Text-based inference of moral sentiment change

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 4654–4663, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 4654 Text-based inference of moral sentiment change Jing Yi Xie*, Renato Ferreira Pinto, Jr.*, Graeme Hirst , and Yang Xu Department of Computer Science, University of Toronto, Toronto, Canada jingyi.xie@mail.utoronto.ca, {renato,gh,yangxu}@cs.toronto.edu Abstract We present a text-based framework for inves- tigating moral sentiment change of the pub- lic via longitudinal corpora. Our framework is based on the premise that language use can inform people’s moral perception toward right or wrong, and we build our methodol- ogy by exploring moral biases learned from diachronic word embeddings. We demon- strate how a parameter-free model supports inference of historical shifts in moral sen- timent toward concepts such as slavery and democracy over centuries at three incremen- tal levels: moral relevance, moral polarity, and fine-grained moral dimensions. We ap- ply this methodology to visualizing moral time courses of individual concepts and analyzing the relations between psycholinguistic vari- ables and rates of moral sentiment change at scale. Our work offers opportunities for apply- ing natural language processing toward char- acterizing moral sentiment change in society. 1 Moral sentiment change and language People’s moral sentiment—our feelings toward right or wrong—can change over time. For in- stance, the public’s views toward slavery have shifted substantially over the past centuries (Old- field, 2012). How society’s moral views evolve has been a long-standing issue and a constant source of controversy subject to interpretations from social scientists, historians, philosophers, among others. Here we ask whether natural lan- guage processing has the potential to inform moral sentiment change in society at scale, involving minimal human labour or intervention. The topic of moral sentiment has been thus far considered a traditional inquiry in philosophy (Hume, 1739; Smith, 1759; Kant, 1785), with con- temporary development of this topic represented ∗Equal contribution. in social psychology (Piaget, 1932; Kohlberg, 1969; Stigler et al., 1990; Fiske and Taylor, 1991; Pizarro and Bloom, 2003), cognitive lin- guistics (Lakoff, 1996), and more recently, the advent of Moral Foundations Theory (Haidt and Joseph, 2004; Haidt et al., 2007; Graham et al., 2013). Despite the fundamental importance and interdisciplinarity of this topic, large-scale formal treatment of moral sentiment, particularly its evo- lution, is still in infancy from the natural lan- guage processing (NLP) community (see overview in Section 2). We believe that there is a tremendous poten- tial to bring NLP methodologies to bear on the problem of moral sentiment change. We build on extensive recent work showing that word em- beddings reveal implicit human biases (Boluk- basi et al., 2016; Caliskan et al., 2017) and so- cial stereotypes (Garg et al., 2018). Differing from this existing work, we demonstrate that moral sen- timent change can be revealed by moral biases implicitly learned from diachronic text corpora. Accordingly, we present to our knowledge the first text-based framework for probing moral sen- timent change at a large …

Content truncated to 500 words