MSCorpus

Weka's SMO classifier for beat gestures. The training set is an annotated corpus of videos. The considered features are:

• Regarding the video:
  • gestures (beat, semantic or absence of gesture)
• Regarding the transcription of the speech:
  • communicative dimensions:
    • thematicity
    • focalization
    • emphasis
    • perspective
  • surface syntax
  • Part of Speech

The corpus is available in Github (https://github.com/carlaTV/MSCorpus/blob/master/wekaTest/CorpusAligned_BinaryBalanced.arff) and in Zenodo (https://sandbox.zenodo.org/record/223199#.W1CdLtgzaqQ)

The java code with the classifier is an adaption of the codes found in https://www.cs.umb.edu/~ding/history/480_697_spring_2013/homework/WekaJavaAPITutorial.pdf and https://stackoverflow.com/questions/33760145/weka-how-to-predict-new-unseen-instance-using-java-code. Originally, several classifiers were tested using the Weka GUI. The Java code is adapted to make predictions with new sentences.

How to run the classifier

Execute the code in MSCorpus/wekaTest/src/wekatesting.java

Processing of the outputs

The output of the classifier implemented in Java can be processed with R using MSCorpus/ProcessOutputPredictor.R