Looking at Interlanguage Processing

Look at interlanguage
processing(1)
Second Language
Acquisition
21/10/16
Unit 8
2011 Fall
Second Language Acquisition
Overview
Connectionist/emergentist
models
Information processing
Knowledge types
Interface of knowledge types
Psycholinguistic constructs
Conclusion
21/10/16
Emergentist models
Referred to as constructivist
approaches with emphasis on usage.
Learning rely on the extraction of
regularities form the input.
Exemplar-based in that it is the
examples that represent in the input
that form the basis of complex
patterns and from which regularities
emerge.
21/10/16
L is not driven by an innate faculty,

rather the complex linguistic
environment provides the information
from which learners abstract regularities.
In this approach, learning is seen as
simple instance learning (rather than
explicit/implicit induction of rules), which
proceeds based on input alone; the
resultant K is seen as a network of
interconnected exemplars and patterns
rather than abstract rules.
21/10/16
Connectionist models
a.k.a. neural networks, neural computing,

parallel distributed processing
Model Overview: Units and connections
cat
c
21/10/16
eat
t
e
5
Learning takes place as the network (

the learner) is able to make
associations, and associations come
through exposure to repeated
patterns.
21/10/16
Features of connectionist models
Neuronal plausibility
Distributed representations and locus of
control
Robustness
21/10/16
Neuronal plausibility
Connectionist models mimic the behaviour of neurons
in the brain, the modification of units representing synaptic
firing that take place in learning. This is strengthening of
connections is described in the Hebbian learning rule, one of
the earliest connectionist learning rules: When an axon of
cell A is near enough to excite a cell B and repeatedly or
persistently takes part in firing it, some growth processes or
metabolic change takes in one or both cells such that As
efficiency, as on of the cells firing B is increased. Hence
connectionist modelling has been described as brainstyle
cognition (sometimes referred to as neural nets)
21/10/16
Distributed representations and locus of control
Connectionist models have been termed "subsymbolic"

in order to contrast them with traditional symbolic
approaches to cognition and language, where it is
assumed that the primary processes are carried out over
symbolic structures like words and propositions.
The distributed nature of the representations has

important ramifications for the approach as a model of
cognition. There is no problem with locus of control, or
what is referred to as the homunculus problem.
21/10/16
Robustness
Distributed representation systems are robust
in that partial inputs to the system can produce
appropriate outputs. The system can recognize
partial words and structures and is thus able to
mimic the filling in that listeners routinely do
in everyday communicative settings where
ambient noise often obscures or distorts what the
speaker says. This leads to what is called
'graceful degradation' in the knowledge system in
less than optimal settings.
21/10/16
10
Competition Model
functionalist model of speech

processing
direct mapping between underlying
meaning and surface forms
surface forms (= cues) include casemarking particles, word order, stress,
etc.
cues compete and converge in realtime processing
21/10/16
11
The Competition Model

In processing meaning in sentences we
use
these cues:
Syntax
Word order ( SVO)
Morphology (Case of nouns, Agreement

between subject and verb)
Semantics
Human
Animate
21/10/16
12

In some cases the semantic and
syntactic cues converge to give the
same interpretation of a sentence.
In other cases the semantic and

syntactic cues conflict.
The interpretation we choose is

based on
competition among the cues.
21/10/16
13

Different languages assign different
weights to syntactic and semantic
cues.
Learners of a second language tend

to transfer the weights associated
with the cues in their first language.
21/10/16
14
Competition Model and SLA

L2 learners are faced with conflicts between
L1 and L2 cues and cue strengths.
Learners first resort to their L1 interpretation

strategies.
When they recognize an incongruity between

the L1 and L2 systems, they resort to a
universal selection of meaning-based cues.
Finally, learners gradually adopt the

appropriate biases and their L2 proficiency
increases.
21/10/16
15
Strengths of the CM
Direct form-function relationship
A unified account of processing and learning

Generalizes well across languages
Models second language transfer processes
21/10/16
16
Limitations of the CM
Narrow empirical base.
Nature of the experimental task.
21/10/16
Cue definition
17
Information processing in
SLA
Key assumptions:
> Humans are limited-capacity processors.
- capacity is a function of organization: quality
and quantity
> SLA is a cognitive skill.
- can be broken down into subskills.
example: tennis grip, footing,
stroke
Controlled
21/10/16
Automatic
Restructuring
18
Skill results from mastery and integration of

component subskills.
*Subskills become automatized.

*Subskills in speaking:
retrieve lexical item
formulate syntax
meet pragmatic needs
perform articulatory movements
21/10/16
19
Controlled versus automatic processing

Controlled processes
Automatic processes
* temporary activation of memory nodes

to input
* habitual activation in response
* short-term working memory
* long term memory
* slow
* effortful
*easy to alter
21/10/16
* fast
* effortless
* difficult to alter
20
Skill automatization &

restructuring
* Automatization is a quantitative
factor: Get better, faster, smoother
on a skill.
* Restructuring: qualitative way in
which knowledge is organized.
21/10/16
21
Restructuring
Restructuring is the process by which

task components are coordinated,
integrated, or reorganized into new
units. Procedures involving old
components are replaced by more
efficient procedures involving new
components.
>> Restructuring attempts to account
for transitions in skill levels.
21/10/16
22
Automaticity vs Restructuring
Automaticity
vs.
Restructuring
*quantitative
* qualitative change in the
way
(speed, memory demand)

information is organized
* isolated subskills
* unifies subskills
*early stage of development* latter stages
*accounts for 'practice effects' * accounts for transition

between
levels
21/10/16
23
Role of restructuring
Restructuring has been proposed as mechanism to

explain discontinuities and shifts in development.
Two are of particular importance:
(1) U-shaped curves. (If only automatization we might expect a
straight line)
Phase 1>>>>>>>
Phase 2>>>>>>
Phrase 3
automaticity
organizing
Integration
data-driven
simplifying internal
of #1 &#2
subskills mastered
representations to
"bottom-up"
gain control
"top-down"
(2) Plateaus and shifts. Feeling of making progress. Harder to
document.
21/10/16
24
U-shaped curve in lexical development
Willingness of Dutch EFL learners to accept English senses of break
100%
80%
He broke his leg.
60%
The cup broke.
______________________________________________________
13-14 15-16
16-17
17-18
18-19 19-20
20-21
age of learners
Kellerman (1984)
21/10/16
25
Input-driven language learning: Matching

learner output with language input.
Input-matching characterizes language development as the interaction

between the learning environment and the cognitive system, with no
reference to a unique language capacity (MacWhinney, 1998).
The input-matching perspective contrasts sharply with approaches

inspired by Universal Grammar, in which more emphasis is placed on the
innate, language-specific capacity of the individual, and correspondingly
less on the characteristics of the learning environment.
Other approaches that capture the primary role of the learning

environment, including data-driven (McLaughlin, 1987), input-driven
(Plunkett, 1998), and exposure-based (Mitchell, Cuetos, & Corley, 1995).
21/10/16
26
Characteristics of the input-matching

approach
Language is learned
Associative mechanisms of learning
Language knowledge is graded
Learning outcomes are a probabilistic

function of input type and quantity
(IOW: The more you are exposed to a form

the more likely you are to learn it.)
21/10/16
The unity of processing and learning

27
Associative learning is:
"the ability of an organism to adapt its

behaviour as a result of acquiring information
about associations or contingencies that exist
between events in its environment".
There are two types of associative learning:
Causal associations link elements in a temporal sequence, in which one

cue in the environment is associated with another cue that immediately
follows.
Structural associations link features of a single entity. The set of individual

links are perceived as a single configuration.
21/10/16
28
Examples of input-matching
approaches
1.
2.
Vocabulary size studies

Competition Model
21/10/16
29
Vocabulary size research
Attempts by researchers to index the size of L2

learner vocabulary by frequency of occurrence
statistics.
Hazenberg & Hulstijn, (1996). Defining a minimal receptive vocabulary

for non-native university students. Applied Linguistics, 17, 145-163.
Nation, I. S. P. & Waring, R. (1997). Vocabulary size, text coverage and

word lists. In Schmitt, N. & McCarthy, M. (Eds.) Vocabulary: Description,
acquisition and pedagogy (pp. 6-19). Cambridge: Cambridge University
Press.
21/10/16
30
L2 vocabulary size by
frequency level
Number of words recognised by frequency level and proficiency
(English L1, n = 42; Advanced ESL, n = 36;
Intermediate ESL, n = 32)
%recognised
100
90
80
70
60
ENGLISH L1
ADVANCED ESL
INTERMED ESL
50
40
30
20
10
0
2000
3000
5000
10000
Word frequency
21/10/16
31
Which noun phrase is the

agent/subject of the sentence?
1. The rock the scissor kissed.
2. Ate the thing the man.
3. The horse hit the tree.
4. The house the doctor met.
5. Saw the boy the girl.
6. The boat the giraffe steered.
7. The husband the wife visited.
8. Played the time the room.
21/10/16
32
Table 3.1 Cues used in Harrington (1987).
The agent of the sentence
(Stress underlined)
favors
The girl kissed the rock
The girl tapped the boy
The turtle the rock kicked.
Word order
NVN
NVN
NNV
Animacy favors
Contrastive stress
first noun
both nouns
first noun
first noun
second noun
neither
21/10/16
33
Problems with input-based approaches
21/10/16
Input-based models do not go beyond the input

in developing knowledge of grammar, though
they can generalize/analogize within limits.
a. Mary watched television before she had her dinner.
b. Before Mary had her dinner she watched television.
c. Before she had her dinner Mary watched television.
d. *She watched television before Mary had her dinner.
Input-based models have not been able to account for the

development of knowledge representations.
Where do cues come from?
34
TheMonitorModel
The Monitor Model is a general term for

an approach to second language learning
proposed by S. Krashen. The Monitor
Model consists of five interrelated
constructs, which the author labels
hypotheses:
1. The acquisition/learning hypothesis.
2. The monitor hypothesis.
3. The natural order hypothesis .
4. The input hypothesis.
5. The affective filter hypothesis.
21/10/16
35
1. The acquisition/learning
hypothesis.
21/10/16
Adults have two independent ways of

developing competence in SL:
(1) acquisition unconscious process
similar to Chomsky's language acquisition
device (LAD) that is, identical to the way
children develop ability in their first
language.
(2) learning - conscious knowledge of a
second language, knowing, being aware,
and being able to articulate the rules.
36
1.1 The acquisition/learning

hypothesis.
The acquisition/learning hypothesis: two key claims
1. Adults can access same LAD as children
2. Learning and acquisition are separate

processes: learning cannot become acquisition
Learning > practice > acquisition seems a

plausible way to describe the way we learn.
However, there three reasons why the learning
and acquisition are separate:
i. acquisition can occur without learning i.e.

language can be picked up ii. learning may not
become acquisition e.g. fossilisation iii.
impossible to know all the rules
21/10/16
37
2. The Monitor Hypothesis
21/10/16
Learning serves as a monitor on

output that is initiated by the
'acquired' system. At best it
provides a 'fine-tuning' of
production.
Implication for teaching: focus on
communication, not rule-learning.
38
2. 1 The Monitor
Hypothesis:
Conditions for use
The Monitor is only used when three conditions are

met:
i. Enough time is available

ii. There is a focus on form
iii. There is knowledge of rule
Monitor is only available in production
The Monitor is used to explain individual differences

and adult/child differences; Children do not use
Monitor, as it appears at onset of formal operations
21/10/16
39
3. The Natural Order

Hypothesis
L2 learners acquire grammatical

structure in a predictable order.
Originally one natural order was
posited but given the difficulty in
assessing comparability of
structures, multiple streams of
development are assumed.
21/10/16
40
3.1 Acquisition Hierarchy (Dulay, Burt & Krashen,

1982)
GROUP I
CASE (I/me) WORD ORDER (simple sentences)
GROUP II
SINGULAR COPULA (s/is) SINGULAR AUXILIARY (s/is)
PLURAL AUX (are)
PROGRESSIVE (-ing)
21/10/16
GROUP III
PAST IRREGULAR (went) CONDITIONAL AUX (would)
POSSESSIVE (s)
LONG PLURAL (-es)
3rd PERSON SINGULAR (-s)
GROUP IV
PERFECT AUX (have) PAST PARTICIPLE (-en)
41
4. The Input Hypothesis
Consists of four parts and assumes NOH is correct.

i. IH concerns acquisition, not learning
ii. Acquisition occurs by understanding language that
contains structure just beyond current level of
competence "i+1". This is done through context or
extralinguistic information, i.e. acquisition is caused
by understanding.
iii. When input is understood and there is enough of
it, i + 1 will be provided automatically (simplified
input important).
iv. Production ability emerges, it is not taught directly
(important implications for teaching).
21/10/16
42
4.1 Comprehensible Input
1. CI is receptive language that is slightly ahead of

the learners current state of grammatical
knowledge, referred to as i & 1.
2. Speech emerges from competence built from
comprehensible input.
3. Grammar knowledge is automatically provided
by input that is both comprehensible and of
sufficient quantity.
21/10/16
43
4.2 Comprehensible Input:

Where does it come from?
Linguistic adjustments to(1)
NonNative Speakers
(NNS)
> foreigner talk ( Ferguson, 1975)
> NS adjustment are generally well-formed but
modified.
* slower, more careful pronunciation, stress on

key words
* shorter grammar structures
* grammatical relations made explicit
* less syntactic or propositional complexity
* restricted vocabulary size
* fewer pro forms
21/10/16
44
4.2 Comprehensible Input:

Where does it come from?
(2)
Interactional adjustments to NNSs
21/10/16
> content: narrower range of topics, here

and now orientation
> interactional structure: acceptance of
topic shifts, confirmation and clarification
checks.
> paraphrase: approximation and
circumlocution
> avoidance: topic avoidance.
45
5. The Affective Filter

Hypothesis
There is an affective filter that
subconsciously screens incoming
language based on learner's motives,
needs, attitudes and emotional states.
The AF is outside the LAD proper.
AF incorporates affective variables

into the model.
21/10/16
46
Affective Filter Hypothesis:

Characteristics
AF incorporates affective variables into
the model.
Main source of individual differences in L2
learning, determining:
21/10/16
> which language models will be attended

to;
> which part of the language will be attended
to;
> when the learner will stop trying to learn;
> and how fast the learner will learn.
47
Problems with the

acquisition/learning
/how do learning and
acquisition differ in behavioural terms?
hypothesis.
/LAD is underspecified
/logical inconsistency: if most L2 is acquired unconsciously
then how can learning, which by definition is conscious,
contribute to the development of L2 competence?
/ learning need not precede acquisition but it is not the case
that it can't
/unconscious and conscious not defined. Does unconscious
means out of awareness at the moment or not accessible to
consciousness? If unconsciousness and consciousness are
mutually inaccessible then the model is true, but trivial. If
the two interact then there is no basis for the claim of
discreteness.
21/10/16
48
Problems with the Monitor

Hypothesis
21/10/16
/no evidence for the claim that it affects

production only, it is counterintuitive to say
that learning cannot aid comprehension
/ the Monitor is considered to be the only
way in which conscious knowledge of rules
can be used, but according to A/L
distinction, conscious knowledge plays no
role in acquisition, thus it is meaningless to
say that there are two ways to gain
competence.
/ the rules learner's use are not the same
as the linguist's rules
49
Problems with the Natural

Order Hypothesis
21/10/16
/Based on morpheme studies which

equate morpheme orders with
structural development. In absence
of linguistic theory it is not possible
to assess what the forms indicate.
/Multiple streams render hypothesis
meaningless.
50
Problems with the Input

Hypothesis
21/10/16
/acquisition is caused by
understanding? Simplified codes are
important but are they necessary?
/output not important?
/ narrow, unbalanced view that relies
only on comprehension and external
factors.
51
Problems with the

Affective Filter Hypothesis
/underspecified
/how is it developed?
/adolescents should be the worst
language learners; they arent
/why doesn't it apply to L1?
21/10/16
52
Current status of the

Monitor Model
The Monitor Model has largely been rejected as a testable

theory of SLA, however, it has had a major impact on SLA
research. The following quote captures the prevailing view
shared by most SLA researchers today:
It remains true none the less that Krashens ideas have
been highly influential in shaping many research agendas
and projects, and in doing so, considerably advancing our
understanding of SLA. Krashens main overall weakness
was the presentation of what were just hypotheses that
remained to be tested, as a comprehensive model that
had empirical reality. He then used his hypotheses
prematurely as a basis for drawing pedagogical
conclusions.
Mitchell, R. & Myles, S. (1998). Second language learning theories. P. 39.

London: Arnold.
21/10/16
53

Looking at Interlanguage Processing

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Looking at Interlanguage Processing

Uploaded by

Copyright:

Available Formats

Look at interlanguage

Second Language Acquisition

Second Language Acquisition

Second Language Acquisition

L is not driven by an innate faculty,

Second Language Acquisition

a.k.a. neural networks, neural computing,

Second Language Acquisition

Learning takes place as the network (

Second Language Acquisition

Features of connectionist models

Second Language Acquisition

Features of connectionist models

Second Language Acquisition

Features of connectionist models

Distributed representations and locus of control

Connectionist models have been termed "subsymbolic"

The distributed nature of the representations has

Second Language Acquisition

Features of connectionist models

Second Language Acquisition

functionalist model of speech

Second Language Acquisition

The Competition Model

Word order ( SVO)

Morphology (Case of nouns, Agreement

Second Language Acquisition

The Competition Model

In other cases the semantic and

The interpretation we choose is

Second Language Acquisition

The Competition Model

Learners of a second language tend

Second Language Acquisition

Competition Model and SLA

Learners first resort to their L1 interpretation

When they recognize an incongruity between

Finally, learners gradually adopt the

Direct form-function relationship

A unified account of processing and learning

Second Language Acquisition

Second Language Acquisition

Skill results from mastery and integration of

*Subskills become automatized.

Second Language Acquisition

Controlled versus automatic processing

* temporary activation of memory nodes

* habitual activation in response

* short-term working memory

* long term memory

Second Language Acquisition

Skill automatization &

Second Language Acquisition

Restructuring is the process by which

Second Language Acquisition

(speed, memory demand)

*accounts for 'practice effects' * accounts for transition

Second Language Acquisition

Restructuring has been proposed as mechanism to

Second Language Acquisition

U-shaped curve in lexical development

Willingness of Dutch EFL learners to accept English senses of break

He broke his leg.

Second Language Acquisition

accounts for 'practice effects' accounts for transition