Taraldsen, Taraldsen Medová & Langa (2018): class prefixes as specifiers #
[TTML18] (NLLT 36, 1339–1394) analyzes Southern Bantu noun-class prefixes as morphemes lexicalizing phrasal Specifiers built around a silent classifier-like noun Nₓ — [Nₓ] for singulars, [# Nₓ] for plurals ((48)–(50), pp. 1357–1358) — rather than single functional heads. Agreement with conjoined singular subjects (§2) shows which singular/plural class pairs share one N (Xhosa: 1/2, 7/8, 9/10) and which pair distinct Ns (3/4, 5/6). Where the plural entry's N is distinct and the language cannot merge it directly with the root, the Foot Condition blocks direct spellout and Starke-style backtracking yields prefix stacking: Changana/Rhonga mi-mu-twa 'thorns', ma-rhi-tu 'words', ti-yi-n-dlu 'houses' ((62), p. 1361, from [Bac06]). Xhosa instead first-merges the plural N ((82)–(83), p. 1366) and never stacks.
Main declarations #
NCFeature: number (#) and classifier (Nₓ) featuresxhosaSg/xhosaPl,rhongaSg/rhongaPl: the Xhosa and Changana/Rhonga lexica ((41), (48)–(50), (76)–(80))SharesClassifierN: a singular/plural entry pair shares its classifier NderivePlural: cyclic spellout with last-resort backtracking (§4.2, (70)–(75))derivePluralFirstMerge: the first-merge option ((82)–(83)) — Xhosa's routeisStacked_iff_not_sharesClassifierN: stacking iff distinct Ns, for any lexicon where direct spellout succeeds exactly on shared-N targetsnot_isStacked_of_firstMerge: the first-merge option makes stacking underivable — the paper's account of stacking-free Xhosa
The bridge to [Car26]'s interpretable-gender agreement diagnostic lives
in Linglib.Studies.Carstens2026 (the later paper hosts the comparison). The
paper's DM comparison (§3.4, portmanteaux and Fusion) is not yet formalized.
Feature inventory #
Equations
- TaraldsenEtAl2018.instDecidableEqNCFeature.decEq TaraldsenEtAl2018.NCFeature.num TaraldsenEtAl2018.NCFeature.num = isTrue ⋯
- TaraldsenEtAl2018.instDecidableEqNCFeature.decEq TaraldsenEtAl2018.NCFeature.num (TaraldsenEtAl2018.NCFeature.cls a) = isFalse ⋯
- TaraldsenEtAl2018.instDecidableEqNCFeature.decEq (TaraldsenEtAl2018.NCFeature.cls a) TaraldsenEtAl2018.NCFeature.num = isFalse ⋯
- TaraldsenEtAl2018.instDecidableEqNCFeature.decEq (TaraldsenEtAl2018.NCFeature.cls a) (TaraldsenEtAl2018.NCFeature.cls b) = if h : a = b then h ▸ isTrue ⋯ else isFalse ⋯
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- TaraldsenEtAl2018.instReprNCFeature = { reprPrec := TaraldsenEtAl2018.instReprNCFeature.repr }
The Xhosa lexicon #
Exponents are the paper's analytical units — the post-augment prefix (p. 1340 restricts "prefix" to exclude the augment; surface u-m-, a-ba- etc. add the augment vowel). Entries by gender pair:
| gender | sg | tree | source | pl | tree | source |
|---|---|---|---|---|---|---|
| A (1/2) | m | [N₁] | (48a) | ba | [# N₁] | (48b) |
| B (3/4) | m | [N₃] | implied, cf. (25a), p. 1358 | mi | [# N₄] | (41a) |
| C (5/6) | li | [N₅] | (50a) | ma | [# N₆] | (50b) |
| D (7/8) | si | [N₇] | (49a) | zi | [# N₇] | (49b) |
| E (9/10) | n | [N₉] | the homorganic nasal, set aside in fn. 39 | zi | [# N₉] | implied by §2.3, fn. 40 |
Class-10 zi is accidentally syncretic with class-8 zi (fn. 11); it must contain N₉ because conjoined class-9 singulars allow class-10 formal agreement in Xhosa (§2.3, (10)–(12); fn. 40). The plural shape [# Nₓ] is not exceptionless across Bantu: Shona mi lexicalizes bare [N₄] ((89), fn. 42), which is how Shona double plurals like ma-mi-sha arise (§4.3).
Xhosa singular class-prefix entries: each lexicalizes a bare classifier [Nₓ]. See the table in the section docstring for per-entry sources.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Xhosa plural class-prefix entries: each lexicalizes [# Nₓ]. Genders A, D, E share their N with the singular; B and C contain distinct Ns (N₄ ≠ N₃, N₆ ≠ N₅) — the §2 conjoined-agreement finding.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The five Xhosa genders, in Fragment order.
Equations
Instances For
The full Xhosa class-prefix lexicon.
Equations
Instances For
The Changana/Rhonga lexicon #
The stacking languages ((76)–(80), pp. 1365–1366). Here the plural-specific classifiers N₄, N₆, N₁₀ are distinct from every singular's N, and — unlike in Xhosa — may not be first-merged with the root, so pluralization of classes 3, 5, 9 must stack ((62), p. 1361). Entries by gender pair:
| gender | sg | tree | source | pl | tree | source |
|---|---|---|---|---|---|---|
| A (1/2) | mu | [N₁] | (79a) | va | [# N₁] | (79b) |
| B (3/4) | mu | [N₃] | (76a) | mi | [# N₄] | (77a) |
| C (5/6) | rhi | [N₅] | (76b), printed ri/rhi | ma | [# N₆] | (77b) |
| D (7/8) | xi | [N₇] | (80a) | swi | [# N₇] | (80b) |
| E (9/10) | yi | [N₉] | (76c) | ti | [# N₁₀] | (77c) |
The five Changana/Rhonga gender pairs, indexed like Xhosa's A–E (no Tsonga Fragment exists yet, so the index is study-local).
- gA : RhongaGender
- gB : RhongaGender
- gC : RhongaGender
- gD : RhongaGender
- gE : RhongaGender
Instances For
Equations
- TaraldsenEtAl2018.instDecidableEqRhongaGender x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Changana/Rhonga singular class-prefix entries ((76), (79a), (80a)).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Changana/Rhonga plural class-prefix entries ((77), (79b), (80b)).
Equations
- One or more equations did not get rendered due to their size.
Instances For
The singular's classifier index per gender — the N the noun structure is built on in the Tsonga languages, which lack the first-merge option.
Equations
- TaraldsenEtAl2018.rhongaBaseN TaraldsenEtAl2018.RhongaGender.gA = 1
- TaraldsenEtAl2018.rhongaBaseN TaraldsenEtAl2018.RhongaGender.gB = 3
- TaraldsenEtAl2018.rhongaBaseN TaraldsenEtAl2018.RhongaGender.gC = 5
- TaraldsenEtAl2018.rhongaBaseN TaraldsenEtAl2018.RhongaGender.gD = 7
- TaraldsenEtAl2018.rhongaBaseN TaraldsenEtAl2018.RhongaGender.gE = 9
Instances For
The plural entry's classifier index per gender — the backtrack target.
Equations
- TaraldsenEtAl2018.rhongaPlN TaraldsenEtAl2018.RhongaGender.gA = 1
- TaraldsenEtAl2018.rhongaPlN TaraldsenEtAl2018.RhongaGender.gB = 4
- TaraldsenEtAl2018.rhongaPlN TaraldsenEtAl2018.RhongaGender.gC = 6
- TaraldsenEtAl2018.rhongaPlN TaraldsenEtAl2018.RhongaGender.gD = 7
- TaraldsenEtAl2018.rhongaPlN TaraldsenEtAl2018.RhongaGender.gE = 10
Instances For
The five Changana/Rhonga genders.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The full Changana/Rhonga class-prefix lexicon.
Equations
Instances For
Spellout sanity checks #
[N₁] spells out as m: the singular entry wins over ba ↔ [# N₁] (which also contains [N₁]) by the Elsewhere Condition — smallest match.
[# N₁] spells out as ba ((48b)).
[# N₉] spells out as class-10 zi.
No Xhosa entry lexicalizes a bare N₂ — class 2 has no singular form.
Shared vs distinct classifier Ns #
The conjoined-subject diagnostic (§2.2–2.7): a conjunction of two class-X singulars triggers matching plural class-Y agreement iff Y's prefix contains the same N as X's prefix.
Foot Condition: why distinct-N plurals cannot spell out #
Rhonga mi ↔ [# N₄] cannot spell out [# N₃]: its foot N₄ is absent from the target (the Foot Condition).
Rhonga va ↔ [# N₁] can spell out [# N₁]: its foot N₁ is present.
No Rhonga entry spells out [# N₃] — the lexicalization failure that forces backtracking and stacking ((70)–(75), p. 1364).
Pluralization: direct spellout vs stacking #
Result of pluralizing a Bantu noun (§4.2): direct — one plural prefix
forms the sole Specifier; stacked — the plural prefix stacks on top of
the singular prefix ((75)).
- direct : String → PluralizationResult
- stacked : String → String → PluralizationResult
Instances For
Equations
- One or more equations did not get rendered due to their size.
- TaraldsenEtAl2018.instDecidableEqPluralizationResult.decEq (TaraldsenEtAl2018.PluralizationResult.direct a) (TaraldsenEtAl2018.PluralizationResult.stacked a_1 a_2) = isFalse ⋯
- TaraldsenEtAl2018.instDecidableEqPluralizationResult.decEq (TaraldsenEtAl2018.PluralizationResult.stacked a a_1) (TaraldsenEtAl2018.PluralizationResult.direct a_2) = isFalse ⋯
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
A stacked outcome.
Equations
- (TaraldsenEtAl2018.PluralizationResult.direct a).IsStacked = False
- (TaraldsenEtAl2018.PluralizationResult.stacked a a_1).IsStacked = True
Instances For
Derive a plural form by cyclic spellout with last-resort backtracking (§4.2, (70)–(75)):
- The plural structure [# N_baseN] is built, where
baseNis the classifier merged with the root at the first step. Which classifier that is, is the cross-linguistic parameter ((82)–(83), p. 1366): Xhosa may (and prefers to) first-merge the plural entry's N (derivePluralFirstMerge); Changana/Rhonga must use the singular's N. - If some entry spells out [# N_baseN], the derivation is
direct. - Otherwise backtracking builds a second Specifier [# N_plN] inside the
first, spelled out on top of the singular prefix [N_baseN] —
stacked. By construction the outer prefix always lexicalizes a # -containing target, deriving §4.5's observation that singular prefixes never stack on top of plural ones.
Which [# N_W] entry the backtrack targets (plN) is not derived by the
spellout calculus: the paper attributes prefix–noun pairing to semantic
compatibility with the silent N or to idiom listing (§5.3, §5.5), so
plN is a parameter here.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The first-merge option ((82)–(83), p. 1366): the plural entry's N merges with the root directly, so the built structure and the backtrack target coincide. Xhosa takes this option in preference to stacking; the Tsonga languages cannot.
Equations
- TaraldsenEtAl2018.derivePluralFirstMerge entries plN = TaraldsenEtAl2018.derivePlural entries plN plN
Instances For
Changana/Rhonga gender B: class 3 nouns stack — mi-mu-twa 'thorns' ((62a), [Bac06]).
Rhonga gender C: class 5 nouns stack — ma-rhi-tu 'words' ((62b)).
Rhonga gender E: class 9 nouns stack — ti-yi-n-dlu 'houses' ((62c)).
Rhonga gender A pluralizes directly: va ↔ [# N₁] shares N₁.
Rhonga gender D pluralizes directly: swi ↔ [# N₇] shares N₇.
Xhosa first-merges the plural N, so even the distinct-N gender B pluralizes directly: the plural of a class-3 noun is built on N₄ from the start and mi spells it out — no stacking.
Xhosa gender C: direct ma, same first-merge route.
Xhosa gender A: direct ba.
The cross-linguistic contrast in one statement: the same gender (3/4), direct in Xhosa, stacked in Changana/Rhonga — the paper's central prediction ((82)–(83) vs (62)).
Stacking iff distinct Ns #
With the first-merge option the built structure and the backtrack target coincide, so a failed direct spellout leaves nothing for backtracking to lexicalize either: stacking is underivable. This is the paper's account of why Xhosa never stacks.
Prefix–concord identity #
The paper's entries for nominal class prefixes and subject concords are identical: (60)–(61) (p. 1360) repeat the nominal entries (48)–(49) as the SC entries (the SC's gender head G having been dropped). The paper's own exceptions: SC1 is u, not m (fn. 30), and SC6 is a, not ma ((53c)).
The class-prefix lexicon yields the Fragment's subject-concord forms for classes 5, 7, 2, 8 — one set of entries spells out both nominal prefixes and SCs ((60)–(61)).