We report on our experimental results concerning reasoning with quantifiers “some” and “most” in natural language. We focus on differences between logically correct inferences and the inferences that are accepted or produced by people in everyday reasoning. According to classical approach both “some” and “most” have existential meaning in this sense that they are false in empty domains, on the contrary - a sentence with “all” can be vacuously satisfied. Thus whereas “Some A’s are B” can be inferred from “Most A’s are B”, neither of these sentences can be inferred from a universal sentence, unless we assume that A and B are non-empty. On the other hand, it seems that in natural language the use of informationally weaker “some” or “most” by a speaker strongly suggests that a sentence with the informationally stronger quantifier “all” is false.

What is even more, since “some” is informationally weaker than “most”, the use of “some” suggests that a sentence with “most” is false. However, the stronger items logically imply the weaker, so “all” implies both “most” and “some” (with additional assumption that domains are non-empty). In our experiments we checked whether people’s inferring “some” (resp. “most”) from “all” depends on the character of domains (empty vs. nonempty terms). We analyzed also people’s readiness to accept “not all” as a conclusion from sentences with “some” or “most”, and “not most” as a conclusion from sentences with “some”. In this talk we report on our observations and propose how to explain our results.