Robust Question Answering Through Sub-part Alignment
Jifan Chen and Greg Durrett. Session 4B (Monday 7pm CDT). Talk page
By building explicit alignment structures between questions and contexts, we can improve confidence in QA answers. We model this process and evaluate our model
adversarial and out-of-domain data.
Annotating and Modeling Fine-grained Factuality in Summarization
Tanya Goyal and Greg Durrett. Session 4D (Monday 7pm CDT). Talk page
A number of models have been proposed to determining factuality of summarization systems. We investigate classification-based approaches and show that many of the
commonly-used systems such as FactCC are trained on synthetic data that is mismatched with real errors. Instead, we show that using human annotation and
fine-grained modeling of factuality can lead to better performance.
Did they answer? Subjective acts and intents in conversational discourse
Elisa Ferracane, Greg Durrett, Junyi Jessy Li, and Katrin Erk. Session 5B (Monday 8:20pm CDT). Talk page
Learn about a new dataset of acts and intents annotated on Congressional hearings. Critically, the annotators' subjective judgments about who is
speaking impact their labels!