Bug 2145

Summary:	Downcasing of derived proper nouns broken by hyperminimisation
Product:	smj lexicon	Reporter:	Sjur Nørstebø Moshagen <sjur.n.moshagen>
Component:	stem lexica	Assignee:	Sjur Nørstebø Moshagen <sjur.n.moshagen>
Status:	ASSIGNED ---
Severity:	normal	CC:	elena.j.paulsen, sjur.n.moshagen, thomas.omma, trond.trosterud
Priority:	P5 - Later
Version:	unspecified
Hardware:	All
OS:	All

Description Sjur Nørstebø Moshagen 2016-01-11 14:22:44 CET

./configure --with-hfst --enable-hyperminimisation

gives the following:

[ 1/18][FAIL] Narvijkka+N+Prop+Sem/Plc+Sg+Gen+Der/k+N+Sg+Nom => Missing results: narvijkak
[ 1/18][FAIL] Narvijkka+N+Prop+Sem/Plc+Sg+Gen+Der/k+N+Sg+Nom => Unexpected results: Narvijkak

This is because hyperminimisation inserts an extra symbol at the very beginning of the net: @P.LEXNAME.Root@. This symbol breaks the context requirements of the downcasing regex.

Hyperminimisation is not used very much, but can be turned on for speller optimisations without people being aware of or remembering this bug. It should thus be fixed.

Comment 1 Thomas Omma 2016-01-12 08:54:24 CET

hyperminimisation! :O

Comment 2 Sjur Nørstebø Moshagen 2018-05-09 12:52:55 CEST

This is now finally fixed - mostly. There are still a few regressions for the spellers for words with CmpN/*Left tags, so will keep this open until it is completely fixed.