Lecture: 'A commented history of Open Information Extraction towards Multilingual approaches'

Open Information Extraction (Open IE) is the task of extracting open-domain assertions from natural language sentences. Most Open IE methods are developed for the English language. The first systems were learning-based, followed by a new wave of shallow parsers/dependency analysis and hand-crafted rules, which achieves better accuracy. Recently, new Open IE methods based on deep learning approaches have achieved state of the art performance for the English language.

Languages other than English, however, have received far less attention from the area, and few examples developed methods for languages such as Portuguese, Spanish, German, and Chinese. Considering the low availability of datasets and tools for these languages, developing data-based methods for these languages is a challenging task. Recently multilingual approaches emerge to extract facts for different languages.

In this talk, we would like to present a commented history and evolution of Open IE methods, discussing their purposes and ability to achieve a multilingual system to extract facts from natural language sentences. We provide our contributions and evolutions for the Open IE domain, positioning our methods on the Open IE timeline.