Post-generator

Many languages use a post-generator FST to fix minor orthographical issues. This FST is in lttoolbox format and is run by lt-proc with the -p or --post-generation switch. An example of such an orthographical issue is the "a" vs "an" difference in English. The english generator will output ~a, and the post-generation FST changes that to a or an depending on the following word.

The source dictionary is typically named something like apertium-cat.post-cat.dix, while the compiled file gets a name like spa-cat.autopgen.bin.

Here's a minimal example for turning ~a into an before vowels:

<?xml version="1.0" encoding="UTF-8"?>
<dictionary>
  <alphabet/>
  <sdefs>
    <sdef n="n" c="Noun"/>
  </sdefs>
  <pardefs>
    <pardef n="vocals">
      <e>
        <i>a</i>
      </e>
      <e>
        <i>e</i>
      </e>
      <e>
        <i>i</i>
      </e>
      <e>
        <i>o</i>
      </e>
      <e>
        <i>u</i>
      </e>
    </pardef>
  </pardefs>
  <section id="main" type="standard">
    <e>
      <p>
        <l><a/>a<b/></l>
        <r>an<b/></r>
      </p>
      <par n="vocals"/>
    </e>
  </section>
</dictionary>

Debugging

A debugging version of a postgen dictionary can be made by compiling with the -d/--debug flag, such as with

lt-comp --debug lr apertium-cat.post-cat.dix cat-debug.bin

Then the output will include the approximate line number of each rule that applies as in

$ echo "~a apple" | lt-proc -p cat-debug.bin
Line near 29 an apple

Unfortunately, due to a limitation of the XML parsing library that lt-comp uses, the line number reported will often be a few lines past the entry in question. To help with this, a comment can be added to the entry like

<e c="indef+vowel">

and then the output will be

$ echo "~a apple" | lt-proc -p cat-debug.bin
Line near 29 indef+vowel an apple