Transcript Seminar Slides - CSE, IIT Bombay

Chat Bots
Welcome to the
world of living
people and artificial
intelligence entities
called bots!
Ruta Mehta
Mugdha Jain
Jeetendra Mirchandani
Questions We’ll Answer
A Chat Bot…
 What is it? Who wants it? Why?
 Since when is it around?
 How does it work?
 How do you test it?
 Can it answer all my questions?
 Can I make one of my own?
 Where can I put it to work?
Introduction



Chat Bot: A computer program that can talk to humans in
natural language!
Uses Artificial Intelligence Markup Language (AIML) to
represent knowledge.
Can replace a human for monotonous jobs of answering
queries, e.g. E-help desk.
How It All Started


Eliza – the first chat bot made by Joseph Weizenbaum.
Eliza Effect
–

tendency of humans to attach associations to terms from prior
experience.
Working of Eliza is based on
–
–
–
Knowledge Representation
Pattern Recognition
Substitution of key words into known phrases.
How does it respond




Looks for certain patterns of words in the user's input.
Replies with pre-determined output, if the pattern is
matched.
Needs to have an idea of what the user will chat
Has suitable responses defined in the AIML file
Architecture of a Chat Bot
Responder
AIML Interpreter
AIML Objects
A Chat Bot
Knowledge Representation

Types of AIML objects
–
–

Topics
Categories
E.g. :
<aiml>
<topic name=“the topic” >
<category>
<pattern>PATTERN</pattern>
<that>THAT</that>
<template>TEMPLATE</template>
</category>
</topic>
</aiml>
Example AIML Object

AIML Object
<category>
<pattern>HELLO</pattern>
<template>Hi
there!</template>
</category>

Chat Sequence
User: Hello!
Chat Bot: Hi There!
<category>
<pattern>YES</pattern>
<that>DO YOU LIKE MOVIES</that>
<template>What is your favorite
movie?</template>
</category>
Chat Bot: Do you like Movies?
User: Yes
Chat Bot: What is your favorite
movie
Topics and Categories



Topic: an optional top-level element that contains category
elements
Category: consists of an input question (pattern) and an
output answer (template)
Types of Categories
–
–
–
Atomic Category
Default Category
Recursive Category
Atomic Category


Contains patterns that does not have wildcards
“*” or “_”.
Example:
<category>
<pattern>10 DOLLARS</pattern>
<template> wow, what a cheap </template>
</category>

Conversation:
User: This watch is for 10 dollars
Chat Bot: Wow, what a cheap watch!
Default Category

Contains patterns that have wildcards “*” or “_”.
Example:

Conversation:

<category>
<pattern>10 *</pattern>
<template> It is ten.</template>
</category>
User: 10 dollars.
Chat Bot: It is ten.
Recursive category



Template calls the pattern matcher recursively
Uses <srai> tag, that stands for symbolic recursion artificial
intelligence
For example,
In English there are different ways to ask about X:
Describe x?
Tell me about X?
Do you know what X is?
The knowledge is stored in the simplest way.
Whatever the question is, it will be reduced to category like <What is>.
Input normalization

Substitution normalizations
Abbreviations such as "Mr." may be spelled out as "Mister" to avoid
sentence-splitting at the period in the abbreviated form.

Sentence-splitting normalizations
Rule: break sentences at periods. It relies upon substitutions performed
in the substitution phase.

Pattern-fitting normalizations
Remove all characters that are not normal characters; like converting
lowercase letters to uppercase .
Example
<category>
<pattern>HELLO</pattern>
<template> <random>
<li>Well hello there!</li>
<li>Hi there!</li>
<li>Hi there. I was just wanting
to talk</li>
<li>Hello there !</li>
</random></template></category>
<category>
<pattern>_ WHAT IS 2 AND
2</pattern>
<template>
<sr/><srai>WHAT IS 2 AND
2</srai>
</template>
</category>
<category>
<pattern>WHAT IS 2 *</pattern>
<template><random>
<li>Two.</li>
<li>Four.</li>
<li>Six.</li>
<li>12.</li>
</random></template></category
>
<category>
<pattern>HALO</pattern>
<template>
<srai>HELLO</srai>
</template>
</category>
Question :
Halo, What is 2 and 2
_ What is 2 and 2
</sr>
<srai> WHAT IS 2 AND 2 </srai>
HALO
WHAT IS 2 AND *
HELLO
WHAT IS 2 *
Well hello there!
Hi there!
Hi there. I was just
wanting to talk.
Hello there !
Answer :
Hi There! Six
Two
Four
Six
12
Graph-master – an example
interpreter





AIML interpreter: tries to match word by word to obtain the
largest pattern matching which is the best one.
Graph-master: an interpreter that models this behavior
Contains a set of nodes called Node-mappers .
Node-Mappers : map branches from each node where
branches represent the first words of all patterns.
Each leaf node contains a template.
Flowchart for Pattern Matching
yes
Node-mapper
Contains ‘_’?
Search sub-graph
rooted at child
node linked by ‘_’
Try all remaining
suffixes of input
following ‘X’
no
no
Node-mapper
Contains ‘X’?
yes
Search sub-graph
rooted at child
node linked by ‘X’
using input ‘tail’
Match?
no
Node-mapper
Contains ‘*’?
yes
Match?
yes
no
yes
Search sub-graph
rooted at child
node linked by ‘*’
Try all remaining
suffixes of input
following ‘X’
Match?
yes
Some Observations

Priority order at every node:
–
–
–



‘_’ wildcard, followed by
an atomic word, followed by
‘*’ wildcard .
Patterns need not be ordered alphabetically
The matching is word-by-word, not category-bycategory
Highly restricted form of depth-first search
Turing Test



Alan Turing proposed the Turing Test as a replacement for
the question “Can machines think?”
Turing's aim is to provide a method to assess whether or not a
machine can think.
The test
–
–
–
A man (A), a woman (B) and an interrogator (C) chat.
The objective of the interrogator is to determine which of the other
two is the woman
If a machine (bot) chats instead of A or B and fools the interrogator, it
has passed the Turing test.
Can It Answer All My Questions?




A Chat Bot has a limited number of patterns and responses.
The AIML structure supports regular expressions
In AIML, you can define recursive categories.
The bot may sometimes give funny replies, but that depends
on the AIML spec, i.e. Its brain!
How to build a bot of your own





Components – An AIML Object Spec, AIML interpreter, and
a responder!
A Chat Bot represents and models a character.
The brain is the AIML file, that defines patterns and
corresponding replies
A bot can be trained to be an Expert System about a special
theme – large data!
Training data – Yahoo Chat!
What is a Chat Bot useful for?





Commercial chatter bots to help customers
– At web-shops and e-commerce sites.
– Bots to receive complaints from users, online
An interactive (talking) encyclopedia.
Chatter bots administrating IRC-channels and Hotline server.
The Psychiatrist – the famous pronoun reversal trick
Starship Titanic, a game created by the famous writer
Douglas Adams along with Terry Jones
Questions We Answered
A Chat Bot…
 What is it? Who wants it? Why?
 Since when is it around?
 How does it work?
 How do you test it?
 Can it answer all my questions?
 Can I make one of my own?
 Where can I put it to work?
Clarifications

How powerful is JavaCC-built parser?
–

Powerful than a yacc generated parser?
–
–

AIML is connected to a system that allows for learning new patterns and
putting them into a context
Context sensitive
Learning ?
–

As long as one can use JavaCC's look-ahead specification to guide the
parsing where the LL(1) rules are not sufficient, JavaCC can handle any
grammar that is not left-recursive.
Dialogue Corpora can be used to train a Chatbot! (See References)
Intelligent?
–
Even Turing tests have different levels and yet cannot finally decide
whether the system is intelligent or not. Nowadays a system is considered
to be intelligent if it is able to mimic intelligent behaviour (e.g. within a
specified domain).
Thank You
References





The Anatomy of A.L.I.C.E.: Dr. Richard S. Wallace,
http://www.alicebot.org/anatomy.html
Artificial Intelligence Markup Language (AIML), A.L.I.C.E. AI
Foundation, http://alicebot.org/TR/2001/WD-aiml/
AIML Interpreter Overview 2004, http://www.aimlbots.com/en/aimlinterpreters.html
Computing machinery and intelligence, Alan Turing [1950],
http://www.abelard.org/turpap/turpap.htm
Using Dialogue Corpora to Train a Chatbot (Bayan Abu Shawar, Eric
Atwell)
http://www.comp.leeds.ac.uk/andyr/research/papers/techreport2003_02.pdf
Project Proposal



Extension to Chat Bot, that will reply emails
Will maintain a chat session in terms of email-session-id
Can be useful for auto replying to emails that have a
common reply, related to the question in the email
–
e.g. – help@cse - email to register your complaints!