FlexTag: A Highly Flexible PoS Tagging Framework
FlexTag is a PoS tagger which exposes the feature space to the user enabling him/her to fully customize the feature space. Models that perform satisfiyling are furthermore easily persisted and can be used as standalone component.
Everyone who feels that the existing, re-trainable PoS taggers such as Stanford or TreeTagger do not provide the needed flexiblity i.e. allow modifiyng the feature space but neither feels like implementing an own PoS tagger should consider FlexTag.
FlexTag offers everything for creating and own PoS tagger model and evaluating it on any input data format. Additional to training and persisting of trained models each model is easily evaluate by using out-of-the-box available evaluation methods such as Train-Test or N-fold CrossValidation on any data set.
In the example project you find a runnable example for each of those functionalities if you take a look at those demos:
- ExampleCrossValidation
- ExampleTrainTest
- ExampleTrainStore
FlexTag uses DKPro TextClassification in the background and provides additionally file format readers for many formats commonly used in NLP, see ExampleReaders for a pointer to learn more about the supported file formats. Furthermore, the machine learning algorithm used for training a model is exchangeable, too. Users can choose from CrfSuite, SvmHmm and the classifier provided by Weka for training and evaluating their model, see package de.unidue.ltl.flextag.examples.config for changing the classifier.
Features
Adding of user-defined features
Limitations of user-defined features
FlexTag is on
<dependency>
<groupId>de.unidue.ltl.flextag</groupId>
<artifactId>FlexTag</artifactId>
<version>0.2.0</version>
</dependency>