Arnold, Wasow, Losongco & Ginstrom 2000: heaviness vs. newness #
Postverbal constituent order in English — heavy NP shift and the dative alternation — has been attributed to grammatical weight (Behaghel's law of growing constituents) and to discourse status (given before new), but the two are confounded: given referents get short expressions and new ones long. A corpus study of the Aligned-Hansard debates and an elicitation experiment in which directors instruct actors to give objects to toy animals measure heaviness as the difference in word count between the two constituents and newness by prior mention, and find that both factors independently shift the order, the newer and heavier constituent coming later, with neither reducible to the other. Their relative strength varies with the data: heaviness dominates where length differences are large, newness where the given referent has just been mentioned — a system of interacting constraints whose strength grows as the competing constraint weakens. Speakers' early disfluencies favour goal-first orders, so the postponement of new and heavy material serves planning as well as comprehension.
The two factors are read as weighted constraints over the two orders of a constituent pair: a graded penalty for putting the heavier constituent first, sized by the length difference the paper measures, and a penalty for putting a new constituent before a given one. The harmony difference between the orders is the weighted sum of the two signed preferences, which is the additive independence the paper reports; the paper's interaction is a matter of effect size in the data and is not represented.
Main definitions #
Phrase,Pair,Order: a constituent by word count and givenness, a postverbal pair, and which constituent comes last.heavyFirst,newFirst: the two constraints;contheir weighted family.heavyDiff,newDiff: each constraint's signed preference for the theme-last order.
References #
Phrases, orderings, and candidates #
A constituent by the two measures the paper codes: word count and discourse status.
- words : ℕ
- discourse : Discourse.BinaryGivenness
Instances For
Equations
- ArnoldEtAl2000.instDecidableEqPhrase.decEq { words := a, discourse := a_1 } { words := b, discourse := b_1 } = if h : a = b then h ▸ if h : a_1 = b_1 then h ▸ isTrue ⋯ else isFalse ⋯ else isFalse ⋯
Instances For
A postverbal pair: theme and goal for the dative alternation, direct object and PP for heavy NP shift.
Equations
Instances For
Equations
- ArnoldEtAl2000.instDecidableEqOrder x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
Instances For
The two constraints #
The weight penalty: the number of words by which the first constituent exceeds the second — the relative length the paper measures.
Equations
- ArnoldEtAl2000.heavyFirst ((th, gl), ArnoldEtAl2000.Order.themeLast) = gl.words - th.words
- ArnoldEtAl2000.heavyFirst ((th, gl), ArnoldEtAl2000.Order.goalLast) = th.words - gl.words
Instances For
The newness penalty: a new constituent before a given one.
Equations
- ArnoldEtAl2000.newFirst ((th, gl), ArnoldEtAl2000.Order.themeLast) = if gl.discourse = Discourse.BinaryGivenness.new ∧ th.discourse = Discourse.BinaryGivenness.given then 1 else 0
- ArnoldEtAl2000.newFirst ((th, gl), ArnoldEtAl2000.Order.goalLast) = if th.discourse = Discourse.BinaryGivenness.new ∧ gl.discourse = Discourse.BinaryGivenness.given then 1 else 0
Instances For
The two constraints.
Equations
Instances For
The weights: wH for heaviness, wN for newness.
Equations
- ArnoldEtAl2000.weights wH wN = ![wH, wN]
Instances For
Signed preferences #
Heaviness's preference for themeLast: the theme's length minus the goal's.
Equations
- ArnoldEtAl2000.heavyDiff p = ↑p.1.words - ↑p.2.words
Instances For
Newness's preference for themeLast: 1 when the theme is new and the goal given, -1
in the mirror case, 0 otherwise.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The harmony difference between the two orders is the weighted sum of the two signed preferences.
The theme-last order wins exactly when the weighted preferences sum in its favour.
Independence #
Heaviness alone places the heavier constituent last.
Newness alone places the new constituent after the given one.
Weight effects are graded: lengthening the theme raises the harmony of the theme-last order by the heaviness weight per word.
When newness is neutral, heaviness decides; when weight is neutral, newness decides.
The paper's examples #
The number of words in a constituent as printed.
Equations
- ArnoldEtAl2000.words s = (List.filter (fun (x : Char) => decide (x = ' ')) s.toList).length + 1
Instances For
The word counts of a row's two constituents in surface order.
Equations
- ArnoldEtAl2000.constituents r = do let a ← r.feature? "first" let b ← r.feature? "second" pure (ArnoldEtAl2000.words a, ArnoldEtAl2000.words b)
Instances For
In every corpus and text example the paper offers as ordered by weight, the heavier constituent comes last.
The examples the paper sets aside — the announcer's That brings to the plate Barry Bonds and (17) — put the lighter constituent last: other factors, planning among them.