Providing Automatic Multilingual Text Generation to Artificial Cognitive Systems

Abstract

This contribution addresses the incorporation of a module for advanced user interaction into an artificial cognitive vision system to include the human-in-the-loop. Specifically, the document describes a method to automatically generate natural language textual descriptions of meaningful events and behaviors, in a controlled scenario. One of the goals of the system is to be capable of producing these descriptions in multiple languages. We will introduce some relevant stages of the whole system, and concentrate on the linguistic aspects which have been taken into account to derive final text from conceptual predicates. Some experimental results are provided for the description of simple and complex behaviors of pedestrians in an intercity crosswalk, for Catalan, English, Italian, and Spanish.