Jardine (2016): Computationally, tone is different #
This file formalizes [jardine-2016a], the computational characterization of a typological
asymmetry: unbounded circumambient processes, whose application depends on unboundedly distant
material on both sides of the target, (2), are common in tonal phonology and rare in segmental
phonology, and they are the attested maps beyond weak determinism. The witness is unbounded
tonal plateauing, every TBU between two H-toned TBUs surfacing H ([hyman-katamba-2010]'s rule,
(7)), as the string map (36) of Section 4.1 over H and Ø, which is Tone.utp and reproduces
the plateaus of Section 2.2 (utp_map_rows). The map is neither left- nor right-subsequential,
Section 4.2 (utp_not_isSubsequential), by bounded delay and the reversal symmetry; it is
regular, since the mark-up decomposition (43), a left-to-right pass writing ? after a H and a
right-to-left pass resolving ?, is a bimachine ([elgot-mezei-1965], utp_eq_resolve_mark,
utp_isBimachineComputable); and it is not weakly deterministic, Section 5.2, no union of
one-sided rules expressing a map that requires both sides (utp_not_weaklyDeterministic), so it
is fully regular, the class Section 5.3 places tone in and bars segmental phonology from
(utp_fullyRegular). Read back into autosegmental representations by (40), the OCP-merged
output is one H linked to the plateau, Section 4.4.
Implementation notes #
- Weak determinism is rendered as computability by a non-interacting bimachine
(
IsNonInteractingBimachineComputable), the class of [meinhardt-mai-bakovic-mccollum-2024]; the paper conjectures, after [heinz-lai-2013], that no mark-up-free decomposition of UTP exists, and under this rendering the conjecture is a theorem. - The rows write the paper's string representation one symbol per mora, a long vowel counting two; the Digo, Xhosa and Yaka data, whose plateaus interact with tone shift or an accentual analysis, are not encoded.
References #
- [jardine-2016a]
- [hyman-katamba-2010]
- [heinz-lai-2013]
- [elgot-mezei-1965]
- [meinhardt-mai-bakovic-mccollum-2024]
The plateauing data #
The paper's string representation of Section 4.1, one symbol per mora, H a TBU associated to a
H tone and Ø, written O, an unspecified one.
A row of Section 2.2: the underlying and surface TBU strings.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
A TBU string from its H/O spelling.
Equations
- Jardine2016a.tbuString s = List.map (fun (c : Char) => if c = 'H' then Tone.TBU.H else Tone.TBU.O) s.toList
Instances For
A row from the paper's features.
Equations
- Jardine2016a.Row.ofExample e = do let u ← e.feature? "underlying" let s ← e.feature? "surface" some { underlying := Jardine2016a.tbuString u, surface := Jardine2016a.tbuString s }
Instances For
The plateauing data of Section 2.2: Luganda (8) to (12), Zulu (18b) and Saramaccan (21).
Equations
- Jardine2016a.rows = List.filterMap Jardine2016a.Row.ofExample Jardine2016a.Examples.all
Instances For
The map (36) reproduces every row: no change with at most one H, a plateau between the outermost Hs otherwise.
UTP is not subsequential #
By bounded delay: a left machine reading H Øⁿ⁺¹ has emitted at most one symbol, since
utp.map (H Øⁿ⁺¹) = H Øⁿ⁺¹ and utp.map (H Øⁿ⁺¹ H) = Hⁿ⁺³ already differ at position 1;
so it withholds n + 1 symbols.
UTP is not left-subsequential (§4.2, online appendix).
UTP is not right-subsequential: by the reversal symmetry, a right machine faces the mirror-image unbounded look-ahead.
UTP is subsequential in neither direction.
UTP is not weakly deterministic #
Under the non-interacting-bimachine rendering of weak determinism, the conjecture of Section 5.2,
that UTP has no mark-up-free decomposition, is a theorem: UTP RequiresBothSides, which no union
of one-sided rules expresses.
UTP is not weakly deterministic (§5.2).
The (43) mark-up decomposition #
With one extra symbol the two-pass decomposition exists: a left pass marks every toneless
TBU after a H with ?; a right pass resolves ? by whether a H follows. The mark is
exactly the alphabet enlargement weak determinism disallows, so with the impossibility
theorem this locates UTP precisely.
Equations
- Jardine2016a.instDecidableEqMark x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- Jardine2016a.instReprMark = { reprPrec := Jardine2016a.instReprMark.repr }
Equations
- Jardine2016a.instReprMark.repr Jardine2016a.Mark.H prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Jardine2016a.Mark.H")).group prec✝
- Jardine2016a.instReprMark.repr Jardine2016a.Mark.O prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Jardine2016a.Mark.O")).group prec✝
- Jardine2016a.instReprMark.repr Jardine2016a.Mark.Q prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Jardine2016a.Mark.Q")).group prec✝
Instances For
Left pass of (43): mark every toneless TBU after a H with ?.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Right pass of (43), run right-to-left: resolve ? to H when a H follows, else to Ø.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The (43) decomposition computes UTP: mark left-to-right, then resolve right-to-left.
Both passes run finite Mealy machines, so this exhibits UTP as a right-subsequential map
after a left-subsequential one (Mealy.isLeftSubsequential_run,
Mealy.isRightSubsequential_runRight).
UTP is regular (§4.2): the (43) decomposition is a right-to-left Mealy pass after a left-to-right one, and such a composite is computed by a bimachine — one deterministic pass per direction ([elgot-mezei-1965]). So what fails above is one-directional determinism, not finite-state computability.
UTP is fully regular (§5.3): regular but not weakly deterministic — the class the paper places tone in and bars segmental phonology from.
The autosegmental reading (§4.4) #
The string representation reads back into autosegmental representations by TBU.toAR
((40)); the OCP-merged representation of the output has one H, linked exactly to the
plateau — [hyman-katamba-2010]'s rule as given in (7).
The merged output's links are the fused H over the surfacing positions.