Using complexity asymmetries to uncover preferences among contexually equivalent meanings

Abstract

Phenomena such as Hurford disjunctions suggest that natural language pragmatics involves a ban on „needless structural complexity“ (see e.g. Meyer 2014, Katzir & Singh 2014): An utterance of a sentence S  is blocked if S has an alternative with contextually equivalent truth conditions that is less complex in the structural sense defined by Katzir (2007 et seq.).

(1)
(a) #I’m in Paris or in France.
(b) I’m in France.

This talk explores what looks like a systematic class of counterexamples to this generalization. In (2)-(4), the (b)-sentences differ from the (a)-sentences in exhibiting imprecision, a kind of context-dependency driven by the salient issue or implicit QUD (see e.g. Krifka 1996, Malamud 2012, Križ 2015, Križ & Spector 2020 among many others). Still, for *some* choices of the implicit QUD parameter, the (a)- and (b)-sentences appear to convey the same truth conditions. What makes these cases different from (1)?

(2)
(a) All the squares are black.
(b) The squares are black.

(3)
(a) They always play soccer if the sun shines.
(b) They play soccer if the sun shines.

(4)
(a) The flag is completely blue.
(b) The flag is blue.

One potential answer is that imprecision precludes the (b)-sentences from competing with the (a)-sentences for the purposes of Manner. I will argue instead that there is pragmatic competition in (2)-(4), but the preference for lower complexity is counterbalanced by a preference for sentences without the potential for imprecision. More generally, the idea is that the grammatical implementation of Maxim of Manner involves an interaction between several preference orderings, and an utterance is blocked if it is not Pareto-optimal with respect to this set of orderings.

In the first part of the talk, I will motivate this proposal in more detail and argue that it provides a principled account of a cross-linguistic markedness asymmetry between definite plurals and *all*-type quantifiers. More generally, it predicts that imprecision has to correlate with lower structural complexity across constructions, so that the surface syntactic complexity asymmetries in (2)-(4) are not accidental. I will discuss several apparent counterexamples to this prediction, such as *every*-type distributive quantifiers and weak necessity modals.

The assumption that we have a preference for sentences without the potential for imprecision raises the question of whether there are any other non-truth-conditional semantic properties that are preferred for the purposes of Manner. The second part of the talk attempts to reframe Maximize Presupposition effects (Heim 1991, Sauerland 2008) in these terms. The resulting system has the counterintuitive consequence that sentences with presuppositions are preferred over their non-presuppositional alternatives of equal complexity, but the (b)-sentences in (2)-(4), which also exhibit truth-value gaps (so-called homogeneity gaps) are dispreferred compared to their gap-less alternatives. The question arises whether something explanatory can be said about this contrast between two types of truth-value gaps. I speculate that both preference asymmetries can be viewed as instances of a broader principle according to which sentences that impose more restrictions on the values of discourse-based contextual parameters are preferred.