The Importance of Modeling Social Factors of Language: Theory and Practice

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 588–602 June 6–11, 2021. ©2021 Association for Computational Linguistics 588 The Importance of Modeling Social Factors of Language: Theory and Practice Dirk Hovy Bocconi University Via Sarfatti 25 20136 Milan, Italy dirk.hovy@unibocconi.it Diyi Yang Georgia Institute of Technology CODA Tech Square Atlanta, GA, 30308 dyang888@gatech.edu Abstract Natural language processing (NLP) applica- tions are now more powerful and ubiquitous than ever before. With rapidly developing (neural) models and ever-more available data, current NLP models have access to more infor- mation than any human speaker during their life. Still, it would be hard to argue that NLP models have reached human-level ca- pacity. In this position paper, we argue that the reason for the current limitations is a fo- cus on information content while ignoring lan- guage’s social factors. We show that current NLP systems systematically break down when faced with interpreting the social factors of language. This limits applications to a subset of information-related tasks and prevents NLP from reaching human-level performance. At the same time, systems that incorporate even a minimum of social factors already show re- markable improvements. We formalize a tax- onomy of seven social factors based on lin- guistic theory and exemplify current failures and emerging successes for each of them. We suggest that the NLP community address so- cial factors to get closer to the goal of human- like language understanding. 1 Introduction “[T]he common misconception [is] that language use has primarily to do with words and what they mean. It doesn’t. It has primarily to do with people and what they mean.” Clark and Schober (1992) Until the 1970s, economics assumed that in- dividuals, markets, and firms always acted ratio- nally, based on all the available information. This assumption allowed researchers to use linear mod- els and worked well for several applications. How- ever, it came at the cost of ignoring essential as- pects of human decision making, which oversim- plified an inherently complex matter in a way that limited possible insights and applications. The seminal work by Tversky and Kahneman (1973) showed that people would make irrational deci- sions, time and again, even with full information, and that simple models could not account for this behavior. By introducing the human factor into the equation, they opened up a new research field: behavioral economics. Like economics in the mid-twentieth century, Natural Language Processing (NLP) still makes a limiting assumption: language is only about infor- mation, i.e., message content alone. This assump- tion makes it possible to model language statisti- cally and works for several applications. However, it completely ignores the fact that people use lan- guage to achieve (social) goals; like economists before 1973, NLP researchers are oversimplifying an inherently complex matter in a way that lim- its possible insights and applications. And like in- troducing behavior transformed economics, intro- ducing social factors into NLP will similarly trans- form the field: it will open up …

Content truncated to 500 words