Design Opportunities for Explainable AI Paraphrasing Tools: A User Study with Non-native English Speakers

User needs vary by language proficiency level: The study found that non-native English speakers with different proficiency levels preferred different types of explanatory information, with less proficient users starting with global information before moving to detailed analysis. This suggests one-size-fits-all AI explanations may not serve all users equally well.

Multiple explanations work better than single ones: While back-translation was the most frequently used aid, users didn't rely on it alone to decide whether to accept AI suggestions. Instead, they combined multiple information sources (back-translation, explanations, and usage examples) to make confident decisions, highlighting the value of diverse explanation types in AI tools.

Explainable AI can build user confidence and autonomy: The ParaScope tool demonstrated that providing transparent information about how AI generates paraphrases enhanced users' confidence in their writing decisions and gave them greater autonomy—rather than blindly accepting AI suggestions, users could make informed choices aligned with their needs.

Design must balance helpfulness with cognitive load: While multiple explanatory aids improved decision-making, the study emphasizes that "thoughtful design to prevent information overload" is critical. Too many explanation options could overwhelm users rather than empower them.

Explainable AI has practical benefits for vulnerable user groups: For non-native English speakers specifically, explainable paraphrasing tools improved both writing efficiency and user confidence, suggesting that transparency in AI systems can be particularly valuable for users who may already feel less certain about language choices.