Timeline Counterpoint: Sub-Saharan Africa

Companion to Chapter 2: Sub-Saharan Africa. This page assumes you know what the gankogui bell and E(7,12) are, and states the counterpoint rules explicitly. Everything below is summarised from the published scholarship listed under Sources — prior research restated in Poly’s terms, not original theory. See About This Guide for what this guide is, what it is not, and what it leaves out.

The System

West African ensemble drumming — this page draws principally on Ewe practice (Ghana/Togo) and Manding practice (Mali/Guinea) — is organised around a timeline: a short, asymmetric ostinato, usually on a bell or other cutting timbre, that functions as the ensemble’s shared clock (Arom 1991). The timeline is not a melody and not a solo; it is a referent. Every other part is defined, learned, and heard by its relationship to the timeline, not by its relationship to an abstract “beat one” (Arom 1991; Agawu 2006).

Beneath the timeline sits a second referent that Western transcription usually omits — the analytical move is Arom’s (Arom 1991): the dance beat — the regular footfalls of the dancers, typically four even beats per 12-pulse cycle. The characteristic tension of the music is the continuous cross-rhythm between the ternary and binary readings of the same 12 pulses — the 3:2 relationship that Novotney argues is the generative core of the whole system (Novotney 1998).

The Rules

1. The timeline never varies. Not in pattern, not in dynamics, not in placement. The bell player’s virtue is constancy; variation in the timeline is not expressive, it is an error. (In Poly: timeline mode, mutation 0, no phrase gating, no envelope on the bell lane.)

2. Every part has a fixed point of entry against the timeline. Parts are learned as “start here relative to the bell,” and that phase relationship is part of the part’s identity. Shifting a support drum’s entry point does not create a variation — it creates a different (usually wrong) part. (In Poly: rotation is identity, not decoration. Set it once per lane and leave it.)

3. Support parts are placed off the dance beat, systematically. Locke’s analysis of southern Ewe drumming shows support patterns consistently displaced onto offbeats relative to the dance beat, creating sustained cross-rhythm rather than occasional syncopation (Locke 1982). The displacement is structural: the part lives offbeat; it does not visit.

4. The 3:2 cross-rhythm is continuous, not an effect. At any moment the texture supports both a ternary (12 = 3×4) and binary (12 = 4×3, or 6×2) hearing. Idiomatic parts commit to one stratum each; the composite carries the ambiguity. A single part that flip-flops between strata sounds amateur.

5. The variation budget is concentrated in the lead drum. Timeline: zero variation. Support drums (kidi, sogo; sangban, dununba): small sanctioned variations from a known set, at low rate. Lead drum (atsimevu; djembe soloist): large vocabulary of phrases, freely sequenced — but drawn from a shared lexicon, cued to dancers, and cadencing onto the cycle’s convergence point (Chernoff 1979). Improvisation means phrase selection and timing, not free invention.

6. Lead phrases are cycle-shaped. A lead phrase occupies one, two, or four timeline cycles and resolves at the cycle boundary. Mid-cycle entries are common; mid-cycle endings are not — the phrase lands where the cycles agree (Anku 2000 models this formally as rotation within a circle that must close).

7. Register and rate separate the voices. Each part occupies a distinct combination of timbre, register, and note-rate: bell high and sparse-but-cutting, dunun low and sparse, accompaniment djembes mid and continuous, solo djembe high and bursty. Two parts in the same stratum at the same rate is doubling, and doubling is avoided — every voice must be individually audible.

8. Subdivision is felt, not isochronous. Measured Malian jembe playing shows a stable short-medium-long subdivision profile that persists across tempi and players — a feel that is part of the style’s identity, not expressive noise (Polak 2010; Polak & London 2014). Straight-grid renditions are recognisably foreign to practitioners.

9. The ensemble converses. Call-and-response operates at every scale: bell calls, drums respond; lead calls, chorus drum answers; drummers call, dancers answer. Structurally: when the lead speaks, support parts do not simultaneously vary. One voice varies at a time.

Construction

A procedure for building an Ewe-style texture that respects the rules:

1. Lay the timeline. 12 steps, E(7,12), rotation 0, 1/8 subdivision, timeline mode on, velocity high. This is Rule 1’s fixed clock.
2. Lay the dance beat. A low drum at 12 steps, 4 hits (E(4,12) = every third pulse), moderate velocity. You now have the 3:2 matrix (Rule 4): the bell articulates the ternary hearing, the feet the binary one.
3. Add one support drum per stratum. Each at 12 steps (same cycle as the bell — Ewe support parts usually share the timeline’s cycle), each with a rotation chosen so its hits fall between dance beats (Rule 3), each in a distinct register (Rule 7). Ghost values low; mutation 0–10%.
4. Add the lead. Different cycle length is permissible here (7 or 5 steps for a roaming, phrase-like quality) or same-cycle with high mutation. Give this lane the entire variation budget: mutation 15–30%, phrase gating (Length one or two cycles, Gap one cycle) so it speaks and falls silent (Rules 5, 6, 9).
5. Apply the feel. No swing. Small per-lane timing offsets (support lanes a few ms apart), and a light Humanize (≤0.15) as an approximation of Rule 8 — but be clear what kind of approximation it is. Humanize applies random jitter around each onset. Rule 8’s profile is a systematic profile: particular positions, particular ratios, stable across tempi and players. Jitter reproduces the presence of non-isochrony, not its shape, and a straight grid plus noise is still not the feel. Poly ships one. The subdivision profile on each lane of the Manding Djembe preset encodes the long-first beat subdivision Polak 2010 measures, as ratios rather than offsets — which matters, because a ratio is what survives a tempo change. Per-step micro-timing is a different tool and a worse fit for this: it is absolute milliseconds, clamped to ±20, so a profile entered there would hold only at the tempo it was measured at. Use micro-timing for a player’s own lean against the beat; use the profile for the meter’s shape. The shipped values render the shape Polak reports, not a transcription of his published ratios.

Rule-Checked Ewe Texture

Lane	Role	Steps	Hits	Rotation	Subdivision	Note	Velocity	Ghost	Mutation	Phrase Len	Gap
1	Gankogui (timeline)	12	7	0	1/8	58	110	0	0%	0	0
2	Dance beat (low)	12	4	0	1/8	41	95	0	0%	0	0
3	Kidi (support, offbeat)	12	4	1	1/8	50	80	30	8%	0	0
4	Sogo (support, offbeat)	12	4	2	1/8	45	85	25	8%	0	0
5	Atsimevu (lead)	7	5	0	1/8	55	90	45	25%	12	6

Check the rules against the patch: lanes 1–2 fixed (Rules 1–2), lanes 3–4 rotated off the dance beat with a small budget (Rules 3, 5), lane 5 holds the budget and is phrase-gated so it converses rather than chatters (Rules 5, 6, 9), and every lane occupies its own register via distinct MIDI notes (Rule 7).

For a Manding-style texture, keep all lanes on a shared cycle length and express the hierarchy through density and register instead (dunun sparse/low, sangban mid, djembe dense/high) — hocketing within one metric frame rather than stacked cycles.

What Breaks the Idiom

• A varying timeline. Mutation, density modulation, or phrase gating on the bell lane. The single most damaging error.
• Western downbeat accenting. Slamming velocity onto pulse 0 of every cycle. The cycle has a convergence point, not a backbeat; weight comes from coincidence of parts, not from one loud hit.
• Uniform variation. Mutation spread evenly across all lanes. The result is texture soup — no fixed ground for the varying voice to speak against (violates Rules 5 and 9).
• Stratum collisions. Two lanes at the same rate in the same register. Merge them or separate them.
• Swing. The 12-pulse frame’s interest is the 3:2 ambiguity; binary swing flattens it. The correct feel is the non-isochronous profile of Rule 8, which is not swing.

Sources

Primary: Arom 1991 (referent theory), Locke 1982 (offbeat timing), Locke 1998 (complete transcriptions of a full Ewe repertoire piece), Chernoff 1979 (ensemble roles and conversation), Anku 2000 (circular structure), Novotney 1998 (3:2 foundation), Polak 2010 / Polak & London 2014 (jembe timing), Agawu 2006 (a critical check on over-formalising the “standard pattern”). See also refs 4, 6, 7, 8, 9.

Attribution: Rules 1–6 and 8 restate claims made directly in the cited sources. Rules 7 (register/rate separation) and 9 (one voice varies at a time) are this guide’s practical distillations of the ensemble practice described across Chernoff 1979 and Locke 1998, rather than statements any single source makes in that form. All Poly parameter mappings and patch values are ours.