Article Preview
TopIntroduction
The relevance of this paper is in the lack of elaboration of a comprehensive solution to the problems in artificial intelligence (AI) systems, a low degree of integration of the results obtained in various areas of research in the AI sphere. The complexity of the integration is related to the heterogeneity of methods and the terminology that is used in a variety of ways. This work is devoted to the complex solution of the problems of natural language processing (NL), automatic problem solving based on applied ontology and interactive visualization. The emphasis in the paper is made on the linguistic processor because eventually, it is particularly the NL that describes a complicated problem in the most understandable form. Of course, full integration involves a more complete account of the sources of information and the output of the results obtained by the system (analysis and synthesis of images, audio and video sources), but the complexity of the problem requires a narrowing of the research area.
The first integrated systems appeared in the early stages of AI research. In the traditional works by Vinograd and Shank (Winograd, 1972; Schenk et al., 1975; Seitova et al., 2018). The first one demonstrated the understanding of NL limited by the world of simple bodies (cubes, prisms, pyramids). The program had a graphical interface that displays the results of the interpretation of the NL user’s request. In the second system, after the translation of the text into the internal representation, it started the process of reasoning and synthesized the NL-description of the result. The systems demonstrated both the great potential of integrated systems and significant difficulties in its implementation.
With the emergence of advanced systems of knowledge representation, ontologies, and advances in computational linguistics, it was expected that integrated systems would receive a new powerful impetus for its development. However, while these expectations are not met even with significant restrictions on the input language, the systems are fragile and difficult to adapt to the new subject areas. Nevertheless, in the application aspect, some results were achieved, namely, in the implementation of the NL interface to relational databases.
In the last two decades, the center of gravity of its processing has shifted towards the analysis of large amounts of text information from the global network and the decision-making algorithms based on cognitive methods. This is due to the constant growth of information on the Internet, which requires rapid analytical processing both in public institutions and in large commercial structures. The review (PullEnti-SDK extract named entities…, 2019) provides a list with brief characteristics of about 40 systems of linguistic analysis of unstructured texts. However, it remains of interest to the less ambitious but more advanced in terms of ideological systems, in particular, to systems and question/answer. In recent years, the following can be noted from the relevant publications (Zhang et al., 2018; Gan and Yu, 2018; Seo et al., 2015; Shi et al., 2015; Sundaram and Khemani, 2015). The overview of the systems is provided in (Morton and Qu, 2013). Their features in comparison with the described integrated system are briefly discussed in the related works section.