Event-centered semantics: Using clustering algorithms to group events that typically occur together
Travis Nguyen, Jeane Wenzel · Texas ScholarWorks (Texas Digital Library) · 2014
There is evidence that in processing sentences, people use their real-world knowledge about how events unfold, including what participants an event typically has, and which events follow other events. Our research takes event-based approaches to problems in computational semantics, or automatic natural language understanding, that traditionally use single-word-based approaches, such as determining word similarity. In our research, the first task is to determine the similarity of events computationally. To do so, we process large bodies of text to extract event mentions and context information. Then, our team members convert the events to vectors whose components indicate their similarity to other events. Using this vector space, we can cluster the events, where events in the same cluster may indicate events that occur together in a real-world situation. Our research has several applications, including determining similarity of words or events and event paraphrasing.