INTRODUCTION

Within the field of pragmatics, a large body of experimental research has been devoted to implicatures. The majority of this research has focused on the category of conversational implicatures. A conversational implicature is “a component of speaker meaning that constitutes an aspect of what is meant in a speaker's utterance without being part of what is said” (Horn, Reference Horn, Horn and Ward2004, p. 3). A widely investigated subtype of conversational implicatures is the scalar implicature. The scalar implicature arising from a sentence like some children are naughty is that ‘some but not all children are naughty’. In contrast to this pragmatic interpretation, this utterance can also be interpreted logically: ‘some and perhaps all children are naughty’. However, the general assumption is that, whenever a weaker term (e.g. the quantifier some) is used, a stronger term from the same scale (e.g. all) does not hold, or that the speaker does not know whether it holds, because otherwise they would be underinformative. This interpretation can be applied to a range of different scales (e.g. <always, often, sometimes>, <three, two, one>, etc.).

In this paper, however, we will focus on the other category of implicatures, conventional implicatures, which have received much less attention in the literature. Unlike conversational implicatures, conventional implicatures are related to the conventional meaning of words. A typical example is that but conventionally implies a contrast. In an utterance such as he is short, but he is healthy, the use of but suggests that there is a contrast between ‘short’ and ‘healthy’, even though this contrast is not explicitly expressed.

Grice (Reference Grice1989) introduced the concept ‘conventional implicature’ to describe conventional aspects of meaning that are not truth-conditional. For example:

1. (1)

   1. a. He is short but he is healthy.

   2. b. He is short and he is healthy.

Truth-conditionally, (1a) and (1b) have the same meaning. However, but is not a truth-conditional connector because but provides an additional meaning to the utterance in (1a). It suggests that there is a contrast between the first and the second clause. This additional feature of but is what Grice describes as a conventional implicature. However, within the context of relevance theory, Blakemore (Reference Blakemore2002) proposed a procedural analysis of but, which can be summarized as follows:

Thus, according to Blakemore, but encodes a specific procedure. This procedural analysis conflicts with Grice's conceptual analysis in which but encodes the concept ‘contrast’.

The experiment discussed in this paper focuses on the conventional implicature induced by the conjunction but. This work builds on Janssens and Schaeken (Reference Janssens and Schaeken2013), but focuses on children instead of adults. Janssens and Schaeken (Reference Janssens and Schaeken2013) investigated ‘p but q’ utterances constructed as distancing-contrastive connections. In a distancing contrast, but connects two parts of a complex speech act (Van Belle & Devroy, Reference Van Belle and Devroy1992) and the second part is disassociated from the first part, without denying what the latter expresses (Haeseryn, Romijn, Geerts, De Rooij & van den Toorn, Reference Haeseryn, Romijn, Geerts, de Rooij and van den Toorn1997). In the ‘p but q’ construction, the speaker endorses that p is true (Van Belle, Reference Van Belle2003). However, because but is used, the possible inference from p is cancelled in favour of the conclusion stemming from q. The but construction indicates that the q-argument should be considered more relevant than the inference made from p. For example:

1. (2) I really like chocolate, but it's almost dinner time.

In a context where someone is offered a piece of chocolate, the two arguments of (2) lead to opposite conclusions. The first part (p) of this sentence elicits the conclusion that the speaker will eat the piece of chocolate, whereas the q-argument yields the opposite conclusion, i.e. that the speaker will not eat the chocolate. In this construction, but not only indicates that there is a contrast between the two arguments, it also provides more weight to the q-argument. The conclusion from the q-argument is therefore the pragmatic conclusion that follows from but. In other words, by uttering (2) the pragmatic conclusion is that the speaker will not eat the chocolate. Note that when the arguments are reversed (i.e. it's almost dinner time, but I really like chocolate), the opposite conclusion will follow, i.e. that the speaker will eat the chocolate.

The conclusion that follows from the ‘p but q’ construction can be introduced by words like so or nevertheless. These two words elicit opposite conclusions. When the word so follows a ‘p argumentative but q’ utterance, it introduces the expected conclusion from q (e.g. I really like chocolate, but it's almost dinner time. So I won't eat chocolate.). In contrast, Lepère (Reference Lepère2008) argued that, according to Van Belle (Reference Van Belle2003), the word nevertheless,Footnote ¹ used as a conjunctive adverb, reverses the argumentative orientation of a ‘p argumentative but q’ sentence. It overrules the conclusion from q and redirects the reader towards the conclusion implied by p (e.g. I really like chocolate, but it's almost dinner time. Nevertheless, I will eat chocolate.). Note that the argumentative strength of but, so, and nevertheless should be considered separate from the content of the arguments, which they are supposed to take precedence over (Van Belle, Reference Van Belle2003).

Janssens and Schaeken (Reference Janssens and Schaeken2013) presented adult participants with short stories, each ending with a ‘p but q’ sentence. The p- and q-arguments were either sensible or irrelevant and always led to opposite conclusions. For instance, ‘I really like chocolate’ and ‘it's almost dinner time’ are both sensible arguments (for eating and not eating chocolate, respectively). However, in this context – in which someone is offered a piece of chocolate – ‘I have blonde hair’ would be an example of an irrelevant argument. An irrelevant argument has no relation with the story and in no way indicates which conclusion should follow. Such irrelevant arguments were included in order to investigate whether people might be influenced by the content of the arguments rather than the structure of the sentence (i.e. the pragmatic meaning of but).

After participants read the ‘p but q’ sentence, they were asked to indicate the appropriate conclusion introduced by so (‘so conclusion from p’ or ‘so conclusion from q’) for one half of the stories, and the one introduced by nevertheless (‘nevertheless conclusion from p’ or ‘nevertheless conclusion from q’) for the other half. The ‘appropriate’ pragmatic conclusion introduced by so is the one stemming from the q-argument, while for nevertheless this is the one from the p-argument. The results showed that, in general, people do make the inference induced by but. This was clear from the items with two sensible arguments for which the so-conclusion was asked (82% pragmatic answers). In contrast, for the nevertheless-conclusions, the expected conclusions from p were given in only 48% of the cases. This could indicate that the meaning of nevertheless does not give as much precedence to the p-argument as was expected. On the other hand, this latter finding could also be explained by the fact that the inference stemming from nevertheless is opposite to the one from but. In a ‘p but q’ construction, but leads the reader towards the conclusion from q, but nevertheless overrules this conclusion in favour of the conclusion from p. This understandably requires more effort than a so-conclusion, which explains why the number of appropriate answers was lower for nevertheless.

The content of the arguments was also found to play an important role. Whenever an irrelevant argument was combined with a sensible argument, most participants favoured the conclusion from the sensible argument, regardless of the conventional meaning of but, so, or nevertheless. The importance of the content was confirmed in a second experiment in which participants were asked to justify their answer. As expected, whenever their conclusion was not the pragmatic one, participants mostly explained this by referring to the content of the arguments.

This paper investigates whether the pragmatic meaning of but, combined with so and nevertheless, is also clear to children. In conversational implicature research, it has been shown that children are less able to make pragmatic inferences than adults. For example, Noveck (Reference Noveck2001) found that 89% of the seven- to eight-year-olds in his study agreed with statements such as some giraffes have long necks, compared to only 41% of the adults. Similarly, with respect to propositional connectives, Braine and Rumain (Reference Braine and Rumain1981) presented evidence that deductively competent seven- and nine-year-old children favour a logical interpretation of or (‘p or q and perhaps both’) over an implicit one (‘p or q but not both’). Adults given the same task were equivocal, although they tended to favour exclusive interpretations (Braine & Rumain, Reference Braine and Rumain1981). Such observations that children are less able to make pragmatic inferences than adults have led to further research into factors that enhance children's pragmatic competence. For example, in an experiment with five-year-old children, Papafragou and Musolino (Reference Papafragou and Musolino2003) found that a training session prior to the presentation of the test sentences in order to enhance the children's awareness of pragmatic anomalies caused the number of pragmatic answers to increase.

For conventional implicatures, we also expect that children are less able to make pragmatic inferences than adults. This is because there is a certain similarity between conversational implicatures and the conventional implicatures investigated in our experiment. A specific feature of conventional implicatures – in contrast with conversational implicatures – is that they are not cancellable. However, but sentences seem to carry two implicatures. First, there is the implicature that but creates a contrast between p and q (e.g. he is short, but he is healthy), which indeed seems to be a classic, non-cancellable conventional implicature. However, this is not the case for the second implicature, i.e. that but provides more weight to the q-argument (e.g. I really like chocolate, but it's almost dinner time). The use of nevertheless, for example, can lead to a cancellation of this implicature as it provides more weight to the p-argument in a ‘p but q’ utterance. Accordingly, the conventional implicatures that we investigate in this paper may not be purely conventional, but they share certain features with conversational implicatures. That is why we expect similar results for the conventional implicature task with but, i.e. that children are less able to make these pragmatic inferences.

The typical pragmatic development in children for conversational implicatures is often explained by the fact that drawing the implicature requires effort and children have less cognitive resources available than adults. Cognitive effort in adults has been tested by De Neys and Schaeken (Reference De Neys and Schaeken2007), among others, who found that burdening working memory (WM) with a secondary task decreases pragmatic processing. Together with other observations (e.g. Bott & Noveck, Reference Bott and Noveck2004; Noveck & Posada, Reference Noveck and Posada2003), this was taken as evidence that scalar implicature processing is effortful. Since the conventional implicatures we investigate in this paper have certain properties in common with conversational implicatures (i.e. they are cancellable), we also examined whether WM plays a role in processing these implicatures. In addition to testing children's understanding of the pragmatic meaning of but, so, and nevertheless, we also measured whether a higher WM span corresponds with a better understanding of the pragmatic meaning of but, so, and nevertheless.

Adults had been found to clearly grasp the pragmatic meaning of but. However, they also proved to be influenced by the content of the arguments, in spite of their pragmatic competence. For this reason, it is plausible to expect that children will definitely show great sensitivity to the content of the arguments in ‘p but q’ constructions. This expectation is based on the similarity between conversational implicatures and the conventional implicatures focused on in this paper, but this expectation can also be deduced from the findings of Schaeken, Sevenants, and Madruga (Reference Schaeken, Sevenants and Madruga2011), who observed a clear effect of content in children. For nine- to thirteen-year-old children, who were given a reasoning task with unless, abstract problems proved to be much more difficult to grasp than concrete problems. The concrete, meaningful problems were about daily life situations that could easily be imagined in a realistic context (e.g. you will have to go to bed early, unless you make no more mistakes) whereas the abstract problems concerned cards with letters on one side and numbers on the other side (e.g. there is a ‘2’, unless there is not an ‘A’). This suggests that for children to be able to reason with unless, they need to be presented with a meaningful context that is imaginable in real life. Children seem to need meaning to reason. We therefore expect the children in our experiment to be greatly influenced by the type of sentence, and more specifically by the content of the arguments. We expect that when children see an argument that they judge as a very strong argument in its context, the content of the arguments will often prevail over the conventional meaning of but, so, and nevertheless. In particular, the combination of sensible arguments with irrelevant arguments is likely to make a difference. Furthermore, the interaction between the type of argument (sensible or irrelevant) and the type of conclusion (so or nevertheless) observed in adults can also be expected in children. We expect the irrelevant arguments to facilitate the so-conclusions in the ‘irrelevant but sensible’ contexts, and the nevertheless-conclusions in the ‘sensible but irrelevant’ contexts. Moreover, since adults were found to have great difficulty with nevertheless, we expect these conclusions to be especially difficult for children.