A Canonical Context-Preserving Representation for Open IE: Extracting Semantically Typed Relational Tuples from Complex Sentences

Modern systems that deal with inference in texts need automatized methods to extract meaning representations (MRs) from at scale. Open Information Extraction (IE) is a prominent way of extracting all potential relations given text comprehensive manner. Previous work this area has mainly focused on the extraction isolated relational tuples. Ignoring cohesive nature where important contextual information spread across clauses or sentences, state-of-the-art IE approaches are thus prone generating loose arrangement tuples lack expressiveness needed infer true complex assertions. To overcome limitation, we present method allows existing enrich their output additional meta information. By leveraging semantic hierarchy minimal propositions generated by discourse-aware Text Simplification (TS) approach presented Niklaus et al. (2019), propose mechanism semantically typed source sentences. Based novel type output, introduce lightweight representation for form normalized and context-preserving It extends shallow predicate-argument structures capturing intra-sentential rhetorical hierarchical relationships between In way, context extracted preserved, resulting more informative coherent which easier interpret. addition, comparative analysis, show benefits second dimension: canonical structure simplified sentences process analyze, facilitates tuples, an improved precision (up 32%) recall 30%) large benchmark corpus.