Tagging words according to their syntactic categories represents an important problem in natural language processing with several applications to speech synthesis, speech recognition, character recognition, automatic translation, etc. Due to the presence of homographs any word-labeling algorithm must incorporate some mechanism for coping with ambiguity. A number of methods have been proposed for the disambiguation task, including connectionist models, dynamic programming techniques, Markov models. In this paper we introduce a novel approach based on probabilistic relaxation labeling, an iterative method quite popular in the area of image processing and pattern recognition. Experiments are described and preliminary results are given for the Italian language. Keywords: natural language processing, probabilistic relaxation, syntactic category disambiguation, speech processing.
Bibliographic reference. Pelillo, Marcello / Refice, Mario (1991): "Syntactic category disambiguation through relaxation processes", In EUROSPEECH-1991, 757-760.