Asher & Lascarides 2003: SDRT Right Frontier Constraint worked example #
@cite{asher-lascarides-2003}
Example (21) from p. 149 — the discourse where the book explicitly
enumerates which attachment sites are available under the Right
Frontier Constraint, and which are excluded. We replicate it
against the SDRT substrate to verify that
availableAttachmentPoints returns exactly the book-predicted set.
The example (21), p. 149 #
π₀:
├─ π₃, π₄
│ ├─ π₃: K_π₃
│ ├─ ⇓(π₃, π₄) ← subordinating topic relation
│ └─ π₄:
│ ├─ π₁, π₂
│ ├─ π₁: K_π₁
│ ├─ π₂: K_π₂
│ └─ Background(π₁, π₂) ← coordinating
Reading the labelling F:
- F(π₀) ⊇ ⇓(π₃, π₄) — the topic relation lives at the top level
- F(π₄) ⊇ Background(π₁, π₂) — the coordinated pair lives inside π₄
- F(π₁), F(π₂), F(π₃) are atomic clausal contents K_πᵢ
What the book asserts (p. 149) #
"assuming that π₂ in (21) is LAST, the available attachment sites are π₂ (by condition 1), π₄ (by condition 2a), π₃ (by condition 2b and 3) and π₀ (by condition 3). But π₁ is not available."
Why π₁ is not available #
π₁ and π₂ are siblings under Background, which is coordinating
(not subordinating). After attaching to π₂ (LAST), the only way to
reach π₁ would be through a subordinating relation — but
Background doesn't qualify. Walking up via outscoping reaches π₄
(which contains both π₁ and π₂), and from π₄ we reach π₃ via the
subordinating ⇓ relation. But π₁ is on the wrong side of a
coordinating sibling boundary.
Substrate note #
@cite{asher-lascarides-2003}'s ⇓ topic-relation is not in the
substrate's RhetoricalRelation enum (it's a structural relation
the book uses informally; the headline truth-conditional inventory
is Narration / Elaboration / Explanation / Result / Background /
Contrast / Parallel + Consequence / Alternation / Correction). We
encode ⇓ as .elaboration in the example below — both are
subordinating, and the RFC's condition 2b only sees the
isSubordinating projection. The example would behave identically
with a hypothetical .topic constructor.
Equations
Instances For
Equations
Instances For
Example (21) from @cite{asher-lascarides-2003} p. 149, with LAST = π₂.
Labels: 0 = π₀, 1 = π₁, 2 = π₂, 3 = π₃, 4 = π₄.
Edges encode the labelled-discourse-structure conjuncts:
⟨0, 3, 4, .elaboration⟩represents⇓(π₃, π₄) ∈ F(π₀)(using.elaborationas a stand-in for ⇓; both subordinating).⟨4, 1, 2, .background⟩representsBackground(π₁, π₂) ∈ F(π₄).
Plus container-marking edges so that iOutscopes recovers the box-containment structure:
- π₀ contains π₃ and π₄
- π₄ contains π₁ and π₂
These containment relations are conjuncts in the parent box's F-content (the book's "labels A and labelling F" representation — being a label inside a box IS a content fact about the parent). We encode them as additional outscoping conjuncts using the same edge structure.
For Example (21) specifically, the ⇓(π₃, π₄) and Background(π₁, π₂) edges already encode the containment (since these relations have π₃, π₄, π₁, π₂ as arguments and the relations themselves live in F(π₀) and F(π₄) respectively). No additional outscoping edges needed.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Book p. 149 headline claim: π₁ is NOT in the available attachment set from π₂ (Example 21).
The reason: π₁ and π₂ are siblings under Background, which is
coordinating (not subordinating). RFC's condition 2b only opens
up new attachment sites via subordinating relations, so the
sibling-via-coordinating π₁ is unreachable from π₂'s right
frontier.
The full available set from π₂ matches the book's prediction: {π₀, π₂, π₃, π₄}.