These are some guidelines for building "standard" KATR encodings of
morphological theories.

Authors:  Gregory Stump, Raphael Finkel

Date: 3/2000

This document is a draft; we modify it and clean it up occasionally.

-----

Use Sandhi to handle spelling rules and elision rules, but don't make the
Sandhi depend on grammatical conditioning.

Name internal nodes like VERB_AA, with the morphological category first.

Write long vowels as a doubled ordinary vowel: "a a".

Within a node, if a lhs has a known value, when that value is needed for some
rhs, use the known lhs, not the known rhs.  Example: <themeVowel> ==
a a, so if I need a a on some rhs, I should instead call it <themeVowel>.
Reason: It is easier later to hoist such a rule to a higher node.
We hope to modify KATR to introduce a rhs that is identical with the lhs except
as stipulated.  Such a rhs is even better than repeating some other lhs.

When in doubt whether to make one or another situation the default,
choose the one that appears more often in the entire paradigm.

In Sandhi, let there be only one rule that extracts characters from the
path; the others should have rhs in brackets.

It is better to introduce a new node than a set of path rules starting
with a nonmorphological tag.  The labels for affix slots should be nodes,
not attributes in paths.

It is a good idea to show examples that justify unexpected rules.

The name of a leaf node representing a word should be the English equivalent of
the word.
