Home

Treat is a toolkit for natural language processing and computational linguistics. It provides a common API for a number of existing tools in C, Ruby and Java for document retrieval, parsing, annotation, and information extraction.

Resources

• Do a one-minute tutorial.
• Read the latest documentation.
• See how to install Treat.
• Learn how to use Treat.
• Help out by contributing to the project.
• View a list of papers about tools included in this toolkit.
• Open an issue and get a quick answer.

Current features

• Text extractors for PDF, HTML, XML, Word, AbiWord, OpenOffice and image formats (Ocropus)
• Text retrieval with indexation and full-text search (Ferret)
• Text chunkers, sentence segmenters, tokenizers, and parsers for several languages (Stanford & Enju)
• Word inflectors, including stemmers, conjugators, declensors, and number inflection
• Lexical resources (WordNet interface, several POS taggers for English, Stanford taggers for several languages)
• Language, date, time and named entity extraction, as well as coreference resolution
• Topic extraction (LDA or Reuters-trained model)
• Simple text statistics (frequency, TF*IDF)
• Serialization of annotated entities to YAML or XML format
• Visualization in ASCII tree, directed graph (DOT) and tag-bracketed (standoff) formats
• Linguistic resources, including full ISO-639-1 and ISO-639-2 support, and tag alignments for five treebanks.

New in version 0.2.4

Here's how to access the Ferret API:

c = Collection 'folder'                                    # Recursively searched for supported file types.
c.index                                                    # Indexes are stored under foler/.index
c.search(:ferret, :q => 'hungary').each do |doc|           # Search for a document with the word "hungary".
  # Do processing/annotation/extraction
  puts doc.file         
end
c.serialize(:yaml, :file => 'test.yaml')                   # Save annotated entities to reopen later.

Pretty sweet, no?

Caveats/Planned features

• The few native Ruby statistics algorithms are slow. Some of the highly recursive code in the core Tree and Entity classes will be ported to inline C.
• XML unserializer is currently broken; it will need to be fixed.
• The API to the Stanford Coreference Resolver and the NER system will need to be integrated with the parser to allow retrieval of coreferences/tags at the same time as the parse tree. Currently, it is only possible to retrieve them separately.
• Tests need to be improved for extractors and processors.
• A faster WordNet API in Java will be interfaced.

License

This software is released under the GPL License and includes software released under the GPL, Ruby, Apache 2.0 and MIT licenses.