Proposal for PORQUE, a Polylingual Hybrid Question Answering System

Organizations can benefit from integrating multilingual information from both textual and structured sources, and from its retrieval by means of Question Answering (QA) systems. Hybrid QA approaches, capable of finding answers in both documents and KGs, usually rely on translating textual sources into KG statements or vice-versa, and are often not leveraging the whole extent of a graph or the richness of the natural language text. Here we propose PORQUE, a hybrid QA system that utilizes multilingual language models, graph embeddings and modern decoder models to generate answers in many languages based on information contained in multilingual textual corpora and multilingual KGs. Of novelty is the hybrid representation of information which, guided by existing work in KG-augmented NLP, allows a more complete exploitation of both KG and documents.