Or what I did in my MSc

advertisement
Psycholinguistics
Or what I did in my MSc
Natasha Dare
This talk



What is psycholinguistics?
The origins of language
Some of the major areas of research



Famous experiments
Terminology
State of the art



Particularly what is being done at Edinburgh
When language goes wrong
Not controversial
What is psycholinguistics?

Psychological processing of language



Part of cognitive science
Input from neuroscience, informatics and
linguistics
Why is this interesting/important?
“No man is an island, entire of itself; every man is
a piece of the continent, a part of the main.”
John Donne, Meditation XVII
Origins of language - Child
language

Infants very quickly learn about language









3-day old neonates prefer the sound of their mother’s voice
(DeCasper & Fifer, 1980)
4-day old neonates prefer listening to their parents’ language
(Mehler & Dupoux, 1994)
1-month old babies can distinguish between speech sounds
(Eimas, Miller, & Jusczyk, 1987)
6 weeks: cooing
6-9 months: babbling
12 months: initial word use
18 months: vocabulary explosion of 40 new words per
week
24 months: short sentences
36 months: 90% intelligible
Origins of language - Animal
communication

Many animals have complex communication systems



But is this true language?



Vervet monkeys: leopard vs snake vs eagle
Chaffinches: combined territorial and mating calls – an initial trill
to deter males, and a final flourish to attract females
Animal communication is holistic
Human language is compositional
Can see origins of human language in animals


e.g. teeth baring ritual = threat -> symbolic acts
courtship gift = attention to third entity -> reference
“It is nothing other than words which has made us human”
Pavlov
How did human language develop?

Robotic agents to simulate evolution, especially
front
back
emergence of regularities




i
u
high
5 vowels most common (Latin)
e
o
a
Maximal acoustic distance
low
De Boer (1997) - regularities emerge spontaneously
after 3,000 games
Pidgins -> creoles


Pidgins are formed when communities are
deliberately mixed e.g. Ruso-Norsk
Creoles develop from pidgins but have full syntax and
native speakers

Tok Pisin: sapos = if, bilong = possessive
The problem
Bilingualism
dvs
Production
Comprehension
lexicon
Spoken word
production
Written word
production
Dialogue
Spoken word
comprehension
Written word
comprehension
/Discourse
Some linguistics terms

Semantics


Syntax


Sounds of language e.g. thin = [qIn]
Phonology


Specialised term for grammar – word order
Phonetics


Meaning
Language-specific phonetics e.g. rhotic [r] in party in USA
Morphology

Words and word formations



Sometimes change the underlying meaning (re-)
Sometimes don’t (-s)
Pragmatics

Language use
Auditory word recognition


What are the sound representations used to
access the lexicon (mental dictionary)?
Very difficult task





We understand 20 phonemes per second
We can recognise words in context 200 ms after
onset (Marslen-Wilson, 1984)
Miller and Jusczyk (1989): invariance (phonemes
sound different in different contexts) and
segmentation (speech slurs words together)
Assimilation of sound properties from other words
Co-articulation of words
Auditory word recognition cont.

Two cues: uniqueness point and context
/t/ /tr/ /tre/ /tres/ /tresp/ /trespass/
uniqueness point

Cohort model (Marslen-Wilson, 1989)


Bottom-up: Access a cohort of possible words,
select one, integrate its properties into current task
TRACE model (McClelland & Elman, 1986)

Interactive: Context (top-down) and acoustic
signal (bottom-up) both cause one candidate to be
selected
Visual word recognition

How do we know if a letter string is a lexical item?


What factors affect the ease of recognition?




Two levels of checks: phonetic constraints on impossible nonwords (mxbt), phonological/semantic constraints on possible
non-words (kstreem/glub)
Frequency: have easier than jade
Neighbourhood: mine easier than much
Length: bank easier than discriminate
Lexical ambiguity



Homophones: knight/night
Homographs: lead
Meanings versus senses: bank versus film
Visual word recognition cont.

Connectionist modelling of orthography (letters) ->
phonology (sounds) e.g. split-fovea model (Shillcock,
/a//o/ /l/ /g/ /d/
Ellison, & Monaghan, 2001)






output
Neurally inspired
Distributed network of layers of nodes
input
Words are represented by patterns of
o d
activation between associated words and sounds
Models are trained with word-pronunciation pairs
They learn by reducing the error between the
actual and desired outcome
a
l g
Dyslexias – developmental (necisary), surface (mint),
phonological (kint), deep (sympathy -> orchestra)
Sentence comprehension


How do we parse sentences (build up syntactic
representations) given the meanings of words?
Phrase structure rules -> tree diagrams of
sentences
S


S -> NP VP
VP -> V NP
NP -> Det N
How do we access this seemingly
effortless task?
NP
Det N
VP
V
NP
Det N

Ambiguity – two or more possible structures The dog bit the man

Reassessment using garden-path sentences
(Frazier, 1987)

“enraged cow injures farmer with axe”
 The horse raced past the barn fell – reduced
relative clause
Sentence processing cont.

Eye-tracking and self-paced reading: longer time spent
looking at a word = greater processing difficulty


Does semantics help with parsing i.e. are we garden-pathed
when there are strong semantic pointers to one interpretation?
Ferreira and Clifton (1986): The defendant/evidence examined
by the lawyer turned out to be unreliable


Although evidence disambiguates examined, eye-movements
showed that semantic information did not stop people from being
garden-pathed
But how much do we actually parse at all?

Christianson, Hollingworth, Halliwell & Ferreira (2001): While Bill
hunted the deer ran into the woods

Participants thought that the deer ran into the woods and that Bill
hunted the deer
Discourse
“One night there flew over the city a little Swallow. His friends had gone away to Egypt six
weeks before, but he had stayed behind, for he was in love with the most beautiful Reed. He
had met her early in the spring as he was flying down the river after a big yellow moth, and had
been so attracted by her slender waist that he had stopped to talk to her.
"Shall I love you?" said the Swallow, who liked to come to the point at once, and the Reed
made him a low bow. So he flew round and round her, touching the water with his wings, and
making silver ripples. This was his courtship, and it lasted all through the summer.”
Oscar Wilde, ‘The Happy Prince’

How do we maintain coherence across longer texts?
 Inferences




Logical: Toby is a bachelor – Toby is a man
Bridging: integrating new with previously given information.
Uses anaphor e.g. John gave Bob the book. He liked it very
much – who do it and he refer to?
Elaborative: extending what is in the text to world
knowledge
Tend to only remember gist/important points e.g.
passages with personal significance
Language production

3 parts: conceptualisation, formulation,
articulation





Pre-verbal message
Concept -> linguistic form
Articulatory planning
Motor execution
Although we are very good at this (1/1000
words is an error), errors are very revealing

Spoonerisms: exchange of initial consonants
“The Lord is a shoving leopard to his flock”
“I’d rather have a bottle in front of me than a frontal lobotomy”

Freudian slips
“A Freudian slip is when you say one thing but mean your mother”
Language production cont.

Tip of the tongue states




To formally renounce the throne


Temporary inability to retrieve word despite ‘feeling of knowing’
Can retrieve partial information e.g. gender, initial phoneme,
number of syllables
Blockers/pop-ups
begins with a, like abduct
Suggests that there is a separation between
syntax/semantics (lemma) and morphology/
phonology (lexeme)


Levelt (1992)
Dell (1986)
The lexicon


How do we know what words mean?
Different methods proposed
Defining attributes (ISA links)
animal breathes/skin
superordinate
/skeleton
wings/flies
gills/swims
bird
fish
/feathers
/fins
canary
robin
salmon
subordinate
 Exemplars/typicality effects

Concepts partly based on perception e.g. brown,
sticky
 But not the whole story e.g. the, him

The lexicon cont.

Semantically similar words are interchangeable
in sentences
The child slept on the bed
The dog walked on the carpet

Semantic networks (e.g., Burgess & Lund’s HAL)


Meanings come from other words (like a dictionary)
Distances between words in network show
relatedness, with 140,000 dimensional space
kitten
lion tooth
oyster
car
face
eye
leg
foot
Dialogue

Two or more people – turn-taking, feedback





Difficult to study as fewer paradigms, noisier data, harder to
control
Commonly use tasks or games to elicit controlled language
Audience design: do speakers tailor their utterances to
the listeners?
Common ground: do listeners use information known
only to themselves?
Perspective-taking: do speakers and listeners
take each other into account?

Schober and Clark (1989) tangram matching task

Matchers 99% accurate, overhearers 78%
Dialogue cont.

Alignment (Pickering & Garrod, 2004)

Linguistics representations used by interlocutors become aligned
at multiple levels via priming






Lexical: use the same referring expressions
Syntactic: ‘At what time do you close?’ ‘At 9’
Accent and speech rate
Alignment permeates throughout levels
All happens automatically
Prosody, disfluency

What information do they give the listener?



When Roger leaves the house is dark/it’s dark
Kjelgaard and Speer (1999): when syntactic and prosodic cues
matched, listeners’ comprehension was facilitated
Does it follow that they are produced intentionally by the
speaker?
Bilingualism

Do we have separate language systems for each
language?



semantics
One system – semantic priming
produces facilitation between languages
Two systems – aphasia can affect one
L1
language only
Does age/proficiency explain these contradictions?




Problem: no standard, widely-used proficiency test
Hard to generalise across results
Country/community/family/colleagues all have effects
Most likely is one semantic store, two lexicons
L2
Final facts




6,912 known living languages in the world
896,190 words in English (correct as of
yesterday)
Mandarin has 1,075,000,000 speakers
Most popular word is ‘ok’
“When ideas fail, words come in very handy”
Goethe
Download