Value conflicts in natural language processing: A systematic literature review

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Despite broad consensus among social scientists that technology embodies values, scholarship is conspicuously lacking in reflexive analysis of the value-based implications of natural language processing (NLP) technologies. To explore embedded values and potential harms in the NLP domain, a systematic literature review grounded in established value frameworks focused on how uninformed design decisions can perpetuate algorithmic biases and lead to discriminatory outcomes. Analysis applying Schwartz's theory of basic values to a sample of 60 articles indexed by the Association for Computing Machinery and Institute of Electrical and Electronics Engineers, published between 2010 and 2023, identified underrepresented values such as stimulation and tradition, and it detected the complete absence of attention to hedonism and benevolence. The omission of scholarly attention to hedonism is particularly surprising, in light of its relevance for social media services; this points to a potential oversight in enhancing personal enjoyment and quality of life. Benevolence's absence indicates unexplored areas of communal application. The study highlights a need for more comprehensive value-based analysis in NLP research, to ensure ethical and inclusive technological development. Highlighting this gap should contribute to cross-discipline dialogue among researchers, technologists, and policymakers, for NLP technologies better aligned with diverse human values.