Unit 4(nlp _neural_network)

1
[Unit 7: Application of AI]
Artificial Intelligence (CSC 355)
Bal Krishna Subedi
Central Department of Computer Science & Information Technology
Tribhuvan University

Artificial Intelligence Chapter- Application of AI
Bal Krishna Subedi
2
Expert Systems:
An Expert system is a set of program that manipulates encoded knowledge to solve
problem in a specialized domain that normally requires human expertise.
A computer system that simulates the decision- making process of a human expert in a
specific domain.
An expert system’s knowledge is obtained from expert sources and coded in a form
suitable for the system to use in its inference or reasoning processes. The expert knowledge
must be obtained from specialists or other sources of expertise, such as texts, journals,
articles and data bases.
An expert system is an “intelligent” program that solves problems in a narrow problem
area by using high-quality, specific knowledge rather than an algorithm.
Block Diagram
There is currently no such thing as “standard” expert system. Because a variety of
techniques are used to create expert systems, they differ as widely as the programmers who
develop them and the problems they are designed to solve. However, the principal
components of most expert systems are knowledge base, an inference engine, and a user
interface, as illustrated in the figure.
Fig: Block Diagram of expert system
Knowledge
Base
Inference
Engine
User
Interface
User

Bal Krishna Subedi
3
1. Knowledge Base
The component of an expert system that contains the system’s knowledge is called its
knowledge base. This element of the system is so critical to the way most expert
systems are constructed that they are also popularly known as knowledge-based
systems
A knowledge base contains both declarative knowledge (facts about objects, events
and situations) and procedural knowledge (information about courses of action).
Depending on the form of knowledge representation chosen, the two types of
knowledge may be separate or integrated. Although many knowledge representation
techniques have been used in expert systems, the most prevalent form of knowledge
representation currently used in expert systems is the rule-based production system
approach.
To improve the performance of an expert system, we should supply the system with
some knowledge about the knowledge it posses, or in other words, meta-knowledge.
2. Inference Engine
Simply having access to a great deal of knowledge does not make you an expert; you
also must know how and when to apply the appropriate knowledge. Similarly, just
having a knowledge base does not make an expert system intelligent. The system must
have another component that directs the implementation of the knowledge. That
element of the system is known variously as the control structure, the rule interpreter,
or the inference engine.
The inference engine decides which heuristic search techniques are used to determine
how the rules in the knowledge base are to be applied to the problem. In effect, an
inference engine “runs” an expert system, determining which rules are to be invoked,
accessing the appropriate rules in the knowledge base, executing the rules , and
determining when an acceptable solution has been found.
3. User Interface
The component of an expert system that communicates with the user is known as the
user interface. The communication performed by a user interface is bidirectional. At
the simplest level, we must be able to describe our problem to the expert system, and
the system must be able to respond with its recommendations. We may want to
ask the system to explain its “reasoning”, or the system may request additional
information about the problem from us.
Beside these three components, there is a Working Memory - a data structure which
stores information about a specific run. It holds current facts and knowledge.

Bal Krishna Subedi
4
Stages of Expert System Development:
Although great strides have been made in expediting the process of developing an expert
system, it often remains an extremely time consuming task. It may be possible for one or
two people to develop a small expert system in a few months; however the development of
a sophisticated system may require a team of several people working together for more
than a year.
An expert system typically is developed and refined over a period of several years. We can
divide the process of expert system development into five distinct stages. In practice, it
may not be possible to break down the expert system development cycle precisely.
However, an examination of these five stages may serve to provide us with some insight
into the ways in which expert systems are developed.
Fig: Different phases of expert system development
Identification:
Beside we can begin to develop an expert system, it is important that we describe, with as
much precision as possible, the problem that the system is intended to solve. It is not
enough simply to feel that the system would be helpful in certain situation; we must
determine the exact nature of the problem and state the precise goals that indicate exactly
how we expect the expert system to contribute to the solution.
Conceptualization:
Once we have formally identified the problem that an expert system is to solve, the next
stage involves analyzing the problem further to ensure that its specifics, as well as it
generalities, are understood. In the conceptualization stage the knowledge engineer
frequently creates a diagram of the problem to depict graphically the relationships between
the objects and processes in the problem domain. It is often helpful at this stage to divide
the problem into a series of sub-problems and to diagram both the relationships among the
pieces of each sub-problem and the relationships among the various sub-problems.
Formulating
Rules that
Embody the
Knowledge
Conceptualization Formalization Implementation
Designing
structures to
organize the
knowledge
Finding
Concepts to
represent the
knowledge.
Validating
The Rules.
Testing
Determining the
Characteristics
of the problems.
Identification

Bal Krishna Subedi
5
Formalization:
In the preceding stages, no effort has been made to relate the domain problem to the
artificial intelligence technology that may solve it. During the identification and the
conceptualization stages, the focus is entirely on understanding the problem. Now, during
the formalization stage, the problem is connected to its proposed solution, an expert
system, by analyzing the relationships depicted in the conceptualization stage.
During formalization, it is important that the knowledge engineer be familiar with the
following:
 The various techniques of knowledge representation and heuristic search
used in expert systems.
 The expert system “tools” that can greatly expedite the development
process. And
 Other expert systems that may solve similar problems and thus may be
adequate to the problem at hand.
Implementation:
During the implementation stage, the formalized concepts are programmed onto the
computer that has been chosen for system development, using the predetermined
techniques and tools to implement a “first pass” prototype of the expert system.
Theoretically, if the methods of the previous stage have been followed with diligence and
care, the implementation of the prototype should be as much an art as it is a science,
because following all rules does not guarantee that the system will work the first time it is
implemented. Many scientists actually consider the first prototype to be a “throw-away’
system, useful for evaluating progress but hardly a usable expert system.
Testing:
Testing provides opportunities to identify the weakness in the structure and
implementation of the system and to make the appropriate corrections. Depending on the
types of problems encountered, the testing procedure may indicate that the system was
Features of an expert system:
What are the features of a good expert system? Although each expert system has its own
particular characteristics, there are several features common to many systems. The
following list from Rule-Based Expert Systems suggests seven criteria that are important
prerequisites for the acceptance of an expert system .
1. “The program should be useful.” An expert system should be developed to meet a
specific need, one for which it is recognized that assistance is needed.

Bal Krishna Subedi
6
2. “The program should be usable.” An expert system should be designed so that
even a novice computer user finds it easy to use .
3. “The program should be educational when appropriate.” An expert system may
be used by non-experts, who should be able to increase their own expertise by
using the system.
4. “The program should be able to explain its advice.” An expert system should be
able to explain the “reasoning” process that led it to its conclusions, to allow us to
decide whether to accept the system’s recommendations.
5. “The program should be able to respond to simple questions.” Because people
with different levels of knowledge may use the system , an expert system should be
able to answer questions about points that may not be clear to all users.
6. “The program should be able to learn new knowledge.” Not only should an expert
system be able to respond to our questions, it also should be able to ask questions to
gain additional information.
7. “The program’s knowledge should be easily modified.” It is important that we
should be able to revise the knowledge base of an expert system easily to correct
errors or add new information.

Bal Krishna Subedi
7
Neural Networks:
A neuron is a cell in brain whose principle function is the collection, Processing, and
dissemination of electrical signals. Brains Information processing capacity comes
from networks of such neurons. Due to this reason some earliest AI work aimed to
create such artificial networks. (Other Names are Connectionism; Parallel distributed
processing and neural computing).
What is a Neural Network?
An Artificial Neural Network (ANN) is an information processing paradigm that is
inspired by the way biological nervous systems, such as the brain, process information.
The key element of this paradigm is the novel structure of the information processing
system. It is composed of a large number of highly interconnected processing elements
(neurones) working in unison to solve specific problems. ANNs, like people, learn by
example. An ANN is configured for a specific application, such as pattern recognition or
data classification, through a learning process.
Why use neural networks?
Neural networks, with their remarkable ability to derive meaning from complicated or
imprecise data, can be used to extract patterns and detect trends that are too complex to be
noticed by either humans or other computer techniques. A trained neural network can be
thought of as an "expert" in the category of information it has been given to analyze. Other
advantages include:
1. Adaptive learning: An ability to learn how to do tasks based on the data given for
training or initial experience.
2. Self-Organisation: An ANN can create its own organisation or representation of the
information it receives during learning time.
3. Real Time Operation: ANN computations may be carried out in parallel, and
special hardware devices are being designed and manufactured which take
advantage of this capability.
4. Fault Tolerance via Redundant Information Coding: Partial destruction of a
network leads to the corresponding degradation of performance. However, some
network capabilities may be retained even with major network damage

Bal Krishna Subedi
8
Neural networks versus conventional computers
Neural networks take a different approach to problem solving than that of conventional
computers. Conventional computers use an algorithmic approach i.e. the computer follows
a set of instructions in order to solve a problem. Unless the specific steps that the computer
needs to follow are known the computer cannot solve the problem. That restricts the
problem solving capability of conventional computers to problems that we already
understand and know how to solve. But computers would be so much more useful if they
could do things that we don't exactly know how to do.
Neural networks process information in a similar way the human brain does. The network
is composed of a large number of highly interconnected processing elements(neurones)
working in parallel to solve a specific problem. Neural networks learn by example. They
cannot be programmed to perform a specific task. The examples must be selected carefully
otherwise useful time is wasted or even worse the network might be functioning
incorrectly. The disadvantage is that because the network finds out how to solve the
problem by itself, its operation can be unpredictable.
On the other hand, conventional computers use a cognitive approach to problem solving;
the way the problem is to solved must be known and stated in small unambiguous
instructions. These instructions are then converted to a high level language program and
then into machine code that the computer can understand. These machines are totally
predictable; if anything goes wrong is due to a software or hardware fault.
Units of Neural Network:
Nodes(units):
Nodes represent a cell of neural network.
Links:
Links are directed arrows that show propagation of information from one node to
another node.
Activation:
Activations are inputs to or outputs from a unit.
Weight:
Each link has weight associated with it which determines strength and sign of the
connection.
Activation function:
A function which is used to derive output activation from the input activations to a
given node is called activation function.
Bias Weight:
Bias weight is used to set the threshold for a unit. Unit is activated when the
weighted sum of real inputs exceeds the bias weight.

Bal Krishna Subedi
9
Simple Model of Neural Network
A simple mathematical model of neuron is devised by McCulloch and Pit is given in the
figure given below:
It fires when a linear combination of its inputs exceeds some threshold.
A neural network is composed of nodes (units) connected by directed links A link from
unit j to i serve to propagate the activation aj from j to i. Each link has some numeric
weight Wj,i associated with it, which determines strength and sign of connection.
Each unit first computes a weighted sum of it’s inputs:
n
ini =  Wj,i aj
J=0
Then it applies activation function g to this sum to derive the output:
n
ai = g( ini) =g(  Wj,i aj)
J=0
Here, aj output activation from unit j and Wj,i is the weight on the link j to this node.
Activation function typically falls into one of three categories:
 Linear
 Threshold (Heaviside function)
 Sigmoid
 Sign

Bal Krishna Subedi
10
For linear activation functions, the output activity is proportional to the total weighted
output.
g(x) = k x + c, where k and x are constant
For threshold activation functions, the output are set at one of two levels, depending on
whether the total input is greater than or less than some threshold value.
g(x) = 1 if x>= k
= 0 if x < k
For sigmoid activation functions, the output varies continuously but not linearly as the
input changes. Sigmoid units bear a greater resemblance to real neurons than do linear or
threshold units. It has the advantage of differentiable.
g(x) = 1/ (1 + e-x
)
Realizing logic gates by using Neurons:

Bal Krishna Subedi
11
Network structures:
Feed-forward networks:
Feed-forward ANNs allow signals to travel one way only; from input to output. There is
no feedback (loops) i.e. the output of any layer does not affect that same layer. Feed-
forward ANNs tend to be straight forward networks that associate inputs with outputs.
They are extensively used in pattern recognition. This type of organization is also referred
to as bottom-up or top-down.
Feedback networks (Recurrent networks:)
Feedback networks (figure 1) can have signals traveling in both directions by introducing
loops in the network. Feedback networks are very powerful and can get extremely
complicated. Feedback networks are dynamic; their 'state' is changing continuously until
they reach an equilibrium point. They remain at the equilibrium point until the input
changes and a new equilibrium needs to be found. Feedback architectures are also referred
to as interactive or recurrent.
Outputs
Inputs
Outputs
Inputs

Bal Krishna Subedi
12
Feed-forward example
Here;
a5 = g(W3;5 a3 +W4;5 a4)
= g(W3;5 g(W1;3 a1 +W2;3 a2) + W4;5 g(W1;4 a1 +W2;4 a2)
Types of Feed Forward Neural Network:
Single-layer neural networks (perceptrons)
A neural network in which all the inputs connected directly to the outputs is called a
single-layer neural network, or a perceptron network. Since each output unit is independent
of the others each weight affects only one of the outputs.

Bal Krishna Subedi
13
Multilayer neural networks (perceptrons)
The neural network which contains input layers, output layers and some hidden layers also
is called multilayer neural network. The advantage of adding hidden layers is that it
enlarges the space of hypothesis. Layers of the network are normally fully connected.
Once the number of layers, and number of units in each layer, has been selected, training is
used to set the network's weights and thresholds so as to minimize the prediction error
made by the network
Training is the process of adjusting weights and threshold to produce the desired result for
different set of data.
Learning in Neural Networks:
Learning: One of the powerful features of neural networks is learning. Learning in
neural networks is carried out by adjusting the connection weights among neurons. It
is similar to a biological nervous system in which learning is carried out by changing
synapses connection strengths, among cells.
The operation of a neural network is determined by the values of the interconnection
weights. There is no algorithm that determines how the weights should be assigned in
order to solve specific problems. Hence, the weights are determined by a learning process
Learning may be classified into two categories:
(1) Supervised Learning
(2) Unsupervised Learning

Bal Krishna Subedi
14
Supervised Learning:
In supervised learning, the network is presented with inputs together with the target
(teacher signal) outputs. Then, the neural network tries to produce an output as close as
possible to the target signal by adjusting the values of internal weights. The most common
supervised learning method is the “error correction method”.
Error correction method is used for networks which their neurons have discrete output
functions. Neural networks are trained with this method in order to reduce the error
(difference between the network's output and the desired output) to zero.
Unsupervised Learning:
In unsupervised learning, there is no teacher (target signal) from outside and the network
adjusts its weights in response to only the input patterns. A typical example of
unsupervised learning is Hebbian learning.

Bal Krishna Subedi
15
Consider a machine (or living organism) which receives some sequence of inputs x1, x2,
x3, . . ., where xt is the sensory input at time t. In supervised learning the machine is given
a sequence of input & a sequence of desired outputs y1, y2, . . . , and the goal of the
machine is to learn to produce the correct output given a new input. While, in unsupervised
learning the machine simply receives inputs x1, x2, . . ., but obtains neither supervised
target outputs, nor rewards from its environment. It may seem somewhat mysterious to
imagine what the machine could possibly learn given that it doesn’t get any feedback from
its environment. However, it is possible to develop of formal framework for unsupervised
learning based on the notion that the machine’s goal is to build representations of the input
that can be used for decision making, predicting future inputs, efficiently communicating
the inputs to another machine, etc. In a sense, unsupervised learning can be thought of as
finding patterns in the data above and beyond what would be considered pure unstructured
noise.
Hebbian Learning:
The oldest and most famous of all learning rules is Hebb’s postulate of learning:
―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 process or metabolic changes take
place in one or both cells such that A’s efficiency as one of the cells firing B is
increased‖
From the point of view of artificial neurons and artificial neural networks, Hebb's principle
can be described as a method of determining how to alter the weights between model
neurons. The weight between two neurons increases if the two neurons activate
simultaneously—and reduces if they activate separately. Nodes that tend to be either
both positive or both negative at the same time have strong positive weights, while those
that tend to be opposite have strong negative weights.
Hebb’s Algorithm:
Step 0: initialize all weights to 0
Step 1: Given a training input, s, with its target output, t, set the activations of the input
units: xi = si
Step 2: Set the activation of the output unit to the target value: y = t
Step 3: Adjust the weights: wi(new) = wi(old) + xiy
Step 4: Adjust the bias (just like the weights): b(new) = b(old) + y

Bal Krishna Subedi
16
Example:
PROBLEM: Construct a Hebb Net which performs like an AND function, that is, only
when both features are “active” will the data be in the target class.
TRAINING SET (with the bias input always at 1):
Training-First Input:
Training- Second Input:

Bal Krishna Subedi
17
Training- Third Input:

Bal Krishna Subedi
18
Training- Fourth Input:
Final Neuron:

Bal Krishna Subedi
19
Perceptron Learning Theory:
The term "Perceptrons" was coined by Frank RosenBlatt in 1962 and is used to describe
the connection of simple neurons into networks. These networks are simplified versions of
the real nervous system where some properties are exagerrated and others are ignored. For
the moment we will concentrate on Single Layer Perceptrons.
So how can we achieve learning in our model neuron? We need to train them so they can
do things that are useful. To do this we must allow the neuron to learn from its mistakes.
There is in fact a learning paradigm that achieves this, it is known as supervised learning
and works in the following manner.
i. set the weight and thresholds of the neuron to random values.
ii. present an input.
iii. caclulate the output of the neuron.
iv. alter the weights to reinforce correct decisions and discourage wrong decisions,
hence reducing the error. So for the network to learn we shall increase the weights
on the active inputs when we want the output to be active, and to decrease them
when we want the output to be inactive.
v. Now present the next input and repeat steps iii. - v.
Perceptron Learning Algorithm:
The algorithm for Perceptron Learning is based on the supervised learning procedure
discussed previously.
Algorithm:
i. Initialize weights and threshold.
Set wi(t), (0 <= i <= n), to be the weight i at time t, and ø to be the threshold value
in the output node. Set w0 to be -ø, the bias, and x0 to be always 1.
Set wi(0) to small random values, thus initializing the weights and threshold.
ii. Present input and desired output
Present input x0, x1, x2, ..., xn and desired output d(t)
iii. Calculate the actual output
y(t) = g [w0(t)x0(t) + w1(t)x1(t) + .... + wn(t)xn(t)]
iv. Adapts weights

Bal Krishna Subedi
20
wi(t+1) = wi(t) + α[d(t) - y(t)]xi(t) , where 0 <= α <= 1 (learning rate) is a positive
gain function that controls the adaption rate.
Steps iii. and iv. are repeated until the iteration error is less than a user-specified error
threshold or a predetermined number of iterations have been completed.
Please note that the weights only change if an error is made and hence this is only when
learning shall occur.
Delta Rule:
The delta rule is a gradient descent learning rule for updating the weights of the artificial
neurons in a single-layer perceptron. It is a special case of the more general
backpropagation algorithm. For a neuron with activation function the delta rule for
's th weight is given by
,
where is a small constant called learning rate, is the neuron's activation function,
is the target output, is the weighted sum of the neuron's inputs, is the actual output,
and is the th input. It holds and .
The delta rule is commonly stated in simplified form for a perceptron with a linear
activation function as
Backpropagation
It is a supervised learning method, and is an implementation of the Delta rule. It requires a
teacher that knows, or can calculate, the desired output for any given input. It is most
useful for feed-forward networks (networks that have no feedback, or simply, that have no
connections that loop). The term is an abbreviation for "backwards propagation of errors".
Backpropagation requires that the activation function used by the artificial neurons (or
"nodes") is differentiable.
As the algorithm's name implies, the errors (and therefore the learning) propagate
backwards from the output nodes to the inner nodes. So technically speaking,
backpropagation is used to calculate the gradient of the error of the network with respect to
the network's modifiable weights. This gradient is almost always then used in a simple
stochastic gradient descent algorithm, is a general optimization algorithm, but is typically
used to fit the parameters of a machine learning model, to find weights that minimize the
error. Often the term "backpropagation" is used in a more general sense, to refer to the

Bal Krishna Subedi
21
entire procedure encompassing both the calculation of the gradient and its use in stochastic
gradient descent. Backpropagation usually allows quick convergence on satisfactory local
minima for error in the kind of networks to which it is suited.
Backpropagation networks are necessarily multilayer perceptrons (usually with one input,
one hidden, and one output layer). In order for the hidden layer to serve any useful
function, multilayer networks must have non-linear activation functions for the multiple
layers: a multilayer network using only linear activation functions is equivalent to some
single layer, linear network.
Summary of the backpropagation technique:
1. Present a training sample to the neural network.
2. Compare the network's output to the desired output from that sample. Calculate the
error in each output neuron.
3. For each neuron, calculate what the output should have been, and a scaling factor,
how much lower or higher the output must be adjusted to match the desired output.
This is the local error.
4. Adjust the weights of each neuron to lower the local error.
5. Assign "blame" for the local error to neurons at the previous level, giving greater
responsibility to neurons connected by stronger weights.
6. Repeat from step 3 on the neurons at the previous level, using each one's "blame"
as its error.
Characteristics:
• A multi-layered perceptron has three distinctive characteristics
– The network contains one or more layers of hidden neurons
– The network exhibits a high degree of connectivity
– Each neuron has a smooth (differentiable everywhere) nonlinear activation
function, the most common is the sigmoidal nonlinearity:
• The backpropagation algorithm provides a computational efficient method for
training multi-layer networks

Bal Krishna Subedi
22
Algorithm:
Step 0: Initialize the weights to small random values
Step 1: Feed the training sample through the network and determine the final output
Step 2: Compute the error for each output unit, for unit k it is:
Step 3: Calculate the weight correction term for each output unit, for unit k it is:
Step 4: Propagate the delta terms (errors) back through the weights of the hidden units
where the delta input for the jth
hidden unit is:
The delta term for jth
hidden unit is:
Step 5: Calculate the weight correction term for the hidden units:
Step 6: Update the weights:
Step 7: Test for stopping (maximum cylces, small changes, etc)
Note: There are a number of options in the design of a backprop system;
– Initial weights – best to set the initial weights (and all other free parameters)
to random numbers inside a small range of values (say –0.5 to 0.5)
– Number of cycles – tend to be quite large for backprop systems
– Number of neurons in the hidden layer – as few as possible

Bal Krishna Subedi
23
Natural Language Processing:
Perception and communication are essential components of intelligent behavior. They
provide the ability to effectively interact with our environment. Humans perceive and
communicate through their five basic senses of sight, hearing, touch, smell and taste, and
their ability to generate meaningful utterances. Developing programs that understand a
natural language is a difficult problem. Natural languages are large. They contain infinity
of different sentences. No matter how many sentences a person has heard or seen, new
ones can always be produced. Also, there is much ambiguity in a natural language. Many
words have several meanings and sentences can have different meanings in different
contexts. This makes the creation of programs that understand a natural language, one of
the most challenging tasks in AI.
Developing programs to understand natural language is important in AI because a natural
form of communication with systems is essential for user acceptance. AI programs must be
able to communicate with their human counterparts in a natural way, and natural language
is one of the most important mediums for that purpose. So, Natural Language Processing
(NLP) is the field that deals with the computer processing of natural languages, mainly
evolved by people working in the field of Artificial Intelligence.
Natural Language Processing (NLP), is the attempt to extract the fuller meaning
representation from the free text. Natural language processing is a technology which
involves converting spoken or written human language into a form which can be processed
by computers, and vice versa. Some of the better-known applications of NLP include:
 Voice recognition software which translates speech into input for word processors
or other applications;
 Text-to-speech synthesizers which read text aloud for users such as the hearing-
impaired;
 Grammar and style checkers which analyze text in an attempt to highlight errors
of grammar or usage;
 Machine translation systems which automatically render a document such as a
web page in another language.

Bal Krishna Subedi
24
Natural Language Generation:
"Natural Language Generation (NLG), also referred to as text generation, is a subfield of
natural language processing (NLP; which includes computational linguistics)
Natural Language Generation (NLG) is the natural language processing task of
generating natural language from a machine representation system such as a knowledge
base or a logical form.
In a sense, one can say that an NLG system is like a translator that converts a computer
based representation into a natural language representation. However, the methods to
produce the final language are very different from those of a compiler due to the inherent
expressivity of natural languages.

Bal Krishna Subedi
25
NLG may be viewed as the opposite of natural language understanding. The difference can
be put this way: whereas in natural language understanding the system needs to
disambiguate the input sentence to produce the machine representation language, in NLG
the system needs to make decisions about how to put a concept into words.
The different types of generation techniques can be classified into four main categories:
 Canned text systems constitute the simplest approach for single-sentence and multi-
sentence text generation. They are trivial to create, but very inflexible.
 Template systems, the next level of sophistication, rely on the application of pre-
defined templates or schemas and are able to support flexible alterations. The
template approach is used mainly for multi-sentence generation, particularly in
applications whose texts are fairly regular in structure.
 Phrase-based systems employ what can be seen as generalized templates. In such
systems, a phrasal pattern is first selected to match the top level of the input, and
then each part of the pattern is recursively expanded into a more specific phrasal
pattern that matches some subportion of the input. At the sentence level, the
phrases resemble phrase structure grammar rules and at the discourse level they
play the role of text plans.
 Feature-based systems, which are as yet restricted to single-sentence generation,
represent each possible minimal alternative of expression by a single feature.
Accordingly, each sentence is specified by a unique set of features. In this
framework, generation consists in the incremental collection of features appropriate
for each portion of the input. Feature collection itself can either be based on
unification or on the traversal of a feature selection network. The expressive power
of the approach is very high since any distinction in language can be added to the
system as a feature. Sophisticated feature-based generators, however, require very
complex input and make it difficult to maintain feature interrelationships and
control feature selection.
Many natural language generation systems follow a hybrid approach by combining
components that utilize different techniques.
Natural Language Understanding:
Developing programs that understand a natural language is a difficult problem. Natural
languages are large. They contain infinity of different sentences. No matter how many
sentences a person has heard or seen, new ones can always be produced. Also, there is
much ambiguity in a natural language. Many words have several meaning such as can,
bear, fly, bank etc, and sentences have different meanings in different contexts.
Example :- A can of juice. I can do it.
This makes the creation of programs that understand a natural language, one of the most
challenging tasks in AI. Understanding the language is not only the transmission of words.
It also requires inference about the speakers’ goal, knowledge as well as the context of the

Bal Krishna Subedi
26
interaction. We say a program understand natural language if it behaves by taking the
correct or acceptable action in response to the input. A word functions in a sentence as a
part of speech. Parts of the speech for the English language are nouns, pronouns, verbs,
adjectives, adverbs, prepositions, conjunctions and interjections. Three major issues
involved in understanding language.
 A large amount of human knowledge is assumed.
 Language is pattern based, phonemes are components of the words and words make
phrases and sentences. Phonemes, words and sentences order are not random.
 Language acts are the product of agents (human or machine).
Levels of knowledge used in Language Understanding
A language understanding knowledge must have considerable knowledge about the
structures of the language including what the words are and how they combine into phrases
and sentences. It must also know the meanings of the words and how they contribute to the
meanings of the sentence and to the context within which they are being used. The
component forms of knowledge needed for an understanding of natural languages are
sometimes classified according to the following levels.
 Phonological
 Relates sound to the words we recognize. A phoneme is the smallest
unit of the sound. Phones are aggregated to the words.
 Morphological
 This is lexical knowledge which relates to the word construction
from basic units called morphemes. A morpheme is the smallest unit
of meaning. Eg:- friend + ly = friendly
 Syntactic
 This knowledge relates to how words are put together or structure
red together to form grammatically correct sentences in the
language.
 Semantic
 This knowledge is concerned with the meanings of words and
phrases and how they combine to form sentence meaning.
 Pragmatic

Bal Krishna Subedi
27
 This is high – level knowledge which relates to the use of sentences
in different contexts and how the context affects the meaning of the
sentence.
 World
 Includes the knowledge of the physical world, the world of human
social interaction, and the roles of goals and intentions in
communication.
Basic Parsing Techniques
Before the meaning of a sentence can be determined, the meanings of its constituent parts
must be established. This requires knowledge of the structure of the sentence, the meaning
of the individual words and how the words modify each other. The process of determining
the syntactical structure of a sentence is known as parsing. Parsing is the process of
analyzing a sentence by taking it apart word – by – word and determining its structure
from its constituent parts and sub parts. The structure of a sentence can be represented with
a syntactic tree. When given an input string, the lexical parts or terms (root words), must
first be identified by type and then the role they play in a sentence must be determined.
These parts can be combined successively into larger units until a complete tree has been
computed.

Bal Krishna Subedi
28
To determine the meaning of a word, a parser must have access to a lexicon. When the
parser selects the word from the input stream, it locates the world in the lexicon and
obtains the word’s possible functions and features, including the semantic information.
Lexeme (Lexicon) & word forms:
The distinction between these two senses of "word" is arguably the most important one in
morphology. The first sense of "word", the one in which dog and dogs are "the same
Input String Parser Output representation
structure
Input String
Figure :- Parsing an input to create an output structure

Bal Krishna Subedi
29
word", is called a lexeme. The second sense is called word form. We thus say that dog and
dogs are different forms of the same lexeme. Dog and dog catcher, on the other hand, are
different lexemes, as they refer to two different kinds of entities. The form of a word that is
chosen conventionally to represent the canonical form of a word is called a lemma, or
citation form.
A lexicon defines the words of a language that a system knows about. This is includes
common words and words that are specific to the domain of the application. Entries
include meanings for each word and its syntactic and morphological behavior.
Morphology:
Morphology is the identification, analysis and description of the structure of words (words
as units in the lexicon are the subject matter of lexicology). While words are generally
accepted as being (with clitics) the smallest units of syntax, it is clear that in most (if not
all) languages, words can be related to other words by rules. For example, English speakers
recognize that the words dog, dogs, and dog catcher are closely related. English speakers
recognize these relations from their tacit knowledge of the rules of word formation in
English. They infer intuitively that dog is to dogs as cat is to cats; similarly, dog is to dog
catcher as dish is to dishwasher (in one sense). The rules understood by the speaker reflect
specific patterns (or regularities) in the way words are formed from smaller units and how
those smaller units interact in speech. In this way, morphology is the branch of linguistics
that studies patterns of word formation within and across languages, and attempts to
formulate rules that model the knowledge of the speakers of those languages.
Morphological analysis is the process of recognizing the suffixes and prefixes that
have been attached to a word.
We do this by having a table of affixes and trying to match the input as:
prefixes +root + suffixes.
– For example: adjective + ly -> adverb. E.g.: [Friend + ly]=friendly
– We may not get a unique result.
– “-s, -es” can be either a plural noun or a 3ps verb
– “-d, -ed” can be either a past tense or a perfect participle
Morphological Information:
• Transform part of speech
– green, greenness (adjective, noun)
– walk, walker (verb, noun)
• Change features of nouns
– boat, boats (singular, plural)
• Bill slept , Bill’s bed

Bal Krishna Subedi
30
– (subjective case, possessive case)
• Change features of verbs
– Aspect
• I walk. I am walking. (present, progressive)
– Tense
• I walked. I will walk. I had been walking. (past, future, past
progressive)
– Number and person
• I walk. They walk. (first person singular, third person plural)
Syntactic Analysis:
Syntactic analysis takes an input sentence and produces a representation of its grammatical
structure. A grammar describes the valid parts of speech of a language and how to combine
them into phrases. The grammar of English is nearly context free.
A computer grammar specifies which sentences are in a language and their parse trees. A
parse tree is a hierarchical structure that shows how the grammar applies to the input. Each
level of the tree corresponds to the application of one grammar rule.
It is the starting point for working out the meaning of the whole sentence. Consider the
following two sentences.
1. “The dog ate the bone.”
2. “The bone was eaten by the dog.”
Understanding the structure (via the syntax rules) of the sentences help us work out that
it’s the bone that gets eaten and not the dog. Syntactic analysis determines possible
grouping of words in a sentence. In other cases there may be many possible groupings of
words. Consider the sentence “John saw Mary with a telescope”. Two different readings
based on the groupings.
1. John saw (Mary with a telescope).
2. John (saw Mary with a telescope).
A sentence is syntactically ambiguous if there are two or more possible groupings.
Syntactic analysis helps determining the meaning of a sentence by working out possible
word structure. Rules of syntax are specified by writing a grammar for the language. A
parser will check if a sentence is correct according to the grammar. It returns a
representation (parse tree) of the sentence’s structure. A grammar specifies allowable
sentence structures in terms of basic categories such as noun and verbs. A given grammar,
however, is unlikely to cover all possible grammatical sentences. Parsing sentences is to
help determining their meanings, not just to check that they are correct. Suppose we want a
grammar that recognizes sentences like the following.

Bal Krishna Subedi
31
John ate the biscuit.
The lion ate the zebra.
The lion kissed John
But reject incorrect sentences such as
Ate John biscuit the.
Zebra the lion the ate.
Biscuit lion kissed.
A simple grammar that deals with this is given below
sentence --> noun_phase, verb phrase.
noun_phrase --> proper_noun.
noun_phrase --> determiner, noun.
verb_phrase --> verb, noun_phrase.
proper_noun --> [mary].
proper_noun --> [john].
noun --> [zebra].
noun --> [biscuit].
verb --> [ate].
verb --> [kissed].
determiner --> [the].
Incorrect sentences like “biscuit lion kissed” will be excluded by the grammar.

Bal Krishna Subedi
32
Semantic Analysis:
Semantic analysis is a process of converting the syntactic representations into a meaning
representation.
This involves the following tasks:
– Word sense determination
– Sentence level analysis
– Knowledge representation
- Word sense
Words have different meanings in different contexts.
Mary had a bat in her office.
• bat = `a baseball thing’
• bat = `a flying mammal’
- Sentence level analysis
Once the words are understood, the sentence must be assigned some meaning
I saw an astronomer with a telescope.
- Knowledge Representation
Understanding language requires lots of knowledge.

Bal Krishna Subedi
33
Parameters in Natural Language Processing:
 Auditory Inputs
 Segmentation
 Syntax Structure
 Semantic Structure
 Pragmatic Analysis
- Auditory Inputs:
Three of our five senses – sight, hearing and touch – are used as major inputs. These are
usually referred to as the visual, auditory and tactile inputs respectively. They are
sometimes called input channels; however, as previously mentioned, the term "channel" is
used in various ways, so I will avoid it.

Bal Krishna Subedi
34
In the fashion of video devices, audio devices are used to either capture or create sound. In
some cases, an audio output device can be used as an input device, in order to capture
produced sound.
 Microphone
 MIDI keyboard or other digital musical instrument
- Segmentation:
Text segmentation is the process of dividing written text into meaningful units, such as
words, sentences, or topics. The term applies both to mental processes used by humans
when reading text, and to artificial processes implemented in computers, which are the
subject of natural language processing. The problem is non-trivial, because while some
written languages have explicit word boundary markers, such as the word spaces of written
English and the distinctive initial, medial and final letter shapes of Arabic, such signals are
sometimes ambiguous and not present in all written languages.
Word segmentation is the problem of dividing a string of written language into its
component words. In English and many other languages using some form of the Latin
alphabet, the space is a good approximation of a word delimiter. (Some examples
where the space character alone may not be sufficient include contractions like can't
for can not.)
However the equivalent to this character is not found in all written scripts, and without it
word segmentation is a difficult problem. Languages which do not have a trivial word
segmentation process include Chinese, Japanese, where sentences but not words are
delimited, and Thai, where phrases and sentences but not words are delimited.
In some writing systems however, such as the Ge'ez script used for Amharic and Tigrinya
among other languages, words are explicitly delimited (at least historically) with a non-
whitespace character.
Word splitting is the process of parsing concatenated text (i.e. text that contains no spaces
or other word separators) to infer where word breaks exist.
Sentence segmentation is the problem of dividing a string of written language into its
component sentences. In English and some other languages, using punctuation,
particularly the full stop character is a reasonable approximation. However, even in
English this problem is not trivial due to the use of the full stop character for
abbreviations, which may or may not also terminate a sentence. For example Mr. is
not its own sentence in "Mr. Smith went to the shops in Jones Street." When
processing plain text, tables of abbreviations that contain periods can help prevent
incorrect assignment of sentence boundaries. As with word segmentation, not all
written languages contain punctuation characters which are useful for approximating
sentence boundaries.

Bal Krishna Subedi
35
Other segmentation problems: Processes may be required to segment text into
segments besides words, including morphemes (a task usually called morphological
analysis), paragraphs, topics or discourse turns.
A document may contain multiple topics, and the task of computerized text segmentation
may be to discover these topics automatically and segment the text accordingly. The topic
boundaries may be apparent from section titles and paragraphs. In other cases one needs to
use techniques similar to those used in document classification. Many different approaches
have been tried.
- Syntax Structure:
Same concept as in the syntactic analysis above
- Semantic Structure:
Same concept as in the semantic analysis above
- Pragmatic Analysis:
This is high level knowledge which relates to the use of sentences in different contexts and
how the context affects the meaning of the sentences. It is the study of the ways in which
language is used and its effect on the listener. Pragmatic comprises aspects of meaning that
depend upon the context or upon facts about real world.
Pragmatics – Handling Pronouns
Handling pronouns such as “he”, “she” and “it” is not always straight forward. Let us see
the following paragraph.
“John buys a new telescope. He sees Mary in the distance. He gets out his telescope. He
looks at her through it”.
Here, “her” refers to Mary who was not mentioned at all in the previous sentences. John’s
telescope was referred to as “a new telescope”, “his telescope” and “it”.
Let us see one more example
“When is the next flight to Sydney?”
“Does it have any seat left?”
Here, “it”, refers to a particular flight to Sydney, not Sydney itself.
Pragmatics – Ambiguity in Language
A sentence may have more than one structure such as

Bal Krishna Subedi
36
“I saw an astronomer with a telescope.”
This English sentence has a prepositional phrase “with a telescope” which may be attached
with either with verb to make phrase “saw something with telescope” or to object noun
phrase to make phrase “a astronomer with a telescope”. If we do first, then it can be
interpreted as “I saw an astronomer who is having a telescope”, and if we do second, it can
be interpreted as “Using a telescope I saw an astronomer”.
Now, to remove such ambiguity, one possible idea is that we have to consider the context.
If the knowledge base (KB) can prove that whether the telescope is with astronomer or not,
then the problem is solved.
Next approach is that; let us consider the real scenario where the human beings
communicate. If A says the same sentence “I saw an astronomer with a telescope.” To B,
then in practical, it is more probable that, B (listener) realizes that “A has seen astronomer
who is having a telescope”. It is because, normally, the word “telescope” belongs to
“astronomer”, so it is obvious that B realizes so.
If A has says that “I saw a lady with a telescope.” In this case, B realizes that “A has seen
the lady using a telescope”, because the word “telescope” has not any practical relationship
with “lady” like “astronomer”.
So, we may be able to remove such ambiguity, by defining a data structure, which can
efficiently handle such scenario. This idea may not 100% correct but seemed more
probable.

Bal Krishna Subedi
37
Machine Vision:
Machine vision (MV) is the application of computer vision to industry and manufacturing.
Whereas computer vision is the general discipline of making computers see (understand
what is perceived visually), machine vision, being an engineering discipline, is interested
in digital input/output devices and computer networks to control other manufacturing
equipment such as robotic arms and equipment to eject defective products.
Machine vision is the ability of a computer to "see." A machine-vision system
employs one or more video cameras, analog-to-digital conversion ( ADC ), and digital
signal processing ( DSP ). The resulting data goes to a computer or robot controller.
Machine vision is similar in complexity to voice recognition . The machine vision
systems use video cameras, robots or other devices, and computers to visually analyze an
operation or activity. Typical uses include automated inspection, optical character
recognition and other non-contact applications.
Two important specifications in any vision system are the sensitivity and the resolution.
Sensitivity is the ability of a machine to see in dim light, or to detect weak impulses at
invisible wavelengths. Resolution is the extent to which a machine can differentiate
between objects. In general, the better the resolution, the more confined the field of vision.
Sensitivity and resolution are interdependent. All other factors held constant, increasing the
sensitivity reduces the resolution, and improving the resolution reduces the sensitivity.
One of the most common applications of Machine Vision is the inspection of
manufactured goods such as semiconductor chips, automobiles, food and
pharmaceuticals. Just as human inspectors working on assembly lines visually inspect
parts to judge the quality of workmanship, so machine vision systems use digital
cameras, smart cameras and image processing software to perform similar
inspections.
Machine vision systems have two primary hardware elements: the camera, which
serves as the eyes of the system, and a computer video analyser. The recent rapid
acceleration in the development of machine vision for industrial applications can be
attributed to research in the areas of computer technologies. The first step in vision
analysis is the conversion of analog pixel intensity data into digital format for
processing. Next, an appropriate computer algorithm is employed to understand the
image data and provide appropriate analysis or action.
Machine vision encompasses computer science, optics, mechanical engineering, and
industrial automation. Unlike computer vision which is mainly focused on machine-based
image processing, machine vision integrates image capture systems with digital
input/output devices and computer networks to control manufacturing equipment such as
robotic arms. Manufacturers favour machine vision systems for visual inspections that
require high-speed, high-magnification, 24-hour operation, and/or repeatability of
measurements.

Bal Krishna Subedi
38
A typical machine vision system will consist of most of the following components:
 One or more digital or analogue cameras (black-and-white or colour) with suitable
optics for acquiring images, such as lenses to focus the desired field of view onto
the image sensor and suitable, often very specialized, light sources
 Input/Output hardware (e.g. digital I/O) or communication links (e.g. network
connection or RS-232) to report results
 A synchronizing sensor for part detection (often an optical or magnetic sensor) to
trigger image acquisition and processing and some form of actuators to sort, route
or reject defective parts
 A program to process images and detect relevant features.
The aim of a machine vision inspection system is typically to check the compliance of a
test piece with certain requirements, such as prescribed dimensions, serial numbers,
presence of components, etc. The complete task can frequently be subdivided into
independent stages, each checking a specific criterion. These individual checks typically
run according to the following model:
1. Image Capture
2. Image Preprocessing
3. Definition of one or more (manual) regions of interest
4. Segmentation of the objects
5. Computation of object features
6. Decision as to the correctness of the segmented objects
Naturally, capturing an image, possible several for moving processes, is a pre-requisite for
analysing a scene. In many cases these images are not suited for immediate examination
and require pre-processing to change certain sizing specific structures etc. In most cases it
is at least approximately known which image areas have to be analysed, i.e. the location of
a mark to be read or a component to be verified. These are called Regions of Interest
(ROIs) (sometimes Area of Interest or AOIs). Of course, such a region can also comprise
the entire image if required.
Machine vision is used in various industrial and medical applications. Examples include:
 Electronic component analysis
 Signature identification
 Optical character recognition
 Handwriting recognition
 Object recognition
 Pattern recognition
 Materials inspection
 Currency inspection
 Medical image analysis

Bal Krishna Subedi
39
Computer Vision:
Computer vision is the science and technology of machines that see, where see in this
case means that the machine is able to extract information from an image that is necessary
to solve some task. As a scientific discipline, computer vision is concerned with the theory
behind artificial systems that extract information from images. The image data can take
many forms, such as video sequences, views from multiple cameras, or multi-dimensional
data from a medical scanner.
As a technological discipline, computer vision seeks to apply its theories and models to the
construction of computer vision systems. Examples of applications of computer vision
include systems for:
 Controlling processes (e.g., an industrial robot or an autonomous vehicle).
 Detecting events (e.g., for visual surveillance or people counting).
 Organizing information (e.g., for indexing databases of images and image
sequences).
 Modeling objects or environments (e.g., industrial inspection, medical image
analysis or topographical modeling).
 Interaction (e.g., as the input to a device for computer-human interaction).
Computer vision is closely related to the study of biological vision. The field of biological
vision studies and models the physiological processes behind visual perception in humans
and other animals. Computer vision, on the other hand, studies and describes the processes
implemented in software and hardware behind artificial vision systems. Interdisciplinary
exchange between biological and computer vision has proven fruitful for both fields.
Computer vision is, in some ways, the inverse of computer graphics. While computer
graphics produces image data from 3D models, computer vision often produces 3D models
from image data. There is also a trend towards a combination of the two disciplines, e.g.,
as explored in augmented reality.
Sub-domains of computer vision include scene reconstruction, event detection, video
tracking, object recognition, learning, indexing, motion estimation, and image restoration.

Unit 4(nlp _neural_network)

More Related Content

What's hot (20)

Similar to Unit 4(nlp _neural_network) (20)

More from Tribhuvan University (20)

Recently uploaded (20)

Unit 4(nlp _neural_network)