[
https://issues.apache.org/jira/browse/LUCENE-2906?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12990739#comment-12990739
]
Robert Muir commented on LUCENE-2906:
-------------------------------------
I'll take it, ive done the unibigram approach already (maybe we can just have
it as a separate filter option), so the bigram should be easy.
My original design, just lets you provide a BitSet of script codes. (this would
be simple i think to parse from say a solr factory).
I think its also useful to have an option, for whether the filter should only
do this for "joined" text or not (based on offsets). For CJK i think it makes
sense to enforce this, so that it won't bigram across sentence boundaries. But
for say the Tibetan language, where you have a syllable separator, you would
want to turn this off.
Separately, if you want it to work "just like" CJKTokenizer, please be aware
that by default, the unicode standard tokenizes Katakana to words (only
hiragana and han are tokenized to codepoints). So in this case you would have
to use a custom ruleset if you wanted katakana to be tokenized to codepoints
instead of words, for later bigramming. I'm not sure you want to do this
though... (in truth CJKTokenizer bigrams ANYTHING out of ascii, including a lot
of things it shouldnt).
For hangul the same warning applies, but its more debatable, you might want to
do this if you don't have a decompounder... but in my opinion this is past
tokenization, and its the same problem you have with german, etc... the default
tokenization is not "wrong".
In either case, if you decide to do that, it would be a pretty simple ruleset!
Let me know if this makes sense to you.
> Filter to process output of ICUTokenizer and create overlapping bigrams for
> CJK
> --------------------------------------------------------------------------------
>
> Key: LUCENE-2906
> URL: https://issues.apache.org/jira/browse/LUCENE-2906
> Project: Lucene - Java
> Issue Type: New Feature
> Components: Analysis
> Reporter: Tom Burton-West
> Priority: Minor
>
> The ICUTokenizer produces unigrams for CJK. We would like to use the
> ICUTokenizer but have overlapping bigrams created for CJK as in the CJK
> Analyzer. This filter would take the output of the ICUtokenizer, read the
> ScriptAttribute and for selected scripts (Han, Kana), would produce
> overlapping bigrams.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
