Antiracist corpora-collections of texts explicitly produced to challenge or oppose racismare generally assumed to be free of discriminatory or assimilationist logic. However, prior qualitative research suggests that even well-intentioned antiracist writing can inadvertently reproduce racist assumptions, particularly through appeals to cultural assimilation that center majority norms. This study proposes a computational approach to systematically detect such hidden discourse. Using a combination of keyword analysis, sentiment classification, and contextual word embedding models (e.g., word2vec or BERT), the method scans antiracist texts for lexical and semantic patterns commonly associated with racial discrimination and assimilationist pressure. The objective is to identify statistical anomalies, contradictory framings, and implicit biases that escape manual reading. Findings from this computational analysis could inform best practices for antiracist writing, editorial review, and corpus curation.