The roles of linguistic, cognitive, and social-pragmatic processes in word learning are well established. If statistical mechanisms also contribute to word learning, they must interact with these processes; however, there exists little evidence for such mechanistic synergy. Adults use co-occurrence statistics to encode speech–object pairings with detailed sensitivity in stochastic learning environments (Vouloumanos, 2008). Here, we replicate this statistical work with nonspeech sounds and compare the results with the previous speech studies to examine whether exclusion constraints contribute equally to the statistical learning of speech–object and nonspeech–object associations. In environments in which performance could benefit from exclusion, we find a learning advantage for speech over nonspeech, revealing an interaction between statistical and exclusion processes in associative word learning.