Suggestion for Final Project: Negation and Lexical Substitution

Negation has been a big open problem for a long time. PMI tends to find words that appear in similar contexts, but that includes both synonyms and antonyms, since it is common for texts to compare this with that, and to contrast this with that. Similar issues hold for more recent methods such as BERT.

Consider these references:

  1. NegBERT
  2. Neural Lexical Substitution
  3. Training with Lexical Resources
  4. Lexical Distribution Similarity

This project could include a survey of papers, benchmarks and tasks in this space, as well as replications of some of the suggestions.