A grammatical inference system based on case grammar and semo-syntactic features
Howard Reed Smith · 1982
SGRD is a system which infers grammatical relations from sentences and their corresponding deep case structures without the use of illegal constructs (near miss inputs) to assist it. These deep case structures include both syntactic and semantic features and relations. The three major operations performed by SGRD are: grouping--how to combine adjacent constituents; folding--when to combine two or more distinct word classes into one word class; and merging--when to combine two or more grammar rules into a single grammar rule. These operations imply that overgeneralization of a grammar rule will not be followed by further specification. SGRD is applied to a sample corpus and infers a grammar which results in a minimum amount of ambiguity when it is used for parsing. The growth of the number of grammar rules as sentences are input to the system resembles a logarithmic growth curve. Isomorphic grammars result from the inference algorithm, regardless of the ordering of the input sentences. This hypothesis is proven to follow from the commutativity of the folding and merging operations.