Some of the psychological abilities that underlie human speech are shared with other species. One hallmark of speech is that linguistic context affects both how speech sounds are categorized into phonemes, and how different versions of phonemes are produced. We here confirm earlier findings that swamp sparrows categorically perceive the notes that constitute their learned songs and then investigate how categorical boundaries differ according to context. We clustered notes according to their acoustic structure, and found statistical evidence for clustering into 10 populationwide note types. Examining how three related types were perceived, we found, in both discrimination and labeling tests, that an "intermediate" note type is categorized with a "short" type when it occurs at the beginning of a song syllable, but with a "long" type at the end of a syllable. In sum, three produced note-type clusters appear to be underlain by two perceived categories. Thus, in birdsong, as in human speech, categorical perception is context-dependent, and as is the case for human phonology, there is a complex relationship between underlying categorical representations and surface forms. Our results therefore suggest that complex phonology can evolve even in the absence of rich linguistic components, like syntax and semantics.bird song | categorical perception | phonology | speech perception A lthough language as a whole is unique to humans, some of its components, particularly in the domains of phonology and phonetics, are shared with other species (1-4). This sharing has often been demonstrated in tests of animals' perception of human speech, but occasionally an animal's own communication system may reveal speech-like traits. One notable case is bird song, which is learned by imitation and the development of which shares several other features with speech (1, 2, 5). Because songs are constructed hierarchically from smaller units (Fig. 1A), birds may possess a "phonology" similar to that of human language (2-4).In speech, words and other linguistic units are composed from smaller units called phonemes, which are themselves categories, each encompassing a range of acoustic variants. Phonemes typically exhibit a degree of categorical perception: that is, individuals label continuously varying stimuli as belonging to discrete categories with abrupt boundaries between them, and discriminate stimuli that span these boundaries more readily than stimuli within one category. Categorical boundaries between phonemes are learned early in life and shared within a speech community (6, 7). It has become clear, however, that entirely categorical perception (where individuals can perceive variation only between categories) rarely, if ever, applies in speech. One key departure from the ideal is that categorical perceptual boundaries between phonemes vary with linguistic context (8-10), such as the position of the sound (11,12).Categorization is essential for the linguistic functions of speech, allowing categorical distinctions between discrete words...