Learning to Classify Natural Language Titles in a Recurrent Connectionist Model

This paper describes a recurrent connectionist model which learns to classify book titles from a library. This task poses several difficult constraints to the recurrent network: learning sequences of words, detecting the context of preceding words, assigning a class, and dealing with variable lenght, syntax, and semantics of available titles. We describe our underlying word representation, the connectionist model, and experiments with titles from an online library classification. The model learned to classify almost perfectly in comparison with the existing library classification. This research shows that a recurrent connectionist model can learn the necessary knowledge for scaling up to "real-world" title classifications in natural language processing.

@InProceedings{Wer91, 
 	 author =  {Wermter, Stefan},  
 	 title = {Learning to Classify Natural Language Titles in a Recurrent Connectionist Model}, 
 	 booktitle = {Proceedings of the International Conference on Artificial Neural Networks},
 	 editors = {Mäkisara, K.; Simula, O.; Kangas, J.; Kohonen, T.},
 	 number = {},
 	 volume = {2},
 	 pages = {1715--1718},
 	 year = {1991},
 	 month = {},
 	 publisher = {Elsevier},
 	 doi = {10.1016/B978-0-444-89178-5.50171-8}, 
 }