Documentation

Linglib.Phenomena.ArgumentStructure.Studies.Adger2003

Adger 2003: C-Selection and Subcategorization @cite{adger-2003} #

Replicates Adger's c-selection account of subcategorization (Core Syntax ch. 3 §3.5–§3.6) on linglib's Minimalism substrate via the Derivation.WellFormed predicate from Theories/Syntax/Minimalism/Selection.lean.

Adger's apparatus (recap) #

Each lexical head bears interpretable categorial features [F] (e.g. [V] for verbs) and uninterpretable c-selectional features [uG] (e.g. [uN] for transitive verbs). Adger's checking system:

Substrate mapping #

Adgerlinglib
interpretable [F]SimpleLI.cat : Cat
uninterpretable [uG]SimpleLI.sel : SelStack (a List Cat)
Full InterpretationDerivation.WellFormed (Selection.lean)
first Merge / complementStep.emR (consumes one pending feature)
second Merge / specifierStep.emL (no consumption)

verbToSelStack (in Theories/Syntax/Minimalism/Derivations.lean) derives the SelStack from VerbEntry.complementType:

complementTypeSelStackEnglish class
.none[]intransitive
.np[.D]transitive
.np_np[.D, .D]ditransitive

Sections #

§0: Lexicon — Fragment-grounded SO building blocks #

Migrated from the deleted Theories/Syntax/Minimalism/Derivations.lean; instances belong in Phenomena per the architecture audit. Each SO is a leaf with Cat and SelStack derived from its Fragment lexical entry.

LIToken id allocation (one-derivation scope) #

Every leaf carries a Nat id used by Step.im/movement to identify chains. The scheme below partitions ids into ranges so that ids are distinct across SOs that may co-occur in a derivation:

RangeUse
1–9Pronouns
10–19Proper-noun DPs
20–29Common-noun bare Ns
30–39Verb leaves
40–49Determiners
100–199§2 ungrammatical-derivation seeds
200–299§5 null-D-wrapped derivation seeds
9000+Silent / null-D heads (nullDWrap)

"John sleeps": [S John sleeps] — intransitive.

Equations
Instances For

    "Mary arrives": [S Mary arrives] — intransitive.

    Equations
    Instances For

      "John sees Mary": [S John [VP sees Mary]] — transitive.

      Equations
      Instances For

        "He sees her": pronouns project D directly.

        Equations
        Instances For

          "Mary eats pizza": transitive with bare common-noun complement.

          Equations
          Instances For

            "John devours pizza": transitive with bare common-noun complement.

            Equations
            Instances For

              "John gives Mary book": ditransitive. (Word order is gives book Mary in linearization — see substantive note in the audit; the Larson 1988 shell would put goal Mary as inner complement and theme the book as outer.)

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

                Phase 1.0 substrate caveat: Derivation.WellFormed is Decidable only via a noncomputable instance ... classical infer_instance because WellFormed routes through checkedFinal? (Selection.lean:178), which is noncomputable via three Phase 1.0 placeholders: SyntacticObject.checkedSel?, leftmostLeaf, and Step.applyChecked (each lifting through Quot.out). Concrete-instance decide checks fail at kernel reduction. TODO Phase 2: once LCA-based head/leftmostLeaf/checkedSel land, restore by decide.

                "John sleeps" — intransitive: empty pending stack consumed trivially.

                "Mary arrives" — intransitive variant.

                "John sees Mary" — transitive: see's [.D] is checked by Mary's D.

                "Mary sees John" — symmetric variant.

                "He sees her" — pronouns also project D, so this checks too.

                "*Mary arrives John" — intransitive with spurious complement. arrive's pending = [] so the emR john step has nothing to check against and applyChecked returns none.

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

                  "*John sees" — transitive missing required complement. see's [.D] is never checked, so the final pending = [.D][].

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

                    Adger's c-selection predictions agree with Phenomena.ArgumentStructure.Subcategorization.data on the proper-noun fragment of the contrast set:

                    FrameGrammaticalUngrammatical
                    intransitive"John sleeps""*Mary arrives John"
                    transitive"John sees Mary""*John sees"

                    The full data file's bare-common-noun examples are addressed in §4.

                    Bare common nouns under strict c-selection #

                    The existing *_pizza / *_book derivations in Theories/Syntax/Minimalism/Derivations.lean use bare common nouns (cat = .N per nounToSO for non-proper entries) as direct objects of verbs that c-select [.D]. Per Adger ch. 7 Functional Categories II — the DP, bare common nouns require null-D wrapping to project as DPs and check verbs' [uD]. linglib's substrate currently encodes common nouns as N without this wrapping, so strict c-selection flags these derivations as ill-formed even though their intended English glosses are grammatical.

                    This is a substrate-encoding artifact, not a pedagogical ill-formedness. Resolving it requires one of: (a) introducing a null-D wrapper for bare common nouns (Adger ch. 7); (b) allowing verbs to alternately c-select N (counter to standard Minimalism); (c) treating bare common nouns as having an implicit kind-denoting D feature.

                    Out of scope for this study; the well-formedness predicate's strictness makes the gap visible (rather than silently passing), which is what linglib aims for. The three examples below pin the current behaviour.

                    "John devours pizza" with bare pizza (.N) fails devour's [uD].

                    "Mary eats pizza" — same pattern with eat.

                    "John gives Mary book" — book (.N) is the inner complement and fails give's first [uD].

                    Null-D-wrapped DPs are well-formed #

                    @cite{adger-2003} ch. 7 (Functional Categories II — the DP) treats bare common nouns as DPs headed by a phonologically silent D. Selection.lean provides nullDWrap : SyntacticObject → Nat → SyntacticObject, which constructs (∅ᴅ, n): the silent D's [uN] feature is checked internally against the noun's .N, yielding a saturated DP with outerCat = .D and checkedSel = some [].

                    The parallel *_DP derivations below mirror the canonical sentences from Derivations.lean but route the common-noun complement through nullDWrap, producing well-formed structures. The phonYield is unchanged because nullD has an empty phonForm.

                    Bare pizza (cat = .N) wrapped under null D, projecting a saturated DP.

                    Equations
                    Instances For

                      "John devours pizza" with null-D-wrapped DP complement. devour's [uD] is checked by the wrapped DP's outer .D.

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

                        "Mary eats pizza" — same pattern with eat.

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

                          "John gives Mary book": ditransitive with book as inner complement (direct object) wrapped under null D, Mary as proper-noun outer complement (indirect object). Both [uD] features on give are checked.

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

                            Surface form #

                            By construction, nullD's phonForm is empty, so the null-D-wrapped derivations linearize to the same surface strings as the canonical bare- noun derivations — only the internal structure differs. PhonYield equality theorems analogous to Derivations.lean's *_phon are deferred: the foldl-over-steps + nested phonYield append exceeds decide's reduction budget and is not closeable by rfl, and the existing Derivations.lean precedent (native_decide) is non-mathlib-compatible per CLAUDE.md. A structured proof via explicit unfolding lemmas is the right next step but out of scope here.

                            Combined: the §4 ill-formed bare-noun derivations and the §5 null-D-wrapped DP derivations form a minimal pair that operationalises Adger ch. 7's claim that argumental nominals are DPs.