Worked cases

What people do with an interlingua: carry an analysis out to every encoding that will take it, and read the corpora research actually ships in without a parser per corpus. Each runs as a tour in the repository, and the output below is what it printed when this page was built.

From a score with no analysis to every encoding

A Bach chorale out of music21's corpus: four voices, a key signature, and not one harmonic label in the file. The chain is analyse, carry, write — and the only part that is HAMON's is the carrying. music21 does the musicology; FlexOHR (DCMLab) is the object model at the far end. HAMON is what lets them be swapped without the harmony quietly degrading on the way.

  4 voices, 165 notes, 0 harmony labels

music21 says: f# minor
----------------------
  III bVII6 i bVII6 III bVII6 III III6 bVII bVII7 III V6

HAMON
-----
  12 labels over one region: F# minor

what each encoding cannot say
-----------------------------
  hamon      nothing
  dcml       nothing
  dezrann    roman fn. x12, 7th/tensions x6, key/region x1
  mei        roman fn. x12, key/region x1
  harte      roman fn. x12, key/region x1
  jams       roman fn. x12
  musicxml   key/region x1
  lilypond   roman fn. x12, key/region x1
  abc        roman fn. x12, key/region x1
  musescore  roman fn. x12, key/region x1
  romantext  nothing
  humdrum    nothing
  ireal      roman fn. x12, key/region x1

RomanText
---------
  Time Signature: 4/4
  m1 f#: III
  m2 bVII6
  m3 i
  m4 bVII6
  m5 III

HAMON → FlexOHR
---------------
  III     → quality=major_triad            inversion=root_position
  bVII6   → quality=major_triad            inversion=first
  i       → quality=minor_triad            inversion=root_position
  bVII6   → quality=major_triad            inversion=first

The split in that table is clean because the input is a pure Roman analysis. The formats built for functional analysis carry every bit of it; the formats built for chord symbols keep no numerals at all, so for them the analysis is the loss. Run the same chain on a lead sheet and the table turns over, which is the argument the ICCCM'26 page makes at length.

The corpora HAMON reads

The case above leans on the same fact: the formats research actually ships in are readable without a bespoke parser per corpus. HAMON reads 22 formats and writes 13. The other 9 are read-only dialects that nothing writes back — each arrived with its own paper, its own table layout and its own parser.

Distant Listening Corpus  dcml            I V64 I6 V7/V V
Jazz Harmony Treebank     treebank        Dm7 G7 Cmaj7 A7 Dm7
Key / Modulation (DDMAL)  key_modulation  I IV V I IV
DiLeMMa pitch array       dilemma         I V7/V V   (4 notes -> 3 harmonies)

22 formats in, 13 out. The other 9 are read-only dialects nothing writes back.

FlexOHR (DCMLab)          A7[of:ii] -> C(SPC(A)) dominant_seventh root_position
                                       P1 M3 P5 m7
                          back -> A major dom7   (the glyphs and [of:ii] stay behind)

Among them: the Distant Listening Corpus (Hentschel, Moss, Neuwirth, Rohrmeier), the Jazz Harmony Treebank (Harasim, Finkensiep et al.), DiLeMMa (Hentschel et al.) whose note-level pitch arrays collapse back to one label per harmony, and the DDMAL key/modulation set. The full list, with authors, licenses and download recipes, is the dataset manifest.