Natural Language Inference Prompts for Zero-shot Emotion Classification in Text across Corpora

  • 2022-09-14 16:07:36
  • Flor Miriam Plaza-del-Arco, María-Teresa Martín-Valdivia, Roman Klinger
  • 7

Abstract

Within textual emotion classification, the set of relevant labels depends onthe domain and application scenario and might not be known at the time of modeldevelopment. This conflicts with the classical paradigm of supervised learningin which the labels need to be predefined. A solution to obtain a model with aflexible set of labels is to use the paradigm of zero-shot learning as anatural language inference task, which in addition adds the advantage of notneeding any labeled training data. This raises the question how to prompt anatural language inference model for zero-shot learning emotion classification.Options for prompt formulations include the emotion name anger alone or thestatement "This text expresses anger". With this paper, we analyze howsensitive a natural language inference-based zero-shot-learning classifier isto such changes to the prompt under consideration of the corpus: How carefullydoes the prompt need to be selected? We perform experiments on an establishedset of emotion datasets presenting different language registers according todifferent sources (tweets, events, blogs) with three natural language inferencemodels and show that indeed the choice of a particular prompt formulation needsto fit to the corpus. We show that this challenge can be tackled withcombinations of multiple prompts. Such ensemble is more robust across corporathan individual prompts and shows nearly the same performance as the individualbest prompt for a particular corpus.

 

Quick Read (beta)

loading the full paper ...