Institute of Formal and Applied Linguistics

People communicate through discourse. We talk to each other, we read the newspaper or interact on social media, and in all these contexts we use discourse. Constructing coherence relations between utterances is a crucial part of discourse processing. Understanding a discourse includes the inference of coherence relations between utterances, such as Cause-Consequence, Temporal Sequence or Contrast. Languages have specific devices to express such relations: Connectives like because, therefore and  however, but also lexical cue phrases like As a result, The problem is, or To the contrary.

In this talk, I will outline a Cognitive approach to Coherence relations (CCR) and show how this approach is corroborated with empirical research, by looking at different types of converging evidence: cross-linguistic analyses of connective use, as well as acquisition data and results from studies on discourse processing and representation.

For instance, languages of the world provide their speakers with means to indicate causal relationships. Causal relations can be expressed by connectives and lexical cue phrases, such as because, since, so and As a result. Striving for converging evidence, we may ask about these phenomena:  What is the system behind the use of such connectives in languages like English, French, Dutch and German, or Mandarin Chinese? How can we describe these systems in a cognitively plausible way? How do children acquire this connective system? And what is the role of these causal relations and connectives in discourse processing? Based on the results, we are able to identify salient categorizing principles.

Finally, we will discuss the implications of such cognitive categories for discourse annotation. In recent years, we have seen how corpora of language use are annotated at the level of coherence relations. Excellent but different annotation systems exist, such as the Penn Discourse Treebank (PDTB) and Rhetorical Structure Theory (RST). So, the question is: are cognitive insights useful for such discourse annotation systems? We will argue that our approach identifies unifying dimensions, will lead to more systematicity and is likely to improve existing relation definitions.