Sentence Processing Pt2 Flashcards
Making Predictions
We use sentences in increments, using the info available to us
- we also anticipate the upcoming input
- we show behaviours suggesting we make predictions
Visual world eyetracking (Altmann & Kamide, 1999)
Research Q: how likely do we predict the upcoming direct object based on the semantic characteristics of verbs?
- looking for anticipatory effects; building expectations based on verb meanings
Method: participants are shown an image depicting a boy sitting on the ground surrounded by 4 objects: “ball”, “toy truck”, “cake” and “toy train set”
- the boy will move… non-restrictive; there are more movable objects than edible objects
- the boy will eat… restrictive condition (cake = only option)
- before “cake” is even mentioned, greater proportion of looks to “cake” after the verb “eat” than “move”
We can use the semantic properties of the verb to anticipate the likely direct object
Predictions:
Subject + Verb Info
(Kamide et al., 2003)
(Kamide et al., 2003)
Research Q: Can we use subject + verb information to anticipate the likely direct object?
Method: image depicting a man wearing a helmet, and a young girl wearing a dress. Around them are 4 objects: “carousel”, “motorbike”, “candy”, “beer”
- listeners showed anticipatory looks to the likely direct object:
- the man will ride … looks to “motorbike”
- the man will taste … looks to “beer”
- the girl will ride … looks to “carousel”
- the girl will taste … looks to “candy”
- people don’t solely rely on verb information — subjects also provide cues (eg: social / gender roles and biases)
Shows that we use linguistic and contextual knowledge (subject + verb combo) to make predictions
Predictions:
Case Marking
Case Marking: identifies the grammatical roles of nouns in a sentence
Example: Japanese; verb-final language
Kodomo-ga | sensei-ni | purozento-o |ageru
- child | teacher | gift | give
- SUBJ | INDIRECT OBJ | DIRECT OBJ
- “The child gives the gift to the teacher”
- (-ga) subject
- (-ni) indirect object
- (-o) direct object
Predictions:
Case Marking
Research Example
Research Q: Can Japanese listeners use case marking info to predict an upcoming direct object, even without hearing the verb information?
- after hearing only nouns with their case markings
Method: participants see a screen with 4 images: “waitress”, “customer”, “hamburger”, “trash bin” and are given three words from the following sentence:
Weitoresu-ga | kyaku-ni | tanoshigeni | hanbaagaa-o | hakobu
- waitress | customer | merrily | hamburger | bring
Hypothesis #1: upon hearing the first three words… participants will anticipate a direct object given a subject and an object marked with (-ni)
- 3 words: weitoresu-ga | kyaku-ni | tanoshigeni …
- waitress (subj) | customer (indirect obj) | merrily …
- more looks to hamburger as direct object than garbage bin
Hypothesis #2: upon hearing three words; participants don’t anticipate a direct object when given a subject and object marked with (-o)
- weitoresu-ga | kyaku-o | tanoshigeni …
- waitress (subj) | customer (direct obj) | merrily …
- similar proportion of looks between hamburger and trash bin
Results: matched the hypothesis; case markers (cues) alone were sufficient for making predictions
- more looks to hamburger than trash bin given a subject and an object marked with (-ni)
- listeners can also use case marking info to anticipate the unfolding input
Predictions:
Surprisal
Surprisal study: measured neural impulses with EEG nodes, and tracked participants’ reading and reaction times for the sentence:
Its a windy day outside, so the child went to fly _ _ _:
- a kite
- an airplane
Surprisal
- measurement of processing difficulty
- predictions occurred before nouns/determiners
- expectations that are met = lower surprisal (more predictable)
- expectation that are unmet = higher surprisal (less predictable)
- we would expect longer reading times and higher surprisal for an airplane than a kite
- predictability is shaped by linguistic experience (context)
Predictions:
of an upcoming word
Prediction success dependent on context
1) prediction of an upcoming word
The author was writing another chapter about the fictional detective. To date, she thinks it will be her most popular _ _ _
- novel (highly predictable) LOW SURPRISAL
- book (slightly predictable) MID SURPRISAL
- work (least predictable) HIGH SURPRISAL
Predictions:
of an upcoming structure
Prediction success dependent on context
2) prediction of an upcoming structure
Jose is looking for either a maid … or a cook.
- shorter reading times
- more predictable
- LOWER SURPRISAL
- the presence of “either” generates an expectation for “or”
Jose is looking for a maid … or a cook.
- longer reading times
- less predictable
- HIGHER SURPRISAL
Predictions:
Garden-Path Sentences
Garden-Path Sentences and Surprisal
While Anna dressed the baby was crying
- there is something highly unpredictable about “was” in this context
- high surprisal on encountering the disambiguating region
- GP sentences induce higher surprisal, especially when the upcoming structure is least predictable from the context
Predictions:
Is the brain a prediction machine?
Predictions: is the brain a prediciton machine?
Scholars disagree
- some say “yes”
- “the goal of your brain is to make predictions about future events (Clark, 2013)
- prediction =
- essential for survival
- important for learning (recognizing patterns)
- mental mechanisms (active inferencing: adjusting priors to better inform and make more precise predictions in the future)
- others say “no”
- variability in prediction across populations
- literacy skills
- language proficiency (native speakers/second language learners)
- even among native speakers of the language, there is evidence that we don’t predict all the time
- maybe some lack certain required resources, eg:
- different attention and memory capacities
- 2nd language learners may be missing some important knowledge/linguistic elements
Complex Sentences
Ex: same set of words, but not the same difficulty
1) The senator [who spotted the reporter] shouted
- less difficulty
2) The senator [who the reporter spotted] shouted
- more difficulty
- these sentences contain relative clauses; informs the relationships between NP and verbs
Relative Clause Types
Relative Clause Types:
1) Subject relative clause
2) Object relative clause
Relative Clause Types:
1) Subject relative clause
Subject Relative Clause: the relative pronoun takes the place of the subject of the clause; followed by a verb
- called “subject relative clause” bc the subject is the element removed
The senator who [ _ _ _ spotted the reported] shouted
- filler: the noun expected to be in the gap (the senator)
- gap: position where NP is removed (the blank)
The senator spotted the reporter
- The senator who [ _ _ _ spotted the reporter]
The senator shouted
- remove relative clause
Relative Clause Types:
2) Object relative clause
Object Relative Clause: the relative pronoun takes the place of the object of the clause; followed by subject + verb
- called “object relative clause” bc the object is the element moved
The senator who [the reporter spotted _ _ _ ] shouted.
- filler: the senator
- gap: the blank (object position)
The reporter spotted the senator.
- The reporter who [ the reporter spotted _ _ _ ]
The senator shouted.
Subject VS Object Relative Clauses
Subject Relative Clause
- The senator who [ _ _ _ spotted the reporter] shouted
Object Relative Clause
- The senator who [ the reporter spotted _ _ _ ] shouted.
- longer reading times for object relative clauses than subject relative clauses
Why?
- prediction
- subject relative clauses are more frequent (MORE PREDICTABLE) than object relative clauses
- memory
- storage of linguistic elements in memory
- distance between filler and gap—how much information has to be processed
