Children’s derivation of scalar implicatures: Alternatives and relevance

Highlights

• We study the role of alternatives in scalar implicature generation on 5-year-olds.

• Accessibility of the stronger alternative is important for children’s SI generation.

• Alternatives lead children to generate SIs only when they are shown to be relevant.

Abstract

Utterances such as “Megan ate some of the cupcakes” are often interpreted as “Megan ate some but not all of the cupcakes”. Such an interpretation is thought to arise from a pragmatic inference called scalar implicature (SI). Preschoolers typically fail to spontaneously generate SIs without the assistance of training or context that make the stronger alternative salient. However, the exact role of alternatives in generating SIs remains contested. Specifically, it is not clear whether children have difficulty with spontaneously generating possible informationally stronger scalemates, or with considering how alternatives might be relevant. We present three studies with English-speaking 5-year-olds and adults designed to address these questions. We show that (a) the accessibility of the stronger alternative is important for children’s SI generation (Experiment 1); (b) the explicit presence of the stronger alternative leads children to generate SIs only when the stronger scalar term can easily be seen as relevant (Experiment 2); and (c) in contexts that establish relevant alternatives, the explicit presence of the stronger alternative is not necessary (Experiment 3). We conclude that children’s considerations of lexical alternatives during SI-computation include an important role for conversational relevance. We also show that this more nuanced approach to the role of lexical alternatives in pragmatic inference unifies previously unconnected findings about children’s early pragmatic development and bears on major accounts proposed to date for children’s problems with SIs.

Introduction

Implicatures are components of speaker meaning that constitute an aspect of what is meant in a speaker’s utterance without being part of what is said. A scalar implicature (SI) is a pragmatic inference triggered by certain lexical items such as quantifiers. Often, the use of a proposition containing a quantifier such as some is taken to implicate that another proposition containing a logically stronger quantifier (all) would not hold. For example, the statement in (2a) below can be used to implicate (2b).

The term scalar comes from the fact that linguistic terms like some and all form an ordered set of alternatives (a scale) based on informational strength¹ (<all, …, most, some, >; Horn, 1972). Informational strength is based on asymmetrical logical entailment where a proposition containing the informationally stronger term (all) logically entails a proposition containing the weaker one (some) but not vice versa.

On this account, the quantifier some has lower-bounded semantics (‘at least some and possibly all’; Horn, 1972). The upper-bounded meaning (‘some but not all’) corresponds to the scalar implicature and is therefore a pragmatic enrichment of the semantic content of the quantifier. The conclusion that the upper-bounded meaning is a pragmatic, not a semantic, contribution is further supported by the fact that this meaning can be explicitly canceled without logical contradiction (“Megan ate some of the cupcakes. In fact, she ate all of them”). Other logical scales are based on logical connectives (<or, and>) or modals (<might, must>). For instance, the statements in (3a) and (4a) below can be taken to implicate (3b) and (4b) respectively.

Scalar implicatures can also be derived from non-logical scales, based on contextual information (Hirschberg, 1985). In some sense the terms “scales” and “scalar” are actually a misnomer: As Hirschberg has convincingly shown (1985) any partially ordered set can give rise to SIs. For instance, the response in (5b) implicates that the action of changing the oil was not completed.

The first account of how scalar implicatures are derived was described by Paul Grice. He suggested that communication is a co-operative effort largely governed by rational expectations about how a conversation should proceed. These expectations were formalized as a number of principles or maxims that are thought to guide the inferences which hearers usually entertain when interpreting utterances (Grice, 1975). When these expectations seem to be violated, the assumption that this was done on purpose creates a variety of effects (see also Horn, 1972). For instance, in (2a), the speaker has violated the Quantity maxim that asks speakers to make their contribution as informative as is required by the current conversational purposes: some is the less informative term within the scale <some, all>. Thus the choice of the weaker term some is reason to believe that the speaker cannot commit to an informationally stronger statement (“Megan ate all of the cupcakes.”). Therefore, the stronger statement does not hold, thus (2b).

The psycholinguistic literature has shown that adults are very adept at deriving scalar inferences (e.g., Bott et al., 2012, Breheny et al., 2013, Breheny et al., 2006, Huang and Snedeker, 2009a). Young children, however, seem to face difficulties. For instance, Noveck (2001) showed that French speakers between the ages of 5 and 10 interpreted the French existential quantifier certains (“some”) in statements such as “Some giraffes have long necks” as compatible with tous (“all”), while adults were equivocal between the logical and the pragmatic interpretations. Similarly, in another study, Greek-speaking 5-year-olds, unlike adults, accepted statements such as “Some of the horses jumped over the fence” as descriptions of story outcomes where all of the horses in the scene jumped over the fence (Papafragou & Musolino, 2003).

Subsequent studies have replicated and confirmed the finding that children typically display non-adult behavior when interpreting scalar statements (Feeney et al., 2004, Foppolo et al., 2012, Guasti et al., 2005, Katsos and Bishop, 2011; cf. also Braine and Rumain, 1981, Smith, 1980). Importantly, children’s difficulties emerge even in studies that used eye movement measures, as opposed to overt pragmatic judgments, to gain insight into comprehension (Huang & Snedeker, 2009b). Furthermore, a variety of factors seems to affect children’s success with scalar implicatures. These include training in detecting pragmatic infelicity and/or a strong supporting context (Foppolo et al., 2012, Guasti et al., 2005, Papafragou and Musolino, 2003, Papafragou and Tantalou, 2004); the type of scale (logical vs. ad hoc; Barner et al., 2011, Stiller et al., 2015) and scalar item (number vs. quantifier; Papafragou, 2006, Papafragou and Musolino, 2003); and the type of response children have to provide (Katsos and Bishop, 2011, Pouscoulous et al., 2007; see Papafragou & Skordos, 2016, for a review).

Several strands of evidence suggest that part of children’s problem with SIs lies in generating scalar alternatives when faced with a weak scalar term. In early studies that examined the interpretation of the disjunction operator or (Chierchia et al., 2001, Gualmini et al., 2001), adults were shown to be sensitive to the scalar implicature from the use of disjunction: when faced with statements like “Every boy chose a skateboard or a bike” to describe the outcome of a story, adults tended to interpret the statement as meaning ‘Every boy chose either a skateboard or a bike’. However, 3–5-year-old children seemed oblivious to this pragmatic interpretation and treated or as being compatible with the stronger term and. In a follow-up task, however, children were presented with two statements and they overwhelmingly preferred stronger/more informative statement with and (“Every farmer cleaned a horse and a rabbit”) over the weaker/less informative statement with or (“Every farmer cleaned a horse or a rabbit”) when the story made the stronger statement true. Thus children could compare alternatives to a weak scalar term and assess their relative informativeness when these alternatives were explicitly presented to them but did not seem to independently access those scalar alternatives and use them to compute implicatures (see also Ozturk & Papafragou, 2015, for similar results with epistemic modals such as may and have to).

A study by Barner et al. (2011) offers further evidence for the role of the accessibility of unspoken lexical alternatives on children’s SI calculation. Barner et al. tested 4-year-old children in a task that involved answering questions about a group of three animals. In critical trials, all three animals (a dog, a cat and a cow) were sleeping and children were asked whether “…some/only some of the animals are sleeping”. Children responded affirmatively about 66% of the time even to the question with only some. This was taken to indicate that children have difficulty generating scalar alternatives even when this is predicted to be triggered by the grammar (only is a focus element requiring the generation and negation of appropriate alternatives). However when a different group of children were simply asked whether “the cat and the dog are sleeping”, children accurately responded with an affirmative answer 93% of the time. More importantly, when asked whether “only the cat and the dog are sleeping”, children correctly gave No-responses 86% of the time. Barner et al. (2011) interpreted these findings as strong evidence that children’s problem with SIs lies mainly in realizing what terms can come together to form a scale: when scalemates are explicitly provided (e.g., when the experimenter listed the animals that were supposed to be sleeping), children’s generation of SIs improved significantly.

Even though these studies suggest that the accessibility of scalar alternatives contributes to children’s difficulties with SIs, the precise role and potency of lexical alternatives in the derivation of SIs at present remain open. One issue is that children’s apparent insensitivity to SIs has been found even in contexts that should make stronger scalar alternatives highly accessible. For instance, in Noveck’s (2001) judgment study, the critical true but infelicitous some-statements (e.g., “Some giraffes have long necks”) were embedded within a larger battery of statements that also included other types of some statements and a variety of all statements (e.g., “All elephants have trunks”): even though this paradigm presumably made the stronger scalar alternatives accessible, children did not seem to benefit from the presence of the stronger term. In another study, when 5-year-olds were asked to evaluate an underinformative some-statement accompanying a story (e.g., “Some smurfs went on a boat” in a story where all smurfs had gone on a boat), they were no better or worse at detecting underinformativeness compared to another task in which the some-statement was presented after children had to evaluate a true all-statement (e.g., “All of the dwarfs are eating a piece of candy”): in both cases, 5-year-olds gave largely logical responses when judging pragmatically infelicitous statements (Foppolo et al., 2012, Exp.3). These findings challenge the hypothesis that the difficulty of spontaneously generating stronger scalar terms is a major factor in children’s computation of SIs.

A related issue is that, to the extent that the accessibility of stronger lexical alternatives facilitates SI computation in children, the mechanisms whereby this effect is achieved are not well understood. One possibility is that children have problems retrieving the stronger alternative when required. This possibility is more naturally aligned with accounts on which lexical scales containing quantifiers, modals and similar expressions provide a pre-defined set of alternatives that feed into the computation of at least certain classes of SIs (what have been termed ‘generalized’ SIs; see Chierchia, 2004, Chierchia et al., 2009, Levinson, 2000). A particular example of this general class of accounts is the “restricted alternatives” hypothesis proposed by Tieu, Romoli, Zhou, and Crain (2015). In the words of Tieu et al.: “Take the case of scalar quantifiers. The child must learn that some and all lie on the same quantifier scale. A failure to compute the implicature could arise either because the child has yet to learn the co-scalar status of some and all, or because the child is unable to retrieve all from the lexicon during the experiment. […].” (ibid., p.25). A related, albeit somewhat broader proposal is the “processing limitation” hypothesis (Chierchia et al., 2001, Gualmini et al., 2001). On this hypothesis, the process of computing the semantic content of an utterance, generating alternatives to the weak scalar term and rejecting them to strengthen the original proposition pragmatically might overwhelm younger children’s processing abilities because of “the processing cost associated with maintaining in memory different representations of the target sentence” or “… involved in comparing different alternative representations of a sentence” Chierchia et al., 2001 p. 167). These general types of account (that we will collectively refer to as the “lexical retrieval” account for brevity) predict that, other things being equal, making the stronger lexical alternative available (e.g., by mentioning it explicitly prior to the weak scalar) should lead children to derive SIs from weak scalars.

Alternatively, children may not have a problem with considering possible lexical members of scales per se (or spontaneously activating them); the problem might lie rather in their failure to recognize that the scalar terms constitute relevant alternatives. For instance in discussing children’s apparent insensitivity to SIs in judgment tasks, Papafragou and Musolino (2003) observed: “If preschoolers, unlike adults, cannot readily infer the pragmatic nature of the task, and are not given adequate motivation to go beyond the truth conditional content of the utterance, they may readily settle for a statement which is true but does not satisfy the adult expectations of relevance and informativeness” (p.269). They went on to propose that “if children are provided with a context where communicative [i.e., relevance] expectations are clear and where the stronger alternative to the weaker statement is made particularly salient, they will be more prone to noticing the implicature” (ibid, p.277; see also Foppolo et al., 2012, Exp.6; Bott and Noveck, 2004, Noveck and Sperber, 2007; cf. also Pouscoulous et al., 2007, where it is hypothesized that children’s problems with SIs are due to difficulties in the optimization process between cognitive gains and processing costs as defined within a relevance-theoretic framework.). Such an account would predict that simply activating the stronger lexical item will not necessarily be enough for children to derive a SI, unless children appreciate that this alternative to the statement offered (out of many different possibilities) is relevant to the goal of the conversation.

Existing developmental studies cannot adjudicate between these two possibilities about the role of lexical alternatives because they have not independently manipulated the contribution of the accessibility vs. relevance of lexical alternatives in children’s SI generation. Our goal is to do so in the studies that follow. Clarifying the role of alternatives in the computation of SIs has broader implications for further theories of children’s pragmatic difficulties with SIs that do not attribute a particular role to the accessibility of scalar alternatives (Katsos and Bishop, 2011, Noveck, 2001). We discuss these theories more fully towards the end of the paper.

In the present studies we explored the role of lexical alternatives in children’s computation of scalar inferences. Our main goal was to throw light on both the potency of lexical alternatives on children’s derivation of SIs and the theoretical machinery whereby such alternatives exert their effects.

We focused on the quantificational scale <some, all>. In Experiment 1, we tested whether the presence of the stronger lexical member of the quantifier scale (all) in the course of the experiment can encourage children to generate a SI from the use of a weak alternative (some). If considering stronger scalar candidates is a limiting factor in children’s computation of SIs, then children’s pragmatic performance should improve when the stronger scalar term (all) is provided for them.

In Experiment 2, we explored the nature of the mechanism that uses lexical alternatives to generate SIs. If lexical alternatives lead to SI generation simply by virtue of providing children with the stronger lexical scale member that they typically fail to consider or activate, then children’s computation of SIs should improve when all is provided for them regardless of whether all is shown to be a relevant scalar alternative to the weak scalar term used or not. If, however, the relevance of lexical alternatives plays a role in SI generation, then the availability of the stronger alternative all should have an effect on children’s computation of SIs only if the lexical scale member can be shown to be a relevant alternative.

Finally, in Experiment 3, we asked whether the generation of SIs from some-statements in children can be achieved even in the lexical absence of the stronger scalar term all if another quantifier (e.g., none) can be used by children as a cue to access the scale <some, all>. This prediction is unexpected on lexical retrieval accounts, according to which supplying the stronger scalar alternative (all) should be a privileged way of helping children recover otherwise inaccessible scalar structure.