Plural marking systems in natural languages follow an Animacy Hierarchy such that although it is common for animate nouns to receive plural markers while inanimate nouns are left unmarked, the reverse pattern is never attested. A hypothesis that has received wide empirical support is that underlying biases of learners during language learning shape language typology: linguistic patterns easily learned by language learners are promoted, while those that are unlearnable are eliminated. This study asked whether learners have underlying biases which would contribute to the settlement of the typologically universal Animacy Hierarchy. With an artificial language learning paradigm, this study showed that learners biases do not align with the universally-attested Animacy Hierarchy: learning a language that violates the Animacy Hierarchy is in fact easier for adult learners and they view shifting a language with a probabilistic plural-marking system towards a pattern that violates the hierarchy as a viable option as well. However, adult learners are more successful in probability-matching on animate trials than inanimate trials, which suggests high salience of animate tokens during the rule learning process. Our preliminary data on children suggest that such an animacy-conditioned grammatical system is hard for children aged 5-8 to fully acquire in a single-session learning task and that for those children who have acquired the pattern, no convincing evidence suggests a learning bias favoring the language that follows the typological pattern either. Taken together, it is observed in the current study that patterns that violate Animacy Hierarchy is not only not disfavored by learners, but sometimes preferred. More investigation is underway to whether native language influence or animacy manipulation gave rise to the opposite pattern. Our results suggest that, in the case of Animacy Hierarchy, other factors during the process of language transmission and language use, such as cognitive and pragmatic salience animate objects, may have played stronger roles in shaping linguistic typology, potentially overriding the learning bias observed in the opposite direction.