Britbot

Britbot (2018), by Libby Heany, is an online chatbot experience that asks the user to engage in questions about Britishness. Through its conversations it develops ideas about what being British means.
The bot was originally trained on the United Kingdom government’s citizenship test and the corresponding text-book: Life in the UK: a guide for new residents, 2017.
Britbot was commissioned in 2017 by Sky Arts as part of the Art 50 programme. Britbot was live from 2018 to 2019 learned from the people that interacted with. 
I developed the computational algorithms behind the Britbot. The system uses techniques in natural language processing for speech-to-text voice recognition and text-to-speech voice synthesis. This allows the bot to hear and speak to the users.
For dialog management, the system used a combination of rule-based and purely-generative based methods. There was a defined conversational structure, but the system was able to improvise inside of the defined constraints.
The language modeling was performed with Google TensorFlow Hub’s Universal Sentence Encoder and the response generation uses a combination of semantic matching, word-level language generation, and post-generation re-ranking.
Here are some videos describing the Britbot in action: