Statistical learning (SL) approaches to reading maintain that proficient reading requires assimilation of the rich statistical regularities in the writing system. Reading skills in developing first- and second-language readers in English have been shown to be predicted by individual differences in sensitivity to regularities in mappings from orthography to phonology (O-P) and semantics (O-S), with good readers relying more on O-P consistency, and less on O-S associations. However, SL and its relation to reading has been primarily studied in English readers in high-income Western countries. We examine individual differences in sensitivity to regularities in emergent French readers in rural agricultural communities in Côte d'Ivoire (N = 134). We show that, in contrast to previous studies, in this cohort better readers are leveraging semantic associations more strongly, while individual differences in sensitivity to orthographic consistency did not predict reading skill. Relatively little variance in reading skill was explained by sensitivity to regularities, and we discuss these findings in terms of literacy acquisition in low-literacy and low-exposure contexts. This showcases the importance of cross-linguistic and cross-cultural research to back up universal theories of literacy, and suggests that current SL accounts of reading must be updated to account for this variance in reading skills.