Skip to content


Folders and files

Last commit message
Last commit date

Latest commit



62 Commits

Repository files navigation

A simple transition-based system for the identification of the verbal multword expressions.

This system has participated in the competition of MWE Shared task - EACL 2017 (Valencia, Spain).

Corpora: Data sets proposed by MWE Shared task - EACL 2017 (Valencia, Spain).

Results: results of experiments on 18 languages with multiple reports reflecting the progress of the identification process.

SourceCode: The folder Src

Overview of the source code:

The identifier runs using identify function (

The input of the system is a set of configurations:

1- the language you want to test and its feature group (This should be represented as a json file in "Experiments/xp").

NB configuration files with our preferred feature group are available in 'Experiments/Langs'.

2- The mode of evaluation:

A- Debug mode: used during development for faster execution time;

B- Test Mode: using the full version of Shared task train and test data sets;

C- Cross validation mode: A 5-fold cross validation evaluation over the Shared task train data (No shuffle used).

The Output is list of F-scores for the identification and the categorization (over each category) of VMWEs.

The actual version is very light (No reports, no analysis and no code for additional experiments).

A good start to understand the code is to start with the identifier script and to follow the code.

**NB:**The folder script contain multiple secondary scripts used for 1- creating a baseline system for identifying Sharedtask MWE using string matching techniques 2- and which were used to generate the automatic annotation (POS tags and syntaxic annotations) for languages whose annotations where manulally annotated (FR, CS, HU, and PL).


Al Saied, H., Constant, M., & Candito, M. (2017). The atilf-llf system for parseme shared task: a transition-based verbal multiword expression tagger. In Proceedings of the 13th Workshop on Multiword Expressions (MWE 2017) (pp. 127-132).