Reasoning and inference are central to both human and artificial intelligence (AI). Modeling inference in natural language is notoriously challenging but is a basic problem towards true natural language understanding. In this talk, I will introduce the state-of-the-art deep learning models for natural language inference (NLI). The talk will also discuss a more fundamental problem: learning representation for semantics and composition. We will focus not only on how deep learning models achieve the state-of-the-art performance but also on their limitations.


Human-like AI [Lake, 2016]: Build machines thinking and learning like people

  • Compositionality
  • Causality models/reasoning/inference
  • Commonsense
  • Learning-to-learn

Potential Solutions

  • convert to LFs (hard and lossy)
  • Data-driven (definitely limited)

Cross-sent-att based

ESIM and Tree-based [Chen et al., ACL2017]

Sent-embed based

no cross-sentence attention here.

semantic composition

refer to tutorial at ACL 2017

RepEval model


NLI with external knowledge … [Chen et al. ACL 2018]

external knowledge may not be learnt from training data. purely from training data, copy-mechanism might be insufficient

new inference dataset [Glockner et al. 2018]

Generalized pooling (??) for NLI