The nature of discourse predictions with Merel Scholman, Ted Sanders and Vera Demberg

Connectives are often called ‘processing instructions’, informing the reader how to integrate two sentences. Previous research has shown that connectives facilitate the processing of the words directly following the connective. However, there are several explanations for why this might be the case. In this project, we develop and test two cognitive models to examine whether readers make predictions at a discourse-structural level or only at a semantic level. The findings of this study will inform not only theories on the nature of predictions in language, but also allows for more specific guidelines for when a connective will help the reader.

Cross-linguistic differences in connective processing with Freya Hewett, Merel Scholman, Sara Shahmohammadi, Vera Demberg and Manfred Stede

Connectives help readers to process a sentence, but it is unclear whether this effect is consistent across languages. A recent study by Blumenthal-Dramé (2021) shows that connectives facilitate reading more in German than in English. This effect is argued to be due to the typological nature of the two languages. Synthetic languages, like German, encode more information linguistically (e.g. in different word forms). Speakers of those languages might therefore be more strongly affected by a linguistic cue for a discourse relation than speakers of an analytic language (e.g. English). For speakers of analytic languages, on the other hand, linguistic information is usually more ambiguous and they might therefore rely more on contextual cues and world-knowledge during language processing. In this project, we aim to replicate and extend these findings.

Individual differences in discourse prediction with Merel Scholman and Vera Demberg

When reading a sentence, one can often predict how it will end. But how long can a reader maintain such predictions? In this project we investigate whether readers can maintain predictions about upcoming material across sentences. In addition, we examine whether readers differ in their ability to do this, and what factors might contribute to maintaining predictions.

A discourse corpus of Naija-Pidgin with Merel Scholman, AriaRay Brown and Vera Demberg

Nigerian Pidgin is a contact language spoken by over 100 million speakers. In previous work, we constructed a lexicon of Nigerian-Pidgin connectives: Naija-Lex. We are now extending this to a parallel discourse-annotated corpus of Nigerian Pidgin and English, containing implicit and explicit discourse relation annotation. This will bring new insights into how discourse structure is marked in Nigerian Pidgin and facilitate the development of automatic discourse parsers for this low-resource language. In addition, our methodology can inform corpus construction for other low-resource languages.

Comprehending connectives with Merel Scholman and Vera Demberg

Connectives play an important role in understanding a text, but this requires that readers know the meaning of the connective. In this project, we examine how well readers actually understand connectives. We investigate what factors influence the comprehension of connectives, looking both at connective-internal factors (frequency and lexical transparency) as well as reader-internal factors (language experience and cognitive skills). Update: Paper published in Discourse Processes.