Transcript Language acquisition

35 years of Cognitive Linguistics
Session 10: Language acquisition
Martin Hilpert
your questions
•
•
•
•
Want other one spoon, Daddy.
It noises.
I becamed to be Spiderman.
She unlocked it open
• buzz, crash, splash, bang, noise
• It buzzes, crashes, splashes, bangs, ...
• * It noises.
• pull, crack, pry, unlock
• She pulled / cracked / pried / it open.
• *She unlocked it open.
overgeneralization errors
• The child has acquired more than a string of
word that is repeated.
• The error shows that the child has acquired an
regularity.
• It also shows that the child has not yet
mastered all constraints on that regularity.
• If children make overgeneralization errors, this
must mean that they have acquired an
abstract rule, right?
• She unlocked it open.
SUBJ V-ed OBJ ADJ
The dictionary-and-grammar model
dictionary-and-grammar model vs. CxG
• Both models assume that children learn
generalizations.
• The models have different hypotheses about
how the process of language learning unfolds.
two hypotheses
• dictionary-and-grammar: the continuity
hypothesis
– children acquire the syntactic categories (SUBJ, ADJ,
etc.) and rules of adult grammar
– at first their language output is imperfect because not
all rules are in place
• CxG: the item-based learning hypothesis
– children acquire knowledge of lexical items and fixed
strings of lexical items (more juice)
– as they acquire more and more of these, they begin to
form generalizations across them
the continuity hypothesis
• The language of children is mentally represented by the
same syntactic rules and categories as adult language.
• Innate knowledge facilitates the learning process.
• POS categories are fundamental.
S. Pinker
not S. Pinker
the continuity hypothesis
• ‘Once a child is able to parse an utterance
such as « Close the door », he will be able to
infer from the fact that the verb « close » in
English precedes its complement « the door »
that all verbs in English precedes their
complements.’
• >> This only works if children are born with
the idea that there are verbs, complements,
and relative clauses.
item-based learning
• Children start out by memorizing and repeating
concrete words and phrases.
• As a child recognizes similarities across different
phrases, a process of schematization sets in.
• POS categories and syntactic constructions
emerge as generalizations over concrete phrases
• Generalizations get increasingly abstract, until
they resemble adult grammar.
foundations of language learning
five socio-cognitive abilities
1.
2.
3.
4.
5.
joint attention
intention reading
schematization
role reversal
pattern recognition
1. joint attention
• Words become meaningful in situations in which
both the child and a caretaker focus on an object
and are mutually aware of this.
• Before 9 months of age, children can only engage
in dyadic joint attention (mutual eye gaze).
• Triadic joint attention means inspecting an
object together.
• With the ability to engage in triadic joint
attention comes word learning.
2. intention reading
• Young children interpret other people’s
actions as purposeful and goal-directed.
• Theory of mind: other people too have ideas,
feelings, knowledge.
• Toddlers imitate actions of others, but only
those that they see as ‘successful’, not those
that are ‘accidental’.
3. schematization
•
•
•
•
all done, all wet
where’s daddy?, where’s cookie?
let’s go!, let’s find it!
I’m holding it, I’m pulling it
>> all X
>> where’s X
>> let’s X
>> I’m X-ing it
• the ability to form pivot schemas
– Pivot: the fixed part of a schema
– Open slot: the variable part of a schema
4. role reversal
• In linguistic interaction, speakers are also
hearers, and vice versa.
• the capability of conceptual blending
– What would it be like if I were in the position of
my interlocutor?
– Creating a model of other people’s current
knowledge.
5. pattern recognition
• the ability to recognize regularities in speech
• 8-month old infants listened to nonce words
– bidaku, padoti, bidala, tupiro, gobida, ...
– bi always followed by da
• exposure to new words
– bidaka >> in line with previous words
– dabiko >> different from previous words
• infants showed greater interest towards words
that violated established patterns
five socio-cognitive abilities
1.
2.
3.
4.
5.
joint attention
intention reading
schematization
role reversal
pattern recognition
generalizing across pivot schemas
• easier with English nouns than with English
verbs
• noodles in there, my noodles, noodles hot
– __ in there
– my __
– __ hot
• verbal pivot schemas
– introducing ‘meeking’: 2-year olds hesitate to use
the form, even though they have pivot schemas
with ing-forms (I’m holding it, I’m pulling it)
the verb island hypothesis
• verbs in early language acquisition form ‘islands’
of grammatical organization
–
–
–
–
each verb is limited to a single syntactic pattern
tickle me, tickle it, tickle doggie
put it there, put it up, put it down
I swinging, baby swinging
• adults use these verbs in several syntactic
patterns
– Stop tickling, John was tickled by Mary, John tickled
me silly, John tickled Mary out of the room, …
Evidence for item-based learning
1. Children are conservative about
verbal pivot schemas
(Brooks and Tomasello 1999)
study questions
1. Can very young children be trained to use the
passive?
–
If they can, it is likely that input frequency, not overall
complexity, explains the absence of passives.
2. Can these children then use the passive
productively?
–
If they learn a new transitive verb, do they immediately
conclude that it can be ‘passivized’?
• 28 younger children (~2.5 years), 28 older
children (~3.5 years)
• both groups saw transitive actions described
as meeking and tamming
meeking
meeking
tamming
tamming
training phase: two conditions
• active training group
–
–
–
–
Look, the fraggle is meeking the banana!
Oh wow, the robot just tammed the banana!
Who’s meeking the banana?
Did you see who tammed the banana?
• passive training group
–
–
–
–
Hey, the banana got meeked by the fraggle!
See how the banana got tammed by the robot?
What’s going to get meeked?
I think something is getting meeked again…
test phase: elicitation
• neutral question
– What happened?
• patient-focused question
– What happened to the banana?
• agent-focused question
– What did the fraggle do?
• purpose
– Can children adjust their use of novel verbs to the
discourse needs (patient-focused, agent-focused)?
results: passive formations
100
***
60
40
Main effect of training group:
active-trained kids – very few passives
20
percent of children using a passive
80
neutral
patient
agent
young passive
old passive
young active
group
old active
results: passive formations
***
100
***
60
Main effect of question type:
agent-focused question - fewer passives
40
***
***
20
percent of children using a passive
80
neutral
patient
agent
young passive
old passive
young active
group
old active
100
results: passive formations
40
60
Interaction between training
group and question type:
question type makes a much
larger difference in the
passive-trained group
20
percent of children using a passive
80
neutral
patient
agent
young passive
old passive
young active
group
old active
results: passive formations
n.s.
100
n.s.
60
40
No effect of age
20
percent of children using a passive
80
neutral
patient
agent
young passive
old passive
young active
group
old active
2. Children’s linguistic creativity has
been overestimated
(Lieven et al. 2003)
How creative are children in their
language use?
• Low creativity with gradient increase
– evidence for a usage-based model that starts with
items
• Initially low and then suddenly high creativity
– evidence for a rule that is acquired and can be
freely applied after that
How do children acquire constructions?
• continuity hypothesis
– start with lexically fixed phrases
– acquire an adult syntactic rule
– productively apply that syntactic rule
• item-based learning hypothesis
– start with lexically fixed phrases
– vary individual items in slots
– arrive at a broader generalization
How can these hypotheses be assessed?
• Evidence
– A high percentage of all recorded constructions in
a dense child language corpus have been
produced before or are found in the input.
assessing creativity
•
•
•
•
•
•
change
substitution
add-on
drop
insertion
rearrangement
Target
I got the butter
Let’s move it around
And horse
finished with your book?
Away it goes
Predecessor
I got the door
Let’s move it
And a horse
finished your book?
It goes away
• Each new utterance can be related to earlier
utterances in terms of how many steps of change are
necessary.
data
• two children
– each recorded for 6 weeks at age 2, then again for 6 weeks
at age 3
– four high-density corpora, divided into main and test
How creative are young children?
0 changes
0
50
100
1 change
150
200
total number of multi-word utterances
2
250
3
300
types of ‘failed’ derivations
• inappropriate filler
– Do you want to football?
– inappropriate filler in the PROCESS slot
• inappropriate add-on
– Which ones go by here?
– inappropriate prep added to Which THING go here?
• constituent omission
– And what that done?
– Omission of has in And what has that done.
failed derivations
• A high proportion of the problematic
utterances are ill-formed by adult standards
– When children use language creatively, they go
beyond what they know, rather than applying
abstract rules to create novel utterances.
conclusions
• The creativity of later child speech has been
overestimated:
– Small variations account for the lion’s share of all
utterances.
– The variation that does exist does not point to the
application of rules.
– Rather, item-based learning continues as the main
process of language acquisition, is not abandoned
in favor of learning abstract principles.
3. The collocational properties of
constructions facilitate acquisition
How are constructions learned?
• Children must learn that there are
correspondences of syntactic form and
meaning:
– John emailed me the report.
• Learners must be able to interpret novel
utterances:
– ‘fast mapping’ – children learn lexical items at a
stunning rate
– is this also possible with syntactic constructions?
How are constructions learned?
• Idea: Learning might be easy if a syntactic
structure often occurs with one particular
lexical element.
Many constructions have a most frequent verb:
experiment
• a new English phrasal pattern with a new meaning
– NP1 NP2 VERB-o ‘appearance’
– The frog the sock moopoed
• Training: Participants heard sentences and
watched video sequences in which an animal/toy
spontaneously appeared
• Task: Participants were given one sentence and
had to choose between two possible video clips
training
“The frog the sock moopoed”
critical trial
“The cow the hat moopoed”
experiment
• Procedure: subjects saw eight clips
• Stimuli:
visual display
sound
experiment
• Three groups:
– The skewed frequency group (4-1-1-1-1)
– The balanced frequency group (2-2-2-1-1)
– The control group (no sound)
results
• The control group did not
perform better than chance
• Both the skewed input
group and the balanced
input group performed well,
the skewed input group
outperformed the balanced
group.
summing up
a usage-based view of language
acquisition
• Cognitive Linguistics tries to explain language
acquisition without positing an innate language
faculty
• Instead, language learning is explained in terms
of domain-general cognitive abilities:
–
–
–
–
–
joint attention
intention reading
schematization
role reversal
pattern recognition
evidence for item-based learning
1. Children extend verbal pivot schemas
conservatively.
2. Children’s linguistic creativity has been
overestimated.
3. Skewed frequencies of collocates facilitates
the learning of constructions.
See you next time!
[email protected]