On the anti-exhaustive inference of ya

<p>Sudo 2014 analyzes the connective <em>ya</em> in Japanese as an inclusive disjunction that triggers<br>two scalar inferences, (i) a conjunctive inference and (ii) an anti-exhaustive inference, based<br>on alternatives involving other connectives. Sauerland et al. 2017 propose a competing<br>analysis of (i) that is based on domain alternatives, but they make no claim about (ii). We<br>report on an experiment with an inference task whose results indicate that <em>ya</em> does indeed<br>trigger (ii) but it is not as robust as (i). We argue that this is problematic for Sudo’s (2014)<br>analysis, but at the same time shows that Sauerland et al.’s (2017) analysis needs to be<br>augmented with a mechanism for generating (ii). In order to account for the difference in<br>robustness, we propose that (i) is a grammatically derived scalar inference, while (ii) is a<br>pragmatically derived scalar inference.</p>