[ 
https://issues.apache.org/jira/browse/LUCENE-4955?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13640657#comment-13640657
 ] 

Jack Krupansky commented on LUCENE-4955:
----------------------------------------

I think that ngram filter and edge-ngram filter are rather different cases.

With edge-ngram it is abundantly clear that all of the edge ngrams "stack up" 
at the same position (at least for "front" edge ngrams!). But embedded ngrams 
seem more like a stretching out of the token, from one token to a sequence of 
tokens. Actually, it is k overlayed sequences, where k = maxGramSize minus 
minGramSize plus 1.

I think the solution should be to have a "mode" which indicates whether the 
"intent" is merely variations (sub-tokens) for the token at the same position 
vs. a stretching the token into a sequence of tokens. Maybe call it 
"expansionMode": "stack" vs. "sequence".

But even for the latter, I would definitely recommend that each of the k 
sequences should restart the position at the original token position.

                
> NGramTokenFilter increments positions for each gram
> ---------------------------------------------------
>
>                 Key: LUCENE-4955
>                 URL: https://issues.apache.org/jira/browse/LUCENE-4955
>             Project: Lucene - Core
>          Issue Type: Bug
>          Components: modules/analysis
>    Affects Versions: 4.3
>            Reporter: Simon Willnauer
>             Fix For: 5.0, 4.4
>
>         Attachments: highlighter-test.patch, LUCENE-4955.patch
>
>
> NGramTokenFilter increments positions for each gram rather for the actual 
> token which can lead to rather funny problems especially with highlighting. 
> if this filter should be used for highlighting is a different story but today 
> this seems to be a common practice in many situations to highlight sub-term 
> matches.
> I have a test for highlighting that uses ngram failing with a StringIOOB 
> since tokens are sorted by position which causes offsets to be mixed up due 
> to ngram token filter.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org

Reply via email to