This system has participated in the PARSEME MWE shared task 1.1 on automatic identification of verbal multiword expressions (VMWEs). See shared task results here

The system is a modified arc-standard transition model for the identification of the verbal multi-word expressions. It's a modified version of ATILF-LLF System of Saied, H. A., Constant, M., and Candito, M. (2017), we combining a data-independent dimension reduction with convolutinal neural networks.

Corpora: Data sets were provided by PARSEME MWE Shared task 1.1.

Requirements

• python 2.7
• keras with a tensorflow backend
• scikit-learn 0.19 or higher
• numpy

SourceCode: The source code is in the Src folder

Run To run the code pick one of the files in the Scripts folder. There are multiple options to run the code, you can run the code with only SVM, only CNN, a CNN with a SVM and also changing much parameters, e.g. number of epoches, number of dimensions after feature reduction, ...

Results: results of experiments on 19 languages with multiple reports reflecting the progress of the identification process.