Insights from metrics (e.g. Kiparsky 1977, Hayes 1988, i.a.) andother artistic adaptations of language have long contributed to phonologicaltheory. This talk demonstrates how musical surrogate languages offer yetanother unexpected source of data for both phonological representations and thecognitive reality of grammatical architecture. I draw on the case study of theSambla balafon, a resonator xylophone with variations throughout WestAfrica. The Sambla are a small Mande ethnicity in Burkina Faso, whose complextonal language, Seenku (Western Mande, Samogo), is spoken by about 17,000people. The balafon surrogate language is a so-called “abridging system” (Stern1957), encoding certain phonological aspects (tone, vowel length, and wordstructure) to the exclusion of others (segmental information) to convey informationto listeners without ever “uttering” a word. Even amongst the encoded aspects,we find a division between lexical/morphological and postlexical toneprocesses, with the latter only variably encoded; this suggests that aseparation between grammatical components is accessible to musicians intransposing speech to musical form. Further, coda nasals, which are weaklyand/or variably encoded in the spoken language, are also variably encoded inthe surrogate language. I show how drawing together evidence from both thesurrogate and spoken language provides deeper insights into phonologicalstructure than either source of data alone. In sum, cases like the Samblatalking balafon highlight the symbiotic nature of theoretical linguistics,fieldwork, and language documentation, where seemingly peripheral genres ofspeech practices turn out to be invaluable tools for analyzing the language anddriving theory forward.