Echoicity and contrast in Spanish conditionals

2.1 Properties

As we have mentioned, eccs have not received much attention in the literature, with two notable exceptions: Narbona (1991) and Schwenter (1999). In his article on “colloquial syntax” of Spanish, Narbona points out that some conditionals are used “[…] with no other purpose besides that of effecting parallelisms or oppositions [by] taking advantage of their two-member protasis-apodosis structure […]” (Schwenter’s translation).^[1] The second author brings up another important property, namely echoicity. To these properties (contrast, lack of conditional interpretation and echoicity), we add other, in our opinion, crucial properties of the construction: scalarity and expression of disdain. We discuss each property in turn.

1. Echoicity

As noted by Schwenter, the antecedent p of the ecc^[2] echoes a previous utterance (“[…] the first clause in B’s response echoes the content of A’s prior assertion […]”, p. 73).^[3]

For instance, both examples (1) and (2) are part of a dialogue and the proposition in p echoes a proposition just uttered by the participant of the dialogue. While eccs are more frequent in dialogue, they are also possible outside the dialogue realm. See as an example (4-b), which could appear in a textbook about dinosaurs, provided that p has either been asserted or can be inferred form a previous assertion, as the one in (4-a).

To see that echoicity is a necessary condition for the ecc interpretation to obtain, consider the following example:

Here we have introduced an explicit question as the preceding discourse move to ensure that Speaker A does not assert the proposition that is then resumed by Speaker B. Hence, there is no echoicity. The result is a distributive interpretation of the conditional, which can be paraphrased as “x did this, while/whereas y did that.” As we will see in Subsection 2.2, this is not an ecc interpretation, since in eccs antecedent and consequent are part of an argumentative competition. Also, notice that the acceptability of both (5-B1) and (5-B2) suggests that scalarity (as described in 3. below) is not a necessary ingredient. Thus, (5) is a felicitous example, but not a case of ecc.

2. Contrast.

Narbona notes that the biclausal structure of the conditional is used to emphasize two opposed propositions. Regarding this opposition, Schwenter (1999) adds the observation that the contrastive function of the construction is evidenced by the obligatoriness of the overt subject pronouns. That is, example (1) is only acceptable with overt pronouns as subject for the two clauses of the conditional; its counterpart with null subject pronouns, as in (6), is not acceptable.

This requirement is noteworthy in a null-subject language like Spanish in which the null pronoun is the default, non-marked option. Overt pronouns in Spanish are used for information-structure purposes: they either mark the focus (Luján 1999) or a Contrastive Topic (Mayol, 2010).

We take the contrast between (1) and (6) to show that eccs need Contrastive Topics, which must be realized overtly, either as subjects or otherwise. For instance, (7) is another example of an ecc; since the subjects of the two clauses refer to the same individual, null pronouns can be used and the Contrastive Topics are realized through temporal adjuncts.

More specifically, the relationship between the two conjuncts needs to be adversative; a very natural way to paraphrase the meaning of eccs is using the adversative conjunction pero (‘but’) as shown in (8).^[4]

3. Scalarity.

On top of expressing two opposed propositions, eccs often contain scalar items in the focus position of the two clauses. When this happens, the scalar term in q is higher than the one in p on a contextually relevant scale. For instance, in (1) and (7) the two scalar items are tired < exhausted, in (2) 100 < 150 and in (4) be scary < be terrifying.

eccs are not acceptable if the stronger scalar item is located in p, even if the two propositions are contrastive, as shown in (9) and (10).^[5]

As noted, scalar terms are frequently present in eccs, but are not required, as shown in the following naturally-occurring examples:

While the examples in (11) do not contain any scalar items, they do convey that the speaker considers her source of tiredness more important than the source of tiredness of her interlocutor, which she disregards as irrelevant (for more discussion of this point, see property 5).

4. Lack of hypothetical interpretation.

In the fairly standard restrictor (Lewis-Kratzer) account (Lewis 1975; Kratzer 1981), conditionals are analyzed as restrictors of a modal operator, which may or may not be overt. It follows that depending on the flavor of the modal (epistemic, generic frequency) the hypothetical relation is of a different type. Consider the two examples provided by von Fintel (2011):

That is, in (12-a) we are evoking a set of worlds compatible with what we know, and stating that in all such worlds where he left at noon, he’s home. On the other hand, in (12-b) we are picking out generic worlds where usually/always, when he leaves work on time, he has dinner with his family.

When the consequent describes a future event, we can think of it as restricting a modal operator whose modal base is metaphysical and whose ordering source is inertial (on this see e.g. Copley 2009).

Roughly, (13) conveys that in all metaphysically accessible worlds which continue as current facts allow and in which it rains, I stay at home.^[6] The general idea remains that p is the restriction of a universal quantifier and, more specifically, the schema that all p worlds (maximally similar to the actual world) are q worlds (Stalnaker 1968; Lewis 1973).

Note that, under this account, causality is not directly encoded in the semantics. However, there is a dependence relation between p and q, which follows from the fact that q is not entailed to be true. In fact, its truth relies on p also being true. Hence, in cases such as (14), which we call hypothetical conditionals, the truth of q depends on the truth of p.

The dependence can be epistemic, as in (15). That is, in all worlds compatible with what I know in which it rains, we have arrived in Bilbao.

In Spanish, epistemic conditionals can be identified because q can be preceded by es que ‘it is’ or esto significa que ‘this means that’. Or else, the modal may be overtly realized, as in (16-c).

Let us now go back to eccs. They are not hypothetical conditionals since there is no obvious connection (either causal or epistemic) between the truth of p and the truth of q. In eccs the speaker is committed to the truth of q. That is, the truth of q does not seem to depend in any way on the truth of p, and q is implied irrespective of the truth of p. Me being exhausted does not depend on you being tired.^[7] Mary having corrected 150 exams does not necessarily / usually depend on John having corrected 50 (but see below for a situation in which this could be the case). Both events can be true at the same time, or one could be true while the other be false at the same time. In this critical sense, eccs differ from hypothetical conditionals.

Certainly, as pointed out by an anonymous reviewer, sentences such as (2), repeated below for convenience, can also convey a hypothetical relation.

Imagine that it is established by contract that Speaker B will always correct three times as many exams as Speaker A. Then, the sentence (19-B) does express a hypothetical relation and, so, the ecc interpretation need not arise. In this hypothetical interpretation, there is no scalarity or disdain (see property 5 below). In fact, we could also imagine a contract that establishes that Speaker A will always correct 10 exams fewer than Speaker B (in which case scalarity would not be observed). Then, example (20) would be possible under a hypothetical reading.

5. Expression of disdain.

eccs often convey that the speaker is either challenging the truth or dismissing the relevance of her interlocutor’s utterance. This is particularly the case in the dialogue examples. While this is an inference that may or may not arise, how it may come about is something that needs explanation.

2.2 What ECCs are not

We would like to point out the existence of some conditionals which look similar to eccs but should not be confused with them.

The first one is illustrated in (21), which was the slogan of the Catalan socialist party for the 2008 Spanish elections. This is a conditional which, like eccs, is marked with a Contrastive Topic in each clause, and the two clauses are opposed in some sense. However, unlike eccs, in this case there is a causal relation between p and q; the antecedent need not be echoic or contain scalar items, and the Contrastive Topic marking is not required (i.e. the pronouns are optional).

Second, eccs are also different from so-called backhanded conditionals (Muñoz 2013), (22), also called monkey’s uncle conditionals in Franke (2007a, fn 1), (23), or indicative counterfactuals in Dancygier (1999, 116), (24).

Dancygier’s comment is that, by offering a blatantly false q as being a consequence of p, the speaker is presenting p as also false. According to Franke, these conditionals convey that the speaker does not believe in the truth of p in virtue of two factors: modus tollens and the commonsense assumption that q is false. That is, since q is obviously not true, p is also implied not to be true. In Muñoz’s account, these conditionals imply that p is false by presenting an outlandish proposition in q, which is clearly known by all participants to be false in the actual world. As noted by an anonymous reviewer, p and q stand in a scalar relation, such that q is more absurd than p. At any rate, though, the potential truth of q is epistemically dependent on the truth of p. That is, at least from the point of view of the literal meaning of the construction, the speaker is presenting being the Empress of China as conditional on the speaker being the Pope. In contrast, in eccs, the speaker is committed to the truth of q irrespective of the truth of p. Furthermore, as opposed to eccs, in backhanded conditionals, two propositions are not necessarily contrastive.

Third, we want to mention another type of conditional in Spanish, pointed out by Narbona (1990, 94) and illustrated below (our glosses and translation), in which an opposition is established between p and q by means of Contrastive Topic:

An appropriate way to translate (25) would be as follows: ‘While economical matters where discussed yesterday, today the senators preferred to focus on political issues.’ That is, the bi-clausal structure is used to distribute the description of two parallel eventualities, as we mentioned previously in the discussion of example (5).

Just like in eccs, this conditional type does not seem to convey a hypothetical relation between p and q, as already acknowledged by Narbona (1991) for similar examples. However, they do not involve echoicity, and scalarity is absent, even in its weak form (it is not conveyed that the second conjunct is more important than the first one). In fact, while p and q are Contrastive Topic marked, the rhetorical relation between p and q is not the one displayed by eccs (we will elaborate more on this in Section 3.4).

Finally, we would like to devote some space to compare eccs with apparently identical conditionals, which are possible in a larger number of languages. Let us present again (1) as our point of departure, repeated below for convenience.

Apart from the ecc interpretation we have been discussing, (26-b) has another reading–-the one rendered by Schwenter (1999), as pointed out in footnote 5–-which we will call metalinguistic. This is one in which what Speaker B does is negotiate about where the standard of tiredness lies. Depending on where it is established, then q is going to be true. In other words, it could be paraphrased as “if this is what counts as being tired, then this means I should describe myself as exhausted”. In this metalinguistic interpretation, the speaker is manipulating the vagueness of the scalar predicates. This is not an ecc interpretation for there is an epistemic connection between p and q (all the worlds compatible with what we know in which this is the standard for being considered tired, are worlds in which I am exhausted).

Comparable constructions are the ones analyzed in Hinterwimmer (2010), illustrated in (27).^[8]

Hinterwimmer observes that (27) can be paraphrased as (28), where a superlative reading of being drunk obtains.

What counts as being drunk depends on a contextually-determined standard. The speaker can look at Mary and utter ‘Mary is drunk’, because she wants to introduce into the Common Ground new information about Mary’s degree of intoxication (namely, that she meets a standard for drunkenness). This is what Barker (2002) names the descriptive mode of use of the vague predicate. Now, we can utter the same sentence in a different context. Imagine Mary’s degree of intoxication is common knowledge, but we do not know what counts as being drunk (in this area, culture, etc.). We are surrounded by people who are drinking and behaving in different ways according to how much (or how long) they have been drinking. Mary is among them, telling jokes and laughing, and not being able to stand still. Then, uttering ‘Mary is drunk’ can be a way of introducing into the Common Ground information about the standard of being drunk. This is Barker’s metalinguistic mode of use.

Hinterwimmer argues that (27) is ambiguous in the same way. In the descriptive reading, he argues, the sentence states that all worlds compatible with what is known and most similar to the actual world where there is someone drunk to a standard degree are worlds in which this person is Mary. To account for the metalinguistic mode of use, Hinterwimmer appeals to Barker’s dynamic analysis, and its key notion of delineation (a function on worlds that maps gradable adjective meanings to a degree). d(c)([[drunk]]) yields the standard for drunk in c. The sentence ‘Mary is drunk’ has the following denotation.

This is a context update whereby the worlds are removed from the context set on the basis of Mary’s degree of drunkenness in that world and the standard of drunkenness in that world. If Mary’s degree of drunkenness is common knowledge, then the worlds that will be excluded will be those in which standards are not low enough to make Mary drunk. In Barker’s terms, in this kind of update, we do not learn anything new about Mary’s drunkenness, but regarding what counts as drunk (in this area, culture, etc.).

If we extend this approach to conditionals, standards can vary across worlds, while the degree of the vague predicate as applied to the relevant individual can be very similar. This is what we need to play with to achieve the metalinguistic mode of use. Hinterwimmer adds to this that the ordering source, which picks out the best worlds, can filter out those worlds that are most strict in the ascription of the vague predicate. With this background information in mind, the sentence conveys that in all relevant accessible worlds c′ (compatible with what is known, potentially different with respect to the standard for drunkenness, where someone got drunk), the unique entity that got drunk to the degree corresponding to the standard in c′ is identical to Mary.

A similar approach can be offered to explain the metalinguistic (non-ecc) reading of (26). In this case, an epistemic dependence can be assumed between p and q, where p is a restrictor of the covert epistemic MUST, but its update does not exclude worlds in which A is not tired, but those in which d(c)([[tired]]) yields a degree that is lower than A’s well-known degree of tiredness. What this conditional amounts to is the statement that all worlds compatible with what is known, which are maximally similar to the actual world, and where A counts as tired, are worlds in which B counts as exhausted (i.e. more than tired). That is, the worlds that will be excluded in each update are those whose standards of tiredness are below A’s degree of tiredness.

Although a particular construction may be ambiguous between the metalinguistic and the ecc interpretation (for instance, see example (18-a) discussed in footnote 8), these two interpretations are different in critical respects. First and foremost, as has been assumed, the metalinguistic construction displays an epistemic dependence between p and q, which is absent in eccs. Recall that in the latter, p and q are independent (to be defined more precisely in Subsection 3.3.3). In example (2), repeated below for convenience, whether or not B has corrected 150 exams does not depend in any way on the possible event of A having corrected 50. Note that the use of numerals rather than vague predicates is decisive in this respect.^[9] There will be languages that accept (26), but not (30).

Observe that, in (26), it is possible to insert entonces ‘then’ preceding q, which enforces the dependence between p and q. By contrast, this is impossible in (30), unless we want to convey that there is a causal or epistemic relation between p and q, (31). ^[10],

^[11]

Finally, under a metalinguistic analysis of (26), it is not necessary that the scalar predicate in q be higher than the one in p. That is, we can be wondering whether I count as exhausted or any degree of tiredness below the standard as specified by the if-clause. This is borne out. We use the degree phrase very tall to enforce felicity, and two possible wordings of the sentence, to rule out the ecc reading.

In these sentences, it is clear that the speaker is trying to place herself on a scale of tallness by assuming that the antecedent is conveying information on what counts as very tall. What is most important for our purposes is that the speaker’s height is below the addressee’s height. Therefore, there is no need for the focus of q to lie on a higher position on the relevant scale, which is one of the properties of eccs described in Section 2.1.^[12]

As far as we have observed, eccs are possible in at least some Romance languages (Spanish, Catalan and Portuguese), but not in languages such as English, German or Hungarian. In these latter languages, (30) is infelicitous and (26) can only be interpreted with the metalinguistic interpretation just explained.

2.3 Interim summary

We have described ecc as a construction with a conditional form in which: p is echoic, the meaning of q is not conditional on the meaning of p, p and q contrast with each other and, if they include scalar items in focus position, they are ordered in terms of their position on a contextual scale, such that the one in q is higher than the one in p. Moreover, a speaker usually uses an ecc to convey her disdain towards p.

In the next section we will propose an analysis of this construction focusing on two main aspects:

1. Why do eccs have the shape of a conditional? Can we reconstruct some kind of conditional relationship between p and q?

2. How are the different properties we identified related to each other?

3.1 Resuming an asserted proposition in the antecedent of a conditional

Let us go into so more detail about the consequences of p echoing a previous utterance. By doing that, the speaker gives rise to various meaning effects.

Following, among others, Gunlogson (2001) and Farkas and Bruce (2010), we assume that each participant has its own set of discourse commitments: that is, propositions to which participants publicly adhere to. The intersection of the commitment sets of the participants in a conversation yields the Common Ground.

Now, by uttering p, speaker A expresses her commitment towards p (she publicly adheres to p, so p is placed on her commitment set) and, simultaneously, proposes to add this proposition to the Common Ground. To this proposal, speaker B can react in several ways. The default, non-marked move for speaker B would be to accept such an addition to the Common Ground. Such acceptance can be explicit (by the speaker uttering ‘OK’, for example) or tacit. Speaker B could also reject p, by explicitly denying it, which would create a ‘conversational crisis’: either one of the participants retracts to the expressed commitment or they agree to disagree. eccs take a different route: what happens in eccs is that Speaker B places p in the antecedent of a conditional. This is not a rejection of p, but it is not an acceptance, either. In fact, ?p, which was supposed to be a settled question, is presented as open for discussion again (or brought to the table, in terms of Farkas and Bruce 2010).^[13]

Let us now consider the inferences of indicative conditionals. It can be shown that a speaker who utters an indicative conditional considers p to be at least possible. This is sometimes called the nontriviality presupposition (see Isaacs and Rawlins 2008, on this) and it can explain the unacceptability of the discourse in (33), from Francez (2015, 16).

From here, a scalar implicature can be calculated such that p is only possible; that is, p is not probable or necessary (◊p⇝¬□p), or else the speaker would not have placed it in the antecedent of a conditional. That is, while the speaker does not necessarily indicate that she considers p to be false, she is clearly indicating her lack of (public) commitment towards p.

As we will see in the following subsections, echoicity is a necessary ingredient to interpret an indicative conditional as expressing an adversative relation, precisely because, by echoing a previously asserted p as the antecedent of a conditional, the speaker can indirectly belittle the value of p, by rejecting to commit to it. As we hope to make clear soon, she is presenting p as a weaker argument for a certain conversational goal.

3.2 Information structure of ECCs

As mentioned, eccs need two Contrastive Topics (CT, henceforth), one in each clause of the conditional. In this section, we will discuss why there is such a need and the consequences of the CT marking.

We assume a model of discourse based on the notion of Question Under Discussion (QUD) (Ginzburg 1994; Roberts, 1996), in which an utterance U, uttered in a context C1, addresses the most salient QUD (QUD-maxC1) and introduces a QUD which becomes the most salient in the output context (QUD-maxC2) (Vallduví 2016), as illustrated in (34).

The answer ‘fish’ in (34-B) is the rheme of the utterance, its actual update potential, which is the locus of progression in discourse. In this utterance, there is no theme, which is the part of the utterance that does not address the QUD, but replicates material in the QUD. A theme-containing utterance can be observed in (35).

Such an answer would be possible in this context, but it is clearly marked, the fragment answer being more natural in a question-answer context. According to Vallduví (2016), “theme-containing utterances prepare the input context by promoting a given QUD to QUD-maximality prior to being elaborated on.” That is, a theme-containing utterance signals that the speaker is not addressing the expected QUD, but some other QUD. For instance, a theme-containing utterance can change the explicit QUD by broadening or narrowing it. (36) shows a well-known example by Büring (2003), in which the explicit QUD is split into several QUDs, and the CT^[14] is necessary to identify the more specific QUDs the speaker has chosen to address: namely, ‘How was the sound?’, ‘How was the audience?’, and ‘How was the band?’, respectively.

(37) shows the opposite case. The speaker chooses not to address the explicit QUD but rather a broader QUD (shown in (37-c)) and, to signal this change, she uses a theme-containing utterance, in which only the verb is rhematic.

Let us go back to eccs and examine the discourse structure of example (1), repeated below as (38).

Speaker A utters ‘I’m tired’ and is, therefore, addressing the QUD in (39-a) and introducing a new QUD: the one in (39-b).

Crucially, Speaker B chooses not to answer the most salient QUD in the output context, but a different one. The two CTs in the ecc indicate that the speaker has decided to change the QUD. Assuming a mechanism à la Büring, in which CT-utterances trigger multiple questions, we end up with the question in (38-B). This QUD explicitly calls for opposed propositions, with different values for the slots ‘who’ and ‘how tired’, and allows the speaker to compare the uttered proposition (that is, p) with the one she wants to vindicate (that is, q).

Thus, the particular information structure of eccs with a CT in each clause serves to indicate that the speaker is changing the current QUD to another one that she is more interested in discussing. This move alone already hints that the speaker does not find p worth discussing and that, instead, she wants to move to a QUD that would allow her to compare p with another proposition. To discuss in more detail the relationship between p and q let us move to the discourse function of an independent conditional.

3.3 The discourse function of an independent conditional

A speaker may decide to use a conditional to express that there is a causal or epistemic dependence between antecedent and consequent. However, as shown in the previous sections, not all conditionals express a hypothetical relation between p and q. One case in point is the so-called biscuit conditionals (cf. DeRose and Grandy 1999; Siegel 2006; Franke 2007a; Franke 2007b; Predelli 2009; Biezma and Goebel 2017, a.o.) (also called relevance or speech act conditionals in the literature), illustrated in (40).

In this paper we will argue that eccs are a subtype of biscuit conditional in the sense that these are conditionals where there is a lack of dependence between p and q, and q seems to be entailed irrespective of the truth of p.

In this subsection, we will first give an overview of biscuit conditionals, followed by a discussion of our assumptions about the semantics of conditionals and of the concept of epistemic independence.

3.4 An adversative rhetorical relation

We have so far seen that, by uttering an ecc, the speaker is expressing simultaneously her lack of commitment towards p and her commitment towards q. In this section, we show that echoicity, together with CT marking, has the rhetorical effect of an adversative relation between p and q, i.e. the one that is denoted by conjunctions such as English but or Spanish pero. More specifically, (49-a) is equivalent to (49-b).^[19]

Remember from Subsection 3.2 that p and q address the same multiple wh-QUD. Let’s take, for instance, (50-A). In terms of discourse relations, there are two options in which the two propositions can be related, either sequentiality, (50-B1), or adversativity, (50-B2).

Sequentiality does not pose any restrictions regarding scalarity. In principle, one could reply (50-A) by saying (50-a). However, this is not possible for the adversative relation, (50-b), which suggests that eccs convey a relation such as (50-B2).

In what follows, we first discuss the pragmatics of adversative but and then address the question why a conditional can have the same rhetorical effect.