Topic chains in dialogues

This paper analyzes topic chains (TChs) in spontaneous speech dialogues. The chain-building property of topical elements is an essential means for managing prominence on the discourse level, creating thematic coherence across sentences. TChs relate the sentential and the discourse aspect of information structure by extending the sentence-internal division between prominent and non-prominent information to the cross-sentential level, where we can distinguish between categorical sentences that continue a TCh and those that begin a new TCh. The paper examines the role of the speaker and addressee feature (Harley and Ritter, 2002) on chain building. My assumption is that prominence management differs between 3rd and 1st/2nd person topics, i.e. according to [−local] or [+local] (Ritter and Wiltschko, 2009). I assume that common assumptions on TChs (Chafe, 1976; Reinhart, 1981; Brunetti, 2009) are plausible with 3rd person referents, but much less so with local persons: 1st and 2nd persons are never new referents in a dialogue situation. I therefore propose an analysis with two parallel TChs: one in the local domain and one in the nonlocal domain. The paper concentrates on topical overt and null subject pronouns in Spanish. I will demonstrate that a detailed analysis of topicality differentiating between two parallel chains allows us to understand hitherto unexplained variance with regard to the variation between sentences with realized subject pronouns (e.g. yo canto esta canción ‘I sing this song’) and their counterparts with zero subjects (e.g. ∅ canto esta canción ‘I sing this song’). •We propose that sentence topics have three chain-related features: [±shift], [±local], and [±switch].•Speaker and hearer operate with two parallel topic chains, one for referents in the local domain, one for referents in the nonlocal domain.•Both domain-switching and the type of chain have specific effects on topic continuity.•Topic shifts can be more convincingly interpreted in a double chain model.•The double chain model accounts for hitherto unexplained variance with regard to the use of null subjects.