ABSTRACT
Formalism used in computational linguistics1 for analyzing (and generating) sentences, which was developed around 1970 as an alternative model to transformational grammar that could be easily implemented on computers. Instead of phrase structure rules (PS rules), augmented transition network grammar uses an equivalent set of finite state automata ( finite state automaton, formal language) that are called up recursively. Corresponding to the expansions of PS rules are permissible transitions between automata states; the working of transformations (e.g. in word order, congruence, active-passive-converse, control, etc.) is modeled by checking and modifying the register contents of the computer (through auxiliary functions). The latter represent augmentations to the simpler (recursive) network grammars that are equivalent to context-free (PS) grammars. Moreover, it is possible to associate any kind of actions-for example, ones which form tree diagrams, semantic representations, etc.— with the transitions between states. In this way, the augmented transition network grammar is not only a recognizing automaton, but also a transducer. Since the use of registers is, in principle, not subject to any limitations and all the possibilities of a conventional programming language can be used, the augmented transition network grammar is as powerful as the universal Turing machine. For the application of augmented transition network grammars to psycholinguistics, see Halle, Bresnan, and Miller (1978).
