[
https://issues.apache.org/jira/browse/OPENNLP-1520?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17790394#comment-17790394
]
Jon Marius Venstad commented on OPENNLP-1520:
---------------------------------------------
Hi Martin. Unfortunately, I was not able to produce such a data sample. We run
stemming on behalf of our users, and I happened upon this stack trace when
debugging something else. We generally have no access to our users' data, so I
don't know exactly what text triggered it. I suppose I could try to obtain a
sample, but that would take some time.
Perhaps someone with knowledge of the Finnish language would be able to
construct a sample? I tried concatenating lots of Finnish stem endings (found
in those tables in the generated code), but had no lock with that. I also tried
a few text samples from the internet, but also to no avail.
> Generated Java code for stemmers is broken, and should be re-generated
> ----------------------------------------------------------------------
>
> Key: OPENNLP-1520
> URL: https://issues.apache.org/jira/browse/OPENNLP-1520
> Project: OpenNLP
> Issue Type: Bug
> Components: Stemmer
> Affects Versions: 2.3.0, 2.3.1
> Reporter: Jon Marius Venstad
> Assignee: Martin Wiesner
> Priority: Major
> Fix For: 2.3.2
>
>
> The recursive stemming, which seems hard to actually trigger, but which is
> the intended usage of the {{methodObject and method}} in the {{Among}} class
> (called reflectively) is completely broken. First off, it tries to invoke a
> private method from outside the class (from a parent class, the
> {{{}SnowballProgram{}}}), which fails with an illegal access exception; if
> that worked, it would also have invoked _all_ such method calls on the
> {_}same, shared, static object{_}—not on the relevant stemmer instance.
> This was fixed 8 years ago, but it looks like the generated code in the
> opennlp-tools is 10 years old. I would urge you to re-generate that code.
>
> Commit that fixed the Java code generation:
> [https://github.com/snowballstem/snowball/commit/0f9d3d64ab965447a7f638b8ededc924f3efca75]
>
> Relevant sample stemmer with broken Java:
> [https://github.com/apache/opennlp/blob/main/opennlp-tools/src/main/java/opennlp/tools/stemmer/snowball/finnishStemmer.java]
>
> Stack trace showing illegal reflection access:
>
> {noformat}
> 2023-10-26 23:21:44.200 class opennlp.tools.stemmer.snowball.SnowballProgram
> cannot access a member of class opennlp.tools.stemmer.snowball.finnishStemmer
> with modifiers "private"
> exception=java.lang.IllegalAccessException: class
> opennlp.tools.stemmer.snowball.SnowballProgram cannot access a member of
> class opennlp.tools.stemmer.snowball.finnishStemmer with modifiers "private"
> at
> java.base/jdk.internal.reflect.Reflection.newIllegalAccessException(Reflection.java:392)
>
> at
> java.base/java.lang.reflect.AccessibleObject.checkAccess(AccessibleObject.java:674)
>
> at java.base/java.lang.reflect.Method.invoke(Method.java:560)
> at
> opennlp.tools.stemmer.snowball.SnowballProgram.find_among_b(SnowballProgram.java:353)
>
> at
> opennlp.tools.stemmer.snowball.finnishStemmer.r_case_ending(finnishStemmer.java:480)
>
> at
> opennlp.tools.stemmer.snowball.finnishStemmer.stem(finnishStemmer.java:1003)
> at
> opennlp.tools.stemmer.snowball.SnowballStemmer.stem(SnowballStemmer.java:131)
> at
> com.yahoo.language.opennlp.OpenNlpTokenizer.processToken(OpenNlpTokenizer.java:64)
>
> at
> com.yahoo.language.opennlp.OpenNlpTokenizer.lambda$tokenize$0(OpenNlpTokenizer.java:54)
>
> at
> com.yahoo.language.simple.SimpleTokenizer.tokenize(SimpleTokenizer.java:74)
> at
> com.yahoo.language.opennlp.OpenNlpTokenizer.tokenize(OpenNlpTokenizer.java:54)
>
> at
> com.yahoo.vespa.indexinglanguage.linguistics.LinguisticsAnnotator.annotate(LinguisticsAnnotator.java:76)
> ...{noformat}
>
>
> Best, Jon Marius Venstad, developer at [vespa.ai|http://vespa.ai/]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
