Incorporating pragmatic knowledge in models of online conversations

Annie Louis - University of Edinburgh

Oct. 4, 2017, 1 p.m. - Oct. 4, 2017, 2:30 p.m.

MC 103

Hosted by: Jackie Cheung


Online conversations are commonplace today and used for various purposes: product
support and troubleshooting, opining about events and people, and student interaction
on MOOC platforms. These conversations get long, involve posts from multiple users,
and the chronological order of the posts in threads does not represent a continuous
flow of dialog. Hence developing models which learn some structure of these
conversations is crucial for facilitating information extraction, search, and
summarization. Apart from applications, we also care, but currently know little,
about the nature of these conversations, what makes them successful, and what
semantics and attributes can be inferred from them.

In this talk, I will present models of forum conversations which are aimed towards
such goals. A primary feature of these attempts is incorporating some notion of
pragmatics in the model development. For example, a user conversation aimed at
troubleshooting her computer display should be interpreted in a different context
compared to one where a user wants to learn about different graphic cards. Pragmatics
refers to such inferred, implied and task knowledge which is external to what is
said in the language itself, but obviously important for conversation interpretation.
We want models to be able to capture such rich contexts, and at the same time to be
learnable on real-world data with little dependence on costly annotations. I will
present two models in this line of work, one which learns the structure of online
troubleshooting conversations, and the other, some initial work at inferring latent
event and character attributes from role-playing game transcripts.

Bio:

Annie Louis is a Research Associate at the Institute for Adaptive and Neural Computation
at the University of Edinburgh. Her research interests are in natural language processing,
machine learning, and the application of language processing technology to solve problems
in information retrieval, social media analysis and software engineering. Previously,
Annie was a Research Associate and Newton International Fellow at the Institute for
Language, Cognition and Computation at the University of Edinburgh. She obtained her PhD
from the University of Pennsylvania in 2013, and is the recipient of a Best Student Paper
Award in 2010 (SIGDIAL) and a Best Paper Award in 2012 (EMNLP-CoNLL).