[Paper Review] The Call for Socially Aware Language Technologies

Diyi Yang, Dirk Hovy+2 more|arXiv (Cornell University)|May 3, 2024

Innovative Teaching and Learning MethodsPsychology3 citations

TL;DR

This paper calls for integrating social awareness into natural language processing (NLP) to address persistent issues like bias, toxicity, and fairness by embedding understanding of social factors, interactions, and implications into language models. It proposes a foundational shift toward socially aware NLP, emphasizing emotional intelligence, cultural context, and ethical design to create more inclusive, trustworthy, and human-centered AI systems.

ABSTRACT

Language technologies have made enormous progress, especially with the introduction of large language models (LLMs). On traditional tasks such as machine translation and sentiment analysis, these models perform at near-human level. These advances can, however, exacerbate a variety of issues that models have traditionally struggled with, such as bias, evaluation, and risks. In this position paper, we argue that many of these issues share a common core: a lack of awareness of the factors, context, and implications of the social environment in which NLP operates, which we call social awareness. While NLP is getting better at solving the formal linguistic aspects, limited progress has been made in adding the social awareness required for language applications to work in all situations for all users. Integrating social awareness into NLP models will make applications more natural, helpful, and safe, and will open up new possibilities. Thus we argue that substantial challenges remain for NLP to develop social awareness and that we are just at the beginning of a new era for the field.

Motivation & Objective

• Address the growing gap in NLP systems' ability to handle social context, cultural nuance, and ethical implications despite high performance on formal linguistic tasks.
• Highlight that current NLP models lack social awareness, leading to biased, unsafe, or exclusionary outcomes in real-world applications.
• Advocate for a paradigm shift in NLP research toward embedding social awareness as a core component, not an afterthought.
• Promote interdisciplinary collaboration across linguistics, social sciences, and AI to build models that reflect diverse human experiences and values.
• Call for ethical development frameworks to prevent misuse, over-reliance, and privacy risks in socially aware systems.

Proposed method

• Introduce the concept of 'social awareness' as a core capability encompassing social factors, social interaction, and social implications in language use.
• Frame social awareness as an extension of emotional intelligence, particularly Theory of Mind, to enable models to understand others' emotions and intentions.
• Propose that socially aware NLP must account for cultural norms, values, power dynamics, and contextual variation across languages and communities.
• Advocate for redefining evaluation metrics and benchmarks to include fairness, trust, and inclusivity beyond traditional accuracy and fluency.
• Call for integrating insights from sociology, pragmatics, and human-computer interaction into NLP model design and training.
• Emphasize the need for continuous monitoring and adaptive updates to ensure systems remain relevant and ethical across time and contexts.

Experimental results

Research questions

• RQ1 How can NLP models be designed to recognize and respond appropriately to social context, cultural differences, and power dynamics in language?
• RQ2 What are the key components of social awareness that must be embedded in language models to improve fairness, trust, and inclusivity?
• RQ3 How can social awareness be operationalized and evaluated in NLP systems, given its complex, multidimensional nature?
• RQ4 What ethical risks emerge when NLP systems become more socially aware, and how can they be mitigated during development?
• RQ5 In what ways can socially aware NLP transform high-stakes domains like healthcare, education, and human-robot interaction?

Key findings

• Many issues in modern NLP—such as bias, toxicity, and unfairness—stem from a lack of social awareness, not just linguistic or computational shortcomings.
• Current large language models perform near-human levels on formal NLP tasks but fail to account for social context, leading to exclusionary or harmful outputs.
• Social awareness is not limited to NLP but is essential across AI modalities, including vision and robotics, to enable safe and effective human-AI interaction.
• Integrating social awareness into NLP can enhance trust, inclusivity, and accessibility, especially for underrepresented languages and communities.
• The absence of standardized metrics and evaluation frameworks for social awareness remains a major barrier to progress.
• There is a critical need for interdisciplinary collaboration and ethical design principles to responsibly develop socially aware systems that avoid manipulation and over-reliance.