Memory and Cognition
PSY 324
Topic: Language
Dr. Ellen Campana
Arizona State University
What is language?


“a system of communication using sounds or
symbols that enables us to express our feelings,
thoughts, ideas, and experiences”
Human (animals have simpler systems)
Bees signal through “waggle dance”
 Chimps have simple calls

Snakes
 Eagles
 Leopards

Essential Properties of Language






Semanticity
Arbitrariness
Flexibility of Symbols
Naming
Displacement
Productivity
Productivity of Language
(“Creativity” in the text)

Language is hierarchical
Made up of different parts that can be combined
 Parts form a hierarchy


Larger and larger units
Hierarchical Structure
S
NP
Det
VP
N
the mouse
V
saw
NP
Det
N
the
cat
Productivity of Language
(“Creativity” in the text)

Language is hierarchical
Made up of different parts that can be combined
 Parts form a hierarchy



Larger and larger units
Language is governed by rules

Some things are permissible and others are not
OK: What is my cat saying?
 NOT OK: Cat my saying is what?

Universality of Language

Everyone with normal capacities develops a
language and learns its complex rules implicitly


Not talking about prescriptive grammar, but
knowledge that produces permissible language
Language occurs in all cultures
Over 5000 different languages
 No culture without a language
 Even isolated communities have language

Universality of Language

Urge to communicate is powerful


Children who are deaf but not exposed to sign
develop their own languages (homesign)
Language development consistent across cultures
Babbling @ 7 months
 First words @ 1 year
 Multiword utterances @ 2 years


All languages are “unique but the same”
Unique but the same

Languages are unique
Different sounds, words, rules
 All different, but sometimes similar


Languages are the same
Have words that serve as nouns / verbs
 Have a system to make things negative (e.g. “not”)
 Have a way to ask questions
 Have a way to refer to past and present
 Have a way to refer to things that are not present

Study of Language


Wundt wrote about language in 1900 but detailed
study started with cognitive revolution
Skinner: Verbal Behavior


People are trained to speak through conditioning
Noam Chomsky

Syntactic Structures described similarities and differences
across languages (Zellig Harris work)


Humans are genetically programmed for language
Critique of Verbal Behavior introduced the poverty of
the stimulus argument, which shattered behaviorism
Poverty of the Stimulus

Behaviorist argument: children learn on the basis
of feedback from parents
Child says “cat eat”
 Parent corrects with “the cat will eat”


Poverty of the stimulus argument
Children do not get enough feedback to learn
 Language requires production of things never heard
before


Classic example: “I hate you, Mommy.”
Study of Language

Disciplines that investigate language
Linguistics: like philosophy + anthropology
 Natural Language Processing: computer science / AI
 Psycholinguistics: cognitive psychology


Psycholinguistic study of language
Comprehension – how we understand language
 Production – how we produce language
 Representation – how we represent or code language
 Acquisition – how we learn language

Language
Comprehension
Levels of Language






Pragmatic Level (use in the real world)
Semantic Level (meaning)
Syntactic Level (sentences)
Lexical Level (words)
Morphological Level (meaningful parts)
Phonological Level (sounds)
Understanding Words
Comprehension of Words

Lexicon: All of the words a person understands
Often called a “Mental Dictionary”
 Adults have over 50,000 different words
 Contains meaning, grammatical category,
phonemes, and rules about combining with
morphemes

Comprehension of Words

Phonemes: The sounds of a language

Shortest segment of speech that, if changed, alters
the meaning of the word within a language


Different languages have different sounds (and
therefore different phonemes)


Not quite the same as our letters, but close
Number of phonemes varies by language
Morphemes: in between words and sounds

Smallest unit with a definable meaning OR
grammatical function

Examples: truck, -ed, -s, banana
Perceiving Words


Perception of (spoken) words is about how we
link the sounds we hear to our lexicon
As in the perception unit, it’s useful to think
about top-down and bottom-up processes

Top and bottom determined by levels of language
Levels of Language






Pragmatic Level (use in the real world)
Semantic Level (meaning)
Syntactic Level (sentences)
Lexical Level (words)
Morphological Level (meaningful parts)
Phonological Level (sounds)
Perceiving Words


Perception of (spoken) words is about how we
link the sounds we hear to our lexicon
As in the perception unit, it’s useful to think
about top-down and bottom-up processes
Top and bottom determined by levels of language
 In the book “context” is usually top


HUGE topic, will only be able to give you a bit
of an overview
Perception of Words

Meaning of a word create a context that helps us
actually hear the sounds of the word


Top-down effect on phoneme perception
Phonemic Restoration Effect (Warren, 1970)
Researchers edited coughs into sentences (actually
replacing phonemes with the cough, not mixing)
 Participants heard sentences, had to say what they
heard and where the cough occurred

Warren (1970)
Sentence:
The state governors met with their respective
legislatures convening in the capital city.
Results:
 People couldn’t report where the cough was
 People didn’t know the /s/ was missing
Explanation: People seemed to “fill in” missing info based
on context provided by sentence (and lexicon)
Warren (1970)
Sentence:
It was time to ave goodbye to the family.
Results:
 People couldn’t report where the cough was
 People didn’t know the /w/ was missing
Explanation: People seemed to “fill in” missing info based
on context provided by sentence (and lexicon)
Warren (1970)
Sentence:
It was time to ave up for a new roof.
Results:
 People couldn’t report where the cough was
 People didn’t know the /s/ was missing
Explanation: People seemed to “fill in” missing info based
on context provided by sentence (and lexicon)
Phonemic Restoration Effect

People use context to “fill in” missing or
degraded sound information
Can be context before or after the sound itself
 Very quick, people aren’t aware


It is a demonstration of a top-down effect
(context effect) on word perception
Speech Segmentation

Ever notice how…



Language you don’t know – words blend together
Your language – words seem separate
You are using your knowledge of the language to find
the word boundaries
Meaning and Segmentation

Same signal segments differently in different
sentences
Be a big girl and eat your vegetables
 The thing Big Earl loved most was his truck


Fun demo: Mad Gab game. For this you need to
find a friend to help you.
Mad Gab Instructions

Take turns with the following roles:
Person A: Close your eyes, listen, and try to figure
out what the other person is saying
 Person B: Read the slide out loud (you may have to
repeat a few times).


Talk about it – Person A will hear something
different than what Person B was saying
“Mad Gab” demo
Ask
Rude
Arrive
Her
“Mad Gab” demo
Eight
Ape
Reek
Quarter
“Mad Gab” demo
Amen
Ask
Hurt
“Mad Gab” demo
Eye
Mull
Of
Mush
Sheen
“Mad Gab” demo
I’ve
Hailed
Ink
Lush
Mad Gab

What was the point of doing this for class?

Meaning affects which phonemes you hear and
where the word boundaries are (top-down)
This meaning comes from your previous experience with
Language
 Based on actual sentences or phrases in English, rather
than just words


There are also bottom-up aspects of speech
segmentation
Transitional Probabilities and
Segmentation

Transitional probabilities: the chances that
one sound will follow another sound
Sounds “pretty” more likely than “tyba” in English
 When we hear “prettybaby” we segment it into the
words “pretty” and “baby”


We can learn to do this even if the words are not
meaningful, through statistical learning
Tested with 8-month-olds (Saffran&colleagues, 1996)
 Still knowledge, just knowledge of the language

Statistical Learning

Study with 8-month-olds

Training: infants heard a stream of nonsense
“words” for two minutes
…bidakupadotigolibutupiropadotibidaku…
 No pauses, random order, flat intonation


Testing: infants chose how long to listen to test
stimuli by turning their heads
Whole word: padoti…padoti…padoti
 Part word: libutu…libutu…libutu


Infants could tell which one (preferred part)
Perceiving Letters

So far we’ve been talking about understanding
the sounds in spoken language.


Top-down and bottom-up effects
Understanding written letters is similar
Visual perception of the letters themselves is both
top-down and bottom-up (from chapter 3)
 Word superiority effect shows that words affect
processing of letters (top-down)

Perception of Letters
Word Superiority Effect

Coglab / Example in the book….
Stimuli: word (FORK), nonword (RFOK), or letter
(K), followed by a target & distractor
 Task: choose the target


Finding: People are more accurate and faster at
picking the target when the stimulus is a word,
compared to when it is *alone* or part of a
nonword
Word Superiority Effect

What’s the point of this study?


Demonstrates that the top-down context (the word)
helps with processing
There’s a model of this that my help you
visualize this.
Picture “spreading activation” going from bottom to
top and then back down
 Only some features are shown – the model would be
complete
 If this doesn’t help, it’s OK! (not on exam)

Interactive Activation Model
Word Level
Letter Level
ROOF
FORK
F
K
Feature Level
Stimulus
K
O
R
Understanding Words
Word Frequency Effect


We respond more quickly to high frequency
words (home vs. hike).
Supported with lexical decision studies
Task = word or nonword
 Findings = faster to say a high-frequency word is a
word than to say a low-frequency word is a word


Supported by faster reading times
Eye movements
 Overall reading times (story with pretty vs. demure)

Lexical Ambiguity

Same word has different meanings
“Bugs” = insects OR recording devices
 “Bank” = river bank OR financial institution


When reading, both meanings are accessed right
away, but then context overrides one of them

Shown with lexical priming at “bug” in a story
Simultaneous presentation: “spy” / “ant” equally fast
 200 ms delay: context-specific meaning faster

Understanding
Sentences
Levels of Language






Pragmatic Level (use in the real world)
Semantic Level (meaning)
Syntactic Level (sentences)
Lexical Level (words)
Morphological Level (meaningful parts)
Phonological Level (sounds)
Syntax vs. Semantics

Meaning / Semantics
Semantic violation: “The cats won’t bake.”
 It’s English but it doesn’t seem meaningful


Form / Syntax
Syntactic violation: “Cat bird the chased.”
 You can guess what it might mean, but it doesn’t
follow the rules of English


Recall discussion from Chapter 2 – different
brain areas for these two levels of language
Parsing

Most of the time language is successful
No syntactic or semantic violations
 Syntax and semantics work together


Meaning of whole sentence depends on syntax
How words are grouped together, or parsed, can
have a major effect
 Example: Ambiguous sentences
 Example: Garden Path sentences

Ambiguous Sentences

Ambiguous means that it has more than one
interpretation
We saw ambiguous man-rat pictures in the
perception chapter
 Some sentence ambiguities depend on structure


Example: “I saw the spy with the telescope”
Option 1: Phrase “with the telescope” is grouped
with “spy”
 Option 2: Phrase “with the telescope” is grouped
with “saw”

Ambiguity
SPEAKER
ACME
TELESCOPE
JOHN
“I saw the spy with the telescope.”
Ambiguity
SPEAKER
ACME
TELESCOPE
JOHN
“I saw the spy with the telescope.”
Ambiguity
SPEAKER
ACME
TELESCOPE
JOHN
“I saw the spy with the telescope.”
Temporary Ambiguity

In the spy example we never really know which
parse is the right one (it remains ambiguous) but
for others the structure becomes clear over time

Time flies like an arrow. Fruit flies like a banana.
Garden Path Sentences


We process language as it unfolds (sound by
sound, word by word)
Part of this involves parsing (grouping words)
Sometimes we have to revise these groupings as we
get new information
 Feels very confusing!

Garden Path Sentences

Read this and pay attention to how confusing it
feels:
The man who whistles tunes pianos.

You were likely confused when you got the
word “pianos”
First parse: “tunes” grouped with “whistles”
 Revised parse: “tunes” grouped with “pianos”

Approaches to Parsing

Syntax-first Approaches – parsing is based on
syntax and later compared with semantics

Late Closure


Minimal Attachment (not in our book)


Attach to current constituent if possible
Group words in the simplest way
Interactionist Approaches – parsing is based on
interactions between syntax and semantics

Meaning affects parse from the earliest moments
We never ever think of this one!
(as in: I saw wood with the bandsaw)
SPEAKER
ACME
TELESCOPE
JOHN
“I saw the man with the telescope.”
Evidence for Interactionist
Approach to Parsing

Lexical Semantics
Man with binoculars vs bird with binoculars
 Meaning of the first word (man/bird) affects parses


Pragmatics / Environment

Tanenhaus et al studies
Head-Mounted Eye Tracker
• Like looking into someone’s thoughts
• As they happen, in a real environment!
600.465 - Intro to NLP - J. Eisner
59
slide courtesy of M. Tanenhaus (modified)
Videotape
• From Mike Tanenhaus’s lab
– University of Rochester
Eye camera
Scene camera
600.465 - Intro to NLP - J. Eisner
60
slide courtesy of M. Tanenhaus (modified)
PP Attachment Ambiguity
Put the apple on the towel in the box.
Only one apple  Garden Path
Put the apple on the towel in the box.
Two apples  use PP to clarify which apple, no garden path
600.465 - Intro to NLP - J. Eisner
61
slide courtesy of M. Tanenhaus
One referent context
garden path
oops!
backtrack
slide courtesy of M. Tanenhaus
One referent context
slide courtesy of M. Tanenhaus
Two-referent context
amazing
lack of
oops
Tanenhaus Study

What????
If there are two apples, people don’t get confused
because “on the towel” is needed to distinguish
between the two apples -- this means that only one
possible parse makes sense
 If there is only one apple, people are confused
because both parses are OK, but the one that groups
“on the towel” with “put” is more natural.


Garden path effect
Understanding Text
and
Stories
Constructive Nature of Language

In previous chapters we’ve discussed cases
where we construct aspects of experience
Top-down aspects of vision
 Gestalt processes
 Scripts / Schemas
 Memory
 Eyewitness Testimony


Language understanding is constructive too!

Construction = “inference”
Inferences

Anaphoric inference
What do “he” / “she” / “it” / “they” mean in
context?
 Which cat is “the orange cat” or “my favorite cat”?


Instrument inference
Was an instrument used to perform an action?
 Use our scripts to make these inferences


Causal inference

Are sentences linked by causal relationships?
Situation Models

We imagine the details of situations we read
about using our knowledge of the world
Scripts, schemas, inference
 Visual imagery, spatial relationships
 Simulations (like imagery but including other sense,
and actions)


Studies demonstrating these
Eagles and nails
 Vampire TV show

Physiology of Situation Models

When reading
Areas of the brain related to performing an action
are active when reading about the action
 Areas of the brain involved in sensing (visual,
auditory, smell, etc.) are active when reading about
sensed environments


Distributed Activity
Reading involves the whole brain
 Changes of different types associated with different
areas

Producing Language
Conversations
Language Production in General

How do people understand each other in
conversations?
Conversations involve simultaneous language
comprehension and production
 Each of these is an incredibly complex topic, so the
book can only give a little taste of the area


Coordination – staying “on the same page”
Semantic Coordination
 Syntactic Coordination

Semantic Coordination

Recall that language is constructive
Anaphoric inferences allow us to link referring
expressions (i/we/ they / the beer) to objects in the
real world
 Speakers and listeners need to coordinate to make
sure they are always making the same inferences


Example: Given / New contract
Definite noun phrases (“the x”) refer to previously
introduced objects
 Followed by both speakers and listeners

Syntactic Coordination

Ambiguity can make it harder to communicate
quickly (e.g. garden path sentences)
Speakers and listeners coordinate syntax to reduce
ambiguity
 Example: Syntactic priming… I say “Sue gave the
boy the ball” you say “Sally gave the girl the bat”
(same syntax) not “Sally gave the bat to the girl”


Often syntactic coordination (including priming)
happens automatically and unconsciously
Culture, Language and
Cognition
The Sapir-Whorf Hypothesis

Basic idea: The language people use affects the
way they perceive the world around them
Example: color terms in a language determine how
well we can discriminate (i.e. tell the difference
between) different colors
 Also: Objects, numbers, space, mathematical
concepts


Left vs right brains process concepts differently
(left more likely to show Sapir-Whorf patterns
THE END


This chapter just scratches the surface of the
exciting world of language, but I may be biased
because that’s what I do…..
If you are interested in language as a topic and
would like to become involved in this kind of
research, get in touch with me
PGS 399 credit
 PGS 499 credit
 Honors thesis credit

Descargar

Cognition and Perception in Hybrid Environments