There is substantial variation in language experience between learners, yet there is surprising similarity in the language structure they eventually acquire. While it is possible that this canalisation of language structure may be due to constraints imposed by modulators, such as an innate language system, it may instead derive from the broader, communicative environment in which language is acquired. In this paper, the latter perspective is tested for its adequacy in explaining the robustness of language learning to environmental variation. A computational model of word learning from cross-situational, multimodal information was constructed and tested. Key to the model’s robustness was the presence of multiple, individually unreliable information sources that could support learning when combined. This “degeneracy” in the language system had a detrimental effect on learning when compared to a noise-free environment, but was critically important for acquiring a canalised system that is resistant to environmental noise in communication.