WNJN

WordNet JNI Java Native Support

Bernard BOU
Lycée Champollion
bbou@ac-toulouse.fr

THIS PAGE WAS REVISED ON FEB 20, 2006

CONTENTS

• License
• The purpose of WNJN
• Pure Java or Native methods
• This and other implementations
• NMI vs JNI
• Requirements
• Platforms
• Targets
• Build
• Files in package
• WordNet 2.0 support
• WordNet 2.1 support
• Author
• Contributors
• Download
• Common pitfalls
• Project management

LICENSE

This package is is under a GPL GNU Licence (See file GPL). Some of the code is under the WordNet license agreement.

top of page

THE PURPOSE OF WNJN

This package allows you to use WordNet from Java. Some of the code is under the WordNet license agreement.

top of page

PURE JAVA OR NATIVE METHODS IN C

I weighed the pros and cons of writing a pure java solution (with no C code) and decided against it on the following grounds :

top of page

THIS AND OTHER IMPLEMENTATIONS

There are other implementations with the same functionality. I'm referring to Jorge Gomez' ( jjgomez@sip.ucm.es ) implementation of a Java native interface. It came first and suffices in most situations but, as as I started using it, I wanted to improve on on a number of points. So I started a rewrite and the result has evolved away from the original. 

However we do have points in common : 

• write a minimum wrapper around the C interface. Additions and further development can be carried out in Java. 
• leave OOP (object oriented programming) aside, which could could bear my view of how things should be. So there's hardly any OOP to be found but for the Synset and Index classes which naturally come to mind , matching the cousin C structures. 
• stick to the usual naming WordNet conventions, however arcane and inconsistent they may be
• do not retain garbage-collection and string fuctions, in addition to internals-oriented functions

The points I wanted to improve on  :

• conform to JNI standards and not NMI standards which are still supported but deprecated (see below Sun's JDK2 warning)
• support recursive searches
• use C++ instead of C to make it more legible and error-prone (automatic freeing of resources for instance)
• use a global naming convention for classes : I used edu.princeton.wordnet as a prefix to the java classes. I hope they don't mind.

top of page

NMI VERSUS JNI

This is taken from the JDK 1.2 documentation :

Java Native Interface is a standard programming interface for writing Java native methods and embedding the Java virtual machine into native applications. The primary goal is binary compatibility of native method libraries across all Java virtual machine implementations on a given platform.

JDK 1.2 extends the Java Native Interface (JNI) to incorporate new features in the Java platform.

JDK 1.2 still supports the old native method interface (NMI), which was originally implemented in JDK 1.0. The old NMI is not part of the Java platform standard, and will not be supported by the upcoming Java HotSpot^TM virtual machine. Programmers are strongly encouraged to migrate their existing NMI code to the JNI.

 

top of page

REQUIREMENTS

You need :

• a C++ compiler
• the JDK http://www.javasoft.com
• the WordNet API C sources available from http://wordnet.princeton.edu with the bugs fixed (see bug list there)
• this package
• (optionally but strongly recommended) the Ant 1.5 build tool available from http://apache.org

top of page

PLATFORMS

I have carefully avoided features which could not be portable. You simply need a standard C++ tool.

However I developed this on two specific platforms :

• a Linux Box with  Borland's JBuilder 8.0. or 9.0 and ant 1.5.4
• a Windows XP box with Microsoft's Visual 6.0 Developer Studio and Borland's JBuilder 8.0. or 9.0 and ant 1.5.3.

I include the build files attached to these tools for convenience togehter with redundant batch or shell files.

But I didn't test it on other platforms as I don't have any other platform handy. Please refer to the build process below.

top of page

TARGETS

BUILD

You have to have Apache Jakarta Ant and run (in the folder where the build.xml file is to be found):

ant -Dtarget.wnx="21"

Valid wnx targets are "16", "17" or "171" or "20" or "21"

Alternatively you can run the different steps :

We owe the Unix makefiles to Ola Åkerberg and Hans Svensson (e94hsv@efd.lth.se), two masters thesis students from Lund, Sweden.
 

top of page

FILES

• core files
  
  Index.java
  Synset.java
  WordNet.java
  Wordnet.cpp
  WnFactory.cpp
  WnFactory.h
  
  
• doc files
  
  wnjn.html
   
• platform-dependent files or tool dependent files
  (_ prefixed)
  

top of page

WORDNET 2.0 support

The tests compile and run fine with the WordNet 2.0 C API files. One major change with versions as of 1.7 is the addition at the beginning of the C-structure of a idxoffset, which could mean trouble if the proper defines and header files are not included.
The distribution has been compiled to WordNet 2.0

top of page

WORDNET 2.1 support

The tests compile and run fine with the WordNet 2.1 C API files. One major change with versions as of 1.7 is the addition at the beginning of the C-structure of a idxoffset, which could mean trouble if the proper defines and header files are not included.
The distribution has been compiled to WordNet 2.1

top of page

AUTHOR

Bernard BOU
Lycée Champollion,
Figeac, France.

This wnjn interface is needed as part of a data provider for my Treebolic engine which will be released later this year. I also authored WordNet TreeWalk (for which I have received very positive feedback):

• W O R D N E T  W E B  A P P L I C A T I O N
  J2EE web application for WordNet
  http://wnwa.sourceforge.net
  http://jws-champo.ac-toulouse.fr:8080 (online jws)
• W O R D N E T   T R E E W A L K
  Browser for WordNet
  http://wntw.sourceforge.net
• W N J N
  Java Native Interface (JNI) library for WordNet
  http://wnjn.sourceforge.net
• W N C O M
  COM component for WordNet
  http://wncom.sourceforge.net
• W N W S
  Java Web Services for WordNet
  http://wnws.sourceforge.net
• T R E E B O L I C
  Hyberbolic tree viewing- engine,browser, generator and
  Web services on Apache Tomcat (GPL)
  includes a WordNet data provider
  http://treebolic.sourceforge.net
   

as well as other various software programs ( http://www.ac-toulouse.fr/bbou)

top of page

CONTRIBUTORS

Ola Åkerberg and Hans Svensson
Barry Hathaway

top of page

COMMON PITFALLS

When the output yields wninit=-1, the trouble can originate in the following:

• The WNHOME environment variable is not set. Check this with echo %WNHOME% (Windows) or echo $WHNHOME (Unix). It must point at the WordNet folder (above the dict folder that contains the lexicographic files)
• The Java runtime cannot find the native library because the path is unknown or incorrect. Use the -Djava.library.path="<path>" on the command line and ensure the path contains the proper libwnjn.so or wnjn.dll. Alternatively you can place the library in the $JAVA_HOME/jre/lib/i386 folder.
• The Java runtime cannot find the native library because it does not expect a version-tagged library : the valid names are wnjn.dll for Windows or libwnjn.so for Unix.
• The version of the library does not match that of the database.
• Ant 1.5.3 and lower does not work properly with J2SDK 1.4.2 and stumbles on the javah task. Please update to Ant 1.5.4 or comment out the task in the build.xml file.
• [Barry Hathaway] You will find a couple a couple of Makefiles with the extension of ".vc80" which can be used to build a Windows version using Visual C++ 8.0 (Visual Studio 2005). While using these would be recommended by Microsoft they make deployment more difficult. With VC 6.0 the wnjn.dll gets built with the C run-time libraries statically linked in. This makes deployment consist of just copying the wnjn.dll file to the target machine. With VC 8.0 Microsoft decided to try to eliminate "dll hell" and not really allow statically linked libraries.  In our situation the wnjn.dll will get built without the C run-time libraries. The dll contains a manifest describing which libraries it needs to run.  Windows/XP (and 2003) will use policies in WinSxS to map to the proper libaries in WinSxS.  So deployment now requires the installation of the VC 8.0 run-time libraries on the target machine.  This is accomplished by running the redist.exe that is part of a VC 8.0 installation.  (In my case is used the free Visual Studio Express that does not include the redist.exe so I could not deploy it on another machine). Also, note that the makefile has to embed a manifest into the dll using the undocumented options of the "mt" utility.
  

top of page

DOWNLOAD

The project page at SourceForge is http://sourceforge.net/projects/wnjn . There you will be directed to the CVS or to distribution (Zip file).

top of page

PROJECT MANAGEMENT

If anyone is interested in managing the project I'd also be happy to hand it over.

top of page