Donkey sentences

Project outcomes: The influence of the ontology of the domain of quantification on the interpretation of so-called donkey sentences. Grant N N101 188840.

The immediate subject of the research was the logical form of donkey sentences, such as “Every/some farmer who owns a donkey beats it”. A further, equally important aim was to support a degree of autonomy for logical syntax and semantics by clarifying the boundaries between the logical philosophy of language and linguistics, building on M. Tałasiewicz’s Philosophy of Syntax (Springer, Berlin–New York, 2010). Both parts of the project produced the expected outcomes, adding non-trivially to existing knowledge and bearing on the methodology of research on language more generally.

The dispute over donkey sentences concerns whether they have universal or existential readings. We clarified the methodological status of the discussion in the earlier literature. Arguments appealing to what seems linguistically obvious transform the logical question about the form of donkey sentences into two empirical questions: EQ1: Does ordinary speakers’ linguistic competence determine their logical form? EQ2: If the answer to EQ1 is affirmative, which form does that competence determine?

Closer examination showed that ordinary speakers had rarely been studied properly: researchers generally imagined what such speakers would say if asked. An affirmative answer to EQ1 was also taken for granted, leaving only EQ2 for investigation. B. Geurts’s Donkey Business differs favourably in methodological care: although it too assumes an affirmative answer to EQ1, it investigates EQ2 empirically. Nevertheless, we identify several problems in its reasoning. The most important concerns the interpretation of studies supposedly supporting a widely accepted rule: donkey sentences with an existential quantifier over farmers also receive an existential interpretation of the quantifier over donkeys. Properly interpreted, those studies do not support this claim, and other grounds for accepting it are also unreliable. One outcome of our work is therefore to show that, at least for now, the rule lacks an adequate basis.

More importantly, we argue that Geurts’s observation that participants manipulate the ontology of farmers actually removes the need to appeal to universal readings when explaining his experimental results. In participants’ mental representations, ontologically weak entities occupying the farmer position are “split” so that each farmer has only one donkey. The next step was to test whether a genuine tendency towards universal readings might nevertheless coincide with, and be concealed by, this tendency to split farmers. We devised a different experiment in which the two tendencies yield different results, making a universal reading observable if present and allowing pragmatic factors to be controlled.

During the study we observed that participants not only tend to split ontologically weak farmers but also to merge ontologically weak donkeys. This observation proved important for the work. Since we found no precedent in the literature, it can be regarded as the discovery of a further possible motivation for answers in such tests.

Our study supported the hypothesis that universal readings are unnecessary to explain the experimental results. It did not so much show that donkey sentences have only existential interpretations as show that they lack universal interpretations. The distinction matters because many participants rejected both logical interpretations on pragmatic grounds, involving implicature. In addition to the uniqueness implicature frequently discussed in the literature, scalar implicature was important in sentences with existential quantification: it suggests that “some” means “not all”, although that restriction does not follow from logic. Depending on independently established preferences for logical or scalar readings of the existential farmer quantifier, two interpretations predominated: one with an existential donkey quantifier and a pragmatic interpretation allowing neither an existential nor a universal logical form.

These findings support our initial hypothesis, on a considerably broader empirical basis than Geurts’s studies, that linguistic competence does not determine the logical form of donkey sentences. We do not claim that the answer to EQ2 is an existential reading; rather, the answer to EQ1, previously tacitly taken to be affirmative, is negative. Ordinary speakers do not clearly determine which logical form these sentences should have. Their logical form must therefore be established using considerations beyond linguistic competence, for example broader logical considerations about syntax. Within the area studied, this supports the autonomy of logical and philosophical inquiry from empirical findings in linguistics.