ABSTRACT

Moral Foundations Theory (MFT) has attracted a great deal of attention in Natural Language Processing (NLP) and social media analysis, including both applications that attempt to infer moral values inference from text, or otherwise use moral foundations information to perform another downstream task. In this work, we address the issue of moral foundations inference from text data according to two perspectives, hereby called text- and author-dependent classification, by presenting a number of experiments to compare traditional text classifiers with more recent approaches based on pre-trained language models in both English and Portuguese languages. Results suggest that moral foundations classification relies heavily on lexical information, and that different models may be more suitable to each task, and leave a number of opportunities for further research in the field.

摘要

道德基础理论 (MFT) 在自然语言处理 (NLP) 和社交媒体分析中引起了极大的关注，包括试图从文本中推断出道德价值或以其他方式使用道德基础信息来执行另一项下游任务的应用程序。在这项工作中，我们通过提出一些实验来比较传统文本分类器与基于预训练的最新方法英语和葡萄牙语的语言模型。结果表明，道德基础分类严重依赖词汇信息，不同的模型可能更适合每项任务，