The increasing amount of machine-readable data available in the context of the Semantic Web creates a need for methods that transform such data into human-comprehensible text. In this paper we develop and evaluate a Natural Language Generation (NLG) system that converts RDF data into natural language text based on an ontology and an associated ontology lexicon. While it follows a classical NLG pipeline, it diverges from most current NLG systems in that it exploits an ontology lexicon in order to capture context-specific lexicalisations of ontology concepts, and combines the use of such a lexicon with the choice of lexical items and syntactic structures based on statistical information extracted from a domain-specific corpus. We apply the developed approach to the cooking domain, providing both an ontology and an ontology lexicon in lemon format. Finally, we evaluate fluency and adequacy of the generated recipes with respect to two target audiences: cooking novices and advanced cooks.
展开▼
