Documentation

Linglib.Phenomena.Classifiers.Studies.TaraldsenEtAl2018

Taraldsen, Taraldsen Medová & Langa (2018) #

@cite{taraldsen-et-al-2018}

"Class prefixes as specifiers in Southern Bantu." Natural Language & Linguistic Theory 36:1339–1394.

Core claims #

  1. Bantu class prefixes are phrasal specifiers, not heads.
  2. Each prefix lexicalizes a nanosyntactic tree: [# Nx] (plural) or just [Nx] (singular), where # is the number head and Nx is a classifier-like nominal head.
  3. "Strong" classes (1/2, 7/8) share the same N between singular and plural; "weak" classes (3/4, 5/6, 9/10) have distinct Ns (N₄ ≠ N₃, N₆ ≠ N₅). This is the paper's central empirical finding from agreement with conjoined subjects (§2.1–2.6).
  4. The Foot Condition constrains prefix insertion and derives nP stacking (double prefix constructions in Changana/Rhonga).
  5. The specifier analysis unifies Bantu class prefixes with classifiers on the @cite{aikhenvald-2000} continuum.

Formalization #

Nominal features on the Bantu nanosyntactic fseq. num = number head (#); cls n = classifier head Nn.

@cite{taraldsen-et-al-2018}: class prefixes spell out phrasal trees built from these features. Singular prefixes lexicalize just Nx; plural prefixes lexicalize [# Nx].

Instances For
    Equations
    • One or more equations did not get rendered due to their size.
    Instances For

      Singular class prefixes: each lexicalizes just Nx. @cite{taraldsen-et-al-2018} (60)–(61): singular prefixes are specifiers of nP, spelling out the classifier head N alone.

      Equations
      Instances For

        Plural class prefixes: each lexicalizes [# Nx]. @cite{taraldsen-et-al-2018} (60)–(61), (77), (83): plural prefixes are specifiers of #P, spelling out [number + classifier].

        Critical: "strong" classes (2, 8) share the same N with their singular counterparts (N₁, N₇). "Weak" classes (4, 6, 10) contain DISTINCT Ns: N₄ ≠ N₃, N₆ ≠ N₅, N₁₀ ≠ N₉. This is the paper's central empirical finding from agreement with conjoined subjects (@cite{taraldsen-et-al-2018} §2).

        Equations
        • One or more equations did not get rendered due to their size.
        Instances For
          Equations
          • One or more equations did not get rendered due to their size.
          Instances For
            Equations
            • One or more equations did not get rendered due to their size.
            Instances For
              Equations
              • One or more equations did not get rendered due to their size.
              Instances For
                Equations
                • One or more equations did not get rendered due to their size.
                Instances For
                  Equations
                  • One or more equations did not get rendered due to their size.
                  Instances For

                    Whether a singular-plural class pair shares the same classifier N. @cite{taraldsen-et-al-2018} §2: strong classes (1/2, 7/8) share Ns; weak classes (3/4, 5/6, 9/10) have distinct Ns. The evidence comes from agreement with conjoined singular subjects: a conjunction of two class X singulars triggers plural class Y agreement iff the plural prefix for Y contains the same N as X.

                    Equations
                    Instances For

                      Classes 1/2 share N₁ — strong class pair. A conjunction of two class 1 singular nouns triggers class 2 plural agreement because ba ↔ [# N₁] and the singular m ↔ [N₁] share the same N₁.

                      Classes 7/8 share N₇ — strong class pair.

                      Classes 3/4 do NOT share Ns — weak class pair. @cite{taraldsen-et-al-2018} §2.1–2.4: a conjunction of two class 3 singular nouns does NOT trigger class 4 agreement. The plural prefix mi contains N₄, distinct from the N₃ in the singular prefix mu.

                      Classes 5/6 do NOT share Ns — weak class pair. @cite{taraldsen-et-al-2018} §2.5: a conjunction of two class 5 singular nouns does NOT trigger class 6 agreement. The plural prefix ma contains N₆, distinct from the N₅ in class 5.

                      Classes 9/10 do NOT share Ns — weak class pair (in Changana/Rhonga; in Xhosa the class 10 prefix = class 8).

                      The foot of cl2Pl ("aba") is N₁.

                      The foot of cl4Pl ("imi") is N₄ — NOT N₃. This is the key structural fact: imi requires N₄ to be present in the syntactic structure. Since N₄ ≠ N₃, imi cannot spell out a structure built from a class 3 noun.

                      Strong class: "aba" CAN spell out [# N₁] because its foot (N₁) is present in the target. No stacking needed.

                      Weak class: "imi" CANNOT spell out a structure containing only N₃. The Foot Condition requires N₄ (the foot of imi's stored tree), but the target [# N₃] contains only N₃. This forces backtracking, producing stacking — e.g., mi-mu-twa 'thorns' (class 4-3) in Changana/Rhonga.

                      Weak class: "ama" CANNOT spell out a structure containing only N₅. Same logic: N₆ ≠ N₅.

                      The stacking prediction: for weak classes, no entry in the lexicon can spell out the target [# N_sg]. The derivation must backtrack, splitting the structure and building a Specifier — producing the stacking pattern observed in Changana/Rhonga. @cite{taraldsen-et-al-2018} §4.2, prediction derived from (77)–(83).

                      Contrast with strong classes (§3) where spellout succeeds directly.

                      Result of pluralizing a Bantu noun. @cite{taraldsen-et-al-2018} §4.2 derives two outcomes:

                      • direct: one plural prefix replaces the singular (strong class).
                      • stacked: the plural prefix stacks on top of the singular prefix (weak class), producing double prefix constructions.
                      Instances For
                        Equations
                        • One or more equations did not get rendered due to their size.
                        Instances For

                          Derive the plural form for a noun with singular class sgCls.

                          Models @cite{taraldsen-et-al-2018} §4.2 cyclic derivation:

                          1. Noun N_Y merges with classifier N_sgCls → [N_sgCls N_Y].
                          2. Number head # merges → target [# N_sgCls].
                          3. Try spellout of [# N_sgCls]. If some entry's tree contains this target, the plural prefix directly replaces the singular (strong class path — e.g., ba replaces m for cl 1→2).
                          4. If no entry matches: the Foot Condition blocks lexicalization (§5). Backtrack (Starke's last resort): form [# N_plCls] as a Specifier merged with N_Y, producing a stacked structure (weak class path — e.g., mi-mu for cl 3→4).
                          Equations
                          • One or more equations did not get rendered due to their size.
                          Instances For

                            Strong class 1/2: direct pluralization. ba ↔ [# N₁] matches the target [# N₁], so the derivation succeeds in one step.

                            Strong class 7/8: direct pluralization.

                            Weak class 3/4: stacking. No entry matches [# N₃] (because the only class 4 entry has [# N₄] and N₄ ≠ N₃), so the derivation backtracks: mi ↔ [# N₄] stacks on top of um ↔ [N₃]. Produces the Changana/Rhonga form mi-mu-twa 'thorns'.

                            Weak class 5/6: stacking. ama ↔ [# N₆] on top of ili ↔ [N₅]. Produces Rhonga ma-rhi-tu 'words'.

                            Weak class 9/10: stacking. iin ↔ [# N₁₀] on top of in ↔ [N₉]. Produces Rhonga ti-yi-n-dlu 'houses'.

                            Whether an entry is a plural prefix (has # at root). Structurally, plural entries are [# N_X] (node with num label); singular entries are [N_X] (leaf).

                            Equations
                            Instances For

                              The paper's central prediction, proved structurally: for any lexicon where direct spellout of [# N_s] succeeds iff s = pN, stacking occurs iff the classifier Ns are distinct.

                              The three class-number parameters:

                              • s: the singular class N (e.g., N₁ for class 1)
                              • pN: the N inside the plural entry's tree (e.g., N₁ for strong class 2, N₄ for weak class 4)
                              • plCls: the plural class number used in derivePlural (e.g., 2, 4)

                              For strong classes, pN = s (shared N): cl2Pl has N₁ = cl1Sg's N. For weak classes, pN ≠ s (distinct Ns): cl4Pl has N₄ ≠ cl3Sg's N₃.

                              The proof chains three structural insights:

                              • Algorithm: derivePlural stacks iff direct spellout fails (derivePlural_isStacked_eq)
                              • Lexicon: direct spellout fails iff s ≠ pN (hDirect)
                              • Entries: sharesClassifierN = (s = pN) for standard-form entries (definitional from NanoTree.foot)

                              @cite{taraldsen-et-al-2018}: the correlation between agreement failure and stacking is not stipulated — it is derived from distinct Ns in the lexical entries.

                              Strong class 1/2: no stacking (shared N₁).

                              Strong class 7/8: no stacking (shared N₇).

                              Weak class 3/4: stacking (N₄ ≠ N₃).

                              Weak class 5/6: stacking (N₆ ≠ N₅).

                              Weak class 9/10: stacking (N₁₀ ≠ N₉).

                              A singular-plural class pairing in Bantu.

                              Instances For

                                All five Xhosa class pairings: 2 strong + 3 weak.

                                Equations
                                • One or more equations did not get rendered due to their size.
                                Instances For

                                  In stacked forms, the outer prefix is always from a plural entry (has # at root). Singular prefixes never appear as the outer layer in stacking. @cite{taraldsen-et-al-2018} §4.5: "while plural prefixes may stack on top of singular prefixes, we have found no cases where a singular prefix stacks on top of a plural prefix." This follows from the derivation: stacking is triggered by failure to lexicalize #, so the outer Specifier always contains #.

                                  Every class pairing produces a valid derivation (no failures).

                                  DM Vocabulary Items for Xhosa SC prefixes. Each class maps to a single feature. The Subset Principle selects the item whose features are all present in the target (largest match).

                                  Equations
                                  • TaraldsenEtAl2018.scPrefixDM = [{ features := [1], exponent := "u" }, { features := [2], exponent := "ba" }, { features := [7], exponent := "si" }, { features := [8], exponent := "zi" }]
                                  Instances For

                                    Nanosyntax entries for the same SC prefixes using tree-based spellout. Each class feature is a leaf on the nanosyntactic tree.

                                    Equations
                                    • One or more equations did not get rendered due to their size.
                                    Instances For

                                      Combining @cite{carstens-2026}'s agreement diagnostic with the classifier analysis: matching agreement (resolve succeeds) iff the gender has a semantic core — identifying it as a semantically motivated, classifier-like noun class.

                                      This theorem bridges two phenomena: genders where conjoined singulars allow matching plural agreement (@cite{carstens-2026}) are exactly those whose class prefixes have classifier-like semantic content (@cite{taraldsen-et-al-2018}).

                                      Xhosa sits at the noun-class pole of the @cite{aikhenvald-2000} classifier-to-noun-class continuum.

                                      Derive an NPStack from a tree-based analysis. The visible class is the outermost N; the core class is the innermost (= foot). When visible = core, the noun is canonical (strong class); when they differ, stacking has occurred because the plural prefix contains a different N (@cite{taraldsen-et-al-2018} §4).

                                      Equations
                                      Instances For

                                        Stacked [human] noun in class 3: visible = 3, core = 1. Stacking occurs because N₃ ≠ N₄ (cl3_cl4_distinct_N): the class 4 plural prefix cannot directly spell out [# N₃] (no_spellout_forces_stacking_cl3), so derivePlural backtracks (cl3_plural_stacked), placing class 3's prefix on top.

                                        theorem TaraldsenEtAl2018.stacked_implies_not_canonical (vis core : ) (st : Fragments.Bantu.GenderStatus) (hStack : vis core) :
                                        (treeToNPStack vis core st).isCanonical = false

                                        Stacking (from derivePlural) implies non-canonical NPStack. If derivePlural produces .stacked, the noun cannot be in its canonical class (visibleClass ≠ coreClass).

                                        Direct pluralization (from derivePlural) is consistent with canonical NPStack (visibleClass = coreClass).

                                        The Xhosa NounClass enum produces the same class numbers used in the nanosyntactic entries. This ensures that Fragment data and the nanosyntactic analysis share a common namespace.