Production and Interpretation of embedded implicatures: Experiments and a Model

Abstract

Gotzner & Benz 2018, Benz & Gotzner 2021 (BG21) showed that sentences with embedded implicatures can reliably communicate a situation if strong Gricean constraints are satisfied. This was demonstrated for a subset of conjunctions of sentences with two Aristotelian quantifiers (all, some, none), one embedding the other. This raised the question of how the set of all such sentences can be characterised. An answer was given by Benz & Gotzner 2022 who proposed an algorithmic production model that generates exactly those sentences that can reliably communicate a state of affairs. The model defines a set of coding strategies that were evaluated with a larger corpus of German data. I present the results of new experiments that replicate and extend the experiments described in BG21 for English. The English data, even more so than the German data, suggest that speakers do not generate implicatures by globally negating stronger alternatives as assumed by Horn 1972. There is also a noticeable tendency to strengthen unembedded “some” to “all”.