ABSTRACT
Anti-Asian hate/toxic/racial speech has recently increased, especially in response to the outbreak of COVID-19. However, heavy focus on the COVID-19 context and subsequent Sinophobia may have over-represented Chinese-attacking phenomena rather than offering insights applicable to pan-Asian communities. To fill this gap, this paper underscores a cross-ethnic contextual understanding by taking a multi-ethnic approach to identifying and describing anti-Asian racism expressed in Twitter. The study examines (1) cross-ethnicity difference (or similarity) of anti-Asian expressions; (2) temporal dynamics of cross-ethnicity difference/similarity; (3) topical contexts underlying anti-Asian tweets; and (4) comparison between Sinophobic tweets and pan-Asian or non-Chinese Asian targeting tweets. The study uses a 12 month-long large-scale tweets that contain ethnicity-indicative search keywords, from which anti-Asian tweets are identified using deep-learning models. Multiple correspondence analysis and topic modeling are employed to address research questions. Results show anti-Asian speeches are fragmented into several distinctive groups, and such groupings change dynamically in response to political, social, and cultural reality surrounding ethnic relations. Findings suggest that China is not a representative entity for pan-Asian ethnicity: Most of the times during the observed period, anti-China tweets are positioned distantly from generically mentioned anti-Asian tweets in the $n$-gram-based multidimensional space. Moreover, generic anti-Asian tweets show greater topical similarities to the combined set of tweets that attack other Asian ethnic groups than to the anti-China tweets. Pan-Asian tweets and China-specific tweets become semantically similar in the aftermath of the pandemic outbreak (from Feb. 2020 to Apr. 2020) yet only temporarily.