What's new in MaltParser 1.9.2
Feb 9, 2024
New in MaltParser 1.8 (May 8, 2014)
- New interface to a concurrent "light-weighted" parser.
New in MaltParser 1.7.2 (Jun 17, 2013)
- Add a new option group "multiplanar" that affect the parsing algorithm planar and 2-planar (thanks to Carlos Gómez-Rodríguez)
- Liblinear is now the default learner.Liblinear is now the default learner.
- Liblinear interface: Abort and presentation of an error message when number of training instances * number of classes > Integer.MAX_VALUE (limitation of the size of Java array)
- Liblinear interface: Improvement of the performance of the memory optimization.
- SystemInfo: MaltParser version and build date are not loaded from the manifest-file. Instead version and build date are read from a new file in appdata.
- SystemInfo: Removed the code that tries to find the path to the maltparser-version.jar. The code is not needed anymore and cause problems for MaltParserService users.
- MaltParserService and OptionManager: Fixed problems when loading /appdata/options.xml from the jar-file
- ConfigurationDir: fixed URL-problems when loading files from the parser model .mco-file (thanks to Michael Schmitz)
- New API examples ParseSentence4: shows how to use MaltParserService with a mco-file nested in a jar-file.
New in MaltParser 1.7.1 (Jun 17, 2013)
- Added deterministic action to the parsing algorithms Covington non-projective and Covington projective.
- Fixed a bug with option containers that cause problem when using several parser models via MaltParserService
- Synchronizing the TreeMap that handles the option values.
- Minor adjustments to the documentation.
New in MaltParser 1.7 (Jun 17, 2013)
- Name of jar-file changed to maltparser-.jar, where is the current version of MaltParser.
- Two dependent Java libraries stax-1.2.0.jar and jsr173_1.0_api.jar removed, because the functionality is included in Java 1.6 or later.
- The machine learning package liblinear 1.7 replaced by liblinear 1.8.
- MaltParser 1.7 (and later versions) made available via the official Maven repository.
- Two new options allow_root and allow_reduce added for the Nivre parsing algorithm. These two options replace the older root_handling option from version 1.7 onwards.
- Minor bug fixes in the pseudo-projective parsing component.
- During parsing MaltParser will also search in the java class path(s) to find the parser model (.mco-file).
- Some minor changes throughout the code.
New in MaltParser 1.6.1 (Jun 17, 2013)
- The release of 1.6 reduced the memory usage of the liblinear model, but one drawback was that the training time increased a lot for complex models. This release tries to reduce this training time.
- MaltParserService has been equipped with few more methods that makes it easier to develop client/server applications.
- Some minor changes throughout the code.
New in MaltParser 1.6 (Jun 17, 2013)
- This update address the problem that the liblinear model use a lot of memory.
- First, all features that have zero weights for all candidates are now removed during training.
- Second, all features that have the same weights for all candidates shares the weight array.
- Third, all ending zeros in the weight array are removed.
- Build target (javac.target) are now 1.6. If you want to run MaltParser with Java 5 you have to change javac.target to 1.5 and rebuild MaltParser. We have also tested MaltParser with Java 7 and it works fine.
New in MaltParser 1.5.3 (Jun 17, 2013)
- Fixed a problem with identifying the correct version of malt-.jar when there are several malt-.jar
- Fixed a problem that the symbol table is increasing with new strings during parsing time.
New in MaltParser 1.5.2 (Jun 17, 2013)
- Problem when using MaltParser within a servlet. By making Util.findURL() and Util.findURLinJars() non-static and using getClass().getResource() instead resolves the problem. thanks to Adriane Boyd.
- Fixed a bug when merging Distance, NumOf and Exists features.
- Fixed some minor errors in the User guide. Legal values of the second arguments of the NumOf feature are LDEPS, RDEPS, DEPS.
- Some minor changes throughout the code.