Fine-grained Interpretation and Causation Analysis in Deep NLP Models

TL;DR

This paper discusses methods for interpreting individual neurons and causation in deep NLP models, introducing tools and techniques for analysis, manipulation, and domain adaptation.

Contribution

It presents a comprehensive overview of fine-grained interpretation and causation analysis in NLP models, along with two supporting toolkits, NeuroX and Captum.

Findings

Neuron analysis techniques for language properties

Causation analysis methods for model decisions

Application of tools in network manipulation

Abstract

This paper is a write-up for the tutorial on "Fine-grained Interpretation and Causation Analysis in Deep NLP Models" that we are presenting at NAACL 2021. We present and discuss the research work on interpreting fine-grained components of a model from two perspectives, i) fine-grained interpretation, ii) causation analysis. The former introduces methods to analyze individual neurons and a group of neurons with respect to a language property or a task. The latter studies the role of neurons and input features in explaining decisions made by the model. We also discuss application of neuron analysis such as network manipulation and domain adaptation. Moreover, we present two toolkits namely NeuroX and Captum, that support functionalities discussed in this tutorial.