Transcript Words in the mind

Words in the mind
FROM THE BOOK:
AITCHISON, JEAN (1987). AN INTRODUCTION TO THE
MENTAL LEXICON
PUBLISHER: OXFORD, UK ; BLACKWELL, NEW YORK,
NY, USA
Main aims
 How do humans manage to store so many words?
 How do they find the ones they want?
 What is the nature of human MENTAL LEXICON?
KEY: ORGANIZATION
 Number of words known by an educated adult:
between 50.000 and 250.000
 In speech, around 6 syllables per sec are produced
 ½ sec needed to detect a non-word or find a real
word
FAST and EFFICIENT word-searching ability:
The large number of words known and the speed with
which they can be located point to the existence of a
highly organized mental lexicon
Our mental dictionary
 Not alphabetical
 Fluid
 Flexible
 Ever-growing
 Detailed (lots of information stored in each entry)
 Interconnected
Clues to the mental lexicon organization
 Word-finding efforts
 “slip of the tongue”
 “tip of the tongue”
 normal speakers
 people with speech disorders
How can we define the meanings of words?
The fixed-fuzzy issue
The fixed meaning assumption The fuzzy meaning assumption
Words meaning is:
Words meaning is:
• Basic
• Fixed
• Firm
• Fuzzy
• Slippery
• Ever-changing
For the MENTAL LEXICON:
the answer will affect our view of HOW people
represent meanings in their minds
The fixed meaning assumption
Words meaning can be identified by defining a set of
core criteria
 The ‘check-list’ theory: for each word we have an
internal list of essential characteristics
IF something has these characteristics, we label it
with the corresponding word
 Aristotle school: words have a hard core of
ESSENTIAL MEANING, surrounded by other
characteristics which are not essential (can be
omitted)
 (ex: the ‘that’s impossible’ test, bachelor)
The fixed meaning assumption: problems
 How can we choose the criteria?
 Which attributes go on to the check list?
Not every word can be easily defined:
 Not obvious way to draw a dividing line between
ESSENTIAL and NON-ESSENTIAL characteristics
The fuzzy meaning assumption
Word meanings are inevitably fluid, for 2 reasons:
1. Fuzzy edges  there is no clear edge where one
word ends and another begins. Ex: vase-cup-bowl
2. Family resemblances  a group of words can
share some characteristics, but not all of them at
the same time.
Ex: GAMES: ring-a-rose/chess/tennis
Prototype theories
How humans cope with this fuzziness?
Does everyone operate the same way with meanings?
How do we organize meanings?
Humans tend to find some INSTANCES of a word
more central/basic than others
We have in mind a prototype with specific
characteristics
we label things IF they have a REASONABLE
amount of those characteristics (≠ check list)
Some examples: Eleanor Rosch’s experiments
 Birds: canary – peacock – penguin
 Colours: red – orange red – purple red
Experiments shown very consistent results, we use
categories when we have to label a concept
Tipicality rating
Category verification
Conclusions
 When people categorize objects, they bear in mind
an IDEAL EXEMPLAR  A PROTOTYPE
 MATCHING is the process used to decide whether
something is a member of that category
 NOT every characteristic has to match, it has to be
sufficiently similar to the prototype
Conclusions
Prototypes help us coping with:
1. UNTYPICAL EXAMPLES We still recognize
penguins as birds
2. DAMAGED EXAMPLES We still recognize a
three legged tiger as a quadruped
Prototype theory is a good way to interpret how
humans deal with LARGE CATEGORIES
Some problems with prototypes
A general problem:
Interference between IDENTIFICATION
CRITERIA and STORED KNOWLEDGE
They often are the same or largely overlap, but
sometimes not: the second influences our judgment
Inevitable link PERCEPTION-CONCEPTION
(ex: the story of the farmer: bull vs. cow)
1. The diversity of the characteristics
The properties of a prototype are heterogeneous
 result: prototypes differ one another in type!
ANIMALS vs. COLOURS
A prototypical BIRD is defined
by some TYPICALITY
CONDITIONS:
- Has feathers, wings, beak
- It flies
- It lays eggs
- It makes a nest
A prototypical RED is defined
by some CENTRALITY
CONDITIONS:
It occupies a central place
within a range of red shades
2. The difficulty of arranging them in order of
priority
 In which order can we list the necessary conditions

-
-
which define a prototype?
How do we choose which one is more crucial than
another?
Ex: BIRD
Has feathers
Has wings  bats have wings, they are not birds!
It flies  ducks and penguins don’t fly!
It makes a nest  other animals too
3. The problem of knowing where to stop
 We cannot put on to the list everything we know
about an object/concept/etc. (i.e. our
ENCYCLOPAEDIC KNOWLEDGE)
 Sometimes it’s difficult to separate out the meaning
of a word from the FRAME where it occurs
Ex: define the word zebra  someone will activate a
whole zoo frame, others won’t.
In conclusion:
words cannot be dealt with in isolation,
they are stored in relation to one another
WORDS INTERACTION
Words cannot be treated as separate items.
They are interdependent and we understand them as
interdependent items:
red  orange  yellow
Hot  warm  cold
How does the
mind cope
with these
relationships?
The “atomic
globule” viewpoint
Two broad
viewpoints
The “cobweb”
viewpoint
How does the mind cope with these relationships?
Two broad viewpoints
• Words are globules,
made of “meaning
atoms”
• Related words have
atoms in common
• Words are considered as
related because of the links
that the speakers build
between them
1. The “atomic globule” viewpoint
 There is a universal SET of basic atoms of meaning:
SEMANTIC PRIMITIVES
Innate part of human Organized differently in
mind
Biologically given
notions
different languages
(different segmentation
of the semantic space)
1. The “atomic globule” viewpoint
 How can we define the basic atoms?
 Is there a definitive list?
 Do they suffice to describe every meaning?
Schank (1972) on Verbs:
MOVE, INGEST, CONC,
MTRANS, ATRANS…
Es:
breathe, dink, eat
INGEST
Buy, give, steal  ATRANS
Miller&Johnson-Laird (1976) on
Objects:
Linking primitives to perception.
PLACE, SIZE, HORIZONTAL,
VERTICAL, BOTTOM, TOP…
! Not everything is
PERCEIVABLE !
Es: to promise, to disagree
2. The “cobweb” viewpoint
Words are linked together in a gigantic multidimensional cobweb NETWORK THEORIES:
A network is an interconnected system,
made of nodes and links.
Collins&Quillian
model
“Retrieval time
from semantic
memory»
(1969)
HOW DO THESE INTERCONNECTIONS WORK?
1. Linguistic habits: pen +pencil, moon + stars.
Strong ties, easily revealed by free association tests.
DIFFERENT TYPES
OF
ASSOCIATIONS!
1. Co-ordination
2. Collocation
3. Superordination
4. Synonymy
HOW DO THESE INTERCONNECTIONS WORK?
1. Linguistic habits
CO-ORDINATION seems to be the strongest link.
Hypothesis confirmed by:
The most common slips of the tongue: leftright/today-tomorrow
Difficulties in distinguishing co-ordinate items in
aphasic subjects:
lemon-orange vs. lemon-boot experiments
2. Recognition tasks experiments (squeezing a ball)
1.
Some CO-ORDINATES are so strongly linked that
brain-damaged people tend to confuse them in
production/comprehension, but, at the same time, their link
is recognized faster then others
HOW DO THESE INTERCONNECTIONS WORK?
2. Learning from children
Finding out how babies build up a store of words =
Additional clues as to how interconnections are created.
3 essential steps
1st step LABELLING:
Symbolization, a string of sounds is
used to name something
2nd step PACKAGING: what can
be packed together under the same
label?
3rd step NETWORK BUILDING:
how do I link words one another?
LABELLING
Some clarifications:
 To symbolize = to realize that a particular
combination of sounds means od symbolize a certain
object.
 It emerges around the ages of 1 and 2
 Before: Babbling is an muscle exercise. At this
stage, infants have not attach yet the actual meaning
to the word. They are still experimenting, making
noises.
PACKAGING
 Children package meanings in a different way
 They do not necessarily focus on the same
characteristics/priorities
UnderThey assume a word refers to a NARROWER
extension range of things
Ex: white page/ white snow
(Leopold)
OverThey assume a word refers to a LARGER range of
extension things
Ex: quack=duck, cup of milk, eagle on a coin
(Vygotsky)
NETWORK BUILDING
 A six year old child has a passive vocabulary of about
14.000 words  they need to be put together into a
semantic network
 Network building is a slow process: children tend
to link a word only with the specific context where
they learned it
 Children use COLLOCATION principles more than
CO-ORDINATION principles in free association:
Children associations
Adults associations
TABLE - eat
TABLE - chair
DARK - night
DARK - light
SEND - letter
SEND - receive
Word production & Word recognition
Production
• From the
meaning
to the
sounds
Recognition
• From the
sounds to
the
meaning
Production &
Recognition =
mirror images
of one another
Language information
processing:
Semantic  syntactic 
phonological info
Words production: Hypothesis & Theories
1st process: DECISION MAKING!
Underneath process, even when we are not conscious.
 blends: an evidence that alternative words are often
considered during speech
Blends examples
Buggage  baggage + luggage
Tummach tummy + stomach
Evious evident + obvious
Aphasic speech:
“I forget (don’t remember)
seeing you, sir. I remember
the other document (doctor
+ gentlemen) and was
plazed (pleased + glad) to
see the other document.”
(Butterworth, 1979)
Words production: Hypothesis & Theories
Blends suggestion: we activate a number of words in the area of
the required word and then suppress the words we don’t want.
Spreading/Interactive activation principle:
The initial
input activates
words
spreading
along the
connections
Relevant links
get more and
more activated
Unwanted
links tend to
fade away
connectionist
models of
language
perception/
production
Words recognition
 Basic problems in normal speech:
Sounds are altered by their neighbours
2. Sound segments cannot be separated, they tend to
merge
3. Noises cover the sounds.
1.
So, how do we understand these sounds?
Guessing
 Matching the portion heard with
their mental lexicon
 Choosing the best fit
 Filling in gaps
Words recognition
The role of context in detecting sounds:
 Paint the fence and the ?ate.
 Check the calendar and the ?ate.
 Ho freddo, chiudi la ?orta.
 L’ho fatta ieri, assaggia la ?orta.
 GATE
 DATE
Words recognition
Summing up the process:
1. To split up the stream of sounds into words
2. To identify the words
1. SOUNDS  PHONEMES
• Words frequency: common words
are recognized faster. Why? 
Network theory preferential
attachment.
• The role of context: same as
guessing for phonemes
2. PHONEMES  MORPHEMES
3.  Then fitted to words
An example of connectionist model.
TRACE (McClelland & Elman, 1986)
Connectionist model of spoken
word recognition/production
INTERACTIVE ACTIVATION
among processing units
unit = GUESSING about the
input
Three levels organization:
1. Feature/input
2. Phoneme
3. Words
Three types of connectivity:
1. Bottom-up
2. Lateral (i.e., within-layer)
3. Top-down