STEPS: Semantic Typing of Event Processes with a Sequence-to-Sequence Approach

Enabling computers to comprehend the intent of human actions by processing language is one fundamental goals Natural Language Understanding. An emerging task in this context that free-form event process typing, which aims at understanding overall goal a protagonist terms an action and object, given sequence events. This was initially treated as learning-to-rank problem exploiting similarity between processes action/object textual definitions. However, approach appears be overly complex, binds output types fixed inventory for possible word definitions and, moreover, leaves space further enhancements regards performance. In paper, we advance field reformulating typing generation put forward STEPS, end-to-end producing user objects only, dispensing with need their addition this, eliminate several dataset constraints set previous works, while same time significantly outperforming them. We release data software https://github.com/SapienzaNLP/steps.