Natural language based knowledge representation
September 16 - 18, 2015, We/Th/Fr
Organizers: Ido Dagan and Sebastian Riedel
Original Topic Description: NLKR might be thought of as an extension of the Open-IE paradigm. In "formal" knowledge representation schemes, like those in semantic web ontology or FreeBase style, the vocabulary of relations, and to a certain extent of concepts, is predefined. In Open IE, on the other hand, you represent propositions based on the NL vocabulary found in the text. The problem of Open IE is that the represented knowledge isn't canonicalized - you can have the same proposition appearing in different ways. The idea of NLKR is to consolidate NL-based propositions via entailment and other semantic relationships between propositions, e.g. causality and temporal, that will put a useful structure over these propositions (kind of a knowledge graph). Integrating NLKR and formal-language based KR (NLKR for NLP): If we succeed to represent large amounts of knowledge in a useful Open-IE based rich structure this has the potential to improve NLP tasks, particularly where semantic inference is needed. In relationship to MIC, this direction can be viewed as interpreting meaning in a global context of broadly acquired knowledge, in analogy to how people interpret meanings in the context of their global knowledge, not just in the context of the current discourse.
Example sentences for consolidation