Italian Event Detection Goes Deep Learning

  • 2018-10-04 14:09:20
  • Tommaso Caselli
  • 14

Abstract

This paper reports on a set of experiments with different word embeddings toinitialize a state-of-the-art Bi-LSTM-CRF network for event detection andclassification in Italian, following the EVENTI evaluation exercise. The net-work obtains a new state-of-the-art result by improving the F1 score fordetection of 1.3 points, and of 6.5 points for classification, by using asingle step approach. The results also provide further evidence that embeddingshave a major impact on the performance of such architectures.

 

Introduction (beta)

None

 

Conclusion (beta)

None