Chatbots

Chatbots are computer programs simulating human conversation partners. An early fictitious Chatbot is Robby the Robot who appeared first 1956 in the film Forbidden Planet. ELIZA is a real bot to Chatbotwith. She (or it?) was created between 1964 and 1966 by Joseph Weizenbaum at MIT. ELIZA is amazingly capable of maintaining small talk conversations without having domain knowledge. Since then, the range of applications of Chatbots has been extended significantly.

There are Chatbots which are assistants who can send mail and SMS: Send my wife an SMS that I'm late! Other Chatbots can lookup information: When is the next train leaving for Geneva? More difficult tasks are solved by question answering bots: How much is the square of 743 times 743? Moreover, there are bots used as instructors or trainers, e.g. to learn a foreign language. Finally, there are data acquisition bots which are aimed at gathering information from users, for instance for compliance purposes. Sophisticated data acquisition Chatbots may need quite bit expert know how to formulate the next question depending on the answers to previous questions and depending on information stored in a know how database.

There are two main advantages of Chabots with respect to conventional applications:

1. From the user’s point of view, Chatbot user interfaces are intuitive and easy to use from the very first time.
2. From the developer’s point of view, Chabot user interfaces are flexible. The interactions of the bot can be modified or even extended without changing the front-end.

The Chatbot must cope with the complexity of natural language. The bot shall understand as many as possible formulations provided by the user. This is usually achieved by

• constraining the domain,
• focusing on simple and well-defined goals,
• predicting the user’s situation to increase the probability of a simple yes answer.

In addition, the Chatbot shall …

• achieve a meaningful conversation flow by posing the questions in a proper order.
• maintain representations of the user’s states to take into account his or her context.
• limit the conversation to short and precise answers and questions which are easy to understand.
• avoid the repetition of the same answer and questions respectively.
• provide a way out in cases where the user is lost.
• switch in the right moment from automatic to human interaction.

Javascript version of ELIZA was originally written by Michal Wallace and significantly enhanced by George Dunlop.