Language models: new methods for conversational agents

Conversational agents require a series of reading capabilities. They need to be able to identify references to persons, organisations, events and concepts – comparable to automatically creating Wikipedia links. Despite the fundamental ambiguity and context sensitivity of our language, they must be able to understand who or what is being referred to. The facility to recognise and extract the content of statements is also essential. Recent advances in deep learning provide ways of understanding the meaning of sentences independently of the choice of words and sentence structure. The ultimate aim is to develop suitable discourse models which, in combination with situational understanding, permit effective interaction with humans.

Numerous different algorithms help us to procure and evaluate information, including in the context of search engines or social media. The same applies in the areas of entertainment or e-commerce. Intelligent, conversational systems to which we can direct spoken queries in natural language are being used to an increasing extent. The economic and social consequences of these new technologies are enormous.

Conversational software must be able to understand text and language automatically. This means more than just handling spoken queries. A deeper understanding of the documents and texts containing the world’s knowledge is also called for. The aim of this project was to develop new methods of text comprehension: What is the subject matter and what is being said? What could be relevant or interesting to a user? These are questions that demand an algorithmic solution.

This research project intended to develop technologies and methods to be integrated directly into dialogue systems. Such systems are already widely used in smartphones and on the Internet, and their significance will increase as the quality of interaction with humans improves. Whether at home, at work or on the go: intelligent agents will replace today’s interfaces to the digital world of knowledge and entertainment.

The project achieved results in four areas that are central and foundational to the rapidly developing areas of natural language processing and conversational agents.

First, in entity detection and linking, the project contributed a novel entity-linking system that combined advanced entity embeddings, a neural attention mechanism over local context windows, and differentiable joint disambiguation inferencing. Notably, the system combines entity detection and linking. In addition, work in this area has spurred innovative follow-on work.

Second, in the area of language generation models that are foundational to conversational systems, results were obtained that address different limitations caused by biases and teacher forcing when training unconditional language models.

Third, in relation to the use of deep neural networks for generative models, the project achieved advances in learning algorithms for, and evaluation of, generative adversarial networks (GAN). It was hoped that GANs could be used for text production and conversational exchanges, which remains a challenge.

Fourth, the project achieved results related to the design of reinforcement learning agents in the setting of text-based games. Focus was on how to contend with the compositional and combinatorial nature of language that make it hard to optimise policies, and an agent was designed that was able to perform well across a family of games, rather than in only a single game.

Overall, the project made influential contributions to machine learning methodology, most notably in the areas of geometric embeddings and generative models. The results are documented in a dozen papers that encompass already highly cited papers in top conferences in Machine Learning or AI, such as NeurIPS, ICML, and AISTATS.

Conversational Agent for Interactive Access to Information