The regular expression compiler can be extended with new regular expression operators by providing one or more files defining these operators. The definitions are essentially of two types. In both cases, the actual definitions are written in (often very simple) Prolog. On the one hand, operators can be defined in terms of existing regular expression operators. On the other hand, regular expression operators can be defined by providing a direct implementation on the underlying automata. Many researchers prefer the first style. For instance, Kaplan & Kay [12] (p. 376) argue:
The common data structures that our programs manipulate are clearly states, transitions, labels, and label pairs--the building blocks of finite automata and transducers. But many of our initial mistakes and failures arose from attempting also to think in terms of these objects. The automata required to implement even the simplest examples are large and involve considerable subtlety for their construction. To view them from the perspective of states and transitions is much like predicting weather patterns by studying the movements of atoms and molecules or inverting a matrix with a Turing machine. The only hope of success in this domain lies in developing an appropriate set of high-level algebraic operators for reasoning about languages and relations and for justifying a corresponding set of operators and automata for computation.
Paradoxically, Mohri & Sproat improve upon Kaplan & Kay's algorithm by taking precisely the opposite approach. Their algorithm is primarily presented in terms of manipulations upon states and transitions within automata. One could perhaps translate Mohri & Sproat's algorithm into a high-level calculus, but a great deal of efficiency would be lost in the process. It is a testimony to the flexibility of FSA5, that these two approaches can both be implemented and combined (cf. section 4.3).