RSA Question Embedding #
@cite{geurts-2010} @cite{hamblin-1973b} @cite{groenendijk-stokhof-1984}
Models scalar implicatures embedded in questions.
The Phenomenon #
"Did some students pass?"
Questions have a unique status regarding scalar implicatures:
- Not clearly upward-entailing (UE) like assertions
- Not clearly downward-entailing (DE) like negation
- Implicatures may arise, but weaker than in assertions
Theoretical Background #
Questions are often analyzed as sets of propositions or partitions of logical space. The key insight:
A question "Did some students pass?" asks the hearer to choose between:
- "Yes, some passed" (weak: >=1)
- "No, no one passed"
If the implicature "not all" is computed:
- "Yes, some-but-not-all passed"
- "No, either none or all passed" (??)
The second option doesn't make sense, suggesting local SI in questions is problematic.
Predictions #
Unlike assertions (where local SI strengthens) or DE contexts (where local SI weakens), questions create an asymmetry:
- The "yes" answer benefits from SI (more precise)
- The "no" answer becomes weird with SI (disjunctive)
RSA should predict: Local SI is dispreferred in questions because it makes one answer pragmatically odd.
Student outcome for question scenario.
- noneP : StudentResult
- someP : StudentResult
- allP : StudentResult
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- RSA.QuestionEmbedding.instBEqStudentResult.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Worlds represent the actual state of affairs.
Equations
Instances For
A yes/no question partitions the world into "yes" and "no" answers.
For "Did some students pass?":
- Yes-worlds: where some (>=1) passed
- No-worlds: where none passed
- yesWorlds : List StudentResult
- noWorlds : List StudentResult
Instances For
Interpretations of "Did some students pass?":
- Global: "some" = at least one
- Yes: someP, allP (some passed)
- No: noneP (none passed)
- Local: "some" = some-but-not-all
- Yes: someP (some-but-not-all passed)
- No: noneP, allP (?? disjunctive/weird)
- global : QuestionInterpretation
- local_ : QuestionInterpretation
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- RSA.QuestionEmbedding.instBEqQuestionInterpretation.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
The partition induced by each interpretation.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Global interpretation gives a natural partition:
- Yes = some passed (continuous region on scale)
- No = none passed (complementary region)
Local interpretation gives a strange partition:
- Yes = some-but-not-all (specific point)
- No = none OR all (disjunctive!)
This disjunctive "no" answer is pragmatically odd.
The "no" answer under local interpretation is disjunctive. This is marked by the worlds being non-contiguous on the scale.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Exhaustive interpretation: the "yes" answer conveys the MAXIMAL true proposition consistent with "yes".
For global "Did some pass?":
- If someP: answer conveys "some but not all"
- If allP: answer conveys "all"
This makes local SI redundant - exhaustivity handles it.
Equations
- RSA.QuestionEmbedding.exhaustiveAnswer RSA.QuestionEmbedding.QuestionInterpretation.global RSA.QuestionEmbedding.StudentResult.noneP = "No, none passed"
- RSA.QuestionEmbedding.exhaustiveAnswer RSA.QuestionEmbedding.QuestionInterpretation.global RSA.QuestionEmbedding.StudentResult.someP = "Yes, some (but not all) passed"
- RSA.QuestionEmbedding.exhaustiveAnswer RSA.QuestionEmbedding.QuestionInterpretation.global RSA.QuestionEmbedding.StudentResult.allP = "Yes, all passed"
- RSA.QuestionEmbedding.exhaustiveAnswer RSA.QuestionEmbedding.QuestionInterpretation.local_ RSA.QuestionEmbedding.StudentResult.noneP = "No (none passed)"
- RSA.QuestionEmbedding.exhaustiveAnswer RSA.QuestionEmbedding.QuestionInterpretation.local_ RSA.QuestionEmbedding.StudentResult.someP = "Yes, some-but-not-all passed"
- RSA.QuestionEmbedding.exhaustiveAnswer RSA.QuestionEmbedding.QuestionInterpretation.local_ RSA.QuestionEmbedding.StudentResult.allP = "No (all passed)"
Instances For
Under local interpretation, the "all" world gets a "no" answer, which is pragmatically bizarre.
Under global interpretation with exhaustivity, all answers are natural.
RSA predicts: global interpretation preferred for questions.
Reason: Local creates pragmatically odd partition where "all passed" is a "no" answer.
Questions differ from all other embedding contexts:
- Not about entailment direction (like DE vs UE)
- About pragmatic naturalness of the question-answer structure
- not_de_like : ∃ (w : StudentResult), List.elem w (questionPartition QuestionInterpretation.global).yesWorlds = true ∧ List.elem w (questionPartition QuestionInterpretation.local_).yesWorlds = false
Global doesn't entail local (unlike DE): allP is in global-yes but not local-yes
- not_attitude_like : ∃ (w : StudentResult), List.elem w (questionPartition QuestionInterpretation.local_).yesWorlds = true ∧ List.elem w (questionPartition QuestionInterpretation.global).yesWorlds = true
Local doesn't entail global (unlike attitude verbs): someP is in both
Instances For
Questions are indeed unique - neither DE-like nor attitude-verb-like in their entailment pattern, yet still prefer global.