The online processing of semantic and pragmatic content Brian - - PDF document
The online processing of semantic and pragmatic content Brian Dillon LINGUIST510 Psycholinguistics Comprehension: How do we compute the meaning of a sentence in real time? What are the online computations that we use to map word strings to
LINGUIST510
Comprehension: How do we compute the meaning of a sentence in real time? What are the online computations that we use to map word strings to structured meanings when reading or listening? words grammatical knowledge (extra-)linguistic context memory attention [[words]]
Some students enjoy psycholinguistics.
mechanism that achieves (scalar) implicature?
The horse raced past the barn fell BEVER (1970)
The horse raced past the barn fell BEVER (1970)
preferred over the ‘reduced relative’ parse
The horse raced past the barn fell
because it is structurally simpler: there are fewer syntactic nodes/phrases to postulate on MV analysis. So it is computed more rapidly, and becomes the parse we initially adopt.
The horse raced past the barn fell
sense “pragmatically preferred” over reduced relative analysis (in the so-called ‘null context’). Such a theory would imply that pragmatic computations
RR: [NP The [RC horse raced … ] ] MV: [NP The horse] raced …
generated by this (partial) parse?
generated by this (partial) parse? Suppose that comprehenders incrementally interpret sentences, including presuppositions, implicatures. Then…
RR: [NP The horse raced … ] … MV: [NP The horse] raced … There’s a (unique) horse There is a (unique) horse which +1) is a member of some larger set of horses. +2) This horse was raced somewhere and +3) the other horses weren’t raced to this place…
RR: [NP The horse raced … ] … MV: [NP The horse] raced … Principle of Parsimony (Crain & Steedman 1985): Parse which carries fewer unsatisfied presuppositions/ entailments (implicatures-BWD) is adopted by hearer, all else being equal. Explains Garden Path effect: RR analysis carries ~3 more unsupported implicatures (in ‘null context’) than MV parse, so is dispreferred.
RR: [NP The horse raced … ] … MV: [NP The horse] raced … Prediction: The probability of experiencing a garden path should decrease with the number of implicatures associated with a reduced relative parse of input string.
Def: The children taught by the Berlitz Method passed the test. Bare plurals, existential: Children taught by the Berlitz method passed the test Bare plurals, generic: Children taught by the Berlitz method pass the test
Experiment: Use Rapid Serial Visual Presentation (300ms/word), and ask for rapid grammaticality judgment.
CRAIN & STEEDMAN (1985)
CRAIN & STEEDMAN (1985) +
CRAIN & STEEDMAN (1985) YOUNGSTERS
CRAIN & STEEDMAN (1985) PUSHED
CRAIN & STEEDMAN (1985) INTO
CRAIN & STEEDMAN (1985) THE
CRAIN & STEEDMAN (1985) COULDN’T
CRAIN & STEEDMAN (1985) MOVE
CRAIN & STEEDMAN (1985) ????
CRAIN & STEEDMAN (1985) +
CRAIN & STEEDMAN (1985) YOUNGSTERS
CRAIN & STEEDMAN (1985) PUSHED
CRAIN & STEEDMAN (1985) INTO
CRAIN & STEEDMAN (1985) THE
CRAIN & STEEDMAN (1985) CROWD
CRAIN & STEEDMAN (1985) COULDN’T
CRAIN & STEEDMAN (1985) MOVE
CRAIN & STEEDMAN (1985) ????
CRAIN & STEEDMAN (1985) % “UNGRAMMATICAL” Responses: Definite heads > Indefinite Existentials > Indefinite Generics (29%)
CRAIN & STEEDMAN (1985) “Syntax proposes, semantics disposes”: The parser incrementally generates a semantic interpretation+presuppositions+implied meaning in parallel for some (maybe all) possible parses, and uses the latter to select a parse for further computation.
Can other linguistic operators introduce contexts that support RR parses fast enough to prevent garden pathing? Focus operator only requires contrast set (viz. focus alternatives). Can a focus operator that needs a contrast introduce bias towards postmodifier parse of ambiguous material?
CRAIN, NI & SHANKWEILER (1996) Bare, Def: The businessmen loaned money at a low interest rate were told to record their expenses. Only, Def: Only the businessmen loaned money at a low interest rate were told to record their expenses.
CRAIN, NI & SHANKWEILER (1996) /(Only) the businessmen/ loaned/ money at a low/ interest rate/ were told/ to record their/ expenses/ Experiment: Self-paced reading. Readers see one section of sentence at a time, and they press a button to advance through the sentence. Reaction times to the button press are recorded; processing difficulty is expected to surface as long reaction times.
CRAIN, NI & SHANKWEILER (1996) /(Only) the businessmen/ loaned/ money at a low/ interest rate/ were told/ to record their/ expenses/ ms
CRAIN, NI & SHANKWEILER (1996) /(Only) the wealthy businessmen/ loaned/ money at a low/ interest rate/ were told/ to record their/ expenses/ ms
Recognize “apple” Fixate “apple” 150ms
Ambiguous Instruction: “Put the apple on the towel in the box” Parse 1: “Put [the apple on the towel] in the box” Parse 2: “Put [the apple] [on the towel] in the box”
Unambiguous Instruction: “Put the apple that’s on the towel in the box”
2-Referent Context 1-Referent Context “Put the apple on the towel in the box”
1-Referent Context In 1-referent context with ambiguous instructions, listeners look to incorrect goal (B), suggesting they’ve been garden- pathed.
2-Referent Context In 2-referent context with ambiguous instructions, listeners do not look to incorrect goal (B).
i) uniqueness presupposition for determiner calculated immediately, ii) presupposition immediately evaluated against extra-linguistic (visual) context… iii) … biasing listeners to postmodifier parse of ambiguous PP
Implication: The need to satisfy the uniqueness presupposition of definite the 2-referent context causes comprehenders to anticipate a PP postmodifier!
So far: Listeners rapidly compute presuppositions and implicatures off of partial input; Use this information to information syntactic parsing decisions; Continuously evaluate presuppositions and implicatures against (extra-)linguistic context
Question: Why should comprehenders attempt to minimize unsatisfied implicatures a la Parsimony? Why not make ALL the implicatures and just deal with it? Hypothesis: The calculation of an implicature is a cognitive operation with measurable cost; it is to be avoided, all else being equal.
Some students enjoy psycholinguistics.
Lower-bounded reading: some, maybe all, students like psycholx Upper-bounded reading: some, but not all, students like psycholx
HUANG & SNEDEKER (2009)
Some students enjoy psycholinguistics.
HUANG & SNEDEKER (2009) Hypothesis: Derivation of upper-bounded reading involves computation of scalar implicature, which requires an extra processing step. Some Some, maybe all
access literal meaning compute scalar implicature
Some, not all
Some students enjoy psycholinguistics.
HUANG & SNEDEKER (2009) Prediction: We should observe that comprehenders i) are slower to access upper-bounded reading than lower- bounded and ii) should show evidence lower-bounded reading at some point in processing stream. Some Some, maybe all
access literal meaning compute scalar implicature
Some, not all
Point to the girl that has some/two of the socks…
Two: Point to the girl that has two …
Only compatible with this referent.
Some: Point to the girl that has some …
If lower-bound reading computed first, reference is ambiguous here.
Some: Point to the girl that has some…
If upper-bound reading computed immediately, sentence is only compatible with this referent.
Three/all: Point to the girl that has three/all of …
Control conditions
Some Some, maybe all
access literal meaning compute scalar implicature
Some, not all Hypothesis: Derivation of upper-bounded reading involves computation of scalar implicature, which requires an extra processing step.
Tantalizing possibility: We know that computing implicatures from string takes time. So maybe Parsimony preferences reflect a more general parsing principle: we always prefer the parse that takes the least time to
computed) implicatures gives it a disadvantage in the ‘race’ to be the interpretation. The horse raced past the barn fell.
Pick up the tall glass… SEDIVY ET AL (1999)
Pick up the tall glass… SEDIVY ET AL (1999)