LIN389C topic list Fall 2020

Core topics

Computational pragmatics, and metaphor


Inference, including background and common-sense knowledge and symbolic structures

  • latent-variable neural models, for example recent work by Frank Ferraro and Niranjan Balasubramanian

  • knowledge graphs: knowledge graph triple embeddings, Trans-E and friends, also universal schema

  • common-sense reasoning and background knowledge, and its integration in neural models. For example, KnowBERT, self-talk (UW)

  • Survey on knowledge graphs https://arxiv.org/abs/2002.00388

  • graph-based reasoning: we discussed graph convolutional networks previously. Graph attention networks, graph transformer networks

  • attention supervision for graph-shaped information, including trees in BERT

  • temporal relations and temporal reasoning, starting with Greg Durrett's work


Language and bias

  • Social bias frames (UW plus Dan Jurafsky)

Natural language generation



Ethics

  • Ethics of data collection for NLP
  • Ethics and algorithms
  • “Lessons from the archives” Timnit Gebru

Formal distributional semantics

  • Aurelie Herbelot, EVA model
  • Guy Emerson, Functional distributional semantics

Additional topics

Language learning

multi-agent artificial language learning.

Language modeling

  • The incredible versatility of recent language models, and their recent successes
  • Integration with background knowledge and common-sense reasoning (see above)
  • Coherence issues

Evaluation

Human evaluation of NLP systems, in particular natural language generation