CMU ARK Twitter Part-of-Speech Tagger
Pull request Compare This branch is 176 commits behind brendano:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


CMU ARK Twitter Part-of-Speech Tagger v0.2

Basic usage

Requires Java 6.  To run the tagger:

    ./ -input example_tweets.txt -output tagged_tweets.txt

The output should match tagged_tweets_expected.txt.
Advanced usage

We include a pre-compiled .jar of the tagger so you hopefully don't need to
compile it.  But if you need to recompile, do:

To train and evalute the tagger, see:
  scripts/ and scripts/

 *  is the script you probably want
 * lib/          has runtime dependencies and trained model
 * lib_build/    has buildtime dependency
 * lib_twitter/  has Twitter's text commons annotations API
 * scripts/      helps you build and run
 * src/          has the actual source code (mostly java, and one bit of scala)

The lib_twitter stuff is optional; if you want to use it, you may wish to
update to a newer version of the library.  A demo can be invoked as:
  scripts/ -Xmx1g edu.cmu.cs.lti.ark.tweetnlp.twitter_anno.CMUPOSTagger

This tagger is described in the following paper.  Please cite it if you write a
research paper using this software.

  Part-of-Speech Tagging for Twitter: Annotation, Features, and Experiments
  Kevin Gimpel, Nathan Schneider, Brendan O'Connor, Dipanjan Das, Daniel Mills,
  Jacob Eisenstein, Michael Heilman, Dani Yogatama, Jeffrey Flanigan, and Noah A. Smith
  In Proceedings of the Annual Meeting of the Association for Computational
  Linguistics, companion volume, Portland, OR, June 2011.

The software is licensed under Apache 2.0 (see LICENSE file).

Version 0.2 of the tagger differs from version 0.1 in the following ways:

* The tokenizer has been improved and integrated with the tagger in a single Java program. 

* The new tokenizer was run on the 1,827 tweets used for the annotation effort and the 
annotations were adapted for tweets with differing tokenizations. The revised annotations 
are contained in a companion v0.2 release of the data (twpos-data-v0.2).

* The tagging model is trained on ALL of the available annotated data in twpos-data-v0.2. 
The model in v0.1 was only trained on the training set.

* The tokenizer/tagger is integrated with Twitter's text commons annotations API.

Please contact Brendan O'Connor ( and Kevin Gimpel ( 
if you encounter any problems.