2018 Singapore Symposium on NLP

GQ77I3rYT%SMNo3Uy+DJ6w.jpg 2RULVK9WRbiBJBV+OBOdig.jpg

Consolidating Textual Information

Prof Ido Dagan, Bar Ilan University, Iseral

Consolidation information motivation:

• redundancy
• scattered information

Framework

input texts -> natural knowledge representation ->

structured knowledge graphs/ontology

Open IE
QA-SRL

coreference

Summary

IMG_4352.jpg

Syncretizing Strcutured and Learned Representation

By Noah A. Smith

NL semantics: two operationalizations

1. Dependencies
2. Spans

Linguistic Strcuture Prediction

FliU5dlNTfanxtO7NuWLsw.jpg ObqLqEmKSWi+C2xHbiX%pQ.jpg 58KTs%LsSKuwfWiW9wQOPg.jpg

Towards Content-based Essay Scoring

Vincent Ng

Holistic scoring
corpora:

• Automated Student Assessment Prize
• TOEFL

Approaches

• Regression. classification, preference ranking
  • Argument features
  • prompt-independent features
• Neural models

Main idea: instead of a holistic score, split into many dimensions

• Persuasive argument scoring