Chatbots and dialog systems are currently experiencing a great deal of hype in business and everyday life. Speech-based assistants are now used in smartphones, cars, and smart speakers such as Alexa and others. Speech-based interfaces are the big topic of the future as they make operating computers and smartphones significantly easier than a keyboard, mouse, and touchscreen. Similarly, they enable immense improvements in call centers at lower costs; chatbots prevent long waiting times and save call center operators a great deal of work while also being available constantly.
In research, such systems have been an important topic for much longer. All major scientific conferences are currently recording an immense increase in publications on dialog systems. The German Research Center for Artificial Intelligence has been researching chatbots since its foundation over 30 years ago. The Cognitive Assistants research department explores aspects of human-machine interaction, while the Speech Technology & Multilingualism and Speech and Language Technologies research departments investigate the underlying algorithms. Applications of chatbots can be found across all research departments at DFKI, for example in Educational Technologies or the Robotic Innovation Center. We also work closely with the Quality and Usability Lab at the TU Berlin, which also has a strong focus on dialog systems.
Systems commonly known as chatbots in colloquial language are usually referred to as dialog systems in research. The term dialog system refers to software with which users interact in natural language, i.e., written or spoken language. Dialog systems must therefore both be able to understand the user and respond with a speech output.
The research department Speech and Language Technologies has a special research focus on dialog systems that are composed of several dialog systems. The reasons for such a combination are manifold: Amongst others, this allows to combine dialog systems that consist of different technologies that would otherwise be incompatible. For example, one partial dialog system can handle restaurant bookings (technical term: Task-driven Dialog), another partial system can answer questions about the menu (technical term: Question Answering), while a third system is particularly well suited for small talk (technical term: Open Domain Dialog). Further possible applications for this combination can be found in an improved language understanding of the chatbot (technical term: Natural Language Understanding) or in the application when existing chatbots are to be combined in a meta-chatbot.
This video of the ZDF science program PUR+ asks: "Can computers think?". To this end, we conduct an experiment with students from the Emanuel Lasker School in Berlin.
In this video, we show a dialog system that we developed at DFKI for use at a Deutsche Bahn trade fair:
Relevant keywords are:
Text-based dialog systems
Modular chatbots that are composed of multiple chatbots.
Multitask bots that combine the different subfields of the dialog systems topic, such as task-driven dialog, question answering, and chit-chat.
Frameworks for dialog systems
Natural Language Understanding
Evaluation of dialog systems
Practical work as a consultant
Dialogs & Chat projects at the Speech and Language Technology Lab:
Various industry projects