Till startsida
University of Gothenburg
Sitemap
To content Read more about how we use cookies on gu.se

Vera Demberg "Challenges in Discourse Relation Labelling"

Research profile seminar

Vera Demberg "Challenges in Discourse Relation Labelling"

Discourse relations are an important aspect for natural language understanding beyond the sentence; while discourse relation labelling has received increased attention in recent years, it remains a highly challenging task with accuracies for state-of-the-art systems at 43%-45% accuracy for 11-way classification. In my presentation, I will talk about current work on automatic discourse relation classification using neural networks, and point to the challenges in using such powerful models with many parameters, while the actual training and test data set size is limited.
Additional training data can be acquired in a number of ways. I will discuss our findings for (i) jointly using training data from various corpora annotated according to different schemes; (ii) crowd-sourced discourse relation annotation (iii) exploitation of explicitation of connectives during translation. Along the way, I will reflect on insights regarding different levels of description of discourse coherence, as well as regarding individual biases and individual differences in human discourse interpretation.

Lecturer: Vera Demberg

Date: 3/7/2018

Time: 1:15 PM - 3:00 PM

Categories: Linguistics

Organizer: CLASP

Location: T307, Olof Wijksgatan 6

Contact person: stergios chatzikyriakidis

Page Manager: Webbredaktionen|Last update: 5/23/2016
Share:

The University of Gothenburg uses cookies to provide you with the best possible user experience. By continuing on this website, you approve of our use of cookies.  What are cookies?