Discourse referent prediction is a task where the next noun phrase referent is predicted, given a context of entities and events. In order to model this intuition that humans have in predicting this future content, the script format is used.
A script is "a standardized sequence of events that describes some stereotypical human activity such as going to a restaurant or visiting a doctor" (Barr and Feigenbaum, 1981).
This model was trained on the InScript Corpus , which contains 1,000 total texts based on 10 different scenarios, including:
To perform this task, the generative RNN described in Dynamic Entity Representations in Neural Language Models was trained. The model augments a vanilla RNN by constructing dynamic entity representations that get updated as the text progresses.
To use this demo, either select an example script from the InScript corpus test set below, or write your own script in the text box.
The script should be written in the first-person, and should conclude one word before the next entity referent.
Then, hit 'submit' and see the predictions from the coreference and NP referent EntityNLM models.
Check out the code for the API and model training here.
Waking up API on Heroku...