Annotating the Focus of Negation in terms of Questions Under Discussion

Blanco & Moldovan (Blanco and Moldovan, 2011) have empirically demonstrated that negated sentences often convey implicit positive inferences, or focus, and that these inferences are both human annotatable and machine learnable. Concentrating on their annotation process, this paper argues that the focusbased implicit positivity should be separated from concepts of scalar implicature and negraising, as well as the placement of stress. We show that a model making these distinctions clear and which incorporates the pragmatic notion of question under discussion yields κ rates above .80, but that it substantially deflates the rates of focus of negation in text.