ABSTRACT
The GPARS system is a multilingual suite of computer-assisted language teaching programs built around Generalized Transition Network parsers. GPARS tried to optimize backtracking speed through maximal use of TLC Lisp's native-code control stack and binding operators. The GPARS project adopted the augmented transition network (ATN) formalism because, among other considerations, Grossberg had demonstrated that language learning must be a case of massively parallel pattern learning. Although any serial computer parsing system is ultimately rule based, the ATN formalism seemed especially well suited to analyzing grammatical development as patterns of learned behavior. Accordingly, a series of ATN parsers was developed. The ATN parse analysis used in the English version of GPARS (ENGPARS) defines a parse path through the parse rules, or grammar, for each sentence. The parse paths are annotated to represent data about the frequency of usage of noun phrase kernel constructions found in essays in the Artificial Intelligence to Teach English to Deaf People (AITED) project.
