Scalar implicature depends on the activation of alternatives. For instance, in English, finger implicates 'not thumb', suggesting that thumb is an activated alternative. Is this because it is more specific (Quantity) and equally short (Manner)? Indeed, toe doesn't imply 'not big toe', perhaps because big toe is longer. As L. Horn points out, this Quantity/Manner explanation predicts that if English had the simplex Latin word pollex meaning 'thumb or big toe', then the asymmetry would disappear. But would it suffice for that word to exist in the language, or would the word also have to be sufficiently salient? We explore this question in four languages that are sometimes said to lack a single-word alternative for thumb: Spanish (which does have pulgar 'thumb or big toe' (< pollex), though it is a non-colloquial form), Russian, Persian, and Arabic. To gauge the salience of various ways of describing digits, we use a fill-in-the-blank production task. We then measure the availability of implicatures using a forced choice comprehension task. We find cross-linguistic differences in implicature, and moreover that implicature calculation tracks production probabilities more closely than structural complexity of the alternatives. A comparison between two Rational Speech Act models—one in which the speaker replicates our production data and a standard one in which the speaker chooses based on a standard cost/accuracy trade-off—shows that comprehension is more closely tied to production probability than to the complexity of alternatives.