Augmented Transition Networks (1970) › 1970: Augmented Transition Networks
History of Language AIAll chapters Augmented Transition Networks (1970) › 1970: Augmented Transition NetworksOne section · Sections in this chapter1 / 7 ↓
Reading tools Definitions & inline links
Choose your expertise level to adjust how many terms are explained. Beginners see more tooltips, experts see fewer to maintain reading flow. Hover over underlined terms for instant definitions.
Article links
Make inline references clickable
By the late 1960s, natural-language parsers had to account for ambiguity, optional constituents, and nested phrases. A finite-state network could encode a sequence of categories, but it had no stack for an embedded clause and no convenient place to retain facts discovered earlier in a parse.
William Woods at Bolt Beranek and Newman (BBN) developed transition-network grammar as a programming notation for this problem. A recursive transition network could call a network for a noun phrase or clause and then return to its previous state. Recursion handled nesting, but a useful parser also needed to inspect linguistic features and construct an output as it moved through the graph.
Woods's 1970 Augmented Transition Network (ATN) added that procedural machinery. Registers stored information about a partial analysis. Conditions on an arc decided whether it could be traversed, and actions updated registers or built structure. The grammar writer could therefore mix category recognition with agreement checks and semantic interpretation.
ATNs were widely used in natural-language systems during the 1970s. Woods and colleagues used the formalism in LUNAR, which answered questions about Apollo lunar samples. Later ATN variants supported analysis and generation in other research systems. Their history provides a clear example of grammar expressed as an executable procedure rather than only as a set of declarative rules.