Application Of Models in Child

Language Disorders
MODELS:
1. Dual-Route Cascaded model

2. Hierarchical model

3. Connectionist model
DUAL-ROUTE CASCADED MODEL
• Early 1970s
• Developed extensively by Coltheart et.al in 1990s

• Computational model of reading

• How skilled readers perform certain basic reading tasks

• Two separate mental mechanisms/Cognitive routes

• Output of both contributing to the pronunciation of a
written stimulus
 Reading Aloud

Lexical Route Non- Lexical Route

Written Input

Speech
LEXICAL ROUTE
• Recognize known words by sight alone (Dictionary Procedure)
• Represented in a mental database (Internal Lexicon)

Visual Access Internal Retrieve

Recognition Lexicon Pronunciation
 (Family, Man, Dog)

Regular
Learned Non-Words
Words
Irregular (Zuck, Pilvo, legotim)

(Listen, Psychology)
NON-LEXICAL ROUTE
• Identifying the word's constituent parts
• How these parts are associated with each other

String neighboring Build a phonological

letters sound together representation
 Irregular
Regular
Words

Non-
Words
Two ways in which the units of
different layers interact:
• One is through inhibition,
where the activation of a unit
makes it more difficult for the
activation of other units to
rise.
• Other is through excitation,
where the activation of a unit
contributes to the activation
of other units.
Reading disorders

Children with reading disorders rely primarily on the

non-lexical route while reading
Research shows that children can decode non-words,
letter by letter, accurately but with slow speed.
Children with reading disorders (RD) have both slow
reading speeds and impaired lexical routes, there are
suggestions that the same processes are involved in
fast naming of words.
Developmental/Acquired surface dyslexia (Sherman et.al, 2000)

 Damage to the Left temporal/parietal lobe in a previously

literate person
 Pronunciation errors that indicate impairment of the
lexical route.
 Cannot recognize words as a single unit
 Patients can accurately read words and nonwords that
comply with the letter-sound rules of the sub-lexical
system but mistakenly pronounce irregular words.
Ex: No silent “t” in “Listen”
Developmental dyslexia
(Franklin R Manis and Alan Peterson)

Developmental Surface dyslexia

Developmental Phonological dyslexia

Case Findings Correlated

Studies With Adults

Developmental Surface Dyslexia and Dysgraphia Orthographic Processing

(Hanley et.al) Impairment
Developmental/Acquired phonological dyslexia

Inability to read non words aloud and identify the

sounds of single letters.
Patients can read and correctly pronounce words
(as a whole), regardless of length, meaning, or
how common they are, as long as they are stored
in memory.
Caused by damage in the non lexical route, while
the lexical route, that allows reading familiar
words, remains intact.
HIERARCHICAL MODEL
 How words are associated with

? each other
How we access words in the
mental lexicon

Most modern models of semantic memory are predicated on

the idea that
1. Words are arranged in networks and
2. Concepts are linked on a variety of levels.
Cognitive Economy

• Collin & Quillian Model (1972)

Mental Concepts about a particular

concept is stored in ONE PLACE.
They supported their idea by testing individuals' response to
- A true sentence “Canary is a bird" and
- A false sentence "A chair is a bird”
They assumed that the time it took to verify the sentences
would indicate how far one was travelling mentally.
This was termed the DISTANCE EFFECT.

BIRD

CHAIR
Spread Activation Model

• Collin and Loftu (1975)

Words are arranged in networks of

nodes not always in a Hierarchial order
How does this work?

Name of the Node

concept is representing that
mentioned concept is
activated

Activation spreads 1st to Activation spreads

original nodes & then to from node to node
more remote nodes. (Spread of activation)
Lexical network model
Bock and Levelt (1994)

Conceptual Level

Lemma Level

Lexeme Level

25
Cromer's (1983) claimed that children with language impairments
have a hierarchical planning deficit that affects language as well as
performance on complex construction tasks.

SLI (Kamhi et.al,1995)

•Delayed language skills (Cromer,1981)

•Bilinguals with CLD (Ramos et.al,1985)

CONNECTIONIST MODEL
SIGNIFICANCE: (Rohde and David, 2003)

• Traditional approaches are explicit and discrete representations

that are difficult to learn from a reasonable linguistic environment.
•Structure of language is highly complex ranging from acoustic
processing to semantic analysis with a set of symbolic rules.
•Language processing operations operate extremely rapidly, requiring
a highly parallel, co-operative style of computation. Thus language
processing appears to represent a very difficult challenge for neural
network modelling.
• Connectionist modeling attempts to help bridge the gulf between
psycholinguistics and neuroscience and note what could be learned
from the environment in the absence of detailed innate knowledge.
 Connectionist models consist of several
interconnected neuron like processing units modified
by learning association between input (Stimulus) and
output (response).

Performance is not an imperfect reflection of some

abstract competence, but rather the behavioral
manifestation of the internal representations and
processes of actual language users

Connectionist model attempts to understand how the

neural computation is conducted in the brain by
simulating the cognitive and neural processes of
language, memory, and learning.
 Connectionist
Model

Simple
Connection
processing
s
elements
(Nodes, Units) (Among these processing elements)
A distributed model which assumes that information is not localized
in specific nodes and thus rely on patterns of activation across
models in the network, particularly through inter- connections
between nodes as in real biological synapses.

Postulates that:
 Input layer – Receives information from input patterns
(eg: representations of alphabetic features)

 Output layer – Provides output patterns produced by

the network.
(eg: classifications of alphabets acc. to shapes)

 Hidden Layer – connect the input and output units

(eg: similarities between “O” and “Q”)
These connections take on different values of strengths, which
are adjusted and updated during learning.
Lexical processing:

1) Phonological Development

•Plaut & Kello (1998) proposed connectionist

framework for phonological development in which
phonology was considered as a learned behavior.

•As an internal representation that mediates among

acoustic, articulatory, and semantic representations in
the service of both comprehension and production.
Plaut and Kello(1999)
The framework instantiates two key assumptions:

1. The first is that both comprehension and production are subserved

by the same underlying phonological representations.

2. The second key assumption is that feedback needed to guide the

development of speech production is derived from the comprehension
Morphology

• Relationship between inputs and outputs is

systematic but admits many exceptions.
• Connectionist model - All items co-exist within a
single system whose representations reflect the
degree of mapping for different items.
• Provides explanation for the cross-linguistic
differences that occur.
Sentence Processing (Rhode & David, 2003)

• Parsing
(Structural description of sentence from surface form)
• Word prediction
(Predict the likelihood of the appearance of a particular word)
• Comprehension
(Representation of the meaning of a sentence)
• Production
(Mapping from intended meaning to sequences of words/sounds)
Chater & Christiansen (2001)
SLI (Seidenberg et.al, 2003) Phonology and Syntax

•Delayed language skills (Trevor Harley, 2007)

•Bilinguals with CLD (Li et.al, 2016)

•Developmental Dyslexia (Seidenberg et.al, 1999)

 Orthographic to phonological representation

Word processing difficulties

Transferring from basic to skilled reading

(Word Recognition And Naming)

Pronouncing novel items

Naming and lexical decision tasks

Seidenberg and McClelland's (1989)

ARTICLES
 Bilingual Object Naming: A
Connectionist Model

Shin Yi Fang, Benjamin D, Barbara Malt and Ping

Li (9th May 2016)

Journal: Frontiers in Psychology, Volume 7,

PMCID: PMC4860466
• Understanding the complex mapping relationships between objects and
their names in each language of the bilinguals.
• Aims to test a computational model that captures bilingual word
representations and provides a foundation for future investigations into
the variables that affect the output (name choices) generated by such
representations.
•When a new L2 word form is taught as a translation equivalent of an L1
word, the L2 word will be activated by the same features as the L1 word,
and non-native L2 patterns of production will result.
• Over time these will be modified by L2 experience and will move away
from a uniform L1 pattern and may settle into a pattern that does not
exactly match that of the L2 but reflects the convergence of L1 and L2.
•This convergence reflects the
interactions between lexical
representations for the two
languages.
•In this study, a self-organizing
connectionist model was developed
to simulate semantic convergence in
the bilingual lexicon and investigate
the mechanisms underlying this
semantic convergence, which
compared to the empirical data from
past research to validate the model.
•Semantic, Phonological and
orthographic SOM, Lateral
connections and input-output
relationships.
•Impaired models were created in which components of the
network were removed so as to examine the importance of the
relevant components on bilingual object naming.
•The results demonstrated that connections between two
languages’ lexicons can be established through the
simultaneous activations of related words in the two
languages.
•These connections between languages allow the outputs of
their lexicons to become more similar, that is, to converge.
•The model provides a basis for future computational studies of
how various input variables may affect bilingual naming
patterns.
Developmental dyslexia and the dual route
model of reading: Simulating individual
differences and subtypes

Johannes C, Caroline Castel, Catherine, Florence

George, Alario, Conrad Perry
(April, 2008)
Developmental dyslexia was investigated within a well-understood and
fully specified computational model of reading aloud: the dual route
cascaded model (DRC [Coltheart et.al, 2001).
Four tasks were designed to assess each representational level of the
DRC:
1. Letter level
2. Orthographic lexicon
3. Phonological lexicon
4. Phoneme system
No single cause of dyslexia, but rather a complex pattern of phonological,
phonemic, and letter processing deficits. Importantly, most dyslexics had
deficits in more than one domain.
• Analyses also suggested that both the phonological and surface
dyslexics almost always had more than a single underlying deficit.
• To simulate the reading performance for each individual with the
DRC, noise was added to the model at a level proportional to the
underlying deficit(s) of each individual.
• The simulations not only accounted fairly well for individual
reading patterns but also captured the different dyslexia profiles
(surface, phonological and mixed dyslexia).
• Thus, taking into account the multiplicity of underlying deficits on
an individual basis an accurate description of developmental
dyslexia.
Thank you!