Improving Open Information Extraction for Semantic Web Tasks

Open Information Extraction (OIE) aims to automatically identify all the possible assertions within a sentence. Results of this task are usually a set of triples (subject, predicate, object). In this paper, we first present what OIE is and how it can be improved when we work in a given domain of knowledge. Using a corpus made up of sentences in building engineering construction, we obtain an improvement of more than 18%. Next, we show how OIE can be used at a base of a highlevel semantic web task. Here we have applied OIE on formalisation of natural language definitions. We test this formalisation task on a corpus of sentences defining concepts found in the pizza ontology. At this stage, 70.27% of our 37 sentences-corpus are fully rewritten in OWL DL.