Wug-Test Paradigm #

@cite{berko-1958} @cite{albright-hayes-2003}

Shared vocabulary for the wug paradigm: subjects are presented with nonce (novel) lexical items and asked to produce, judge, or rate forms. Because nonce items are by construction unlisted, a wug response cannot be a rote recall — it must be the output of a productive generalisation.

Architectural role #

Paradigms/ is the contract layer between Theories/ and Phenomena/Studies/. A wug study provides a typed cell whose Attestation factor can be swapped between attested (a real lexeme) and novel (a wug). The paradigm-level predicates in §4 quantify over the lens, so any theory whose predictions ride along the cell's other factors can be tested for the qualitative pattern the original paper reports.

Anchoring on a methodological lineage #

Two papers ground the contract:

@cite{berko-1958} introduced the test as a probe for productive morpho-phonological knowledge: presented with the nonce wug, children produce wugs /wʌgz/ rather than refusing or randomising. The factor that matters is Attestation.
@cite{albright-hayes-2003} is the modern reference for gradient wug responses: subjects rate alternative output forms, and the ratings track how well the input is supported by lexical generalisations of varying scope. This is what makes wug responses diagnostic for theories of how productivity is encoded (rule-and-analogy, MaxEnt grammars, exemplar models, lexical conservatism).

The contract here is deliberately minimal — a single parametric lens class HasFactor (Cell) (F) plus a real-valued Rate observable. The two known dimensions a wug paradigm crosses (Attestation, log-frequency) are exposed as abbrevs HasAttestation/HasFrequency specialising HasFactor at the relevant codomain. Studies that need additional factors (neighbourhood density, paradigm structure, binary IOR membership) instantiate HasFactor at their own codomain and reuse the same predicate machinery.

Anti-UseListed discriminator #

A wug paradigm is the canonical discriminator between theories that locate productivity in the grammar (where novel forms inherit lexical-frequency effects via constraints / weights) and theories that locate productivity in lexical listing (where novel forms cannot inherit anything because they are by definition unlisted). The predicate NovelInvariantInFactor is the UseListed @cite{zuraw-2000} prediction; NovelShowsFactorGradient is the prediction of indexed-constraint @cite{pater-2010} or scaled-weight @cite{coetzee-pater-2008} accounts. The theorem novelGradient_inconsistent_with_invariance proves the two are mutually incompatible on cells with a non-vacuous factor space, so a study that adopts a typed Cell can pose the discrimination as a single bridge theorem at any factor type with [LT].

Bridge to `Theories/Phonology/ItemSpecificity/HasTokenFreq` #

HasTokenFreq (in Theories/Phonology/ItemSpecificity/Defs.lean) is a getter-only class on fragment lexical entries — fragments are immutable data, so there is no setter. HasFactor Cell ℝ is a lens on paradigm cells, which the paradigm-level predicates below need to quantify over swapping a frequency without touching other factors.

Wug cells that wrap a fragment lexical entry typically store the manipulable factor as a separate field (e.g. WugBKKCell.n2LogFreq) that mirrors the entry's tokenLogFreq for attested items and ranges freely on novel items. This is the right architecture for an experimental paradigm: the cell-level factor IS the experimentally manipulated value, not the lexicon-derived one. The connection to HasTokenFreq is by intent (downstream theories test whether the cell-level factor predicts the rate observable, with the cell-level factor originating in the lexicon channel for attested items), not by an automatic instance.

Out of scope (per `CLAUDE.md` Processing scope) #

The specific form of the rating scale (Likert, forced choice, reaction time) — measurement modality, not paradigm contract.
Item-construction methodology (phonotactic legality, frequency matching, neighbourhood control) — study design choice.
Statistical analysis pipelines (mixed-effects, ordinal regression) — analysis decisions, not paradigm contract.
Per-paper item lists — these belong in the relevant Studies/ file.