Syntactic processing (parsing)

advertisement
Exam 3 Lecture Notes:
(page down to bottom for list of most important topics to study)
Syntactic processing (parsing)
I. Syntax: Questions for psychology
A. Do listeners/readers compute syntactic structure during comprehension?
B. If so, how do they do it?
II. Preliminaries: How might a parser work?
deterministic/nondeterministic parsing
parallel/wait and see/serial parsing models
III. What type of parser do people have in their heads?
A. Click studies
B. ERP study
1. What type of parser
a. syntactic ambiguity studies with eye tracking (Fraizer, Rayner, Ferreira, etc)
Evidence of serial parser with preference for:
simple syntactic structures (“garden-path” model)
OR
most frequent structures (“constraint-based” model)
2. Relationship between syntactic parser and semantics
Chomsky pointed out that syntax and semantics might be very different things:
“Colorless green ideas slept furiously” : semantically weird, syntactically fine
“Jane go store buy fish”
syntactically weird, semantically OK
Chomsky’s claim: Syntax and semantics are distinct
a. Evidence of syntactic “modularity”
1. syntactic ambiguity research
Ferreira & Clifton (1986)
2. ERP evidence (Osterhout & Nicol, 1999)
Research question: If syntactic processing and semantic processing involve
different sets of processes, then the brain might respond differently to syntactic
and semantic anomalies.
Materials:
(1) The cat will eat the food.
(2) The cat will bake the food. (semantically anomalous)
(3) The cat will eating the food.
Results:
Semantic anomalies elicit an N400 effect.
Syntactic anomalies elicit a P600 effect.
So, maybe there is a “syntactic parser”
But: ERP Study by Kim & Osterhout (2005) with sentences like “The hearty
meal was devouring” found that the verb devouring showed evidence that in some
cases, semantics, not syntax, “drives” sentence parsing. This is inconsistent with
the claims of Frazier, Ferreira, etc.
-------------------------------------------------------Semantics
A. Definitions
meaning -- agreed-upon interpretation of certain strings of sounds and letters
Great example of "lack of agreed-upon interpretations: British and American
English
My experience with "Way Out" signs
List of British - American differences in meaning of specific words
One problem of meaning: POLYSEMY
polysemy -- words that have more than one meaning
some words have more than one meaning, and the intended meaning is not always
obvious
Marjie's example: Sign at laundromat said, "No Dying in the Washers"
NO DYING IN
THE WASHERS
(she interpreted dying = cease to live)
Ironically -- In English, the word "mean" itself is polysemous:
Meanings of "mean"
Ask for meaning of the word "mean"
(a) That was no mean accomplishment.
(mean = insignificant)
(b) They are so mean to me.
(mean = cruel)
(c) This will mean an end to our regime.
(mean = result in)
(d) This means so much to me.
(mean = is important)
(e) I mean to help if I can.
(mean = intend)
(f) Keep off the grass! This means you!
(mean = refers to)
(g) Those clouds mean rain.
(mean = are a sign of)
(h) She didn't mean what she said.
(mean = believe)
So what do we mean when we talk about meaning? What seems to be so commonsensical is really very difficult to define. We’ll discuss a couple of ideas.
semantics -- the study of the meaning of words and sentences
B. Types of Meaning
semantics
literal
Lexical
semantics
(word
meaning)
nonliteral
Structural
semantics
(sentence
meaning)
metaphors, idioms
sarcasm, irony
C. Lexical semantics (Word meanings)
What do words mean?
One answer -- look up the word in Webster's.
But it turns out that the definition in Webster's is not a sufficient description of word
meaning.
We'll briefly look at two theories of word meaning.
1. Semantic Feature (Decompositional) Theory
claim: the meaning of a word can be decomposed into a set of essential (necessary and
sufficient) semantic features
captures “denotation” but not “connotation” of word’s meaning
denotation – word’s core meaning
connotation – word’s ancillary meaning
underlying assumption: Every word has a fixed, identifiable meaning.
Examples:
Feature
chair
human
older
female
father
boy
son
bachelor
dog
+
+
-
+
-
+
-
+
+
-
?
?
Two problems with Semantic Feature approach:
a. Difficult to identify semantic features for many words
- try to come up with a set of features that defines: "
game
bird
chair
happy
b. Semantic features don't capture relational aspects of definitions
"father" is a relational noun; To be a father is to be the male parent of someone, that is, to
bear the relation father to some other person. The lists of semantic features shown above
do not represent this notion of relation.
Note -- not all words are relational
2. Prototype theory
claim: word meanings are represented as "prototype exemplars"
underlying assumption: word meanings are fuzzy, not fixed
Example:
bird: properties associated with our "prototype" for a bird (e.g., robin)
father: prototype for father (e.g., image of your own father, or of a TV father, etc.)
The decision over whether something is a bird depends on how close the object is to your
prototype for bird, not on whether the object has the semantic features associated with
"bird"
Notion of “Basic Levels” (Rosch, 1978):
Problem with prototype theory:
1. Not all concepts have a prototype (e.g., what I the prototype of “truth”?)
2. No clear explanation or definition of similarity, which is absolutely central to the
theory
D. Structural Semantics (Sentence Meaning)
The meaning of a sentence is not simply the sum of the meanings of the words in the
sentence.
That is, the meaning of a sentence is not simply the sum of the meaning of its parts.
Why not???
Some of the meaning in a sentence is determined by the structure of the sentence!
For example:
(a) John kissed Mary.
(b) Mary kissed John.
These sentences do not mean the same thing, even though they contain the same words.
1. Thematic Roles
thematic roles -- semantic roles played by the noun phrases in a sentence
Linguists generally believe that each noun phrase in a sentence MUST play ONE and
ONLY ONE "thematic role".
Some examples of thematic roles:
Agent: the cause of the action (the "doer")
Theme: the "recipient" of the action
Instrument: the instrument used to cause the action
Goal: where the action is directed
Source: where the action originated
etc.
Examples of thematic roles in sentences:
Let's try to identify the thematic roles played by the NPs in some sentences -(ask for class participation)
(1) Mary put the book on the shelf
Agent
Theme
Goal
(2) Bill sent money to the foundation
Agent Theme
Goal
(3) Bill sent the foundation money
Agent
Goal
Theme
#4 is tricky:
(4) [John] received [the present] from [Mary]
Theme? Goal?
??
Agent? Source?
Question: Where do thematic roles come from? That is, what determines which
thematic roles are available to be assigned to NPs in a sentence?
Answer: Each verb is associated with a specific set of thematic roles, and it assigns these
roles to NPs in sentences.
Examples of the "thematic structure" of verbs:
Have students guess what the set of roles are:
kill <Agent, Theme> "The cat killed the mouse"
die <Theme> "The dog died"
put <Agent, Theme, Goal> "John put the book on the table"
Semantic Disorders
A. Semantic dementia (Breedin & Saffran, 1999)
Patient D.M. –
56 year old male with an advanced degree
suffered from progressive semantic dementia
- can’t name pictured objects well at all
- has lost the meaning of lots of words
- generally confused
damage to anterior part of the left temporal lobe, bilaterally
Despite tremendous loss of semantic knowledge, he’s nearly perfect at
judging whether a sentence is grammatical or not. he’s much worse at
deciding whether a sentence makes sense or not.
Conclusions: Suggests that syntactic processing is separate from ssemantic
processing
B. Category-specific deficits (see text)
-------------------------------------------------------Language Production
Important patterns in speech error data
1. errors usually involve only certain units of language
words, morphemes, and phonemes
inference: these units are the ones that are involved in speech planning
2. there are differences between word/morpheme errors and sound errors
a. word/morpheme exchange errors:
- usually involves morphemes from the same linguistic class
content words swap with content words,
function morphemes swap with function morphemes
- swapped items can span many words in the sentence
b. sound (phoneme) exchange errors:
- usually involves words from different linguistic classes (e.g., nouns and
verbs; content and function morphemes)
- swapped sounds usually come from adjacent words
inference: words and sound involve different stages of planning
3. A morphologically complex content word “leaves behind” its inflectional morphology
(stranding errors)
Intended: I sampled some randomly
Spoken: I randomed some samply
inference: content (root) words and function morphemes are planned
independently
4. The function morphemes are almost always phonologically appropriate for the
sentence as it is actually uttered, NOT as it was intended to be.
Intended: a maniac for weekends
Spoken: a weekend for maniacs
pronounced the s as /s/, not /z/
Intended: a language acquisition
Spoken: an anguage lacquisition
the determiner /an/ was used, not /a/
Intended: I regard that as imprecise
Spoken: I disregard that as precise the negation morpheme used is dis, not im
Intended: easily enough
Spoken: easy enoughly
inference: “sound” is planned at a relatively late stage in speech planning, and
accommodates the errors.
-------------------------------------------------------Disorders of Speech (Aphasia)
I. “Classical model” of aphasia (Wernicke & Lichteim, 1880s)
Aphasic Syndrome
Broca’s
Functional Loss
“Articulatory image”
of words is lost
(spoken word production)
Site of Damage
Broca’s Area
Wernicke’s
“Auditory image” of
Wernicke’s area
words is lost
(spoken word identification)
Conduction
Identity of word to be
repeated cannot make it
Arcuate
Fasciculus
(area on map)
BROCA
SMG,
STS POST,
MTS POST
from Wernicke’s area
(where it is identified) to
Broca’s area (where the
“articulatory image” for
the word is)
II. Evaluating the classical model
A. Problems with claims about functional loss
1. Broca’s aphasia
3 reasons for believing that it’s more than just a problem moving the articulators:
a. Broca’s aphasics often write like they speak.
b. Many Broca’s aphasics sing fluently.
c. Many Broca’s aphasics have a comprehension problem that
mirrors their production problem.
Experimenter reads aloud one of these 2 sentences:
(1) John hit Mary.
(2) John was hit by Mary.
active sentence
reversible passive sentence
Patient points to the picture that matches the sentence:
Picture of John hitting Mary
Picture of Mary hitting John
Functional deficit in Broca’s aphasia (current opinion):
1. Agrammatism (problem in producing AND understanding sentences)
2. some speech apraxia (difficulty moving articulators)
2. Conduction aphasia
These patients have extreme difficulty holding onto phonological
information in their short-term memory
Digit-span test: Ask patient to remember string of 7 spoken digits (or 7 onesyllable words)
Functional deficit in Conduction aphasia (current opinion):
1. Deficit in holding phonological information in hort-term memory.
B. Problems with claims about damaged area that leads to each syndrome
(Dronkers, 2000)
III. Modern neuroimaging evidence
Language tasks almost always activate the “perisylvian” cortex in the left
hemisphere
Phonological processing seems to occur in the association cortex surrounding
auditory cortex
Much less success at localizing syntactic and semantic processing
END OF LECTURE NOTES
_____________________________________________________________________
Topics to study particularly well for Exam 3
1. Sentence processing
a. types of parsers
b. evidence that people compute syntactic structure
c. evidence concerning what type of parser people have in their heads
d. influence of semantics/pragmatics on the parsing process
2. Semantics
a. types of meaning
b. theories of meaning (features, prototypes, etc)
c. structural semantics
d. semantic dementia
3. Language Production
a. understand the Garrett/Levelt model of speech production; understand how the error
data support it
b. identify types of speech errors
c. identify the stage of speech production a particular error is most likely associated with
4. Aphasia
a. know the syndromes and the diagnostic criteria associated with each
b. understand the classic Wernicke-Lichteim model
c. know the problems with this model and current thinking about aphasia, and (in general)
how language is represented in the brain
Download