Hi there,

I have two files ny_MW.aff and ny_MW.dic. In ny_MW.dic, there are words that
MUST have at least a suffix S but may also take a prefix P. In order to
enforce suffixation on those words, I declare a virtual stemming flag
NEEDAFFIX N that is appended to them. I have prefixes defined like this one
"PFX P 0 achi  .", and  and suffixes define like this "SFX S 0 abe/P   .".

Given the following words:
  batiz
  abatiz
  batiza
  achibatiza
I expect the following result:
  batiz          (WRONG)
  abatiz        (WRONG)
  batiza        (CORRECT)
  achibatiza  (CORRECT)

I prefer defining the affix class combinations using AF flag because of the
number of affix classes to be used. However, AF flag doesn't seem to work
fine for me. Does anyone know why the affix class combination works fine
without using AF flag, and why it does not work when using AF flag? (c.f.
case A and case B below). If so, how can I improve AF flagging (case B) to
work fine?

CASE A: Without AF flag
===================

1. ny_MW.aff

  SET ISO8859-14
  TRY aiuneorbkmgywthszdc'
IjAvfUpRKMNlBGLPYHTCSDZEWVJO

  WORDCHARS -'
  CIRCUMFIX M
  NEEDAFFIX N

  # prefix section
  PFX P Y 4
  PFX P 0 a       .
  PFX P 0 achi    .
  PFX P 0 adza    .
  PFX P 0 adzachi .

  #suffix section
  SFX S Y 5
  SFX S 0 a/P     .
  SFX S 0 abe/P   .
  SFX S 0 eno/P   .
  SFX S 0 ko      .
  SFX S 0 po      .

2. ny_MW.dic
  batiz/SN
  mlimi
  yang'an/SN

CASE B: Using AF flag ( This one does not work)
====================================

1. ny_MW.aff

  SET ISO8859-14
  TRY aiuneorbkmgywthszdc'IjAvfUpRKMNlBGLPYHTCSDZEWVJO

  WORDCHARS -'
  CIRCUMFIX M
  NEEDAFFIX N

  # every word must have at least a suffix
  AF 1
  AF SN
  # more compression aliases will be added later

  # prefix section
  PFX P Y 4
  PFX P 0 a  .
  PFX P 0 achi  .
  PFX P 0 adza  .
  PFX P 0 adzachi  .

  #suffix section
  SFX S Y 3
  SFX S 0 a/P  .
  SFX S 0 abe/P  .
  SFX S 0 eno/P  .

2. ny_MW.dic
  batiz/1
  mlimi
  yang'an/1

I'm using these references:

   - http://manpages.ubuntu.com/manpages/dapper/man4/hunspell.4.html
   - http://pwet.fr/man/linux/fichiers_speciaux/hunspell

*Sorry for sending a longer email. May be I would use pastie/pastebin. [?].

*
---
Edmond
Software Developer | Baobab Health Trust (http://www.baobabhealth.org/) |
Malawi

Cell:  +265 999 465 137 | +265 881 234 717

*"Many people doubt open source software and probably don’t realize that
there is an alternative… which is just as good.." -- Kevin Scannell*

Reply via email to