Flexible and Controllable Dialogue Systems
– CONVERSATIONAL AI –
Talkamatic develops and builds dialogue systems, also known as Conversational AI. The starting point for Talkamatic is always human dialogue, and everything we do is based on an analysis of human dialogue with the goal of reproducing optimal human dialogue strategies in our technology.
The only Dialogue Tool You’ll Ever Need
The following key points summarize the core aspects of Talkamatic’s dialogue system and its advantages in conjunction with Large Language Models (LLMs).
Integration with LLMs
Talkamatic develops and builds dialogue systems. The starting point for Talkamatic is always the human dialogue, and everything that is done is based on an analysis of human dialogue and a reproduction of human dialogue strategies in the system.
The Talkamatic dialogue technology is a general-purpose dialogue system based on research from Gothenburg University. It is completely controllable and predictable, yet flexible and able to deal with the complexities of natural dialogue. It allows for detailed design and curation of dialogue interaction. This makes it an ideal complement to Large Language Models (LLMs) such as GPT, which are very versatile but interactions can sometimes be hard to design, curate, control and predict.
Using Talkamatic tech, an LLM can be used to take some content and generate, automatically and without any need for “prompt design”, dialogue “building blocks” which are then used by the Talkamatic system at runtime to provide a flexible, high quality curated dialogue.
This approach provides the generative power of LLMs while simultaneously offering complete control over the dialogue, including the possibility of human curation and design. It also enables minimizing or even eliminating LLM calls and costs at runtime, thus opening up for running “on premises” (without cloud access). Another advantage is that system response times are significantly reduced compared to communicating with an LLM directly, which is important especially for natural spoken dialogue.
Talkamatic have also devised ways of running an LLM in parallel with the Talkamatic system, to take advantage of the power of LLMs to process natural language while keeping control of the flow of the dialogue.
The dialogue engine TDM (Talkamatic Dialogue Manager) controls the dialogue and ensures that the user’s input is handled as the designer has planned, without the user being limited to doing things in a certain predetermined order. For example, you can specify a number of different conversation topics, but the user can freely switch between them. In other words, TDM provides a controlled but at the same time flexible dialogue. Thanks to this flexibility and versatility of TDM, it is able to use LLM-generated dialogue “building blocks” to provide high-quality dialogue without the need to call an LLM at runtime.
Another aspect of TDM is that it handles spoken dialogue in real time, including making sure that system and user have heard and understood each other, and also handling the delays that can sometimes occur when using GPT and similar models. TDM also supports graphical output and haptic input to complement the linguistic dialogue. TDM also supports written dialogue.
The technology can be used in many verticals, and Talkamatic have recently been focusing on EdTech, where the company also gained some customer traction. Starting from a text or problem description, the tech can automatically generate an Educational dialogue structure which can be curated and verified by a human, e.g., a teacher, before students interact with it and have engaging and pedagogically effective interactions using spoken dialogue. This solution is applicable to all areas of education, including math and science.
The tech is also useful outside of education. For example, it can be used to provide installation instructions for technical equipment, Instructional Dialogues, in the form of a spoken dialogue where the user can request more detailed instructions when needed, or ask about concepts they are unfamiliar with.
Educational dialogue and Instructional dialog can also be combined into an “apprentice dialog” where the user receives help from a teacher who instructs, gives help and asks questions while the apprentice performs a practical task.
Basically, any kind of instructional dialogue can be handled by the tech, with the same benefits of control and curation as in the EdTech vertical. And there are several other kinds of dialogues that can be created from the tech such as;
Q&A Dialogue – Questions and Answers about a specific text or content
Database search Dialogue – When the system needs to ask the user a number of questions in order to answer a user question or perform an action requested by the user
Negotiation Dialogue – A user can get help from the system to make a decision by comparing different solutions to a problem and weighing advantages against disadvantages.
In addition to the technology itself, the Talkamatic team has deep expertise and understanding in conversational AI, including dialogue, speech technology, LLMs, such as GPT, and more. The company also has a close connection to research values with three doctoral team members and a professor of computer linguistics as co-founder.