Transcript Slide 1

Linguistic Theory
Lecture 12
Language Acquisition
Outline of lecture
• This lecture will be in three parts:
– What children actually do
– How to model what children do
– Theories about how they do it
What children don’t do
• Common opinion: children are taught language by
their parents
• This cannot be true:
– The vast majority of children learn language perfectly
– Teaching is a skill that has to be learned
• Most parents are not trained teachers
– You have to know something to be able to teach it
• No adult ‘knows’ the language that they have learned
• The first fact stands in direct opposition to the
following two!
What children don’t do
• Maybe, parents can teach like this:
– Wait until child says something ungrammatical
(this often happens)
– Provide the grammatical version as a correction
• This does not require them to know the
•
grammar – just the difference between
grammatical and ungrammatical sentences
But ...
• For this to work:
– Parents must behave consistently
• They must correct every error the child makes
• They must never correct a grammatical sentence
– Children must know:
• When a parent repeats one of their sentences it is
meant as a correction
• What the mistake is in their sentence and how to
alter the grammar according to the ‘correct’
sentence provided by the child
– The first does not happen, the second is
rather much to expect of a child.
What children do do
• From birth:
– They make noise
• Soon after:
– Babbling
• Not reactive/reflexive noises
• From about 24 weeks:
– Reduplicative babbling
• Repetitive CV sequences
• Soon after:
– Non-reduplicative babbling
• Differing CV sequences with intonation contours
• No one knows why:
– It is not essential
• Babies who had tracheotomies during the babbling
period learn language perfectly well
– It is not dependent on environment
• Profoundly deaf babies do it
– It doesn’t differ depending on language of
parents/community
First words
• From about 1 year (±0;2):
– First ‘meaningful’ words appear
– One or two to start
– Slowly increases over next 4-6 months, then ...
• From about 1;6 (±0;2):
– Vocabulary spurt
– First words in combination (2 word sentences)
– Mirrors word order of parent language
• This continues for nearly a year
– Slowly increase in number of words which are
combined
– But not much syntactic variation
– Until ...
• From about 2;6 (±0;2):
– Syntax spurt
• Rapid increase in number of words
• Syntactically varied expressions present
• This continues to slowly develop for
another 2;6 by which time most of the
system is in place
What do children learn?
• Clearly, they do not learn language
– We do not know language
– We know grammar
• That children learn rules is demonstrated
by:
– Wug experiment (Berko 1958):
• Here is a Wug:
• Here are two .................
• All children say [wʊgz]
– Not [wʊgs] or [wʊgɪz]
• They have never heard the word before
• But they know the correct plural form
• So they must be able to generate new
forms from what they have learned
Modelling the acquisition
situation
• With the onset of the mathematical approach
•
•
to language in the 1960s, linguists also
attempted to set up a mathematical model of
language acquisition.
These models were deliberately
simplifications with respect to the actual case
The point being that any negative result for
the simplified model would carry over to the
more complex real situation
Assumtpions
• What is learned:
– Languages (not grammars)
• Thus the ‘learner’ does not have to figure out from
data what the underlying rules are (a more difficult
process)
– We can assume that the hypothesis space is
limited to mathematically defined languages
(context free, context sensitive, etc.)
• The learning situation
– Data from the target language is presented to
the learner one bit at a time
– On the basis of a datum, the learner makes a
guess at the target language
– Given the next datum, the learner either
sticks with the hypothesis or abandons it in
favour of another
– This continues for ever (no one tells the
learner when to stop)
– Learning is successful if in some finite time
the learner correctly selects the target
language and never rejects it on future data
• The learning strategy
– Conservative learning
• Learner rejects a hypothesis only if they receive
data to show that it is wrong
– Non-conservative learning
• A hypothesis can be rejected even if the data is
compatible with it
– Conservative learning is safer – if the target is
selected, it will never be rejected as all data
will be compatible.
• Data presentation
– Negative and positive
• Both grammatical and ungrammatical sentences
are presented properly indicated
– Positive only
• Only grammatical sentences are presented
Gold’s results
• Assuming a conservative learning strategy
and positive data presentation only finite
languages are learnable
• Context free, context sensitive,
unrestricted re-write systems (formally
equivalent to transformational grammars)
are unlearnable
Theories of language acquisition
• Structuralist/behaviourist theory
– Child makes noises
– Child accidentally makes a noise similar to one of
the parent’s language
– Parents react to this positively
– Child’s behaviour is reinforced and so they are
more likely to make the same noise again
– Over time and a series of accidents and
reinforcements, the child only makes the noises
of the language
• It should be obvious what is wrong with
this!
Early Chomsky
• The child has some species and linguistic
specific innate ability which allows her to
learn language
• The Language Acquisition Device (LAD)
– Not much was known about this
Mathematical approach
• Assuming that human language falls into
one of the mathematically defined set of
languages, the LAD orders possible
languages and the child selects and
rejects grammars on the basis of this
1970s - constraints
• With constraints came greater generality
• Generality allows ‘universal’ principles to
be proposed
• Universal Grammar became a reachable
goal
• But how ‘learning’ operated within this
was still unsettled
1980s Principles and Parameters
• Principles are universal parts of UG
• Parameters are variable parts of UG
– Parameters can be set differently for different
languages
– Parameter settings are what needs to be
learned
E.g. X-bar principles and
parameters
• English has heads preceding complements
(object follows V, prepositions, inflections
precede VP, etc.)
• Japanese has heads following
complements (objects precede V,
postpositions, inflections follow VP, etc.)
• Both have heads and complements but
the order varies:
• X-bar principle
– X’  X, YP
(comma indicates no order)
• X-bar parameter
– A) head first
– B) head last
• All the child has to hear is one head initial
phrase and she will know that the
language has prepositions and VO and IVP word order
Problems
• Isn’t this a bit too easy?
– Why does it take so long to learn language?
• Isn’t this too descriptive?
– What counts as a parameter?
– If any difference between languages can be
stated as a parameter we can account for any
logically possible amount of variation
Restricting parameters
• Rich deductive parameters
– A simple underlying parameter is responsible
for a lot of apparently unconnected surface
phenomena because the modules of the
grammar interact with each other in complex
ways
– E.g. Pro-drop
• Unfortunately this could not be empirically
sustained
Restricting parameters
• Languages differ only in the behaviour of
their functional elements (nouns, verbs,
etc. are pretty much universal).
• =the functional parameterisation
hypothesis (Borer 1986)
• This works for some parameters, but it is
difficult to see how it can work for all
(head left/right parameter)
Restricting parameters
• Learners have to learn the lexicon of the
•
•
•
language
Perhaps parameters are a matter of lexical
differences
= the lexical parameterisation hypothesis
(Manzini and Wexler 1986)
If this is so, it is difficult to see why there is
not more variation within a language with
one word conforming to one parameter value
and another to another value.
Maturation
• None of these theories account for the syntax spurt
– Why do children all of a sudden learn a lot of syntax?
• It has been proposed that what happens at this time
is that functional elements enter the child’s
grammar:
– Before they operate only with thematic categories:
nouns, verbs, adjectives, etc.
• So the question is: why do they not learn functional
categories before 2;6?
• The functional maturation hypothesis (Radford
1988)
– The notion of a functional category is not available to
the child until after a genetically predetermined time
(like teeth and puberty)