Reproduce “The Call for Socially Aware Language Technologies” on OpenResearch

About this paper

Diyi Yang, Dirk Hovy, David Jurgens, Barbara Plank

The paper by Yang et al. conceptualizes "socially aware language technologies," arguing that a fundamental lack of such awareness underlies many pervasive problems in modern Natural Language Processing, especially with large language models. It proposes a three-layered framework for integrating social factors, interaction dynamics, and societal implications directly into language technology design and evaluation.

Key insights

• A core deficit in modern NLP, particularly exacerbated by LLMs, is a fundamental lack of social awareness, which is a common underlying cause for issues like bias and toxicity.
• Socially aware language technologies represent a distinct and broader paradigm that integrates understanding of social factors and societal implications directly into model design, differentiating it from related concepts like personalization or computational sociolinguistics.
• Current LLMs, despite advanced linguistic capabilities, are not inherently socially aware and require significant progress in operationalization, alignment with social science theories, and explicit contextual reasoning.

Summary by alphaXiv.