Tunes cling to the biggest names

Claim (verbatim)

The troubadour corpus survived twice over, and the second survival was an order of magnitude harsher than the first: roughly 2,500 lyric texts endure, but melodies for barely a tenth of them, carried by four notated witnesses - the Occitan-made R (Paris, BnF fr. 22543), the Lombard G (Milan, Ambrosiana R 71 sup.), and the French-made W (BnF fr. 844) and X (BnF fr. 20050), where Occitan songs ride inside trouvere books in gallicized dress. Losing a tune was easier than losing a text because notation demanded a rarer scribe and a rarer exemplar. This conjecture claims the second filter was not a lottery across the corpus but canon-shaped: a melody's chance of survival tracked the poet's textual mass, because a tune had to keep meeting music-literate copyists somewhere, and only the heavily anthologized had enough copies in enough places for that meeting to happen even once. At the extremes the gradient is stark - Guiraut Riquier, self-archived into R from his own dated book, keeps 48 melodies against 89 texts; Bernart de Ventadorn eighteen-odd; while the long tail of one- and two-song poets keeps essentially none. Melody survival should therefore be a steeply concentrating function of text survival: the tunes pooled where the names were already biggest, and the missing nine-tenths of the music is missing disproportionately from the small and middling names - the part of the tradition that was already thinnest.

Prediction clause (verbatim)

Prediction: in a BEdT join of per-poet text counts to melody attestations (van der Werf's The Extant Troubadour Melodies, 1984, as the control census), poets in the top decile by surviving text count will hold at least 60 percent of all melodies attributable to named poets, and their melody share will exceed their text share by at least 15 percentage points (primary clause: the 15-point concentration excess of top-decile melody share over top-decile text share; the verdict follows it). Secondary: at least 90 percent of named poets have zero surviving melodies. Melodies transmitted anonymously count toward no poet; the test voids if fewer than 200 melodies can be matched to named poets.

Kill-dataset (verbatim)

Kill: BEdT per-song records joined on attribution to the melody attestations of chansonniers R, G, W and X (van der Werf 1984 as control) - rank poets by text count and compare decile shares of texts and of melodies.

Novelty / leakage triage

already answered in the literature

Van der Werf's edition is arranged poet by poet and tabulates every extant melody per troubadour, and Aubrey gives the same arithmetic by manuscript and by name: some 250 melodies against roughly 2,500 texts, carried by R, G, W and X, with Guiraut Riquier's 48 (all in R) and Bernart de Ventadorn's 18-19 at the head and melodies surviving for only a few dozen named poets in all. That the tunes pooled on the heavily anthologized names, and that the overwhelming majority of named troubadours keep none, is stated ground in both; the decile join to BEdT text counts repackages published per-poet tables, so the primary clause's outcome is guaranteed rather than open.

• H. van der Werf (with G.A. Bond), The Extant Troubadour Melodies: Transcriptions and Essays for Performers and Scholars (Rochester NY, 1984)
• E. Aubrey, The Music of the Troubadours (Bloomington, 1996)

Predictions

No prediction registered yet.