[developers] processing of lexical rules

[Emily M. Bender]

On Tue, May 03, 2005 at 01:58:26PM +0100, Ann Copestake wrote:
> Assuming that irregularity is associated with all lexemes with a
> particular orthography (as opposed to individual lexemes) is roughly
> right for English (hang/hanged/hung being an exception) but what
> about other languages?  Do we need to support a version where it is
> associated with individual lexemes?  

I don't have tons of examples to hand, but it strikes me that
in the general case, irregularity is associated with lexemes and
not orthographies.  Some more or less textbook examples:

fly - flew
fly out (in baseball) - flied out

break - broke
brake - braked (here we have an orthographic difference, but
	just as easily might not have)

Basically any time you get accidental homophony (historical
convergence) it would seem that you have the possibility of one
form being irregular and the other regular.

> If I change the current code as
> suggested above, will that cause any immediate problems?  (I can't
> see anything that will go wrong for the ERG though I think it won't
> actually help - e.g., it won't capture `undo'/`undid' if we assume
> that the bracketing is ((un do) past).)

In languages with more morphological complexity than English,
it's not hard to find irregularity where there is an irregular
(say, suppletive) stem that then goes on to take the `regular'
affixes.  The future tense of the French verb aller (`to go')
is a case in point.  Once you switch the stem from the expected
all- to ir- the endings are just like all other verbs.

Emily