The replace operator.

In [19] a variant of the replace operator is implemented which is more efficient than previous implementations provided by Kaplan and Kay [12] and Karttunen [13]. This improved version crucially depends on the possibility of manipulating the transitions and states of the underlying automata directly. The replacement of expression Phi into Psi in the context of Left and Right is written replace(Left,Phi,Psi,Right). In the left-to-right interpretation, this operator can be defined as the following cascade:

$$\begin{minipage}[t]{.9\textwidth}\begin{verbatim}macro(repl... ...o replace(Phi,Psi) o l1(L) o l2(L))\end{verbatim}\end{minipage}$$ (23)

This definition and the definitions of the auxiliary operators are closely modelled on those given in [19]. The auxiliary operators are defined in the appendix.

A typical example of the use of the replace operator is provided by the past tense endings of Dutch regular verbs. In Dutch, the singular past tense is formed by the -de and -te suffixes. If the previous phoneme is voiced, the suffix -de must be used; in order circumstances the -te suffix is appropriate. This phenomenon can be analysed by assuming an underlying, abstract, -Te suffix. The T is then transformed into a d or t depending on context. The rule can be defined as follows (the + indicates a morpheme boundary):

$$\begin{minipage}[t]{.9\textwidth}\begin{verbatim}macro(tkof... ...'T',t,e) o replace(+,'T',d, [])\end{verbatim}\end{minipage}$$ (24)