<p>Lidz et al. (2011) propose that meanings provide canonical verification
strategies and argue for specifying <em>most </em>in terms of cardinality and
subtraction. We provide cross-linguistic support for these hypotheses by
testing two Cantonese determiners: <em>daai-do-sou</em>, which has a proportional
meaning like <em>most</em>, and <em>zeoi-do</em>, which has a “largest subset” meaning.
When asked to evaluate statements with respect to displays, both
determiners led participants to rely on their Approximate Number System.
But whereas <em>zeoi-do </em>biased a strategy of serial comparison, <em>daai-do-sou</em>,
like <em>most</em>, biased superset subtraction. Even in cases where either
procedure could be successfully deployed, participants nonetheless used