2.3.5 Combining the two Transducers

The second transducer runs on the output of the first one.

If we now let the two transducers for mapping from the surface to the intermediate form and for mapping from the intermediate to the underlying form run in a cascade (i.e. we let the second transducer run on the output of the first one), we can do a morphological parse of (some) English noun phrases. However, we can also use this transducer for generating a surface form from an underlying form. Remember that we can change the direction of translation when using a transducer in translation mode.

Now, consider the input berries. What will our cascaded transducers make out of it? The first one will return two possible splittings, berries and berrie + s, but the one that we would want, berry + s, is not one of them. The reason for this is that there is another spelling rule at work, here, which we haven't taken into account at all. This rule is saying that ``y changes to ie before s''. So, in the first step there may be more than one spelling rules that all have to be applied.

There are basically two ways of dealing with this. First, we can formulate the transducers for each of the rules in such a way that they can be run in a cascade. Another possibility is to specify the transducers in such a way that they can be applied in parallel.

There are algorithms for combining several cascaded tranducers or several transducers that are supposed to be applied in parallel into a single transducer. However, these algorithms only work, if the individual transducers obey some restrictions so that we have to take some care when specifying them.