Parallel Parsing of Spoken Language

Randall A. Helzerman, Mary P. Harper, and Carla B. Zoltowski School of Electrical Engineering Purdue University West Lafayette, IN 47907 Abstract We have extended Maruyama's [4, 2, 3] constraint dependency grammar (CDG) to process a lattice of sentence hypotheses instead of separate text strings. A post-processor to a speech recognizer producing N-best hypotheses generates the word lattice representation, which is then augmented with information required for parsing. We will rst summarize the CDG parsing algorithm and then describe how the algorithm is extended to process the lattice on a single processor machine. Finally, we outline the CRCW P-RAM algorithm for parsing the word lattice, which requires O(n4) processors to parse in O(k + n) time. 1 Constraint Dependency Grammars To develop a syntactic analysis for a sentence using CDG, a constraint network (CN) of word nodes is constructed. Associated with each node is its position and a set of roles, which indicate the various functions the word lls in a sentence. Though two roles are required to write a grammar at least as expressive as a context-free grammar [2], examples depict a single role, governor, which represents the sentence function a word lls when governed by its head. Each role is initially assigned all role values allowed by the word's lexical category, where a role value consists of a label (the function the word can serve, e.g., SUBJ) and a modi ee (the number corresponding to the position of the word which it modi es, or nil). There are p q n = O(n) possible role values (where p, the number of roles per word, and q, the number of di erent labels, are grammatical constants and n is the number of modi ees or words) for each of the n words in the sentence, giving O(n2) role values altogether, which require O(n2) time to generate. Figure 1 shows the initialization of the role values for the sentence A sh eats.