ABSTRACT
This chapter explores the need for an additional level of structural description, how it might be described by a grammar and how a recognisor based on such grammars might be implemented. Since some of the expressions consist of more than one word, this property must be a property of groups of words or phrases. The hierarchical structure of a sentence can be graphically represented with a phrase structure tree. The recognition procedure developed for FSTNs can also form the basis for recognition using Recursive Transition Networks (RTNs). An extension is required, to handle those arcs labelled with network names. Accordingly, when a phrasal label is encountered in a network, control is temporarily passed to the network which specialises in recognising categories of that sort. Using variables is not dissimilar to an extended form of RTN called an augmented transition network (ATN). The expressive power of Prolog means that it is relatively easy to define recognition procedures based on RTNs.
