Incremental Connectionist Architectures for Language Learning

This report describes techniques for designing hybrid connectionist architectures in real-world language environments. We argue that for dealing with arbitrary real-world corpora we can identify the underlying constraints for a task but the integration and interaction of these constraints cannot be predicted in general for arbitrary unrestricted language bases. Therefore flexible automatic learning and incremental parallel integration of various constraints are particularly important for scaling up in real-world environments. As two particular examples we will focus on the development of connectionist incremental architectures for flat syntactic analysis and flat semantic analysis to support flexible automatic learning and parallel constraint integration. The results from the described architectures are used as the basis for building a more general fault-tolerant learning system for semantic and syntactic speech parsing.

@TechReport{Wer94a,
 	 author =  {Wermter, Stefan},
 	 title = {Incremental Connectionist Architectures for Language Learning},
 	 booktitle = {None},
 	 journal = {None},
 	 editors = {None},
 	 number = {VM-Memo 7},
 	 volume = {None},
 	 pages = {None},
 	 year = {1994},
 	 month = {Jan},
 	 publisher = {None},
 	 doi = {}
 	 url = {}
 }