French stemming failed

Question

Hello,
I want to use the stemming capabilities of Oracle Text on french documents, but I don't get the results I expect.
- I have created a table with "boulanger" (baker) and "boulangerie" (bakery) ;
- A query like "$boulanger" returns only one row, I expected two.
Below the script I use to test.
(NLS_LANGUAGE is set to "french" :
SQL> select value from nls_session_parameters where parameter = 'NLS_LANGUAGE';
FRENCH)
Thanks for any hints on what I'm doing wrong.
prompt >>> creation preferences...
begin
ctx_ddl.create_preference('french_lexer', 'BASIC_LEXER');
ctx_ddl.set_attribute('french_lexer', 'PRINTJOINS', '_');
ctx_ddl.set_attribute ( 'french_lexer', 'INDEX_STEMS', 'FRENCH');
CTX_DDL.create_preference('french_wordlist','BASIC_WORDLIST');
CTX_DDL.set_attribute('french_wordlist','STEMMER', '');
-- CTX_DDL.set_attribute('french_wordlist','FUZZY_MATCH', 'FRENCH');
end;
prompt >>> preferences creees.
prompt >>> creation table
create table french_test (texte varchar2(60));
insert all
into french_test values ('boulanger')
into french_test values ('boulangerie')
into french_test values ('restaurant')
into french_test values ('restauration')
into french_test values ('restaurateur')
select * from dual;
commit ;
prompt >>> table creee
prompt >>> creation index
create index french_idx on french_test(texte) indextype is ctxsys.context parameters ('LEXER french_lexer WORDLIST french_wordlist sync (on commit)') ;
prompt >>> index cree
prompt >>> Recherche : boulanger (boulangerie)
select * from french_test where contains(texte, '<query><textquery grammar="CONTEXT" lang="FRENCH">$boulanger</textquery></query>', 0) > 0 ;
prompt >>> fin
prompt >>> Recherche : restauration (restaurant, restaurateur)
select * from french_test where contains(texte, '<query><textquery grammar="CONTEXT" lang="FRENCH">$restauration</textquery></query>', 0) > 0 ;
prompt >>> fin
prompt >>> suppression table
drop index french_idx ;
drop table french_test ;
prompt >>> table supprimee
prompt >>> suppression preferences
begin
ctx_ddl.drop_preference('french_lexer');
ctx_ddl.drop_preference('french_wordlist');
end;
prompt >>> preferences supprimees
Mat'

Answer 1

Boulanger is not the root word of boulangerie and baker is not the root word of bakery. Although they start with the same letters and sound the same, they are not the same thing. It may help to think of stem words in terms of conjugations. Stemming searches only match if words are conjugations of the same root word . They also must match in terms of type of word, such as nouns or verbs or adjectives or adverbs and whether they are masculine or feminine where applicable. Please see the demonstration below that shows first in French, then in English what some of the roots are and are not. The last query of each demo shows the roots and the words that share those roots. Only words that share the same root will be found using stem searches.
SCOTT@orcl_11g> ALTER SESSION SET NLS_LANGUAGE = 'FRENCH'
  2  /
Session altered.
SCOTT@orcl_11g> begin
  2    ctx_ddl.create_preference ('french_lexer',    'BASIC_LEXER');
  3    ctx_ddl.set_attribute      ('french_lexer',    'PRINTJOINS', '_');
  4    ctx_ddl.set_attribute      ('french_lexer',    'INDEX_STEMS', 'FRENCH');
  5    --
  6    CTX_DDL.create_preference ('french_wordlist', 'BASIC_WORDLIST');
  7    CTX_DDL.set_attribute      ('french_wordlist', 'STEMMER', 'FRENCH');
  8  end;
  9  /
PL/SQL procedure successfully completed.
SCOTT@orcl_11g> create table french_test (texte varchar2(60))
  2  /
Table created.
SCOTT@orcl_11g> insert all
  2  into french_test values ('boulanger')
  3  into french_test values ('boulangers')
  4  into french_test values ('boulangerie')
  5  into french_test values ('boulangeries')
  6  into french_test values ('cuire')
  7  into french_test values ('cuit')
  8  select * from dual
  9  /
6 rows created.
SCOTT@orcl_11g> commit
  2  /
Commit complete.
SCOTT@orcl_11g> create index french_idx on french_test (texte)
  2  indextype is ctxsys.context
  3  parameters
  4    ('LEXER        french_lexer
  5        WORDLIST  french_wordlist
  6        sync        (on commit)')
  7  /
Index created.
SCOTT@orcl_11g> select * from french_test
  2  where  contains
  3             (texte,
  4              '<query>
  5              <textquery grammar="CONTEXT" lang="FRENCH">$boulanger
  6              </textquery>
  7            </query>',
  8              0) > 0
  9  /
TEXTE
boulanger
boulangers
SCOTT@orcl_11g> CREATE INDEX french_idx2 ON dr$french_idx$i (token_text)
  2  INDEXTYPE IS CTXSYS.CONTEXT
  3  parameters
  4    ('LEXER        french_lexer
  5        WORDLIST  french_wordlist
  6        sync        (on commit)')
  7  /
Index created.
SCOTT@orcl_11g> COLUMN words FORMAT A30
SCOTT@orcl_11g> COLUMN root  FORMAT A30
SCOTT@orcl_11g> SELECT root, wm_concat (word) words
  2  FROM   (SELECT DISTINCT NVL (i.token_text, d.token_text) AS root, d.token_text word
  3            FROM   dr$french_idx$i d, dr$french_idx2$i i
  4            WHERE  CONTAINS (d.token_text, '$' || i.token_text(+)) > 0
  5            AND    i.token_type (+)= 9
  6            ORDER  BY root, word)
  7  GROUP  BY root
  8  /
ROOT                           WORDS
BOULANGER                      BOULANGER,BOULANGERS
BOULANGERIE                    BOULANGERIE,BOULANGERIES
CUIRE                          CUIRE,CUIT
SCOTT@orcl_11g> ALTER SESSION SET NLS_LANGUAGE = 'AMERICAN'
  2  /
Session altered.
SCOTT@orcl_11g> begin
  2    ctx_ddl.create_preference ('english_lexer',    'BASIC_LEXER');
  3    ctx_ddl.set_attribute      ('english_lexer',    'PRINTJOINS', '_');
  4    ctx_ddl.set_attribute      ('english_lexer',    'INDEX_STEMS', 'ENGLISH');
  5    --
  6    CTX_DDL.create_preference ('english_wordlist', 'BASIC_WORDLIST');
  7    CTX_DDL.set_attribute      ('english_wordlist', 'STEMMER', 'ENGLISH');
  8  end;
  9  /
PL/SQL procedure successfully completed.
SCOTT@orcl_11g> create table english_test (text varchar2(60))
  2  /
Table created.
SCOTT@orcl_11g> insert all
  2  into english_test values ('baker')
  3  into english_test values ('bakers')
  4  into english_test values ('bakery')
  5  into english_test values ('bakeries')
  6  into english_test values ('cooked')
  7  into english_test values ('cooking')
  8  select * from dual
  9  /
6 rows created.
SCOTT@orcl_11g> commit
  2  /
Commit complete.
SCOTT@orcl_11g> create index english_idx on english_test (text)
  2  indextype is ctxsys.context
  3  parameters
  4    ('LEXER        english_lexer
  5        WORDLIST  english_wordlist
  6        sync        (on commit)')
  7  /
Index created.
SCOTT@orcl_11g> select * from english_test
  2  where  contains
  3             (text,
  4              '<query>
  5              <textquery grammar="CONTEXT" lang="ENGLISH">$baker
  6              </textquery>
  7            </query>',
  8              0) > 0
  9  /
TEXT
baker
bakers
SCOTT@orcl_11g> CREATE INDEX english_idx2 ON dr$english_idx$i (token_text)
  2  INDEXTYPE IS CTXSYS.CONTEXT
  3  parameters
  4    ('LEXER        english_lexer
  5        WORDLIST  english_wordlist
  6        sync        (on commit)')
  7  /
Index created.
SCOTT@orcl_11g> COLUMN words FORMAT A30
SCOTT@orcl_11g> COLUMN root  FORMAT A30
SCOTT@orcl_11g> SELECT root, wm_concat (word) words
  2  FROM   (SELECT DISTINCT NVL (i.token_text, d.token_text) AS root, d.token_text word
  3            FROM   dr$english_idx$i d, dr$english_idx2$i i
  4            WHERE  CONTAINS (d.token_text, '$' || i.token_text(+)) > 0
  5            AND    i.token_type (+)= 9
  6            ORDER  BY root, word)
  7  GROUP  BY root
  8  /
ROOT                           WORDS
BAKER                          BAKER,BAKERS
BAKERY                         BAKERIES,BAKERY
COOK                           COOK,COOKED,COOKING
SCOTT@orcl_11g>