change-that-matters-ACL2022

This is the code repository for The Change that Matters in Discourse Parsing: Estimating the Impact of Domain Shift on Parser Error to appear in Findings of ACL 2022.

This research was conducted in conjunction with PAC-Bayesian Domain Adaptation Bounds for Multiclass Learners to appear in UAI 2022.

Some of the code is shared across repositories as detailed below.

Shared Code

Code for summarizing raw results (e.g., producing regression plots) is available in this repository. Additional code for summarizing results as well as code for running all experiments is available in the shared repository here. Further, a python package designed to compute the statistic we propose is available here.

Please consider citing one or both papers if you use this code.

Relevant Links

arXiv (ACL 2022): https://arxiv.org/abs/2203.11317

OpenReview (UAI 2022): https://openreview.net/pdf?id=S0lx6I8j9xq

shared code: https://github.com/anthonysicilia/multiclass-domain-divergence

UAI code: https://github.com/anthonysicilia/pacbayes-adaptation-UAI2022

package: https://github.com/anthonysicilia/classifier-divergence

Notable Versions

Code was run using the following versions (some packages are only used by shared repos):

• python==3.7.4
• matplotlib==3.5.0
• numpy==1.21.2
• pandas==1.3.5
• scipy==1.7.3
• seaborn==0.12.1
• torch==1.10.2 (build py3.7_cuda10.2_cudnn7.6.5_0)
• tqdm==4.45.0
• pillow==8.4.0
• statsmodels==0.13.0

More Papers

This paper is one of a series from our lab using learning theory to study understanding and generation in NLP. Check out some of our other papers here:

• Modeling Non-Cooperative Dialogue: Theoretical and Empirical Insights
• LEATHER: A Framework for Learning to Generate Human-like Text in Dialogue