Differences and similarities between scalar inferences and scalar modifiers: the case of quantifiers

Abstract

We explore a distinction between ‘high’ and ‘low’ readings in counterfactual donkey sentences and observe three open issues in the current literature on these sentences: (i) van Rooij (2006) and Wang (2009) make different empirical predictions with respect to the availability of ‘high’ donkey readings. We settle this question in favour of van Rooij’s (2006) analysis. (ii) This analysis overgenerates with respect to weak readings in so-called ‘identificational’ donkey sentences. We argue that pronouns in these sentences should not be analysed as donkey pronouns, but as concealed questions or as part of a cleft. (iii) The analysis also undergenerates with respect to NPI licensing in counterfactual antecedents. We propose a strict conditional semantics for counterfactual donkey sentences that derives the correct licensing facts.