Seminaria w Instytucie Podstaw Informatyki PAN

Streszczenie (autorskie):

During the presentation, we will outline the objectives of the Universal Discourse project, which aims to create a universal, multilingual model for describing discursive relations. The theoretical basis for the work is the ISO 24617-8 standard, which serves as a starting point for the harmonisation of existing corpus resources. In the first part, we will discuss the issue of text segmentation into discursive units. We will present a comparative analysis of various formalisms (such as RST or PDTB) and our own rule-based heuristic, which demonstrates high consistency in determining the boundaries of units at the constituent sentence level. We will then present the proposed multi-layered model of relations. We will focus on a proprietary decision tree that allows for the systematic classification of semantic links (including causal, conditional and temporal links). Finally, we will present the current status of work on the development of discourse parsers and the results of our first experiments.