CINECA IRIS Institutional Research Information System

IRIS è la soluzione IT che facilita la raccolta e la gestione dei dati relativi alle attività e ai prodotti della ricerca. Fornisce a ricercatori, amministratori e valutatori gli strumenti per monitorare i risultati della ricerca, aumentarne la visibilità e allocare in modo efficace le risorse disponibili.

Syntactic parsing is a key component of natural language understanding and, traditionally, has a symbolic output. Recently, a new approach for predicting syntactic structures from sentences has emerged: directly producing small and expressive vectors that embed in syntactic structures. In this approach, parsing produces distributed representations. In this paper, we advance the frontier of these novel predictors by using the learning capabilities of neural networks. We propose two approaches for predicting the embedded syntactic structures. The first approach is based on a multi-layer perceptron to learn how to map vectors representing sentences into embedded syntactic structures. The second approach exploits recurrent neural networks with long short-term memory (LSTM-RNN-DRP) to directly map sentences to these embedded structures. We show that both approaches successfully exploit word information to learn syntactic predictors and achieve a significant performance advantage over previous methods. Results on the Penn Treebank corpus are promising. With the LSTM-RNN-DRP, we improve the previous state-of-the-art method by 8.68%.

Senay, G., Zanzotto, F.M., Ferrone, L., & Rigazio, L. (2015). Predicting Embedded Syntactic Structures from Natural Language Sentences with Neural Network Approaches. In Proceedings of the NIPS Workshop on cognitive computation: integrating neural and symbolic approaches co-located with the 29th Annual conference on neural information processing systems (NIPS 2015). CEUR-WS.

Powered by IRIS-about IRIS-Utilizzo dei cookie

 Copyright © 2021