Generating Natural Language from Linked Data: Unsupervised template extraction

摘要: We propose an architecture for generating natural language from Linked Data that automatically learns sentence templates and statistical document planning parallel RDF datasets text. have built a proof-of-concept system (LOD-DEF) trained on un-annotated text the Simple English Wikipedia triples DBpedia, focusing exclusively factual, non-temporal information. The goal of is to generate short descriptions, equivalent stubs, entities found in Datasets. evaluated LOD-DEF against simple generate-from-triples baseline human-generated output. In evaluation by humans, significantly outperforms two three measures: non-redundancy structure coherence.