Rhythm is a nonverbal behavior annotator for virtual characters: given text, it will generate appropriate nonverbal behavior for that text, including eye movement, head movement, hand gestures, and postural changes. The resulting annotated text can be used as a basis for hand-edited animation scripts, or could be directly given to character animation. Rhythm can be seen as a clean reimplementation of earlier academic projects, primarily BEAT.
It is currently in an embryonic state, so do not expect it to do much at the moment.
Rhythm relies on a shallow linguistic analysis of the text, performed using a combination of its own heuristics and some open-source libraries: OpenNLP for statistical natural language processing, and Wordnet (via JWI) to look up word relations.
Most of this processing requires various model files: dictionaries and trained statistical models. For space reasons, most of these models are not included with Rhythm, and need to be downloaded.
OpenNLP model files can be downloaded from http://opennlp.sourceforge.net/models-1.5/.
Rhythm will need models for
sentence detection,
tokenization,
detokenization,
part of speech tagging,
and phrase chunking.
Only the Wordnet database files are required, although it will work fine if you install a full distribution; rhythm just needs to be told where to look for them.
A dictionary used to identify discourse markers (e.g., for topic shifts) is included with Rhythm. You shouldn't need to worry about it unless you want to tweak this file, or want to work in a language other than English.