The only one strategy in context

Highlights

• How do adults solve privative ambiguities and interpret “the only one”?

• Do adults always minimally commit themselves to the weaker interpretation?

• We show that context takes precedence over truth conditional considerations.

• We propose a Principle of Maximal Exploitation: if context is provided, use it.

Abstract

This paper investigates domain restriction in the resolution of privative ambiguities of sentences like The orange parakeet is the only one that is hiding itself, which is ambiguous between an anaphoric and an exophoric interpretation. Previous work by Crain et al. (1994) argued for a built-in parsing preference for weak readings of privative ambiguities. Manipulating the amount of the contextual information available, we present results that challenge Crain et al.’s conclusion: in our study, we show that context takes precedence over truth-conditional considerations in the resolution of privative ambiguities in adults.

Introduction

In this paper we investigate how adults resolve a special case of semantic ambiguity, i.e. privative ambiguity (Horn, 1989, Zwicky and Sadock, 1975). We start from the pioneering study by Crain et al. (1994), who tested the phenomenon of “one-substitution” in children and adults by investigating sentences like (1):

Sentence (1) contains the expression “the only one”, which needs a referent to be interpreted. This gives rise to an ambiguity. The ambiguity derives from the fact that one can be interpreted anaphorically, as in (2a), or exophorically, as in (2b):

In (2a), one receives its reference from the linguistic antecedent that is found in the head noun of the subject Determiner Phrase (DP henceforth), i.e. elephant in this case. In (2b), one is interpreted by making reference to a wider discourse domain that, besides elephants, might include other animals or individuals. Crucially, an entailment relation occurs between the alternative readings in (2), and this is what makes the present ambiguity a privative one. In our example, the exophoric interpretation (2b) asymmetrically entails the anaphoric interpretation (2a). This means that, whenever (1) is true under interpretation (2a), it would also be true under interpretation (2b), but not vice versa. To illustrate, if it is true that there is only one animal in the discourse domain that is doing P (exophoric interpretation, (2b)), then it follows that no other elephant (except for the big one) is doing P as well. By contrast, if it is true that there is no other elephant (except for the big one), that is doing P (anaphoric interpretation, (2a)), then there might be an animal of a different species in the discourse domain that is doing P as well. Consistent with the semantic literature, we can say that (2a) is the weak interpretation, i.e. the one that requires fewer commitments about the status of the world in which a sentence like (1) is uttered: by interpreting (1) as (2a) one will only commit to the fact that no other elephant, except for the big one, is playing the guitar, without making any claim about any other animal that might be present in the discourse domain.

Other linguistic phenomena exhibit the same pattern of entailment between alternative readings and have been the source of intense semantic theorizing. For example, the sentence “The girls know each other” might be interpreted as “Every girl knows every other girl” or “Some girls know some other girls”, the former reading being generally preferred (subject to the lexical properties of the items it combines with). Analogously to what happens in case of “the only one”, the alternative interpretations of “each other” spelled out above are not truth-conditionally independent, in that one reading entails the other. Our alleged preference for the strong interpretation of reciprocals has been accounted for in terms of the Strongest Meaning Hypothesis proposed by Dalrymple et al. (1998) and subsequently extended to plural predication by Winter (2001). In a similar vein, Krifka (1998) proposed the Principle of Pragmatic Strengthening to account for our preferences in the resolution of plural predication and the so-called donkey sentences (i.e. sentences like “Every farmer who owns a donkey beats it”) in which, analogously, an entailment relation occurs between alternative readings (every farmer who owns a donkey beats all/some of the donkeys he owns). This principle assumes that, whenever alternative readings are allowed by the grammar, we should favour the stronger over the weaker one (if consistent with general background assumptions). A similar argument is also implemented in Chierchia (2006) to account for the interpretation of sentences involving scalar terms like some and polarity sensitive items like ever and any, whose interpretation involves considerations about informativeness (see also Chierchia et al., 2013). For example, the interpretation “Lyn ate some but not all of the cookies” is more informative than “Lyn ate some cookies” in upward entailing contexts (Ladusaw, 1980) and thus a sentence such as ‘Lyn ate some cookies” is pragmatically enriched and most often interpreted as “Lyn ate some but not all of the cookies”. This some-but-not all inference does not arise in downward entailing contexts, in which the at-least-some meaning is the strongest interpretation and thus the one generally preferred (e.g., consider the conditional sentence “If Lyn ate some of the cookies, she’ll feel sick”, that does not convey the inference that Lyn will not feel sick if she eats all of the cookies). Analogously, considerations about the informativeness of alternative representations are central in theoretical accounts of presupposition projection (cf. Geurts, 2000 for a criticism and references therein) and in the analysis of scope interactions (cf. Gualmini et al., 2008 and references therein).

In this paper, we restrict our attention to the case of sentences like (1) that were investigated by Crain et al. (1994) in two experiments. In a first experiment, Crain and colleagues tested 3–5 year-old English speaking children with a Truth Value Judgement task (Crain and McKee, 1985, Crain and Thornton, 1998). They asked children to evaluate sentences like (1) in a context in which (1) was true under the anaphoric interpretation (2a) but false under the exophoric interpretation (2b). In the scenarios used for the evaluation, the big elephant was the only elephant playing a guitar, but not the only animal that was doing it, since an octopus and a bird were also playing a guitar. The authors found that children consistently rejected the target sentences in this scenario. To motivate their answer, children pointed to the other animal(s) that were doing the same action as the subject DP in the target sentences. Crain and colleagues interpreted children's behaviour by appealing to a Maximal Commitment strategy: when interpreting ambiguous sentences, children would initially select the interpretation that has more chances of being falsified. In the case at hand, children would interpret (1) by selecting the strong, exophoric, interpretation (2b). This result was explained by appealing to the Semantic Subset Principle (Crain, 1992, Crain, 1993, Crain and Philip, 1993), according to which, in order to converge on the target grammar, the first hypothesis that children make in solving semantic ambiguities is the one that can be falsified on the basis of positive evidence from the input (for a criticism, see also Gualmini and Schwarz, 2009, Musolino, 2006).

In a second experiment, Crain et al. tested a group of adults with a sentence falsification task in which participants were given sentences like (1) and were asked to imagine a scenario that made those sentences false. What they found is that adults tended to mention an animal of the same individual-level category as the head noun of the subject DP. In the case of (1), for example, adults imagined a situation in which there was another elephant that was playing a guitar too (besides the big elephant), thereby suggesting that (1) was disambiguated anaphorically, as in (2a). Crain et al. interpreted adults’ results in terms of the Principle of Parsimony: “if there is a reading that carries fewer unsatisfied presuppositions or entailments than any other, then, other criteria of plausibility being equal, that reading will be adopted as most plausible by the hearer, and the presuppositions in question will be incorporated in his or her model” (Crain and Steedman, 1985: 333). This principle was ultimately motivated by the need to minimize cognitive effort. On this basis, Crain and colleagues hypothesized that the adult parser follows a strategy of Minimal Commitment: in order to avoid unnecessary commitments that might need to be changed at a later stage, adult speakers prefer the interpretation that has more chances of being true. This means that, when interpreting ambiguous sentences in which alternative readings stand in a subset–superset relation, as it is the case for (1), the parser would prefer the weak interpretation, the one that is true in a wider set of circumstances. For example, (1) under the anaphoric interpretation (2a) would be true in the scenario presented, in which other animals (besides elephants) were performing the same action as the big elephant. In the same situation, the exophoric interpretation (2b) would be false.

Let us sum up. On the basis of their findings, Crain et al. proposed the following generalizations about the adult parser: first, it is claimed to be parsimonious, in that it selects the interpretation that entails (the postulation of) the fewest possible entities in the domain of discourse. Second, the human parser is claimed to be guided by a Minimal Commitment strategy, in that it has a built-in preference for weak readings of privative ambiguities, which is purely a truth-conditional property. Third, Crain et al. argued for a difference between the adult parser and the language acquisition device, which is claimed to follow a Maximal Commitment strategy instead.

It is interesting to note that the evidence reported by Crain et al. for a Minimal Commitment strategy in adults is at odds with other hypotheses about adults’ preferences in solving other kinds of privative ambiguities, like the Strongest Meaning Hypothesis or the Principle of Pragmatic Strengthening cited earlier. Despite their different orientation, though, all these hypotheses share the assumption that truth-conditional considerations are used in the resolution of semantic ambiguity. According to this view, the system would be blind to anything that doesn’t specifically bear upon the entailment relations among alternative interpretations (like, for example, the properties of the extra-linguistic/visual context) while it would be affected by the semantic features of the linguistic input (like, for example, Downward Entailingness, that reverts the entailment pattern between alternatives). Indeed, different experimental studies have investigated the effect of the polarity of the linguistic environment on the interpretation of other linguistic phenomena such as Scalar Implicatures and, consistently with this hypothesis, found that these inferences were suspended in Downward Entailing contexts while they normally arose in Upward Entailing contexts, at least in adults (Chierchia et al., 2001, Chierchia et al., 2004, Noveck et al., 2002).

In this paper we concentrate on the study by Crain et al. In the next section, we replicate Crain et al.’s second experiment by manipulating a crucial component in the input, i.e. the availability of alternative referents in the context. Our approach is motivated by previous studies on the role of context in ambiguity resolution. For example, it has long been known that garden path effects like the one associated with the main-clause analysis of the sentence “The horse raced past the barn fell” (Bever, 1970, Frazier, 1979, Kimball, 1973; i.a.), that might be ultimately motivated by the Principle of Parsimony, get mitigated when the pragmatic/visual context is modified so as to include two horses (Crain and Steedman, 1985, see also Altmann et al., 1998, Altmann and Steedman, 1988; i.a.). In terms of the Principle of Parsimony, the reduced relative clause analysis is blocked in Bever's classic example because the parser, being parsimonious, does not postulate additional horses in the discourse domain and interprets “the horse” as the subject of a main clause. However, the reduced relative clause analysis is more easily accessed when the sentence is uttered in a context in which two (or more) horses are mentioned, because the presence of a plausible referent makes a relative clause construction easier to process.

Having introduced the Principle of Parsimony, our ultimate goal is to disentangle this principle and the Minimal Commitment strategy in the resolution of privative ambiguities, and we will do so by manipulating the context, as we will detail in Section 2. The implications of our findings extend beyond the particular constructions we will consider. While the Principle of Parsimony might be integrated within modular and interactive accounts of sentence processing, the Principle of Minimal Commitment requires a strictly modular view of the adult parser. In particular, according to this principle the semantic properties of the linguistic input (like, e.g., information about the entailing relation between alternatives in case of semantic ambiguity) are computed by a module that is specifically equipped to deal with semantic features and, as such, blind to external interferences.

By contrast, the Principle of Parsimony can easily account for the interference of extra-linguistic factors in the processing of ambiguous sentences. According to the Principle of Parsimony, adults should avoid postulating the existence of entities that were not previously mentioned. Thus, the application of this principle needs to take contextual information into consideration.