Hofmeister and Sag (2010): Cognitive Constraints and Island Effects #
This file formalizes Hofmeister and Sag's processing account of island effects. The paper traces
island effects to four factors, namely the length of the filler–gap dependency, the referential
load of the material it spans, the clause boundaries it crosses, and the complexity of the
filler. Each factor is read off a Stimulus, the filler with the words between it and the gap,
and stimuli are compared by Pareto dominance over the four factors (Stimulus.profile).
Dominance predicts the paper's contrasts. Whatever the dependency, a which-N filler is easier
than a bare one (which_easier), the complex-NP island of Experiment 1 is harder than its
baseline even with the richer filler (cnpc_harder_than_baseline), and the which-N wh-island of
Experiment 2 is incomparable with its bare baseline (whIsland_which_incomparable), where the
experiment found no reading-time difference after the embedded verb.
Implementation notes #
- Each factor is ordinal. A word's referential load follows the accessibility scale of §3.2, with pronouns below indefinites below definites and names, a clause boundary's cost follows §3.3, with a declarative complement below an interrogative one, a filler's ease is its complexity, and the dependency's length counts the words it spans.
- Pareto dominance is the product order on
ℕ × ℕ × ℕ × ℕᵒᵈ, with ease reversed because more ease makes processing easier. No rating or reading time enters a theorem, and the paper's acceptability and reading-time results stay in prose. - The NP-type factor of Experiment 1 was weak and local in reading times and not significant in
acceptability, so the profiles predict its direction (
cnpc_definite_harder), not a size.
References #
- P. Hofmeister and I. A. Sag, Cognitive Constraints and Island Effects (2010)
- E. Gibson, Linguistic Complexity: Locality of Syntactic Dependencies (1998)
- R. L. Lewis and S. Vasishth, An Activation-Based Model of Sentence Processing as Skilled Memory Retrieval (2005)
- P. Deane, Limits to attention: A cognitive theory of island phenomena (1991)
- J. Sprouse, A program for experimental syntax: Finding the relationship between acceptability and grammatical knowledge (2007)
Stimuli and profiles #
The filler of a dependency is a bare wh-word or a which-N phrase.
Instances For
Equations
- HofmeisterSag2010.instDecidableEqFiller x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
A word between the filler and the gap is, as far as the four factors see, a discourse reference of some accessibility, a clause boundary of some kind, or other material.
Instances For
Equations
- HofmeisterSag2010.instDecidableEqWord x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
The referential load of a word follows §3.2, since a pronoun refers to an old referent, an indefinite creates one, and a definite or a name searches for one.
Equations
Instances For
The cost of the clause boundary a word opens (§3.3) is higher for an interrogative complement, whose alternatives must be considered as well, than for a declarative one.
Equations
Instances For
The retrieval ease a filler affords (§3.4) is higher for the richer filler, which resists interference and rules out early integration sites.
Equations
Instances For
A filler–gap dependency, as the factors describe it, consists of its filler and the words it spans.
Instances For
The profile of a stimulus records the length of the span, the boundaries it crosses, the references it holds, and the ease its filler affords. Its product order, with ease reversed, is Pareto dominance, so one stimulus is harder than another only when it is at least as hard on every factor and strictly harder on one.
Equations
- s.profile = (s.span.length, (List.map HofmeisterSag2010.Word.boundary s.span).sum, (List.map HofmeisterSag2010.Word.load s.span).sum, OrderDual.toDual s.filler.ease)
Instances For
Whatever the span, the which-N filler is easier than the bare one, since the profiles agree on every cost and differ on ease alone.
Experiment 1: the complex-NP island (48) #
Equations
- HofmeisterSag2010.instDecidableEqIslandNP x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
The baseline of (48) has the which-N filler, as in Emma doubted that we had captured __.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A which-N filler eases the complex-NP island for every island NP (§5.2).
Every island condition is harder than the baseline, the which-N ones too, since the island adds a reference and a word to the span (§5.3).
A definite island NP is harder than an indefinite one, the direction of the local reading- time effects at the complementizer and the verb (§5.2).
The plural and singular indefinites carry the same load.
Experiment 2: the wh-island (49) #
whIsland f is the wh-island of (49), did Albert learn whether they dismissed __, with
filler f.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The baseline of (49) has the bare filler, as in did Albert learn that they dismissed __.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A which-N filler eases the wh-island (§6.2).
The bare wh-island is harder than its baseline, since it crosses an interrogative boundary.
The which-N wh-island and the bare baseline trade the interrogative boundary against the richer filler, which Pareto dominance leaves undecided. The experiment found the two read alike after the embedded verb.