Bailey (2013) and Holmberg (2016) propose that the formation of yes-no questions involves a polarity phrase (PolP), the head of which is a polarity variable with open value ([±Pol]), realized as a question particle -ko/-kö in Finnish and A-not-A in Mandarin. The polarity variable moves to the left periphery to obtain sentential scope, overtly in Finnish but covertly in Mandarin. Using novel data and diagnostics from two mutually intelligible Mandarin dialects, Wuhu and Nanjing Chinese, this paper argues that (i) yes-no questions involve at least two distinct functional projections, a question phrase (QuP) and a PolP; and (ii) what moves in yes-no question is a Q-operator base-generated at Spec QuP, whereas the polarity variable is interpreted in-situ throughout the computation and does not move.