mrdgan / zemberek-nlp

Turkish Nlp libraries

Zemberek-NLP

Zemberek-NLP provides basic Natural Language Processing tools for Turkish. Code and API is not compatible with the old Zemberek2 project. Please note that all code and APIs are subject to change drastically until version 1.0.0

Latest version is 0.11.1 (April 13th 2017). Change Log

FAQ

Please read the FAQ for common questions.

Usage

Maven

Add this to pom.xml file

<repositories>
    <repository>
        <id>ahmetaa-repo</id>
        <name>ahmetaa Maven Repo on Github</name>
        <url>https://raw.github.com/ahmetaa/maven-repo/master</url>
    </repository>
</repositories>

And dependencies (For example morphology):

<dependencies>
    <dependency>
        <groupId>zemberek-nlp</groupId>
        <artifactId>morphology</artifactId>
        <version>0.11.0</version>
    </dependency>
</dependencies>

Jar distributions

Google docs page page contains jar files for different versions.

[module-jars] folder contain all zemberek modules as separate jar files. [zemberek-all-VERSION.jar] contains all zemberek modules. [dependencies] folder contains other dependencies suchas Google Guava.

Examples

Turkish-nlp-examples contains a maven java project with small usage examples.

Modules

Core

Core classes such as special Collection classes, Hash functions and helpers.

Maven artifact id : core

Morphology

Turkish morphological analysis, disambiguation and generation. Documentation

Maven artifact id : morphology

Tokenization

Turkish Tokenization and sentence boundary detection. Documentation

Maven artifact id : tokenization

Language Identification.

Allows fast identification of text language. Documentation

Maven artifact id : lang-id

Language modeling

Provides a language compression algorithm. Documentation

Maven artifact id : lm

Normalization

Provides spell checker and suggestion functions. Documentation

Maven artifact id : normalization

Known Issues and Limitations

• Project requires Java 8.
• Currently word and sentence parse module operations generates parse graph with each initialization. So each run in the system takes some seconds.
• Morphological parsing does not work for some obvious and frequent words.
• Morphological disambiguation is working less accurate then expected.
• Morphological generation may not work for some obvious Stem-Suffix combinations.
• Please see issues section for further issues and feel free to create new ones.

License

Code is licensed under Apache License, Version 2.0

Acknowledgements

Please refer to contributors.txt file.