Neuroinformatics and Semantic Representations: Theory and Applications
Abstract. Over the past decade, a new wave of interest in dialogue agents has been observed. This is largely due to the introduction of machine learning in the tasks of automatic natural language processing. Using the tools of distributional and network semantics makes it possible to summarize data from huge corpora of texts. New language models trained on huge corpora can significantly simplify further training of models for new tasks (transfer learning), and sometimes completely avoid further raining (zero-shot learning). The paper considers both well-established neural network architectures and promising approaches to the use of neural networks in the tasks of automatic processing of information at all levels of the language as a whole, and for building dialogue agents in particular. The use of a modular approach to solving these problems and the main types of modules are described.