How can computers conquer the subtleties of the phoneme? Computational phonology has focused on typed feature structures and their representation formalism over sub-phonic feature structures. Here commercial researcher Neugebauer seeks to correct this oversight by examining the discrepancy between hierarchically-structured lexica in most computational phonology and the "flat-structured" feature bundles common in acoustic modeling, developing a framework for solving combinatorial problems through constraint programming. Although he considers other structures, Neugebauer primarily shows how constraint programming works in the computational treatment of paradigmatic phonological representations, covering grammar formalism and subphonetic acoustic modeling, knowledge representations in constraint-based grammar formalisms, elements of state tying and the role of decision trees, and the appropriateness of constraint programming for this application, including the influence of type inheritance hierarchies such as Galois lattices. He closes with an evaluation of his experiment and a description of future work. Annotation ©2008 Book News, Inc., Portland, OR (booknews.com)
