A filter that decomposes compound words you find in many Germanic languages into the word parts.
This example shows what it does:
example input stream
Input token stream |
Rindfleischüberwachungsgesetz Drahtschere abba |
example output stream
Output token stream |
(Rindfleischüberwachungsgesetz,0,29) |
(Rind,0,4,posIncr=0) |
(fleisch,4,11,posIncr=0) |
(überwachung,11,22,posIncr=0) |
(gesetz,23,29,posIncr=0) |
(Drahtschere,30,41) |
(Draht,30,35,posIncr=0) |
(schere,35,41,posIncr=0) |
(abba,42,46) |
The input token is always preserved and the filters do not alter the case of word parts. There
are two variants of the filter available:
- HyphenationCompoundWordTokenFilter: it uses a hyphenation grammar based approach to
find potential word parts of a given word.
- DictionaryCompoundWordTokenFilter: it uses a brute-force dictionary-only based
approach to find the word parts of a given word.
Compound word token filters
HyphenationCompoundWordTokenFilter
The
HyphenationCompoundWordTokenFilter
uses hyphenation grammars to find potential subwords that a
worth to check against the dictionary. It can be used without a dictionary as well but then
produces a lot of "nonword" tokens. The quality of the output tokens is directly connected to the
quality of the grammar file you use. For languages like German they are quite good.
Grammar file
Unfortunately we cannot bundle the hyphenation grammar files with Lucene because they do not use
an ASF compatible license (they use the LaTeX Project Public License instead). You can find the
XML based grammar files at the
Objects For Formatting Objects
(OFFO) Sourceforge project (direct link to download the pattern files:
http://downloads.sourceforge.net/offo/offo-hyphenation.zip
). The files you need are in the subfolder
offo-hyphenation/hyph/ .
Credits for the hyphenation code go to the
Apache
FOP project .
DictionaryCompoundWordTokenFilter
The
DictionaryCompoundWordTokenFilter
uses a dictionary-only approach to find subwords in a compound
word. It is much slower than the one that uses the hyphenation grammars. You can use it as a
first start to see if your dictionary is good or not because it is much simpler in design.
Dictionary
The output quality of both token filters is directly connected to the quality of the dictionary
you use. They are language dependent of course. You always should use a dictionary that fits to
the text you want to index. If you index medical text for example then you should use a
dictionary that contains medical words. A good start for general text are the dictionaries you
find at the
OpenOffice
dictionaries Wiki.
Which variant should I use?
This decision matrix should help you:
comparison of dictionary and hyphenation based decompounding
Token filter |
Output quality |
Performance |
HyphenationCompoundWordTokenFilter |
good if grammar file is good – acceptable otherwise |
fast |
DictionaryCompoundWordTokenFilter |
good |
slow |
Examples
public void testHyphenationCompoundWordsDE() throws Exception {
String[] dict = { "Rind", "Fleisch", "Draht", "Schere", "Gesetz",
"Aufgabe", "Überwachung" };
Reader reader = new FileReader("de_DR.xml");
HyphenationTree hyphenator = HyphenationCompoundWordTokenFilter
.getHyphenationTree(reader);
HyphenationCompoundWordTokenFilter tf = new HyphenationCompoundWordTokenFilter(
new WhitespaceTokenizer(new StringReader(
"Rindfleischüberwachungsgesetz Drahtschere abba")), hyphenator,
dict, CompoundWordTokenFilterBase.DEFAULT_MIN_WORD_SIZE,
CompoundWordTokenFilterBase.DEFAULT_MIN_SUBWORD_SIZE,
CompoundWordTokenFilterBase.DEFAULT_MAX_SUBWORD_SIZE, false);
CharTermAttribute t = tf.addAttribute(CharTermAttribute.class);
while (tf.incrementToken()) {
System.out.println(t);
}
}
public void testHyphenationCompoundWordsWithoutDictionaryDE() throws Exception {
Reader reader = new FileReader("de_DR.xml");
HyphenationTree hyphenator = HyphenationCompoundWordTokenFilter
.getHyphenationTree(reader);
HyphenationCompoundWordTokenFilter tf = new HyphenationCompoundWordTokenFilter(
new WhitespaceTokenizer(new StringReader(
"Rindfleischüberwachungsgesetz Drahtschere abba")), hyphenator);
CharTermAttribute t = tf.addAttribute(CharTermAttribute.class);
while (tf.incrementToken()) {
System.out.println(t);
}
}
public void testDumbCompoundWordsSE() throws Exception {
String[] dict = { "Bil", "Dörr", "Motor", "Tak", "Borr", "Slag", "Hammar",
"Pelar", "Glas", "Ögon", "Fodral", "Bas", "Fiol", "Makare", "Gesäll",
"Sko", "Vind", "Rute", "Torkare", "Blad" };
DictionaryCompoundWordTokenFilter tf = new DictionaryCompoundWordTokenFilter(
new WhitespaceTokenizer(
new StringReader(
"Bildörr Bilmotor Biltak Slagborr Hammarborr Pelarborr Glasögonfodral Basfiolsfodral Basfiolsfodralmakaregesäll Skomakare Vindrutetorkare Vindrutetorkarblad abba")),
dict);
CharTermAttribute t = tf.addAttribute(CharTermAttribute.class);
while (tf.incrementToken()) {
System.out.println(t);
}
}