport strategy up to 2018, although our amendment on including maritime transport in the ETS was rejected by large majority (roll-call vote). b. Ich habe für den Van Dalen-Bericht über die Seeverkehrsstrategie bis 2018 gestimmt, obwohl unser Änderungsantrag bezüglich der Miteinbeziehung des Seeverkehrs in das Emissionshandelssystem (ETS) von einer großen Mehrheit abgelehnt wurde (namentliche Abstimmung). c. He votado a favor del informe Van Dalen sobre la estrategia de transporte marítimo hasta 2018, aunque se rechazó nuestra enmienda sobre la inclusión del transporte marítimo en el ETS por una amplia mayoría (votación nominal). The focus of the study is on the similarities and differences in the distributions of concessive clauses introduced by although, obwohl and aunque. The following parameters are investigated: • The length and position of the subordinate clauses; 2 • the type of relation holding between the main and subordinate clause (canonical concessive, restrictive concessive, modal concessive, adversative); • the level of linking (propositional, illocutionary, textual); • the information status of the subordinate clause (assertive, non-assertive); • the topic-comment structure of the sentence (no topic contrast, vertical/scalar topic contrast, horizontal/non-scalar topic contrast). The study is exploratory, the underlying hypothesis being that the three markers under investigation differ in terms of their distributions, as they form part of different linguistic systems and have different histories. The parameters listed above have been shown to be relevant to concessivity in empirical investigations (see e.g. Rudolph 1996; CarbonellOlivares 2009; Wiechmann and Kerz 2013 for relevant empirical studies of English, and Gast 2016 for an earlier comparative analysis), and the present contribution sets out to determine to what extent they have an influence on the distribution of (clauses introduced by) although, obwohl and aunque. Even though the focus is on original (non-translated) data, translations into the other languages will be provided as well, sometimes pointing out relevant details. The main results converge in showing that obwohl differs significantly from although and aunque in its distribution. First, obwohl-clauses show a strong tendency to follow the main clause, unlike clauses introduced by although or aunque. This tendency is independent of the other variables investigated in this study. Second, obwohl occurs rarely in concessives with a contrastive topic. Finally, obwohl is more ‘conservative’ than although and aunque in its distribution relative to the functional variables, in the sense that it is rarely used in contexts other than ‘canonical’ concessivity. I relate all of these observations to the fact that German uses a specific (verb-final) 3 word order in subordinate clause, which correlates with the absence of illocutionary force. While English and Spanish seem to use the preposed position of concessive clauses to indicate a non-assertive status of the concession, the position of a concessive clause is therefore less relevant to the question of assertiveness in German, and the late positioning of concessive clauses might reflect their ‘strengthening’ function, as concessives clauses often (though not necessarily) emphasize that a claim holds in spite of unfavourable circumstances, or some “obstacle” (Mann and Thompson 1988, 39), thus representing an ‘argumentative climax’. Verb-final word order may also be responsible for the relative rarity of contrastive topic structures in obwohl-sentences, as such structures are typically ‘symmetrical’ in terms of illocutionary force, e.g. insofar as they hold between two main clauses. The functional ‘conservativeness’ of concessives may be related to verb-final word order because specific types of context extension are blocked – at least in the formal register of political speech as investigated in this study (see Günthner 2000 for a study of colloquial German, where obwohl is distributed quite differently, and where it occurs with verb-second order, too). As verb-final clauses (mostly) do not have illocutionary force, obwohl-clauses are more unlikely to develop restrictive and textual uses, for instance (cf. Sects. 2.1.2 and 2.2). As far as the other subordinators are concerned, aunque shows a significant affinity to restrictive uses, which relativize the main clause and stand in a relationship of ‘pragmatic strength’ to it, thus forming a scale (cf. Sect. 2.1.2). I hypothesize that this tendency might be related to the fact that aunque, unlike although and obwohl, contains a scalar component (aun ‘still, even’). Following this introduction Section 2 provides some basic distinctions and definitions 4 relating to the domain of concessivity. Section 3 contains some remarks on the data and the process of annotation. In Section 4, the data is presented and analyzed in quantitative terms. Section 5 contains a summary and some attempts at explanations. 2 The domain of inquiry: Concessivity 2.1 Semantic relations 2.1.1 The Concessive Presupposition The notion of ‘concessivity’ is generally associated with some type of ‘incompatibility’ or ‘dissonance’ between the concessive clause and the main clause (e.g. König 1985, 1988; Iten 1998, 2000; König and Siemund 2000; Mann and Thompson 1988; Knott and Sanders 1998, among many others). A common view is that concessivity comes with a presupposition saying that the two events said to cooccur in a concessive link do not ‘normally’ cooccur (e.g. König 1988, 147, König 1994, 681). Accordingly, (2) can be analysed as shown in (2a) and (2b). Note that the information status of the concession (assertive or non-assertive) is a non-trivial matter to which we will return below (Sect. 2.3.1). In a more general format, the interpretation of concessive clauses can be represented as in (3) (throughout the paper, I will use c for the concession; m is a placeholder for the [semantic] material in the main clause – cf. Section 2.2). (2) Although it’s raining, I’m going for a walk. a. pres: NORMALLY [it is raining → ¬[Speaker goes for a walk]] b. ass.: (It is raining ∧) Speaker will go for a walk (3) m although c 5 a. pres: NORMALLY [c → ¬m] b. ass.: (c ∧) m While the notion of ‘normality’ seems intuitively rather clear, it is hard to define and moreover introduces a generic presupposition, even for episodic event descriptions (cf. also Kim 2002). An alternative way of defining concessivity is by comparing the conjunction of the main and the concessive clause (m ∧ c) with the conjunction of the main clause, and the negation of the concessive clause (m ∧ ¬c). For example, we can compare the conjunction implied by (2) (cf. (4)) with the conjunction in (5). (4) It’s raining and I am going for a walk. (c ∧ m) (5) It’s not raining and I am going for a walk. (¬c ∧ m) While (4) is implied or even asserted by (2), the more ‘normal’, expected or perhaps more likely situation is the one in (5). We can express the presupposition stated in (2a) in terms of probability as shown in (6) (‘p’ stands for ‘probability’; cf. also Kim 2002 for a probability-based analysis of concessivitiy). (6) p(Spk will go for a walk ∧ it is raining) < p(Spk will go for a walk ∧ ¬[it is raining]) More generally, this presupposition can be formulated as in (7), as a conditional probability. I will call this presupposition the ‘Concessive Presupposition’, abbreviated as ‘ConcPres’, and I assume that it has a ‘pragmatic’ status in the sense of Stalnaker (1974). It is taken to be part of the ‘common ground’ (Stalnaker 2002) by any speaker using a concessive marker, and it says that the probability of m conditional on c is lower than the probability of m conditional on ¬c. 6 (7) Concessive Presupposition ConcPres(m, c) := p(m|c) < p(m|¬c) 2.1.2 Canonical and restrictive concessives As is well known (cf. for instance König 1985; Rudolph 1996; Kortmann 1997), concessive subordinators are rarely monosemous. In this and the following sections we will consider semantic domains that are related to concessivity, but differ in some respects from the ‘canonical’ cases discussed in Sect. 2.1.1 (cf. below for the notion of a ‘canonical’ concessive). In ‘restrictive concessives’ “[t]he state of affairs affirmed in the main clause is restricted in its validity by the state of affairs affirmed in the concessive clause” (Rudolph 1996, 411). A pertinent example from my sample is given in (8). (8) EP-10-03-09-013 (original English) a. We have reserves and, although we will never reach zero accidents, we can do a lot to reduce the number of casualties. b. Wir haben Reserven und wir können Einiges tun, um die Anzahl der Unfallopfer zu reduzieren, obwohl wir niemals Null Unfälle erreichen werden. c. Tenemos pegas y, aunque nunca podremos alcanzar los cero accidentes, podemos hacer mucho para reducir el número de víctimas. In such cases the Concessive Presupposition does actually hold, as is illustrated in (9) for (8). Obviously, the likelihood of reducing the number of casualties is higher when there are no accidents at all – it is, basically, 100% – while the probability of reducing the number of casualties is lower than 100% as long as there are accidents. Note that the propositions are rephrased slightly in (9), as I assume that concessives interact systemat7 ically with the context, specifically insofar as they relate to a Question Under Discussion (Roberts 2012) or Quaestio (Klein and Stutterheim 1987) – cf. below. (9) p(we can reduce the number of casualties | we have accidents) < p(we can reduce the number of casualties | we have no accidents) Even though the Concessive Presupposition does hold in examples like (8), these cases have to be kept apart from concessives of the type of (2), which I will call ‘canonical’. One difference is that the Concessive Presupposition is a contingency in canonical concessives, while it is an implication in restrictive clauses. The expression in (9) does not have to be accessible in the common ground for this sentence to be interpretable, and no relevant world knowledge is required, whereas (2) requires the presupposition of incompatibility, and will trigger accommodation if no such presupposition is available. Related to this first difference is a second one: in restrictive clauses, the negation of c entails m, though potentially at a pragmatic level. Put differently, ¬c is ‘pragmatically stronger’ than m in the sense of Gast and Auwera (2011) – cf. the definition of ‘pragmatic strength’ given in (10) (expressions in small capitals are defined independently by Gast and Auwera 2011). (10) A proposition π is PRAGMATICALLY STRONGER Q) than a proposition ρ iff the (relative to a given quaestio RELEVANT CONTEXTUAL IMPLICATIONS of π (with respect to Q) entail the RELEVANT CONTEXTUAL IMPLICATIONS of ρ (with respect to Q) (Gast and Auwera 2011, 9) Relevant contextual implications are (logical) implications of an utterance U that answer the current Question Under Discussion (Roberts 2012) or Quaestio (Klein and Stutterheim 1987). In example (8), the Quaestio Q is ‘To what extent can we reduce the number of 8 casualties?’. The relevant contextual implication of m (relative to Q) is ‘We will be able to reduce the number of casualties to some extent x’, and this proposition is implied by the negation of c, i.e., ‘We will reduce the number of casualties to zero’. For further illustration, consider the example in (11). The entailment relation characteristic of restrictive concessives is shown in (12) for this example. (11) EP-09-04-21-017 (original English) a. Madam President, like my colleagues I would like to say that, although it is not perfect, the third energy package is a very good basis for developing our common market, especially in gas, and for enhancing gas security. b. Frau Präsidentin! Wie meine Kolleginnen und Kollegen halte ist das dritte Energiepaket, obwohl es nicht perfekt ist, für eine sehr gute Basis zur Entwicklung unseres gemeinsamen Marktes, insbesondere für Erdgas und zur Verstärkung der Gassicherheit. c. Señora Presidenta, al igual que mis colegas me gustaría decir que, aunque no es perfecto, el tercer conjunto de medidas relativo a la energía es una muy buena base para el desarrollo de nuestro mercado común, especialmente del gas, y para reforzar la seguridad del gas. (12) ‘the third energy package is perfect’ (¬c) → ‘the third energy package is very good’ (m) In order to keep restrictive cases of concessivity apart from canonical ones, we have to add a condition to the Concessive Presupposition determining the relationship of semantic strength between ¬c and m. I will use the symbol ‘>PrStr ’ as an operator comparing propositions in terms of pragmatic strength as defined in (10). The definitions for canon9 ical and restrictive concessives are given in (13) and (14). The relevant presuppositions are formulated as special cases of the general Concessive Presupposition. (13) Canonical Concessive Presupposition CnclConcPres(m, c) := ConcPres(m, c) ∧ ¬ [¬c >PrStr m] = p(m|c) < p(m|¬c) ∧ ¬ [¬c >PrStr m] (14) Restrictive Concessive Presupposition RstrConcPres(m, c) := ConcPres(m, c) ∧ [¬c >PrStr m] = p(m|c) < p(m|¬c) ∧ [¬c >PrStr m] 2.1.3 Modal concessives In some cases concessive relations holding between clauses do not seem to be based on real-world probabilities, but on views of what the world should be like. In other words, they imply a component of modality or evaluation. Consider (the made-up example) (15). (15) Nowadays most young couples have sex although they are not married. The Concessive Presupposition does obviously not hold in such cases. It would introduce a contradiction. The claim that most young couples are unmarried and have sex is itself a probabilistic statement. All other things being equal, it implies that for any young couple C, the probability that C is unmarried and has sex is higher than the probability for C to be married and to have sex. In the domain of function words, it is not uncommon for probability (as a functional dimension) to overlap with attitude or evaluation (see for instance Jacobs 1983; König 1991 on scales and evaluation in general; for more recent, and more specific studies, cf. Gast and Rzymski 2015 on evaluative scales interacting with even, and Hole 2015, forthcoming on focus particles of other languages). From a diachronic 10 point of view, evaluative readings are often the result of ‘subjectification’ (Traugott 1989) of probabilistic readings. In order to cover such cases, I define a further concessive presupposition, the Modal Concessive Presupposition, in (16). I use the symbol ‘<Eval ’ to indicate that some state of affairs s1 is more negatively evaluated than some other state of affairs s2 . The evaluation is taken to be part of the common ground and thus has the status of a pragmatic presupposition, just like the (probabilistic) Concessive Presupposition. (16) Modal Concessive Presupposition ModConcPres(c, m) := (m ∧ c) <Eval (m ∧ ¬c) The difference between probabilistic and modal concessivity is not always easy to make, as likelihood and evaluation often go hand in hand. My sample contains very few (relatively) clear cases of modal concessives (four, to be precise). For instance, in (17) the speaker seems to make it clear that (s)he would prefer (m ∧ ¬c) to (m ∧ c). The concession – ‘speaking at a very late hour’ – does not seem to stand in any probabilistic relationship to m – ‘Mrs Hedh and Mrs Bauer came here tonight to hear Mr Barrot and the New Spanish presidency use words like “determined” and “ambitious”’; but the ‘late hour’ seems to be evaluated negatively. (17) EP-10-01-19-010 (original English) a. Mr President, the achievement of Mrs Hedh and Mrs Bauer today, although speaking at a very late hour, as Mrs Hedh said, is to come here tonight and hear the outgoing Commissioner, Mr Barrot, and the new Spanish Presidency use words like “determined” and “ambitious” for the Commissioner. b. Herr Präsident, trotz ihrer Redezeit zu dieser späten Stunde besteht der Erfolg von Frau Hedh und Frau Bauer darin, heute Abend vor dieses Haus getreten 11 zu sein und zu hören, wie der scheidende Kommissar Herr Barrot und der neue spanische Ratsvorsitz Ausdrücke wie “entschlossen” und für den Kommissar “ehrgeizig” nutzen. c. Señor Presidente, el logro de la señora Hedh y la señora Bauer hoy, aunque a estas horas de la noche, como dijo la señora Hedh, ha sido venir aquí esta noche y escuchar al Comisario saliente, el señor Barrot, y a la nueva Presidencia española utilizar palabras como “decidido” y “ambicioso” para el Comisario. In other examples, such as (18), a decision whether a concessive is (primarily) motivated by likelihood or by evaluation is hard to make. The fact that the speaker uses a quantifier viele/many/muchas shows that at least the lack of protection mechanisms on websites to which users are linked is not considered unlikely (though certainly undesirable). (18) EP-10-03-08 (original German) a. For example, on many websites, they are seduced into booking further offers through additional links, although these offers are not subject to the same protection mechanisms as the original ones. b. Beispielsweise werden sie heute auf vielen Internet-Plattformen über weiterführende Links dazu verführt, weitere Angebote zu buchen, obwohl diese Angebote dann nicht den gleichen Schutzmechanismen unterliegen wie die ersten. c. Por ejemplo, en muchas páginas web se seduce a los consumidores con ofertas adicionales a través de enlaces diferentes, aunque estas ofertas no estén sujetas a los mismos mecanismos de protección que las primeras. 12 As far as my coding decisions are concerned (cf. Sect. 3), I was conservative insofar as I treated examples only as instances of modal concessivity when a probabilistic reading was very unlikely. As there were only four examples that were classified as ‘modal’, this particular type does not have a significant influence on the results. 2.1.4 Adversatives Another important case that needs to be distinguished from canonical concessivity is adversativity (cf. for instance König 1985; Rudolph 1996; Carbonell-Olivares 2009, among many others). While concessive clauses establish a relationship between the concessive clause (c) and the main clause (m), adversative clauses make reference to a third element, the ‘conclusion’ (König 1985; Iten 2000; Carbonell-Olivares 2009). As I am using c for the concession, I will use σ as a placeholder for conclusions (cf. the Greek word sympérasma ‘conclusion’). König (1985) and Carbonell-Olivares (2009) discuss the example in (19). (19) Although has he is a bit short of breath, he has long legs. A sentence like (19) is only felicitous if both sentences refer to the same Quaestio, e.g. ‘How good a runner is he?’ The two components of (19) – ‘He has long legs’ (m) and ‘He is a bit short of breath’ (c) – lead to conflicting conclusions, cf. (20). In the case of (19), the main clause – occupying the last position – “carries more weight” (König 1985, 6) and triggers the relevant contextual implication – the conclusion – that ‘He is a good runner’. In sentences with but, arguably the most typical adversative connector in English, it is the but-clause that is more important (cf. also Carbonell-Olivares 2009). (20) a. he has long legs → he is a good runner 13 b. he is a bit short of breath → he is not a good runner In a more general format, the structure of (19) can be represented as shown in (24). (21) a. m → σ b. c → ¬σ An example from my data where an adversative relation is expressed with the three subordinators under investigation is given in (22). (22) EP-97-05-13 (original English) a. Although radiological diagnosis provides undoubted benefits for patients, it entails risks, and for this reason we have to try to reduce the use of medical radiation where at all possible by employing alternative methods. b. Obwohl die Röntgendiagnose Patienten zweifellos Vorteile bietet, birgt sie Risiken in sich, und wir sollten deshalb versuchen, den Einsatz von Strahlen in der Medizin nach Möglichkeit zu verringern und durch alternative Methoden zu ersetzen. c. Aunque el diagnóstico radiológico proporciona beneficios indudables para los pacientes, también entraña riesgos, y por esa razón tenemos que esforzarnos por reducir la utilización de las radiaciones con fines médicos en todo lo posible, mediante el empleo de métodos alternativos. In the case of (22), the conclusion is given explicitly in the context – the use of radiological diagnosis should be reduced. The structure of this example is shown in (23). (23) a. radiological diagnosis provides undoubted benefits (c) → it should be used (σ ) 14 b. radiological diagnosis entails risks (m) → it should not be used (¬σ ) In terms of the (probabilistic) approach taken in this study, adversatives can be characterized as follows: The conjunction of m and σ is more likely than the conjunction of c and σ ; put differently, the probability of σ conditional on m is higher than the probability of σ conditional on c, cf. (24). (24) p(radiological diagnosis should be used | radiological diagnosis entails risks) < p(radiological diagnosis should be used | radiological diagnosis provides benefits) The Adversative Presupposition ‘AdvPres’ can be defined as shown in (25). (25) Adversative Presupposition AdvPres(m, c, σ ) := p(σ |c) < p(σ |m) As is well known, adversative clauses often emerge as diachronic extensions of canonical concessives. One way of modeling this type of development is by regarding it as a metonymy of the type ‘reason for conclusion’ (similar to ‘cause for effect’). The sentence in (26a) contains a canonical concessive, and the reason r provides additional information within the main clause. The main clause m can be deduced from the reason r, especially when sufficient contextual information is given (cf. (26b)). (26b) can then be reanalysed as (26c). (26) a. Although radiological analysis provides benefits (c), its use should be reduced (m), because it entails risks (r). b. Although radiological analysis provides benefits (c), (its use should be reduced [m], because) it entails risks (r). 15 c. Although radiological analysis provides benefits (c), it entails risks (m) (and should therefore not be used [σ ]). 2.2 Levels of linking Concessive relations, like conditional and causal relations, may hold at different levels of interpretation. Based on Sweetser’s (1990) three-way distinction made for conditional and causal clauses, Crevels (2000b, 2000a) has argued that four levels of linking can be distinguished for concessives, cf. the examples in (27)–(30) (from Crevels 2000b, 317): (27) Content level Although it’s raining, we’re going for a walk. (28) Epistemic level He’s not at home, although his car is parked in front of his house. (29) Illocutionary level Even though I am calling a bit late, what are your plans for this evening? (30) Textual level I speak and write Serbian, Albanian, Turkish and Dutch, but I cannot express my true feelings in any other language than Romani. Although, now that I come to think of it, I have done it many times . . . I follow Crevels (2000b) in assuming that the first three levels differ in the linguistic type of m (remember that m is a placeholder for the semantic material in the main clause, cf. Sect. 2). It is invariably treated as an expression of type t, which relates to (i) the instantiation of some state of affairs (HAPPEN(e)), (ii) the truth of a propositions π , or 16 (iii) the performance of a speech act Σ (by some speaker Spk). We can thus distinguish these cases by varying the first argument of the Concessive Presupposition, cf. (31)–(33). (31) ConcPrep(HAPPEN(e),c) (32) ConcPrep(π ,c) (33) ConcPrep(PERFORM(Spk,Σ),c) Concessivity at the textual level is arguably a bit different (cf. also Günthner 2000 on ‘corrective’ uses of obwohl). Note first that the Concessive Presupposition does hold – see (34) for example (30). (34) a. m: ‘I can only express my true feelings in Romani’ b. c: ‘I have often expressed my true feelings in other languages’ c. p(I can only express my true feelings in Romani | I have often expressed my true feelings in other languages) < p(I can only express my true feelings in Romani | I have not expressed my true feelings in other languages (very) often) The characteristic feature of textual concessives is that m is relativized or even abandoned, and replaced by c. By contrast, a canonical concession strengthens the claim made in m, which is maintained in spite of unfavourable circumstances (cf. Note 1). In practice, the distinctions between the four levels are obviously not easy to make. Is is questionable whether the two ‘lowest’ levels can be distinguished at all. Concession always implies epistemic activity (reasoning), and often evaluation (cf. Sect. 2). I will therefore subsume the content level and the epistemic level under the category ‘propositional’. Concessives at the illocutionary levels are relatively easy to identify when the 17 illocution in question is not representative in terms of Searle (1975) (cf. the question in (29)). Unsurprisingly, such examples are relatively rare in the Europarl corpus, a corpus of political (and often scripted) speech. There are some instances of this type in my sample, however – cf. the English example in (35a) (note the aber in the German translation in (35b)), and the German example in (36b) (note that in the Spanish version in (36c), the scope of the question operator is indicated by a pair of question marks). (35) EP-97-11-19 (original English) a. Although we have article 5 in the Lomé Convention, what will you do specifically to enact it in response to the countries that I have referred to in the question? b. Wir haben zwar den Artikel 5 des Lomé-Abkommens, aber was werden Sie speziell unternehmen, um diesen Artikel in den Ländern, auf die ich mich in meiner Frage bezogen habe, anzuwenden? c. Aunque contamos con el artículo 5 del Convenio de Lomé, ¿qué va a hacer usted concretamente para aplicarlo en respuesta a los países a que me he referido en la pregunta? (36) EP-09-05-05 (original German) a. Although there is room for improvement, how would the future look without our European social model? b. Wie sähe die Gegenwart ohne unser europäisches Sozialmodell aus – obwohl es verbesserungsfähig ist? c. Aunque podría mejorarse, ¿cómo se plantearía el futuro sin un modelo social europeo? 18 Instances of speech-act level concessives modifying a representative speech act involve an incompatibility between the proposition expressed in the concessive clause, and the fact that the utterance expressed in the main clause is made. In German examples of this type, the two clauses are sometimes syntactically separated, as in (37b), where the main clause contains a Forefiled constituent of its own. The concessive relation can here be reconstructed at the propositional level by embedding the main clause under a predicate of speech or attitude, e.g. I think . . . . The same situation can be observed in the Spanish example (38c). (37) EP-03-03-27 (original German) a. Although in principle I would be the last person to advocate a further transfer of powers to the EU, in this exceptional case it is justified . . . b. Obwohl ich grundsätzlich die Letzte bin, die einer weiteren Aufgabenübertragung an die EU das Wort spricht, in diesem Ausnahmefall ist sie gerechtfertigt, . . . c. Si bien en principio yo sería la última persona en defender una mayor transferencia de poderes a la UE, en este caso excepcional está justificado . . . (38) EP-03-06-18 (original Spanish) a. The situation in European seas is not good even though it has emerged from monitoring and assessment programmes that information on the marine environment is insufficient. b. Der Zustand der europäischen Meere ist nicht gut, auch wenn die Bewertungsund Kontrollprogramme zur Meeresumwelt nur unzureichende Informationen geben. 19 c. La situación de los mares europeos no es buena, aunque los programas de evaluación y control revelan insuficiente información sobre ese medio ambiente marino. Examples located at the textual level as defined above – uses of concessive clauses which are not intended to strengthen another argument or claim, but establish a claim of their own, relativizing and sometimes correcting the preceding argument – are, obviously, also hard to identify. One useful diagnostic is that the concession functions as the topic or point of reference for the following sentence. This is not normally the case with concessives, which canonically function as ‘satellites’ in terms of Rhetorical Structure Theory (Mann and Thompson 1988), rather than ‘nuclei’. Consider the examples of textual concessives in (39). The concession (I have a word of advice for the commissioner) functions as the point of departure for the following sentence. Note that the Spanish translation does not contain a concessive marker at all. The German translation contains the conjunctional adverb allerdings. Given their status as ‘nuclei’, textual concessives cannot normally precede the main clause, as this would lead to incoherent text structure. (39) EP-97-06-10 (original English) a. By and large, I welcome Mrs Lulling’s amendments, although I have a word of advice for the Commissioner. If he intends to reject any of these amendments, I would suggest that he put on his protective clothing. b. Frau Lullings Änderungsanträge begrüße ich im großen und Ganzen, allerdings möchte ich dem Herrn Kommissar noch einen guten Rat geben. c. En general me satisfacen las enmiendas presentadas por la Sra . Lulling, y antes de acabar quisiera prevenir al Comisario: . . . 20 As far as coding decisions are concerned (cf. Sect. 3), in case of uncertainty I gave preference to the lower level, relative to the hierarchy ‘propositional < illocutionary < textual’. 2.3 The information structure of concessive sentences There are two important parameters concerning the information structure of concessive sentences, the information status of the concession (Sect. 2.3.1), and the topic-comment structure of the whole clause (Sect. 2.3.2). 2.3.1 The information status of the concession The concession in ‘real’ (Rudolph 1996) or ‘factual’ (König and Siemund 2000) concessive clauses – as opposed to concessive conditionals, cf. below – are sometimes said to be presupposed (e.g. König and Eisenberg 1984, 320). They do not seem to have the same status as (specific types of) causal clauses, however, as they are not immune to negation, or (necessarily) preserved in conditionals, for instance. One problem in determining their information status is that concessive clauses cannot be in focus (cf. König 1994; Pittner 1999; Breindl 2004), so that presupposition tests are hard to carry out, as they often rely on focus-sensitive constructions, e.g. negation. Compare the causal clause in (40) with the concessive clause in (41). (40) a. [NEG [F I didn’t stay at home yesterday because it was raining]]. b. [NEG I didn’t stay at home yesterday [F because it was raining]]. (41) a. #[NEG [F I didn’t go for a walk yesterday although it was raining]]. b. *[NEG I didn’t go for a walk yesterday [F although it was raining]]. 21 The causal clauses in (40) both seem to imply that it was is raining on the previous day: (i) with broad focus (‘I didn’t stay at home . . . ’, cf. (40a)), and with a focus on the causal clause (‘I stayed at home for another reason’, cf. (40b); note that the scope of negation is wide in both cases). A similar test for although is pragmatically awkward, cf. (41). The reading with the concessive clause in focus, corresponding to reading (ii) of the causal clause, is not available, as the concessive clause cannot be focused (cf. (41b)). Option (i) is possible, but it would be redundant to even mention the concession under negation, and (41a) has the flavour of an echoic utterance. A concession has the function of strengthening the claim made in the main clause by pointing out that it is even true in spite of unfavourable circumstances or ‘obstacles’.1 If the main clause is negated, the concessive clause loses its textual function.2 Given the unnaturalness of (41b), we cannot use it as a diagnostic for a presuppositional status. When embedded within a conditional clause, a concessive clause does not seem to be implied, cf. (42). This seems to speak against its having a status as a semantic presupposition. However, another way of looking at such examples is that the concessive ‘inherits’ the non-veridicality (cf. Zwarts 1995) of the conditional clause. While (42) does not imply that it is raining, it does make a statement about the co-occurrence of rain and the addressee’s walking in any given world, i.e., the rain becomes factive as soon as the walking is instantiated. In a way, the facticity of the concession is a function of the facticity of the superordinate clause. (42) If you go for a walk although it’s raining, you might catch a cold. 1 Cf. for instance Mann and Thompson (1988, 39): “The speaker acknowledges the apparently contrary information, but then advances the nucleus anyway, showing that s/he does not regard the two as genuinely incompatible. This tends to remove the satellite as an obstacle toward favoring the nucleus.” 2 Note that one way fo rescuing (41) is obviously by accommodating the relevant presupposition, but that does not help us to determine the information status of the concession. 22 From a pragmatic point of view, concessive clauses often provide information that is known to the addressee, as in (43), which is stated at the beginning of the talk – they are pragmatically presupposed (Stalnaker 1974). The audience will probably have been aware that the speaker was replacing Günther Oettinger at the time the speech started. Concessive clauses may introduce new information as well, however, as in (44), where the speaker expresses his own views on the matter under discussion. (43) Concessive clause contains given information EP-10-03-11 (original English) a. Madam President, although I am replacing my colleague Günther Oettinger today, it is a real pleasure to discuss with you the future of low-carbon technologies. b. Frau Präsidentin, obwohl ich heute meinen Kollegen Günther Oettinger vertrete, freue ich wirklich sehr, mit Ihnen die Zukunft der kohlenstoffemissionsarmen Technologien zu erörtern . c. Señora Presidenta, aunque hoy ocupo el lugar de mi compañero Günther Oettinger, es para mí un verdadero placer debatir con ustedes el futuro de las tecnologías con baja emisión de carbono . (44) Concession contains new information EP-97-04-09 (original English) a. In that sense I welcome the statement of the Commission, although I am concerned about the comments of the Council. b. In diesem Sinne begrüße ich die Erklärung der Kommission, obwohl ich über die Mitteilungen des Rates besorgt bin. 23 c. En ese sentido, acojo con satisfacción la declaración de la Comisión, aunque me preocupan los comentarios del Consejo. The information status of concessive clauses has been claimed to vary with the position of the relevant clauses relative to the main clause (see for instance Rudolph 1996, 50f.).3 Initial concessives tend to be presupposed, whereas final concessives often have illocutionary force (cf. (43) vs. (44)). This is only a tendency, however, and it is not hard to find counterexamples. (45) is an example of a preposed concessive clause that (probably) contains information that is not known to the audience, and the postposed concessive clause in (46) (probably) contains information that is given. (45) EP-10-04-20-010 (original English) a. So, although I do not have any illusions about the nature of politics in this country and, indeed, in the region, I think that we have to try and give this government a chance to form itself properly, to agree to do the political and constitutional reforms, which are going to be so essential, to hold the elections it says it will have . . . b. Ich denke also, obwohl ich keinerlei Illusionen über die Art der Politik in diesem Land und natürlich in dieser Region habe, dass wir es versuchen müssen und dieser Regierung eine Chance zur eigenen vernünftigen Regierungsbildung, zum Einverständnis der Durchführung politischer und konstitutioneller Reformen, die so unverzichtbar sein werden, und zur Abhaltung der versprochenen Wahlen geben müssen, . . . c. Por tanto, aunque no me hago ilusiones sobre la naturaleza de la política en 3 See also Wiechmann and Kerz (2013) for a study of the position of concessive clauses that takes other (formal) parameters into account. 24 este país y, desde luego, en la región, creo que tenemos que intentar ofrecerle una oportunidad a este gobierno para que se constituya correctamente, para que acepte hacer reformas políticas y constitucionales, que serán tan importantes, para que celebre las elecciones que dice van a tener . . . (46) EP-09-05-05-003 (original English) a. It is an important step towards enhanced security and privacy protection, although at this stage it remains limited to the electronic communications sector. b. Das ist ein wichtiger Schritt hin zu einem verbesserten Sicherheits- und Datenschutz, obwohl er in dieser Phase nur auf den elektronischen Kommunikationssektor beschränkt bleibt. c. Se trata de un importante paso para aumentar la seguridad y la protección contra la piratería, si bien en esta fase se limita al sector de las comunicaciones electrónicas. While factive concessives commit the speaker to the truth of the proposition expressed in the concessive clause (potentially as a function of the truth of some other proposition, cf. the discussion of (42) above), concessive conditionals explicitly introduce hypothetical concessions (cf. König 1986). The difference between factual concessives and concessive conditionals can be seen in (47) (suggesting that John is drunk at the moment of speaking), and (48) (not containing an indication to that effect). (47) Although John is drunk, he is not impolite. (48) Even if John is drunk, he is not impolite. As has been pointed out, the present study focuses on ‘factual’ concessives. The sample of 3×100 examples was therefore created in such a way that only ‘factual’ subordinators 25 were extracted, i.e., although, obwohl and aunque. For Spanish, the sample moreover contains examples in the indicative mood only, as the mood of the main verb indicates the difference between factual and hypothetical concessives in this language. 2.3.2 The topic-comment structure of concessive sentences Most semantic analyses of concessives have focused on the type of presupposition (or conventional implicature) introduced by these elements (cf. Sect. 2.1.1), and the level of linking (Sect. 2.2). One of the questions that has received much less attention is the (internal) information structure of concessive sentences, and the paradigmatic relationship holding between the concession and the main clause. In most of the ‘standard examples’ discussed in the literature on concessivity, the concession and the claim do not share any material. In this, most simple, case, both the concession and the main clause are thetic. In the following, I will represent information structure with bracketing, as in (49) – ‘Com’ stands for ‘comment’ and ‘Top’ for ‘topic’. In addition, I will use a two-dimensional representation, where contrasting elements are vertically arranged, with a subscript indicating whether the relevant material is topical or part of the comment (cf. (50)). In this format, the concession is always at the top. (49) (50) [Com-1 It’s raining] although [Com-2 the sun is shining]. the sun is shining it’s raining Com The components of a concessive sentence can also share a topic, as in (51)/(52). (51) JohnTop [Com-1 went for a walk] although heTop [Com-2 was seriously ill]. 26 (52) JohnTop was seriously ill went for a walk Com Finally, concessive sentences may comprise contrastive topic structures (Büring 2003; Gast 2010), cf. (53). (53) Although JohnTop-1 [Com-1 went to the party], MaryTop-2 [Com-2 stayed at home]. (54) John went to the party Mary stayed at home Top Com As the inspection of the data shows, in some cases it makes moreover sense to distinguish between an ‘external’ topic, functioning as a point of reference (an ‘address’, in terms of Jacobs 2001) for the whole concessive clause, and ‘internal’ (sub-)topics in the concession and the main clause. Consider the examples in (55). (55) EP-10-03-25-003 (original English) a. Although inflation was far above the ECB’s self-imposed ceiling when it peaked at 4% in June and July 2008, inflation rates have since tumbled. b. Obwohl die Inflation viel höher als die selbst gesetzte Obergrenze der EZB lag, als sie Spitzenwerte von 4% im Juni und Juli 2008 erreichte, sind die Inflationsraten seitdem gefallen. c. Aunque la inflación estaba muy por encima del tope que se impuso a sí mismo el BCE cuando alcanzó un máximo del 4% en junio y julio de 2008, las tasas de inflación se han desplomado desde entonces. 27 The sentences are about inflation, the topic. But the individual clauses provide information about inflation at different times, “when it peaked at 4% in June and July 2008”, and “since” that time. This structure can be represented as is shown in (56). (56) InflationTop in summer 2008 was too high since summer 2008 has tumbled Sub−Top Com In order to compare the concessive subordinators under analysis, we can start by identifying two major information structural configurations, those with an ‘internal topic contrast’, and those without such a contrast. Made-up examples of the latter type were given in (49)–(52) above. An English example from my sample with its German and Spanish translations is given in (57). (57) EP-10-01-19 (original English) a. Although [Com-1 the hour is late], [Com-2 we now need the words to be turned into action] . . . the hour is late we need the words to be turned into action Com b. Obwohl es spät ist, müssen unseren Worten nun Taten folgen. c. Aunque sea a estas horas, necesitamos que las palabras se transformen en acción, y hoy las autoras han hecho un buen trabajo para nosotros. (58) is an example of internal topic contrast. (58) EP-97-05-13 (original English) 28 a. So I am astonished that some people were surprised that later on we said that for the automotive industry itself and not components as such, [Top-1 operational aid] [Com-1 can no longer be given] although [Top-2 investment aid] [Com-2 can ] can be given operational aid can no longer be given investment aid Comm Top b. Es erstaunt mich daher, daß einige überrascht waren, als wir später sagten, für die Automobilindustrie selbst und nicht für Komponenten an sich könnten keine Betriebsbeihilfen mehr, wohl aber Investitionsbeihilfen gezahlt werden . c. . . . y por eso me extraña que algunas personas se sorprendieran cuando dijimos que en el caso de la propia industria de vehículos de motor, pero no en el caso de los componentes para dicha industria, no se podría prestar ya ayuda operacional aunque sí que se podría prestar ayuda en forma de inversiones. A further distinction can be made by looking at the specific types of contrast holding between the internal topics of a conditional sentence. In establishing a distinction between canonical and restrictive concessives (Sect. 2.1.2), we already made use of paradigmatic relationships holding between propositions. This distinction can be refined further. I will distinguish between two types of contrast, ‘horizontal’ and ‘vertical’. Another way of thinking of this distinction is between non-scalar (horizontal) and scalar (vertical) contrast. Horizontal contrast holds between elements that are not hierarchically ordered, e.g. heteronyms – cf. (59). (59) EP-97-10-22 (original English) 29 a. Although [Top-1 Latin American countries] would be the [Com-1 most indebted] in global amounts, it is the [Top-1 sub-Saharam African countries] that are [Com-2 in most difficulty]. sub-Saharan countries Latin American countries most indebted in most difficulty Top Comm b. Obwohl in Zahlen ausgedrückt die Verschuldung der lateinamerikanischen Länder weltweit am höchsten ist, haben die schwarzafrikanischen Länder die größten Schuldenprobleme. c. Aunque los países latinoamericanos serían los más endeudados en cantidades globales, los países del África subsahariana son los que se encuentran con mayores dificultades. In a special case of ‘horizontal’ contrast, one internal topic is a function of the other, cf. (60). (60) EP-97-10-20 (original English) a. Although [Com-1 we support and are very appreciative of] [Top-1 the work that Mr Linkohr has done on this], it is clear [Com-2 a lot of disappointment exists about] [Top-2 the eventual outcome of the process]. we support [work done by Mr Linkohr]i outcome of ti is disappointing Top Comm b. Wir unterstützen und schätzen die Arbeit , die Herr Linkohr hier geleistet hat , aber dennoch sind viele von uns auch über das letztendlich erzielte Ergebnis enttäuscht. 30 c. Aunque apoyamos y agradecemos mucho la labor que ha llevado a cabo el Sr. Linkhort [sic] al respecto, es evidente que existe mucha decepción sobre el posible resultado del proceso. Vertical contrast holds between two elements if one of the elements is a special case of the other. In one case, the concessive clause provides information about a more general entity, and the main clause about a special case of that general entity. A relevant example is given in (61) (note that the German version contains the concessive subordinator auch wenn). (61) EP-97-05-13 a. . . . although [Sub-Top-1 the general regime] [Com-1 allows] [Top this kind of compensation for transport], [Com-2 it is not allowed for] [Sub-Top-1 sectors such as steel and the automotive industry]. [this kind of compensation for transport]Top allows Top the general regime Top is not allowed for steel and the automotive industry Sub−Top Comm b. Das gleiche gilt für die Automobilindustrie, und aus diesem Grunde sind solche Ausgleichszahlungen für Transport in Sektoren, wie der Stahl- und der Automobilindustrie, nicht gestattet, auch wenn sie generell zulässig sind. c. Lo mismo cabe decir de la industria de los vehículos de motor y, por lo tanto, aunque el régimen general tolere este tipo de indemnización por el transporte, no se permite cuando se trata de sectores como la industria siderúrgica o la industria de los vehículos de motor. 31 Alternatively, the concessive clause may contain information about the more specific entity. The semantic or pragmatic effect is sometimes almost indistinguishable semantically and concerns the rhetorical text structure. The main difference between a ‘downward contrast’ as in (61), and an ‘upward contrast’ as in (62), consists in what functions as the main claim, and as the anchor for the following discourse. (62) EP-97-09-16 a. I am glad that [Top-1 Parliament as a whole] [Com-1 voted against this proposal], although I note that [Top-2 British Labour MEPs] [Com-2 voted for it]. voted against this proposal Parliament as a whole voted for it British Labour MEPs Comm Top b. Ich freue mich auch, daß das Parlament insgesamt gegen diesen Verschlag gestimmt hat, auch wenn ich feststelle, daß britische Labour-MEP dafür stimmten. c. Me alegro de que el Parlamento en conjunto haya votado contra esa propuesta, aunque veo que los diputados laboristas británicos al PE han votado a favor de ella. Having established the most important parameters of variation, some distinctions and definitions, we can now determine the distribution of these parameters in the data used for the present study. Before the results are presented in Sect. 4, some remarks on the data and annotation process will be made in Sect. 3. 32 3 Notes on the data and annotation I extracted a random sample of 100 (original) examples for each of the concessive subordinators although, aunque and obwohl from the Europarl corpus (version 7; cf. Koehn 2005; Cartoni and Meyer 2012). The data are available online.4 For aunque, I only used examples with indicative mood, as aunque interacts systematically with sentence mood in its interpretation (the subjuntivo mood leads to a hypothetical interpretation of the concessive clause, cf. Sect. 2.3.1). The examples were parsed with the Stanford PCFG-parser (Klein and Manning 2003) and imported into the annotation software GraphAnno (Gast et al. 2015b, 2015a; Gast et al. 2016),5 using the Python interface to GraphAnno, GraphPynt.6 The parses were partially modified manually for further processing, primarily in order to ensure that the data from the three languages are comparable. The syntactic information was not used for the present analysis, however, and the main purpose of the syntactic pre-processing was to provide a structural presentation to which manual annotations could be added. The nodes corresponding to the concessive clause and the main clause were identified manually and subsumed under a common node of category ‘CONC’, assigned to the functional level (in GraphAnno, levels are differentiated by colours). The elements of the concessive sentence were categorized as ‘ARG-1’ (the concession) and ‘ARG-2’ (the main clause). The material dominated by ARG-1 and ARG-2 was subsequently assigned to a layer of its own. A ‘basic’ sentence structure, ready to be annotated further, is shown in Figure 1 (the different levels or arguments are differentiated by colours representing different annotation levels; cf. Gast et al. 2015b, 2015a; Gast et al. 2016). 4 URL: http://www.uni-jena.de/∼mu65qev/data/index.html also the GraphAnno, documentation, https://github.com/LBierkandt/graph-anno/tree/master/doc. 6 https://github.com/VolkerGast/GraphPynt 5 See 33 Figure 1: A pre-annotated sentence in GraphAnno Functional annotations are represented as properties of the corresponding nodes. For example, properties of the concessive clause (such as its information status, cf. Sect. 2.3.1) were assigned to the ARG-1-node (i.e., the functional node dominating the syntactic node corresponding to the concessive clause). The annotations were exported using the native export function of GraphAnno, and analyzed with R (R Core Team 2017). The data was annotated by a single annotator, the author. I am fully aware that this compromises objectivity, but within the exploratory approach taken in the present study a comprehensive annotation procedure with various annotators – implying the establishment of annotation guidelines, and several annotation-validation cycles – would not have been feasible. The annotation decisions are subtle and require a solid background in matters of semantics and pragmatics, i.e., expert annotators. At an initial stage, group annotations tend to introduce inconsistency into the data, and it takes some time to render 34 the annotations consistent.7 While single-annotator annotations are certainly less objective, they therefore tend to be more consistent than (inchoate) group annotations. The data is available online (cf. Note 4) and readers should feel free to take a look at them (and, if they like, reannotate them). The extracted data contain not only the annotation decisions for the variables investigated in the present study, but also ‘auxiliary annotations’ such as the two arguments of the concessive link and the topics as well as sub-topics in each case, which is intended to make the annotation decisions more transparent. 4 Quantitative analyses In this section the quantitative analyses based on manual annotations will be presented, focusing on the following variables: structural properties of the concessive clause (length and position, Sect. 4.1), the semantic relation holding between m and c (Sect. 4.2), the level of interpretation (Sect. 4.3), the information status of the concessive clause (Sect. 4.4), and their topic-comment structure (Sect. 4.5). 4.1 Structural properties of concessive clauses The two most important structural parameters that are potentially of interest are the position of a concessive clause relative to the main clause, and the length of the concessive clause (cf. also Wiechmann and Kerz 2013). The frequencies of preposed, embedded and postposed concessives are shown in Figure 2 in the form of a barchart (left) and a mosaic plot (right, cf. Friendly 1994). All barcharts shown in this contribution were generated with the R-package ‘lattice’ (Sarkar 2008). The mosaic plots were generated with the 7 As I have witnessed myself in two group annotation projects documented in Atayan et al. (forthcoming) and Gast et al. (forthcoming). 35 ‘vcd’-package for R (Meyer et al. 2016). Each rectangle in a cell of a mosaic plot corresponds to a feature combination, e.g. ‘although/pre(posed)’ in the top-left corner of the plot, and the size of each rectangle is proportional to the numer of attested cases. The two plots show the same information, but barcharts merely display absolute frequencies while mosaic plots, in addition to absolute frequencies (reflected in the size of a rectangle), show relative frequncies for each row (the width of the whole plot obviously corresponding to 100%). Moreover, mosaic plots contain an indication of the relative over- or underrepresentation of a cell: cells that are significantly overrepresented are blue, underrepresented cells are red. A cell is assumed to be significantly over- or underrepresented if the standardized residual is higher or lower than 2 (cf. also the scale at the right margin of the diagram). The diagram moreover shows the p-value for the entire distribution (based on a χ 2 -test) in the bottom right corner. Figure 2: Barchart and mosaic plot for the position of concessive clauses by marker The diagrams in Figure 2 show a clear difference between although and obwohl, with aunque occupying an intermediate position: While although-sentences show a tendency to be preposed to the main clause, obwohl-clauses are significantly underrepresented in 36 this position. Moreover, obwohl-clauses are considerably more frequently embedded within the main clause than although- or aunque-clauses. The position of a concessive clause is known to correlate with its length (cf. Wiechmann and Kerz 2013). It is possible that the bias shown in Figure 2 reflects differential sentence lengths associated with the operators. The data do not confirm this suspicion. An Anova shows no significant association between markers and sentence length (F(2,297) = 1.5, p = 0.225). The three markers do seem to differ in terms of clause length when we consider the interactions with the position of the relevant clauses. As the interaction plot in Figure 3 shows, obwohl-sentences are more or less evenly distributed across the three structural positions, while the length of although- and aunque-clauses varies with the position of the concessive clause relative to the main clause: embedded concessives tend to be shorter than either preposed or postposed concessives. However, an Anova does not show a significant interaction between the variables ‘marker’ and ‘position’ (of the concessive clause) as predictors of the length of the concessive clause (F(4,291) = 1.33, p = 0.26). Figure 3: Interaction plot for ‘sentence length’, ‘marker’ and ‘position’ of concessive clause (mean values) The distributions for individual combinations of a marker and a position are shown in 37 the boxplot in Figure 4, generated with the R-package ‘ggplot2’ (Wickham 2009). Note that the black lines within each box do not indicate the mean values, but the medians. Figure 4: Boxplot showing the interactions between the variables ‘marker’ and ‘position’ as predictors of ‘length’ I conclude that the three markers under investigation differ significantly with respect to their position relative to the main clause, while their length does not seem to vary significantly. It remains to be seen whether this effect is independently associated with the three subordinators under analysis, or whether it is a function of other (functional) properties of the relevant sentences. 4.2 Semantic relations In Sect. 2, four types of semantic relations were distinguished for the concessive markers under study: canonical concessives, restrictive concessives, modal concessives and adversative clauses. The frequencies of these relations relative to the subordinators although, aunque and obwohl are shown in Figure 5. Figure 5 shows a clear difference between obwohl and aunque, with although occupying an intermediate position. Obwohl is almost exclusively used in canonical concessive function, whereas aunque is more frequent in restrictive concessives, in absolute terms. 38 Figure 5: Barchart and mosaic plot for the variables ‘marker’ and ‘semantic relation’ The two smallest categories, ‘adversative’ and ‘modal’, are negligible. The question arises how this finding relates to the tendencies concerning the relative ordering of concessive clauses pointed out in Sect. 4.1. Obwohl-clauses were shown to be rarely preposed, and it is conceivable that this is due to these clauses rarely occuring in a function other than canonical concessive. Figure 6 shows the correlations between the position of concessive clauses and the type of semantic relation. Even though the distribution differs significantly from statistical independence (p = 0.03, according to a regression model; note that because of the low expected frequencies in six cells a χ 2 test is not appropriate), there are no major asymmetries in the most frequent categories, though concessive clauses do show a certain tendency to be postposed, in comparison to restrictive clauses. In order to determine the impact of each variable – ‘position’ and ‘semantic relation’ – on the position of a concessive clause, I fitted a logistic regression model to the data, using the ‘rms’-package for R (Harrell Jr 2017). The model shows that only the variable ‘marker’ is a significant predictor for the position of a concessive clause (p < 0.001), 39 Figure 6: Barchart and mosaic plot for the variables ‘position’ and ‘semantic relation’ while ‘relation’ is not (p = 0.25). There is no significant interaction between the predictors (p = 0.24). While ‘relation’ is not a significant predictor of the position of a concessive, it was shown to be a distinguishing factor between the three subordinators above. We can combine the two variables ‘position’ (of concessive clause) and ‘semantic relation’ as distinguishing factors between the three subordinators under analysis in a conditional inference tree. The tree diagram in Figure 7 was generated with the ‘party’-package for R (Hothorn et al. 2006b). The tree model in Figure 7 basically shows a contrast between although and aunque, on the one hand, and obwohl, on the other. In relations other than canonical concessives (‘cnc’), obwohl is rare. Among the canonical concessive cases, it is the most frequent marker in embedded and postposed concessives, while it is comparatively rare in preposed (canonical) concessives. Note that approx. 60% of obwohl-cases exhibit this combination of variables – ‘position: preposed, relation: concessive’. It is fairly chacteristic of obwohl. I (preliminarily) conclude that Germ. obwohl is primarily used in canonical conces40 Figure 7: Conditional inference tree with the predictors ‘position’ and ‘semantic relation’ for the variable ‘marker’ sives, in a position following the main clause, whereas Engl. although and Span. aunque are more broadly distributed. Aunque exhibits a significant overrepresentation of restrictive uses. 4.3 Levels of linking Figure 8 shows the frequencies of the markers under analysis in concessive clauses at the propositional (‘p’), illocutionary (‘i’) and textual (‘t’) levels. Obviously, the propositional level is by far the most frequent one. The distribution of markers relative to the level of linking shows a significant deviation from statistical independence (p < 0.001). Most importantly, obwohl is significantly underrepresented in textual uses. Another, trivial, correlation concerns the type of semantic relation and the level of linking, as textual uses are normally restrictive (since they relate to the same Quaestio as m; cf. Sect. 2). Again, this correlation is confirmed by the data (p < 0.001; the standardized residual for the combination ‘relation: restrictive, level: textual’ is 4.85). Finally, we can also expect a correlation between the level of linking and the information status 41 Figure 8: Barchart and mosaic plot for the variables ‘marker’ and ‘level’ of the concession – and again, this expectation is confirmed (p < 0.001; the standardized residual for the combination ‘level: textual, status: assertive’ is 4.2). Beyond these expected and trivial correlations, the variable ‘level’ (of linking) does not seem to correlate significantly with any other variable. I assume that the sample contains too few instances of textual and illocutionary links, which is probably a register effect: textual uses typically imply self-correction, which is not a common rhetorical device in political speech, and illocutionary links are often somewhat informal, and thus avoided in scripted speech. 4.4 The information status of the concession If we compare the three operators under unvestigation – although, aunque and obwohl – with respect to the information status of the clauses that they introduce, a certain difference between obwohl, on the one hand, and although and aunque, on the other, can be observed, insofar as obwohl is more frequent in concessive clauses that contain given information, which I regard as ‘pragmatically presupposed’ (cf. Sect. 2.3.1). The variable 42 was treated as a Boolean one, with ‘t’ standing for ‘assertive’. The frequencies of the the three markers under investigation in assertive and non-assertive concessive clauses are shown in Figure 9. While the entire distribution deviates significantly from statistical independence (p = 0.04), no cell makes a significant contribution individually (in the sense that the standardized residual is greater or smaller than 2). Figure 9: Barchart and mosaic plot for the variables ‘marker’ and ‘claim’ As was pointed out in Sect. 2.3.1, the information status of concessive clauses is often claimed to correlate with its position, insofar as preposed concessives tend to be nonassertive, while postposed concessives are mostly assertive. This assumption is confirmed by the data used for the present study (p = 0.002) – cf. Figure 10. 43 Figure 10: Barchart and mosaic plot for the variables ‘position’ and ‘claim’ However, while the correlation between ‘position’ and ‘information status’ holds for the entire data set, it varies across the three subordinators under investigation. The frequencies of combinations of a certain position and information status is shown separately for although, aunque and obwohl in Figure 11. 44 Figure 11: Barchart: Position and information status of concessive clause per marker Figure 11 shows that the main differences are found among the presupposed concessives (in the top row). Although and aunque conform to the general tendency pointed out above: if they introduce non-assertive clauses, they tend to precede the main clause, whereas asserted concessives tend to follow the main clause. Obwohl behaves differently: even among the presupposed clauses there is a strong tendency to follow the main clause. The data are shown in the form of a mosaic plot in Figure 12. The most interesting part of the plot is the lower half, showing the frequencies for non-assertive clauses. The bottomright rectangle – representing the combination ‘marker: obwohl, assertive: f, position: post’ – is not only overrepresented in comparison to although and aunque, it is also by far more frequent than preposed and embedded concessives in absolute terms. 45 Figure 12: Mosaic plot: Position and information status of concessive clause per marker The other variables investigated in the present study have not brought to light any interesting results, as far as correlations with ‘information status’ are concerned. As expected, there are clear positive correlations between an assertive status of a concessive clause and the semantic relation (p < 0.001, Cramer’s V = 0.3), as well as the level of linking (p < 0.001, Cramer’s V = 0.33): While canonical concessives tend to be presupposed, restrictive concessives are more commonly asserted; as far as the level of linking is concerned, textual occurrences exhibit the expected tendency towards being asserted. 4.5 The topic-comment structure of concessive sentences In Sect. 2.3.2, a distinction was introduced between concessive links with an (internal) ‘topic contrast’, and concessives without such a contrast. Topic contrasts may either be ‘horizontal’ (non-scalar) or ‘vertical’ (scalar). The frequency distribution of the three 46 options is shown in Figure 13, relative to the three subordinators although, obwohl and aunque. It deviates significantly from statistical independence (p = 0.03). no internal contrast internal contrast horizontal vertical Figure 13: Frequencies of information structural configurations per marker Figure 13 shows that obwohl tends to be used in contexts where there is no (internal) topic contrast, while although and aunque are comparatively common in contrastive contexts. Moreover, there is a certain difference between although and aunque insofar as although is more frequent in cases of horizontal contrast, whereas the figures for vertical contrast are identical for these two subordinators. As far as correlations with other variables are concerned, we can expect a trivial correlation between both of these variables and ‘semantic relation’ – e.g. insofar as restrictive concessives exhibit what I call a ‘vertical’ contrast between the two propositions in question, where m and c provide information about the same Quaestio (Klein and Stutterheim 1987; Gast and Auwera 2011). The data confirm this expectation (χ 2 = 31.6, df = 6, p < 0.001, Cramer’s V = 0.23). No significant correlation could be found between ‘topic contrast’ and the variables ‘level’ (of linking) (χ 2 = 3.08, df = 4, p = 0.52) or ‘information 47 status’ (of c) (χ 2 = 5.8, df = 2, p = 0.055). However, there is a significant correlation between ‘topic contrast’ and ‘position’ (relative to the main clause) (χ 2 = 11.44, df = 4, p = 0.022). Figure 14 shows the distribution of the data. Figure 14: Barchart and mosaic plot for the variables ‘topic contrast’ and ‘position’ If we consider the distribution of the position of a concessive and the topic contrast per marker, we can see that although is strongly associated with horizontal contrast and preposed concessives, while obwohl shows a strong tendency towards occurring in a postposed position, with no internal contrast – cf. the mosaic plot in Figure 15. To summarize, the following generalizations can be made: obwohl is tends to occur in concessives without contrasting topics, while contrastive topics are more commonly found in although- and aunque-clauses. Although is more frequent in cases of horizontal contrast than aunque, and moreover typically preposed in this configuration. 4.6 A multivariate analysis We can finally compare the three markers under analysis in terms of all the variables considered in Sections 4.1–4.5. In order to determine the most distinctive variables, I 48 Figure 15: Mosaic plot for the variables ‘marker’, ‘position’ and ‘topic contrast’ fitted random forest models to the data, using the R-package ‘party’ (Hothorn et al. 2006a; Strobl et al. 2007; Strobl et al. 2008). In a first step, I only included the funtional variables as potential predictors, i.e., ‘semantic relation’, ‘information status’ (of c), ‘level’ (of linking) and ‘topic contrast’. The results are shown in Figure 16. Figure 16 shows that the semantic relation holding between m and c is the most important functional variable distinguishing the three subordinators under analysis. This result is confirmed by a conditional inference tree (Hothorn et al. 2006b), where all potential predictors were included as well, but only ‘semantic relation’ was a significant distinctor – cf. Figure 17. If we include the structural variables as well, the picture becomes a bit more compli- 49 Figure 16: Variable importance according to a random forest analysis Figure 17: Conditional inference tree with ‘marker’ as a response variable and all functional variables as predictors cated. The importance of the variables is shown in Figure 18. Figure 18 shows clearly that the three subordinators under analysis also differ in terms of the position of their host clauses relative to the main clause. Moreover, the variables ‘topic contrast’ has a certain importance. Figure 19 (which can be regarded as an extension of Figure 7) shows the corresponding conditional inference tree. 50 Figure 18: Variable importance according to a random forest analysis (with structural variables) Each of the configurations in Figure 19 can be interpreted as being either typical or atypical of a specific subordinator. The first box (at the left end) shows that obwohl is rarely used in other than canonical concessive clauses. The three rightmost boxes all display distributions within the group of canonical concessive clauses. The rightmost box shows that within the group of canonical concessives, obwohl is underrepresented in preposed position. For embedded and postposed concessives, obwohl is overrepresented when there is no topic contrast. Among those cases that do exhibit a topic contrast, aunque is underrepresented relative to although and obwohl. This is surprising considering that in Sect. 4.5, aunque was shown to be relatively common in cases with internal topic contrast overall. The reason for this seemingly unexpected result is that aunque has an overproportional share of restrictive concessives, which are not represented in the three rightmost boxes. As aunque occurs only rarely in postposed canonical concessives, its frequency in the two boxes in the centre is relatively low independently of the exact type of topic contrast. 51 Figure 19: A conditional inference tree (all variables as potential predictors) 5 Summary and some attempts at explanations To summarize the findings of this exploratory study, we can make the following generalizations: 1. Position of the subordinate clause (a) Obwohl shows a tendency towards not being used in preposed concessives, even for non-assertive concessives, which are otherwise likely to be preposed. (b) Although shows a tendency to be used in non-assertive preposed concessives. 2. Semantic relation holding between m and c (a) Obwohl shows a tendency to be used in canonical concessives only and is rarely 52 used in restrictive concessives. (b) Aunque shows a tendency to be used in restrictive concessives and is comparatively rare in canonical concessives. 3. Level of linking between m and c Obwohl is underrepresented in textual uses. 4. Topic contrast Obwohl tends to be used without a topic contrast. I will now offer brief attempts at explanations for each of these generalizations. Generalizations 1a and 1b may well be related to the fact that German indicates a subordinate status of clauses independently, i.e., through (verb-final) word order (Zifonun 2013). A subordinate status correlates with non-assertiveness, even though this correlation is of course not a perfect one.8 At least probabilistically, German thus indicates a non-assertive status syntactically, and the position of a concessive clause relative to the main clause is of minor relevance to its information status. By contrast, English and Spanish, which do not distinguish between (assertive) main and (non-assertive) subordinate clauses syntactically,9 may use the position of a concessive clause as an indicator of its illocutionary force (preposing being associated with non-assertiveness). The tendency towards postposition characteristic of obwohl, in turn, may reflect the fact that concessions often have a strenghtening effect, ‘removing an obstacle’, as Mann and Thompson (1988, 39) put it (cf. Note 1). One way of looking at the difference between English and Spanish on the one hand, and German on the other, is to assume that there are ‘competing 8 Note that some generative syntacticians (in the tradition of Wechsler 1991) assume that verb-second ordering is a way of indicating illocutionary force, which can be taken to imply that non-V2-sentences do not have illocutionary force. 9 Note that the subjuntivo in Spanish is related to facticity, not assertiveness. 53 motivations’ (DuBois 1985), e.g. with respect to information status and argumentative weight in relation to the positioning of a concessive clause. Generalizations 2a and 2b concern the degrees of specialization of the markers under analysis. According to the data used for my study, obwohl is the most specialized concessive, being largely restricted to canonical concessives. The fact that aunque is strongly attracted by relativizing concessives might be related to its containing a scalar element (aun), unlike although and obwohl, which are based on expressions of universal quantification (Engl. all) and binary choice (Germ. ob), combined with a (modal) particle (Engl. though > þēah,10 Germ. wohl). Restrictive concessives are characterized by what I have called a ‘vertical’ relation holding between m and c, and the scalar component within aunque might reflect this preference. Of course, this hypothesis, too, is mere speculation at this point and requires thorough historical analyses to be tested. Generalizations 3 and 4 can also be regarded as symptoms of specialization. Textual uses of concessives probably represent a diachronic extensions of propositional uses. They seem to imply the change of a subordinative structure into a coordinative one. In German this process seems to be blocked as a consequence of verb-final word order in subordinate clauses – cf. the explanations for Generalizations 1a and 1b above. Like Generalizations 1b, 2a and 3, Generalization 4 seems to point to a certain ‘conservativeness’ of obwohl, in comparison to although and aunque. The latter operators are sometimes used in contrastive contexts, which are typically covered by subordinators such as Engl. while and whereas and Span. mientras. Contrastive contexts are typically ‘symmetrical’, e.g. insofar as they consist of contrasting main clauses. Again, obwohl may thus be less likely to occur in such contexts because of the verb-final word order of 10 Cf. König (1985) for a historical study of English concessives. 54 the clauses that it introduces. Most, but not all of the generalizations made above have been related to a syntactic difference between German, on the one hand, and English and Spanish, on the other: German, unlike English and Spanish, uses a specific word order for subordinate clauses, which is also (probabilistically) associated with the absence of illocutionary force. This association has been speculated to have an influence on the relative ordering of main and subordinate clauses, and it seems to block certain context extensions, e.g. to restrictive and textual uses. Of course, the situation looks very different in colloquial German, where obwohl (as well as many other subordinators) is used with verb-second ordering (see Günthner 2000 for obwohl, and Uhmann 1998; Wegener 1999, among others, on the causal subordinator weil). One way of testing the hypothesis that verb-final order blocks context extensions would be to systematically compare the formal (and often scripted) speech of the European parliament wth informal colloquiual speech. References Atayan, Vahram, Bettina Fetzer, Volker Gast, Danial Möller, and Tanja Ronalter. Forthcoming. “Ausdrucksformen der unmittelbaren Nachzeitigkeit in Originalen und Übersetzungen: Eine Pilotstudie zu den deutschen Adverbien gleich und sofort und ihren Äquivalenten im Französischen, Italienischen, Spanischen und Englischen”. In Translation und Linguistik, ed. by B. Ahrens, S. Hansen-Schirra, M. Krein-Kühle, M. Schreiber, and U. Wienen. Berlin: Frank & Timme. Breindl, Eva. 2004. “Konzessivität und konzessive Konnektoren im Deutschen”. Deutsche Sprache 1:2–31. 55 Büring, Daniel. 2003. “On D-trees, beans and B-accents”. Linguistics and Philosophy 26 (5): 511–545. Carbonell-Olivares, María. 2009. “A corpus-based analysis of the meaning and function of although”. International Journal of English Studies 9 (3): 191–208. Cartoni, Bruno, and Thomas Meyer. 2012. “Extracting Directional and Comparable Corpora from a Multilingual Corpus for Translation Studies”. In Proceedings 8th International Conference on Language Resources and Evaluation (LREC). Istanbul, Turkey. Crevels, Mily. 2000a. “Concession. A typological study”. PhD thesis, University of Amsterdam. — . 2000b. “Concessives. A typological perspective”. In Cause – Condition – Concession – Contrast. Cognitive and Discourse Perspectives, ed. by E. Couper-Kuhlen and B. Kortmann, 313–340. Berlin: Mouton de Gruyter. DuBois, John. 1985. “Competing motivations”. In Iconicity in Syntax, ed. by John Haiman, 343–366. Amsterdam: John Benjamins. Friendly, M. 1994. “Mosaic displays for multi-way contingency tables”. Journal of the American Statistical Association 89:190–200. Gast, Volker. 2010. “Contrastive topics and distributed foci as instances of sub-informativity: A comparison of English and German”. In Comparative and Contrastive Studies of Information Structure, ed. by C. Breul and E. Göbbel, 15–50. Amsterdam: John Benjamins. — . 2016. “Doing corpus-based typology: Concessivity in a cross-linguistics perspective”. Keynote speech given at the Conference on Discourse Relational Devices, Valencia, 24-26 January, 2016. 56 Gast, Volker, and Johan van der Auwera. 2011. “Scalar additive operators in the languages of Europe”. Language 87 (1): 2–54. Gast, Volker, and Christoph Rzymski. 2015. “Towards a corpus-based analysis of evaluative scales associated with even”. Linguistik Online 71. Gast, Volker, Lennart Bierkandt, and Christoph Rzymski. 2015a. “Annotating modals with GraphAnno, a configurable lightweight tool for multi-level annotation”. In Proceedings of the Workshop on Models for Modality Annotation, held in conjunction with IWCS 11, 2015, 19–28. Stroudsburg, PA. — . 2015b. “Creating and retrieving tense and aspect annotations with GraphAnno, a lightweight tool for multi-level annotation”. In Proceedings of the 11th Joint ACLISO Workshop on Interoperable Annotation, ed. by H. Bunt, 23–28. Tilburg: Tilburg Center for Cognition and Communication. Gast, Volker, Lennart Bierkandt, Stephan Druskat, and Christoph Rzymski. 2016. “Enriching TimeBank: Towards a more precise annotation of temporal relations in a text”. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC 2016), ed. by Nicoletta Calzolari (Conference Chair) et al. Portorož, Slovenia: European Language Resources Association (ELRA). ISBN : 978-2- 9517408-9-1. Gast, Volker, Vahram Atayan, Julien Biege, Bettina Fetzer, Sophie Hettrich, and Anne Weber. Forthcoming. “Unmittelbare Nachzeitigkeit im Deutschen und Französischen: Eine Studie auf Grundlage des OpenSubtitles-Korpus”. In Comparatio delectat III. Akten der VIII. Internationalen Arbeitstagung zum romanisch-deutschen und innerro- 57 manischen Sprachvergleich, ed. by Christine Konecny, Carmen Konzett, Eva Lavric, and Wolfgang Pöckl. Fankfurt: Lang. Günthner, Susanne. 2000. “From concessive connector to discourse marker: The use of obwohl in everyday German”. In Cause – Condition – Concession – Contrast. Cognitive and Discourse Perspectives, ed. by E. Couper-Kuhlen and B. Kortmann, 439–486. Berlin: Mouton de Gruyter. Harrell Jr, Frank E. 2017. rms: Regression Modeling Strategies. R package version 5.1-1. https://CRAN.R-project.org/package=rms. Hole, Daniel. Forthcoming. “A crosslinguistic syntax of scalar and non-scalar focus particle sentences: the view from Vietnamese and Chinese”. Journal of East Asian Linguistics. — . 2015. “A distributed syntax for evaluative only sentences”. Zeitschrift für Sprachwissenschaft 34 (1): 43–77. Hothorn, Torsten, Peter Buehlmann, Sandrine Dudoit, Annette Molinaro, and Mark Van Der Laan. 2006a. “Survival Ensembles”. Biostatistics 7 (3): 355–373. Hothorn, Torsten, Kurt Hornik, and Achim Zeileis. 2006b. “Unbiased recursive partitioning: A conditional inference framework”. Journal of Computational and Graphical Statistics 15 (3): 651–674. Iten, Corinne. 2000. “Although revisited”. UCL Working Papers in Linguistics 12:65–95. — . 1998. “The meaning of although: A relevance theoretic account”. UCL Working Papers in Linguistics 10:81–108. Jacobs, Joachim. 1983. Fokus und Skalen. Zur Syntax und Semantik der Gradpartikeln im Deutschen. Tübingen: Niemeyer. 58 — . 2001. “The dimensions of topic-comment”. Linguistics 39:641–682. Kim, Yong-Beom. 2002. “Concession and linguistic inference”. In Language, Information, and Computation: Proceedings of The 16th Pacific Asia Conference, January 31 – February 2, 2002, 187–194. The Korean Society for Language and Information. Klein, Dan, and Christopher Manning. 2003. “Accurate unlexicalized parsing”. In Proceedings of the 41st Meeting of the Association for Computational Linguistics, 423– 430. Klein, Wolfgang, and Christiane von Stutterheim. 1987. “Quaestio und referentielle Bewegung in Erzählungen”. Linguistische Berichte 109:163–183. Knott, Alistair, and Ted Sanders. 1998. “The classification of coherence relations and their linguistic markers: an exploration of two languages”. Journal of Pragmatics 30:135– 175. Koehn, Philipp. 2005. Europarl: A parallel corpus for statistical machine translation. MT Summit X. Phuket. König, Ekkehard. 1994. “Concessive clauses”. In The Encyclopaedia of Language and Linguistics, ed. by R.E. Asher, 679–681. Pergamon Press. — . 1988. “Concessive connectives and concessive sentences: Cross-linguistic regularities and pragmatic principles”. In Explaining Language Universals, ed. by J. Hawkins, 145–166. New York: Blackwell. — . 1991. “Concessive relations as the dual of causal relations”. In Semantic Universals and Universal Semantics, ed. by D. Zaefferer, 190–209. Berlin: de Gruyter. 59 — . 1986. “Conditionals, concessive conditionals and concessives: Areas of contrast, overlap and neutralization”. In On Conditionals, ed. by E. Traugott, A. Ter Meulen, J. Snitzer Reilly, and C.A. Ferguson, 229–246. Cambridge: Cambridge University Press. — . 1985. “On the history of concessive connectives in English. Diachronic and synchronic evidence”. Lingua 66:1–19. König, Ekkehard, and Peter Eisenberg. 1984. “Zur Pragmatik von Konzessivsätzen”. In Pragmatik in der Grammatik, ed. by G. Stickel, 313–332. Schwann. König, Ekkehard, and Peter Siemund. 2000. “Causal and concessive clauses: Formal and semantic relations”. In Cause – Condition – Concession – Contrast. Cognitive and Discourse Perspectives, ed. by E. Couper-Kuhlen and B. Kortmann, 341–360. Berlin: Mouton de Gruyter. Kortmann, Bernd. 1997. Adverbial Subordination: A Typology and History of Adverbial Subordinators based on European Languages. Berlin and New York: de Gruyter Mouton. Mann, W.C., and S.A. Thompson. 1988. “Rhetorical Structure Theory: Toward a functional theory of text organization”. Text 8 (3): 243–281. Meyer, David, Achim Zeileis, and Kurt Hornik. 2016. vcd: Visualizing Categorical Data. R package version 1.4-3. Pittner, Karin. 1999. Adverbiale im Deutschen. Untersuchungen zu ihrer Stellung und Interpretation. Tübingen: Stauffenburg. R Core Team. 2017. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/. 60 Roberts, Craige. 2012. “Information structure in discourse: Towards an integrated formal theory of pragmatics”. Semantics and Pragmatics 5 (6): 1–69. doi:10.3765/sp.5.6. Rudolph, Elisabeth. 1996. Contrast. Adversative and Concessive Relations and their Expressions in English, German, Spanish, Portuguese on Sentence and Text Level. Berlin & New York: Walter de Gruyter. Sarkar, Deepayan. 2008. Lattice: Multivariate Data Visualization with R. ISBN 978-0387-75968-5. New York: Springer. http://lmdvr.r-forge.r-project.org. Searle, John. 1975. “A classification of illocutionary acts”. Language in Society 5 (1): 1–23. Stalnaker, Robert. 2002. “Common ground”. Linguistics and Philosophy 25 (5): 701–712. — . 1974. “Pragmatic presuppositions”. In Semantics and Philosophy, ed. by M. Munitz and P. Unger, 197–214. New York University Press. Strobl, Carolin, Anne-Laure Boulesteix, Achim Zeileis, and Torsten Hothorn. 2007. “Bias in Random Forest Variable Importance Measures: Illustrations, Sources and a Solution”. BMC Bioinformatics 8 (25). http : / / www . biomedcentral . com / 1471 2105/8/25. Strobl, Carolin, Anne-Laure Boulesteix, Thomas Kneib, Thomas Augustin, and Achim Zeileis. 2008. “Conditional Variable Importance for Random Forests”. BMC Bioinformatics 9 (307). http://www.biomedcentral.com/1471-2105/9/307. Sweetser, Eve. 1990. From Etymology to Pragmatics: Metaphorical and Cultural Aspects of Semantic Structure. Cambridge: Cambridge University Press. Traugott, Elizabeth Closs. 1989. “On the rise of epistemic meanings in English: An example of subjectification in semantic change”. Language 65:31–55. 61 Uhmann, Susanne. 1998. “Verbstellungsvariation in weil-Sätzen: Lexikalische Differenzierung mit grammatischen Folgen”. Zeitschrift für Sprachwissenschaft 17:92–139. Wechsler, Stephen. 1991. “Verb second and illocutionary force”. In Views on Phrase Structure, ed. by Katherine Leffel and Denis Bouchard, 177–191. Dordrecht: Kluwer. Wegener, Heide. 1999. “Syntaxwandel und Degrammatikalisierung im heutigen Deutsch? Noch einmal weil-Verbzweit”. Deutsche Sprache 17:92–139. Wickham, Hadley. 2009. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York. ISBN: 978-0-387-98140-6. http://ggplot2.org. Wiechmann, Daniel, and Elma Kerz. 2013. “The positioning of concessive adverbial clauses in English: Assessing the importance of discourse-pragmatic and processingbased constraints”. English Language and Linguistics 17 (1): 1–23. Zifonun, Gisela. 2013. “Adverbial eingeleitete Verbletztsätze”. In Satztypen des Deutschen, ed. by Jörg Meibauer, Markus Steinbach, and Hans Altmann, 301–316. Berlin: de Gruyter Mouton. Zwarts, Frans. 1995. “Nonveridical contexts”. Linguistic Analysis 25:286–312. 62