Augmented Transition Network
Representation for grammars developed from simple finite state transition networks by allowing (a) recursion and (b) augmentation, i.e. the use of arbitrary tests and actions on arcs, giving full Turing machine power. The use of registers for storing constituents, and the use of tests and actions on register contents allow great flexibility in parsing, and in particular permit the construction of sentence representations quite distinct from the surface text e.g. deep as opposed to surface syntactic structures. The form of grammar representation is procedurally oriented, but the grammar itself is separated from the interpretive parser, which is top-down <248> and usually depth-first <55>. ATNs are a popular formalism and can be adapted e.g. to guide parsing by explicit arc ordering. Problems arise with e.g. passing information between subnets, and the treatment of conjunctions.
- [Woods 70]Woods, W. A. Transition network grammars for natural language analysis. Communications of the ACM 13:591–606, 1970.Google Scholar