Development of the Chatterbot

Introduction

The aim of our group project was to design and prototype a Chatterbot. A Chatterbot converses in natural language (in this case, English) and responds to input from the user, with a grammatically correct sentence.

Having carried out extensive research into this particular area of AI, it soon became obvious that we would have to narrow our scope, due to the time constraint. Many of the existing Chatterbots that have competed in the Turing prize have spent more than four years being developed.

For this reason, we decided to develop a program that showed how the Chatterbot�s �brain� functions by parsing a few syntactically allowable sentences. It will hopefully show that the program is truly extensible, allowing every syntactically allowable sentence to be added.

�

How the Chatterbot was Designed

The Chatterbot will be implemented as a C++ program with several different modules for dealing with the various types of input from the user.

�

• Kbase.dat � This is the knowledge base module. If a keyword is found in the input sentence, the response will be randomly selected from a choice of responses for that particular keyword. This allows the Chatterbot to talk on a subject with precision. This is very similar to the way a human may parse common phrases such as, "Hello" or "What is your name?" Their response will nearly always be instinctive, requiring little or no grammar parsing. This is the reason why when asked the question "Do you know the time?" most people would reply with "Yes it�s ten past twelve" as opposed to "Yes thank you". The knowledge base is extensible.

�

• Kbase.tpl � This contains the templates for the generic answers based on the user�s question. The input sentence is parsed and the response is placed within the template.

�

As the Chatterbot is designed to converse on a particular subject of our choice, we decided that the input sentence should be matched with the keyword database first. The keywords are given the highest priority. Common phrases such as "Hello" and "What is your name" are also stored here.

This is known as a "slot and frame" parser. However, it is quite sophisticated. The input string is first "cleaned". This strips the sentence of any punctuation to allow the user to enter question marks, exclamation marks etc. The sentence is converted to lowercase by the function tolower(). This again allows the user to input a sentence using upper or lower case, or both. The number of keywords is quite small, but can be extended to the extent of the computer�s available memory.

The parser is flexible. As a result, phrases such as "I like Al Pacino" and "Tell me about Al Pacino" generate a sensible response.

It is important to note, the slot and frame parser does not allow the program to understand what is being said. Rather it simply makes automated responses to keywords. The sequence for matching a keyword is as follows: