[
https://issues.apache.org/jira/browse/OPENNLP-1525?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Martin Wiesner updated OPENNLP-1525:
------------------------------------
Description:
While working on OPENNLP-1479 and reviewing a PR by [~l-ma], we identified that
{{TokenizerME}} doesn't make use of locale/language specific abbreviations
provided via the corresponding TokenizerModel.
Therefore, terms will get mis-tokenized if they are abbreviated, such as the
German token "S." which represents an abbreviated form of "Seite" (-> page).
Instead of being tokenized as {"S."}, TokenizerME will incorrectly yield: {"S",
"."}
Improvement suggested:
- Make use of the abbreviations dictionary provided by the {{TokenizerModel}}
- Adapt the idea suggested and implemented in OPENNLP-570
({{SentenceDetectorME}}) for {{TokenizerME}}
- Adjust {{TokenizerFactoryTest}} method testCustomPatternForTokenizerMEDeu()
for German abbreviations, see sent-detector test case. It should expect and
result in 14 tokens, instead of 16 - so there is a TODO here.
was:
While working on OPENNLP-1479 and reviewing a PR by [~l-ma], we identified that
{{TokenizerME}} doesn't make use of locale/language specific abbreviations
provided via the corresponding TokenizerModel.
Therefore, terms will get mis-tokenized if they are abbreviated, such as the
German token "S." which represents an abbreviated form of "Seite" (-> page).
Improvement suggested:
- Make use of the abbreviations dictionary provided by the {{TokenizerModel}}
- Adapt the idea suggested and implemented in OPENNLP-570
({{SentenceDetectorME}}) for {{TokenizerME}}
- Adjust {{TokenizerFactoryTest}} method testCustomPatternForTokenizerMEDeu()
for German abbreviations, see sent-detector test case. It should expect and
result in 14 tokens, instead of 16 - so there is a TODO here.
> Improve TokenizerME to make use of abbreviations provide in TokenizerModel
> --------------------------------------------------------------------------
>
> Key: OPENNLP-1525
> URL: https://issues.apache.org/jira/browse/OPENNLP-1525
> Project: OpenNLP
> Issue Type: Improvement
> Components: Tokenizer
> Affects Versions: 2.3.1
> Reporter: Martin Wiesner
> Assignee: Martin Wiesner
> Priority: Minor
>
> While working on OPENNLP-1479 and reviewing a PR by [~l-ma], we identified
> that {{TokenizerME}} doesn't make use of locale/language specific
> abbreviations provided via the corresponding TokenizerModel.
> Therefore, terms will get mis-tokenized if they are abbreviated, such as the
> German token "S." which represents an abbreviated form of "Seite" (-> page).
> Instead of being tokenized as {"S."}, TokenizerME will incorrectly yield:
> {"S", "."}
> Improvement suggested:
> - Make use of the abbreviations dictionary provided by the {{TokenizerModel}}
> - Adapt the idea suggested and implemented in OPENNLP-570
> ({{SentenceDetectorME}}) for {{TokenizerME}}
> - Adjust {{TokenizerFactoryTest}} method testCustomPatternForTokenizerMEDeu()
> for German abbreviations, see sent-detector test case. It should expect and
> result in 14 tokens, instead of 16 - so there is a TODO here.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
