Download STRUCTURAL LIMITS ON VERB MAPPING The role of abstract

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Embodied language processing wikipedia , lookup

Transcript
STRUCTURAL LIMITS ON VERB MAPPING
The role of abstract structure in 2.5 year olds’
interpretations of novels verbs
Article by:
Cynthia Fisher,University of Illinois, 2002
Presentation by:
Denise, Shauna, Kara, Rachelle, Mélanie & Robyn
Let’s set things up a little...
Background, terminology
& previous studies…

Everyday, without noticing, adults take
for granted that there is a link between
syntax and meaning.

The structure of an utterance affects its
interpretation.
– Example:
• The cat ate the mouse.
• The mouse ate the cat.
WHY? Bootstrapping!

The syntactic bootstrapping theory…

The idea that children use information
about the sentence structure to guide
sentence comprehension from an early
point in acquisition.

In other words… semantics builds on
top of syntax.
Many support this claim...

Children between the approx. ages
of 2 and 5 take novel verbs in
different sentence structures to
have difference meanings.

Naigles & Fisher
NAIGLES 1990 - Results

Children who heard TRANSITIVE:
– LOOKED longer at the causal scene
• The duck is blicking the bunny.

Children who heard INTRANSTIVE:
– LOOKED longer at the non-causal scene
• The duck and the bunny are blicking.
FISHER 1996

Same experiment but they HID THE
IDENTITY WITH AMBIGOUS
PRONOUNS

Assessed the interpretation by asking
questions about whose role the verb
described.
– Intransitive:
• Which one pilked the other one fast?
– Transitive
• Which one pilked fast?
Results for both experiments…

Children and adults are more likely
to choose causal agents as the
subject of the transitive sentences.

This result is evidence that the
structure of the sentence and the
arguments was meaningful to the
children (Fisher)
FISHER - Main Ideas...

Syntactic Bootstrapping

Abstract Representation

Predictions:
– Even though children are conservative with new
verbs, syntactic bootstrapping will still play a role
for children under the 3 years old.
SOME DISAGREE…

Abstract representation plays a minimal
role in the early phases of acquisition

Based on these findings…
– Children are quite conservative
– Children produce late errors
• The late errors are followed by a period
when the child combines verbs and
sentences only in patterns that are present
in the input.
Fisher’s Response...

A big part of syntax acquisition must be
completed before productive innovation
is possible

Considerations:
– Children need to figure out the grammatical
functions
– Innovation requires that the child to have a
language reproductive mechanism allowing
him to retrieve or build sentence structure
only on meaning.
FISHER - Let’s recap...

3 - 5 year old children are similar to adults
in regards to their production of novel
verbs.

They have a set of fundamental syntactic
structures (abstract representations)

They use the syntax to interpret new
utterances.
THIS STUDY
HYPOTHESIS
HYPOTHESIS…
 Do
2.5-year-olds use overall
sentence structure to help
them interpret novel verbs?
EXPERIMENT
EXPERIMENT 1- Subjects

24 2.5 yr. olds
– 12 girls
– 12 boys

23 3 yr. olds
– 12 girls
– 11 boys

24 college-aged adults
– 12 female
– 12 male
EXPERIMENT 1- Subjects

Children and adults randomly assigned either
transitive or intransitive condition and to one
of two stimulus orders.

Showed 4 brief videotaped caused-motion
events. (approx. 6-15 sec).

One active participant (agent) directly caused
the motion of passive participant (patient)

Each videotaped event shown 3 times in row
separated by brief intervals of blank screen.
Sentence Structure

The two sentence contexts differed in the number of
noun-phrase agreements

Sentences provided only structural information about
verb by using only ‘she & ‘her to identify participants

There were no direct clues as to the identity of the
subject or object of verb

Nonsense verbs were heard during the “
blank screen” interval:
– preceding first repetition of event
– repeated 9 times during
– between 3 repetitions of each event.
Table 1 - Stimulus events and sentences
Unfamiliar motion events
Stimulus sentences
Participant A pulls B backwards
along a slippery surface by pulling
on B’s backpack.
She stipes (her) over there!
A rolls B toward her on a wheeled
dolly by pulling on a feather boa
tied around B’s feet.
She braffs (her) over there!
A wheels B forward and back in a
red wagon.
She pilks (her) back and forth!
A rotates B on a swivel stool by
pulling on the ends of a scarf
around B’s waist.
She gishes (her) around!
She stipes (her) over there!
She braffs (her) over there!
She pilks (her) back and forth!
She gishes (her) around!

Depending on which condition the child was
assigned to the experimenter asked either:
“Which one (verb)ed the other…? Point!”
(TRANSITIVE CONDITION)
OR
“Which one (verb)ed…? Point!”
(INTRANSITIVE CONDITION)

This experiment was also run with adults
EXPERIMENT 2

An additional 24 two and a half year
olds were tested.

Same procedure was used.
RESULTS
Hypothesis confirmed

Children interpret new verbs in accord
with their number of arguments.

When interpreting new verbs describing
the same motion events, children who
heard Transitive sentences were more
likely to assume that the verb referred
to the actions of the causal agent.

Both children and adults who heard
the Transitive sentences were more
likely to point to the agent of each
causal event

These results suggest that children
use syntax to guide interpretation of
novel verbs.
Table 2: Mean proportion agent choices by age and sentence
context, Experiment 1





Age group
30 months
40 months
Adults
Mean
Intransitive
0.326 (0.110)
0.045 (0.045)
0.313 (0.088)
0.233 (0.054)
Transitive
0.625 (0.104)
0.667 (0.089)
1.00 (0.000)
0.64 (0.053)
Mean
0..476 (0.080)
0.370 (0.083)
0.656 (0.084)
DISCUSSION

The study supports previous
experiments showing children are
sensitive to structure in interpretation

Extends data to younger children
IF...

The syntax-semantics link is present so
early in life

Children use sentence structure to
interpret new verbs at such an early
age…

WHY ARE THEY UNWILLING TO
INNOVATE WITH NEWLY-LEARNED
VERBS???


the requirements for interpretation of
words are very different from those for
innovation
in order to innovate, children need:
– to know the syntax of the language
(including morphology and word order)
– to develop a system that will produce
sentences with these regularities, without
the help of having practiced a specific
structure
Let’s look at another reason that
children are conservative...

Imagine child sees toy Ernie launched
through the air

Child hears
– “Ernie pilked!”
– (Possible translation: “Ernie
flew/soared/fell”)

In this case, verb is used intransitively

How is child to know if the verb can also
be used transitively?
Assumptions of the structuremapping view:

The data obtained in this experiment
imply that early sentence comprehension
is facilitated by abstract mental
representations of sentence structure
– What is the nature of these abstract
representations?
– How do these representations influence
early verb development?

Researchers continue to investigate the
structure-mapping mechanism for early
syntactic bootstrapping because:
– It makes sense given widely held views
about language acquisition (Fisher, 1996,
2000a; Fisher et al., 1994)
– It makes unique and verifiable predictions of
verb interpretation errors
The structure-mapping mechanism
assumes that:

Semantic structures of verbs match up to
conceptual knowledge of events

Children can identify and understand some
nouns in fluent speech
These assumptions have consequences
for early sentence comprehension...

Structural alignment allows children to obtain
a fairly accurate semantic structure for the
sentence

Structural alignment puts constraints on
sentence interpretation

Because they can already link nouns with
referents, children should not have trouble
comparing structural alignment of sentence
structures with relevant conceptual structures
Structure-mapping and crosslinguistic variation:


Structure-mapping is abstract enough to
allow for cross-linguistic variation in the
relationship between a language’s syntax and
its semantics while still providing necessary
constraints on these links
Structure-mapping takes variation into
account because it assumes an incomplete
knowledge of sentence structure on the part
of the infant
Predictions of Structural Alignment:

Fisher’s proposed structure-mapping
mechanism for syntactic bootstrapping
uniquely predicts that nouns in a sentence
are not the same as the arguments of a verb
Future research of this nature will:

Test in detail the predictions of early partialstructure matching in sentence
comprehension

Help us obtain a better understanding of
which information sources, constraints, and
biases children use in language acquisition
Future Directions…

Neuro-imaging
Questions or Comments?