License

LingPipe License

Before downloading, please read the terms of:

LingPipe is also available under a commercial license.

Model Licenses

Note that the models themselves are often subject to additional licensing restrictions inherited from the corpora from which they were trained. These restricted models are themselves often available for commercial purposes by contract to the original creator of the data, or through Alias-i. All models, to the best of our knowledge, allow research use.

Download LingPipe

In order to run these models, you must download LingPipe:

This includes a compiled jar file, complete source, javadoc, demos, etc..

Where to Put Models

Models should be downloaded to the following directory in order to ensure the demos that require them may find them: 

$lingpipe/demos/models

Download LingPipe Models

What are Models?

For each language, genre, and training corpus, LingPipe is able to generate a range of models based on task (e.g. named-entity recognition or part-of-speech tagging) and training parameters. These models are then loaded at run time by LingPipe where they provide language-, genre- and task-specific details to the core LingPipe API.

Model Naming Convention

The models are all serialized objects. A name such as ne-en-news-muc6.TokenShapeChunker indicates that the task is named-entity recognition (ner), the language is English (en), the genre of text it applies to is news (news), and the corpora on which it was trained is the MUC-6 corpus (muc6). The file suffix is the (unqualified) name of the class of object represented in the model.

Part of Speech Tagging

English Biomedical: GENIA

English Biomedical: MedPost

English General: Brown

Named Entity Recognition

English News: MUC-6

English Genes: GeneTag

English Genomics: GENIA

Chinese Word Segmentation

Academia Sinica Version SIGHAN 2005