This study examines the anaphoric status of the sequence et pourtant si/non in French. This sequence displays some properties not only of TP-Ellipsis but also of propositional anaphora. Consequently, the antecedent of this sequence can be recovered by means of either type of anaphoric process. I argue that the salient and relevant antecedent is constrained by the presence of a modalized environment. I claim that the discursive marker pourtant is assimilated to a modal operator (Jayez 1988, Martin 1987) expressing discourse contrast between two propositions anchored in two possible worlds that are not contradictory. Polarity Particles (POLPARTS) involved in this sequence are analyzed as emphasizing the truth of a proposition. As such, they are conveying semantic contrast between two polarities, that of a salient and accessible discourse antecedent and that of the missing part after et pourtant si/non. This is how POLPARTS upgrade the Common Ground. I develop a focus-based account for Verum Focus, building on alternatives along the lines of Hardt & Romero (2004). I suggest that the scope of an epistemic operator (Romero & Han 2004) and the conditions of use are relevant in order to reconstruct the adequate antecedent, which is not possible in an analysis based solely on lexical insertion and upgrading the Question Under Discussion (qud) by conditions governing the felicitous use of et pourtant si/non.

Recent psycholinguistic studies on the reality of alternative sets in processing focus NPs have shown that focus particles like ‘only’ play a special role in activating the mental representation of alternatives to focused nouns. In this paper we present a new corpus study which provides converging evidence to support psycholinguistic findings and suggests that alternatives preceded by a focus particle are not only more activated in experimental contexts, but are also more likely to be discussed in the subsequent context. To this end we develop and evaluate inter-annotator agreement on two novel annotation tasks in naturally occurring German corpus data: recognition of nominal alternatives in general without any context, and recognition of alternatives in the context of sentence pairs. We show that while annotators agree poorly on the first, they agree strongly on the second. We also develop a concept of ‘alternative density’, the number of alternatives realized in a sentence following a target NP, and present a mixed-effects model showing a very significant rise in density after the presence of German nur ‘only’ independently of other factors.