This dissertation investigates scientific progress and epistemic risks in a case in which formal models are transferred from linguistics to biology. Initially constructed in linguistics to study natural language, Formal Language Theory (FLT) is a mathematical theory of computation that has been applied to research in comparative cognitive biology. Consisting of formal models of languages, FLT provides a basis for ranking computational complexity, known as the Chomsky Hierarchy. Based on FLT, Tecumseh Fitch and fellow comparative biologists have designed artificial languages and tested the ability of human and nonhuman animals to learn languages of varying complexity. I argue that even though certain instrumental progress has been made by introducing FLT to biology, explanatory progress has been limited. Moreover, testing for the ability to learn an artificial language requires one to ‘embody’ the language in some manner or other, and choices about how to do so could lead to bias in the results of the test. However, due to discrepancies in the details concerning how the languages are embodied and exposed to human and nonhuman animal subjects, the results of learning need to be taken with a grain of salt.