Events provide the core semantic content of most texts. Event coreference is a key requirement for many NLP applications, especially those that involve reasoning about the context like Topic Detection and Tracking, Question Answer, Information Retrieval etc. Consider the following example

There was boating, crazy adventure sports, and pro-golfing. We also spent the evenings strolling around the park. We were happy but tired at the end of every day. This vacation was a lot of fun and I highly recommend it.

In this example, we come across the event mention ‘‘This vacation’’, and our task is to resolve this and other such event mentions to their corresponding events.

Current approaches for Event Coreference are limited to identifying and resolving identical event mentions that refer to exactly the same event. In this effort, we address the problem of quasi (partially) identical events. In a quasi-identical event coreference the event mention (trigger) encompasses multiple other non-contiguous events in the context.

Traditional single-event coreference techniques aren’t well suited to address this, and we propose a new approach for the same. I would very soon add more details about how we tackle this.