Neural Nets for NLP - Document Level Models

Explore document-level natural language processing models in this lecture from CMU's Neural Networks for NLP course. Dive into coreference resolution techniques, including mention pair, entity-mention, and ranking models. Examine discourse parsing approaches and document-level prediction tasks. Learn about error analysis in coreference systems and recent advances using deep reinforcement learning and end-to-end neural models. Discover how attention-based hierarchical neural networks can be applied to discourse parsing. Gain insights into document-level challenges like discourse unit prediction and story completion tasks.

Intro
Their Counter-part in Documents
Document Problems: Entity Coreference Queen Elizabeth set about transforming her husband. King
Mention(Noun Phrase) Detection
Coreference Models:Instances
Mention Pair Models Queen Elizabeth set about Model Classty the coreference relation
Entity Models: Entity-Mention Models Are the genders all Is the cluster containing
Ranking Models
Latent Tree Models (Bjorkelund and Kuhn, 2014)
Problems in Coreference: revisited Instance Problem We've introduced 4 different modeling methods, many seem to work in their own settings • Feature Problem
Problems in Coreference: revisited Instance Problem . We've introduced 4 different modeling methods, many seem to work in their own settings. • Feature Problem
Error Driven Analysis (Kummerfeld and Klein, 2013)
Easy Victories & Uphill Battles
Deep Reinforcement Learning for Mention-Ranking Coreference Models
End-to-End Neural Coreference (Span Model)
Quality of Mentions
Ablations of modules
Error Type Revisited
Discourse Parsing w/ Attention- based Hierarchical Neural Networks
Document Problems: Discourse Unit Prediction
Predicting Discourse Units are similar to Language Modeling
Story Completion Task