Assessing Out-of-Domain Language Model Performance from Few Examples

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 2385–2397 May 2-6, 2023 ©2023 Association for Computational Linguistics Assessing Out-of-Domain Language Model Performance from Few Examples Prasann Singhal∗, Jarad Forristal∗, Xi Ye, and Greg Durrett Department of Computer Science The University of Texas at Austin {prasanns, jarad, xiye, gdurrett}@cs.utexas.edu Abstract While pretrained language models have ex- hibited impressive generalization capabilities, they still behave unpredictably under certain domain shifts. In particular, a model may learn a reasoning process on in-domain train- ing data that does not hold for out-of-domain test data. We address the task of predicting out- of-domain (OOD) performance in a few-shot fashion: given a few target-domain examples and a set of models with similar training perfor- mance, can we understand how these models will perform on OOD test data? We start from the baseline of looking at model accuracy on the few-shot examples, then investigate how to incorporate analysis of the models’ behav- ior using feature attributions to improve our understanding of generalization. Specifically, we explore a set of “factors” designed to re- veal model agreement with certain pathologi- cal heuristics that may indicate worse gener- alization capabilities. On textual entailment, paraphrase recognition, and a synthetic classi- fication task, we show that attribution-based factors can help rank relative model OOD per- formance. However, accuracy on a few-shot test set is a surprisingly strong baseline, par- ticularly when the system designer does not have in-depth prior knowledge about the do- main shift. 1 Introduction The question of whether models have learned the right behavior on a training set is crucial for gener- alization. Deep models have a propensity to learn shallow reasoning shortcuts (Geirhos et al., 2020) like single-word correlations (Gardner et al., 2021) or predictions based on partial inputs (Poliak et al., 2018), particularly for problems like natural lan- guage inference (Gururangan et al., 2018; McCoy et al., 2019) and question answering (Jia and Liang, 2017; Chen and Durrett, 2019). Unless we use eval- *Equal contribution Training M1 M2 M3 Suite of trained models (different pre-training, data augmenta;on, inocula;on, …) Which one do I use? OOD SeEng label 10 exs How do models perform on the small sample? This work: Can post-hoc explana5ons reveal generaliza5on failures? train labeled data unlabeled data system developer Figure 1: Our setting: a system developer is trying to evaluate a collection of trained models on a small amount of hand-labeled data to assess which one may work best in this new domain. Can baselines / attribu- tions help? uation sets tailored to these spurious signals, accu- rately understanding if a model is learning them remains a hard problem (Bastings et al., 2021; Kim et al., 2021; Hupkes et al., 2022). This paper addresses the problem of predicting whether a model will work well in a target domain given only a few examples from that domain. This setting is realistic: a system designer can typically hand-label a few examples to serve as a test set, but labeling …

Content truncated to 500 words