Dirk Hovy

Dirk Hovy is a leading researcher in computational linguistics and natural language processing, focusing on social bias, ethical AI, and sociodemographic factors in language technology. His work bridges NLP with social science, examining how socioeconomic status, gender, and cultural aspects influence both model performance and human-AI interaction.

• Key Research Themes: Socioeconomic bias in NLP systems, emotion analysis frameworks, hate speech detection, multilingual speech recognition disparities, and safety mechanisms for conversational AI.
• Collaborative Efforts: Works extensively with teams across Europe and North America, co-authoring studies on topics from Italian Twitter demographics (DADIT dataset) to LGBTQIA+ representation in language models.
• Methodological Contributions: Proposes novel evaluation frameworks like XSTest for safety behaviors, BooStSa for model significance testing, and Twitter-Demographer for data enrichment.
• Ethical Advocacy: Promotes inclusive language technology through research on neopronouns, religious bias mitigation, and value-sensitive design principles for safer AI systems.

His recent publications explore the limitations of current demographic adaptation techniques in transformers, the reproduction of gendered emotion stereotypes in LLMs, and the urgent need for socio-economic class inclusion in NLP research. His work consistently emphasizes the intersection between technical innovation and societal impact.