The purpose of this paper is to show how corpus data can contribute to assessing explicit hypotheses about natural language just as experimental protocols can. The particular hypotheses tested concern the source of generalised conversational implicatures with quantifier some. Is the "some and not all"...