On temporality in discourse annotation

One of the features that determine the coherence of a discourse is the temporal ordering of the segments. Language acquisition studies show that temporal relations are among the first types of coherence relations that are explicitly marked by a connective: children start using and, followed by (and) then and because (Bloom et al. 1980; Evers-Vermeul & Sanders 2009). In contrast to this prominence of temporality in acquisition studies, processing studies reveal that a temporal marking of a coherence relation can easily be overruled by other features of the relation. For instance, Mak and Sanders (2013) have shown that immediate effects of causal relatedness on referential processing occur even with a connective that is not explicitly causal (when). Given these insights, the research question of this paper is: What is the status of temporality in language use in general and in discourse annotation in particular?We aim to answer this question by bringing together and comparing different kinds of data. First, we will discuss several outcomes of language acquisition and processing studies. Among other things, these studies underline the importance of the phenomenon of underspecification (Spooren 1997): linguistic markers of coherence relations need not exactly match the type of relation intended by the writer/speaker or perceived by the reader/listener. For instance, and and and then can both be used in cause-consequence relations).Second, we compare four approaches towards the annotation of temporality in discourse: the ones taken by the Cognitive approach to Coherence Relations (CCR, Sanders et al. 1992), the Penn Discourse Tree Bank (PDTB, Prasad et al. 2008), the Rhetorical Structure Theory Treebank (Carlson et al. 2003), and Segmented Discourse Representation Theory (Reese et al. 2007). We will focus on questions such as: What kind of temporal relations are listed? And is the temporal ordering of discourse segments considered a relational feature or a segment-specific feature (e.g. determined by the tense of one or both of the segments)? For example, CCR treats temporal relations as a subtype of additive relations and claims that temporality is not a relational but a segment-specific feature. In contrast, the PDTB presents Temporals as one out of four major classes of discourse relations, distinguishing between three subtypes (Synchronous, Precedence and Succession).In our discussion of these data and research outcomes, we will address how temporality interacts with other properties of discourse relations, such as order, causality, and subjectivity. This will enable us to draw conclusions on whether temporality should be seen as an independent relational feature or whether it is a segment-specific phenomenon and/or a phenomenon that is a by-product of (certain combinations of) other characteristics of coherence relations.