# Do Neural Dialog Systems Use the Conversation History Effectively? An Empirical Study

## TL;DR

This empirical study investigates whether neural dialog systems effectively utilize conversation history by testing their sensitivity to various perturbations, revealing that common architectures are often insensitive to such changes.

## Contribution

The paper provides a systematic empirical analysis of how neural dialog models respond to unnatural context perturbations, highlighting their limited sensitivity and offering diagnostic tools.

## Key findings

- Models are rarely sensitive to most perturbations
- Recurrent and transformer models show limited response to context changes
- Open-sourced code for evaluating dialog systems

## Abstract

Neural generative models have been become increasingly popular when building conversational agents. They offer flexibility, can be easily adapted to new domains, and require minimal domain engineering. A common criticism of these systems is that they seldom understand or use the available dialog history effectively. In this paper, we take an empirical approach to understanding how these models use the available dialog history by studying the sensitivity of the models to artificially introduced unnatural changes or perturbations to their context at test time. We experiment with 10 different types of perturbations on 4 multi-turn dialog datasets and find that commonly used neural dialog architectures like recurrent and transformer-based seq2seq models are rarely sensitive to most perturbations such as missing or reordering utterances, shuffling words, etc. Also, by open-sourcing our code, we believe that it will serve as a useful diagnostic tool for evaluating dialog systems in the future.