Landau (2026): Silent Resumption: A New Test for Ellipsis #
This file formalizes the ellipsis-internal resumption test of [Lan26]. By the ban on vacuous
quantification ([Cho82]) an Ā-operator binds into a null site only if the site has structure
at LF to host a resumptive variable, so a resumptive dependency into a null site diagnoses a surface
anaphor in the sense of [HS76]. A failed extraction has two possible sources, the
paper's two reasons for the star: the site is a deep anaphor with no structure to host the
dependency, or it is a surface anaphor and ellipsis applied in the derivation bled the dependency.
An island around the site blocks extraction for a further reason, independent of ellipsis. Movement
is exposed to both of these confounds and agreement to the first; a resumptive dependency is formed
at LF and is exposed to neither. A test decides depth when the depth of a site factors through its
outcome, and a dependency decides depth exactly when it is exposed to no confound
(Dependency.decides_iff), so of the three only resumption does
(Dependency.decides_iff_eq_resumption). The paper's Hebrew nominal and prepositional ellipses,
whose domains are all islands, and its cross-linguistic mixed anaphors are rows of
Data/Examples/Landau2026.json. The resumption judgments are the test's predictions
(eir_matches), extraction fails on every Hebrew row (hebrew_extraction_fails), and the judgments
alone fix a deep anaphor and an ellipsis in each Hebrew domain (hebrew_domains) and the deep
status of the mixed anaphors (mixed_anaphors_deep).
Implementation notes #
A site is modelled by its depth and the confounds present at it, and a test by the dependency it
forms into the site. The paper does not say whether islands block agreement, and nothing below
depends on it: agreement fails to decide depth on the timing confound alone. The rows' texts and
judgments agree with the August 2025 preprint of the paper; the example numbers the rows carry run
two below the preprint's throughout and have not been checked against the published version. The
domain of a null site is the category of its head, Minimalist.Cat.
References #
Sites and dependencies #
What can block a dependency into a surface site: ellipsis applied in the derivation before the dependency was formed, bleeding it, or an island around the site.
Instances For
Equations
- Landau2026.instDecidableEqConfound x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- Landau2026.instReprConfound.repr Landau2026.Confound.timing prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Landau2026.Confound.timing")).group prec✝
- Landau2026.instReprConfound.repr Landau2026.Confound.island prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Landau2026.Confound.island")).group prec✝
Instances For
Equations
- Landau2026.instReprConfound = { reprPrec := Landau2026.instReprConfound.repr }
A null site, given by the depth of its anaphor and the confounds present at it.
- depth : Anaphor.Depth
The depth of the anaphor at the site.
- confounds : Finset Confound
The confounds present at the site.
Instances For
The dependency a test forms into a null site: extraction moves an operator out of it, agreement probes into it, and the EIR test binds a resumptive pronoun inside it.
- movement : Dependency
- agreement : Dependency
- resumption : Dependency
Instances For
Equations
- Landau2026.instDecidableEqDependency x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- Landau2026.instReprDependency = { reprPrec := Landau2026.instReprDependency.repr }
Equations
- One or more equations did not get rendered due to their size.
Instances For
The confounds a dependency is exposed to. Movement is formed in the derivation and is constrained by islands; agreement is formed in the derivation, so ellipsis can bleed it; a resumptive dependency is formed at LF and is indifferent to islands.
Equations
Instances For
A site hosts a dependency when it has internal structure for the dependency to end in and no confound the dependency is exposed to is present.
Equations
- s.Hosts d = (s.depth.HasInternalStructure ∧ Disjoint d.exposure s.confounds)
Instances For
Equations
- Landau2026.instDecidableHosts s d = { decide := decide s.depth.HasInternalStructure && decide (Disjoint d.exposure s.confounds), reflects_decide := ⋯ }
A test decides depth when the depth of a site is determined by whether the site hosts the test's dependency.
Equations
- d.Decides = Function.FactorsThrough Landau2026.Site.depth fun (x : Landau2026.Site) => x.Hosts d
Instances For
A site that hosts any dependency is a surface anaphor: the success of a test is always conclusive.
A dependency exposed to fewer confounds is hosted wherever one exposed to more is.
Resumption succeeds wherever any dependency does.
Whatever confounds a site contains, the outcome of resumption into it fixes its depth.
A dependency decides depth exactly when it is exposed to no confound. Otherwise a deep site and a surface site where a confound blocks the dependency both fail it.
Of the three dependencies, only resumption decides depth: a failed extraction or a failed agreement is compatible with a surface site, a failed resumption is not.
The analysis predicts surface sites where ellipsis bleeds extraction but a resumptive pronoun
is hosted; the paper reports no such case yet. The Hebrew rows show the same split with an island
in place of timing (hebrew_extraction_fails, eir_matches).
The paper's examples #
Every row is a resumptive dependency into a null site, and its judgment is the outcome of the EIR test. The depth of the site, its domain, and whether the domain is an island are read from the row's paper features.
Equations
- Landau2026.depths = [("deep", Anaphor.Depth.deep), ("surface", Anaphor.Depth.surface)]
Instances For
Equations
- Landau2026.domains = [("nP", Minimalist.Cat.n), ("DP", Minimalist.Cat.D), ("PP", Minimalist.Cat.P), ("VP", Minimalist.Cat.V)]
Instances For
The confounds a row's extractionAvailable feature records: a domain out of which extraction
is unavailable is an island.
Equations
- Landau2026.barriers = [("true", ∅), ("false", {Landau2026.Confound.island})]
Instances For
The site of a row: its recorded depth, and the island confound when extraction out of its domain is unavailable.
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
Every row records a depth and whether its domain is an island, so each theorem below speaks about all the rows.
The judgments are the predictions of the EIR test: binding a resumptive pronoun inside the null site is acceptable exactly when the site hosts the dependency.
Every Hebrew domain tested is an island, so extraction fails on every Hebrew row, surface or deep, and cannot diagnose ellipsis there.
Hebrew has both a deep anaphor and ellipsis in each of the nP, DP and PP domains: for each domain and depth, some row's resumption judgment is accounted for only by a site of that depth.
The mixed anaphors do so, dat doen, det and the Korean null object are deep: their resumption is unacceptable, so any site that accounts for the judgment is deep, whatever confounds it contains.