Q&A: Dr. Michael McTear on Building a Conversational Chatbot for Google Assistant Using Dialogflow
Dr. Michael McTear, Professor, Ulster University, recently answered the following questions about building conversational chatbots:
Q: Tell us about the three hour workshop you will present on April 26 at the SpeechTEK Conference in Washington, DC.
A: This practical hands-on workshop provides an in-depth introduction to the design and development of conversational chatbots using Google’s Dialogflow platform. We start with a brief overview of chatbots and conversational interfaces, looking at relevant technologies and tools. The main part of the workshop is devoted to hands-on design and development of several sample conversational chatbots. Topics to be covered include: basic one-shot queries; slot-filling dialogs; dealing with context and follow-up queries; testing your chatbot on the Google Assistant Simulator; training the chatbot to improve its performance; enhancing system output using SSML (Speech Synthesis Markup Language); using fulfillment to interact with external web services; how to display on surfaces that have visual and well as speech interfaces. Attendees will be provided with instructions for downloading software to their PC laptops prior to this workshop.
Q: How will this workshop be beneficial to developers not using Dialogflow?
A: Many of the techniques that will be described, such as intent recognition, entity extraction, and slot filling, are common to other popular platforms so the general concepts will be re-usable. Other common aspects to be covered include design, training, and fulfillment.
Q: What is context and how can designers leverage context to design better voice user interfaces?
A: In conversational chatbots, context refers to the dialog context. This is important, for example, in follow-up questions where using context enables the system to understand how a follow-up question relates to the preceding dialog. For example: following a question such as "What's the weather in Washington?", a follow-up question might be "How about Boston?" The system needs to know that the follow-up question is asking about the weather in Boston and not asking a general question about the city of Boston.
Q: How many example utterances are necessary to train the system to respond satisfactory to user requests?
A: Usually about 15 training utterances are sufficient but it is also possible to examine the system logs and add further examples and make changes if the system has misunderstood the user's requests.
Q: What skills are necessary to effectively use the Speech Synthesis Markup Language to produce appropriate sounding pronunciations?
A: This is to some extent a matter of trial and error. In Dialogflow there is a tool that allows the developer to experiment with the markup and listen to the output to determine if it is appropriate.
Q: What are the key learnings attendees will take away from this workshop?
A: Key learnings will be:
- The main concepts and components of a conversational chatbot.
- How to use the Dialogflow platform to implement and test a conversational chatbot.
- How to use fulfillment to interact with external web services.
- How to display on surfaces that have visual as well as speech interfaces.
Q: How can the accuracy of a chatbot be measured and quantified?
A: The accuracy of the chatbot's understanding can be measured using an objective measure of accuracy - i.e. was the correct intent matched. The chatbot's responses can be measured using a scale of how appropriate the responses were. The overall performance of the chatbot is usually measured using subjective metrics such as ease of use.
Q: Can (or should) developers mix and match built-in dialog tools and custom dialog code in Google Assistant?
A: The DialogFlow platform provides a range of built-in tools that enable novice developers to create a conversational chatbot quickly and effectively. As requirements become more complex, it will be necessary to supplement the built-in tools with the developer's custom code.
Register for the SpeechTEK Conference and Dr. McTear’s workshop. There are still openings for SpeechTEK University workshops and presentations. Submit proposals by October 11, 2002.
Conversational chatbots may be the future (or even the present) but Bruce Balentine is taking developers back to basics in his workshop at SpeechTEK in April. Learn how to build chatbots that get results.
Poorly designed IVRS have been angering customers for decades, but it doesn't have to be this way. We talked to the Founder and CEO of UJET about how well designed IVRs can improve customer experience.