Inducing rich interaction structures between words for document-level event argument extraction

An apparatus for natural language processing, comprising:

• a document encoder configured to generate a first word representation vector for an event trigger word of a document corresponding to an event and a second word representation vector for an argument candidate word of the document, wherein the document comprising a plurality of words is organized into a plurality of sentences;
• a structure component configured to generate a plurality of document structures including a semantic structure for the document based on the first word representation vector and the second word representation vector, a syntax structure based on a sentence-level dependency relationship, and a discourse structure representing document-level discourse information based on the plurality of sentences;
• a relationship encoder configured to generate a relationship representation vector based on the plurality of document structures;
• and
• a decoder comprising a graph-based neural network configured to predict a relationship between the event trigger word and the argument candidate word based on the relationship representation vector.

A method for training a neural network, comprising:

• receiving training data including documents comprising a plurality of words organized into a plurality of sentences, the words comprising an event trigger word and argument words, and the training data further including ground truth relationships between the event trigger word and the argument words;
• generating word representation vectors for the words;
• generating a plurality of document structures including a semantic structure based on the word representation vectors, a syntax structure representing dependency relationships between the words, and a discourse structure representing discourse information based on the plurality of sentences;
• generating relationship representation vectors based on the document structures;
• predicting relationships between the event trigger word and the argument words based on the relationship representation vectors;
• computing a loss function by comparing the predicted relationships to the ground truth relationships;
• and
• updating parameters of a neural network based on the loss function.