Natural language is often considered fundamental to mathematical thinking, a view supported by research on language-of-training effects in bilinguals. However, these effects have been primarily examined in arithmetic. While previous research has explored language effects in algebra, it has largely focused on well-established rules. This study extends prior work by investigating whether similar effects apply to learning of algebra with unfamiliar rules. Thirty-nine Chinese-English bilingual undergraduates were trained to solve arithmetic and algebraic problems in Chinese or English and were later tested on both old and novel problems in both languages. Consistent with previous findings, results revealed a clear dissociation between arithmetic and algebra. Bilinguals responded faster in the trained versus the untrained language for arithmetic problems and solved old arithmetic problems faster than novel ones. However, these effects were absent in algebra, suggesting that algebraic learning does not necessarily depend on natural language encoding.