Contextualized Embeddings Encode Monolingual and Cross-lingual Knowledge of Idiomaticity

Installation:

• To run the experiments, install python3. Then, you need to install the following packages:

python -m pip install -r requirements.txt

---

Running experiments:

You can find the code for each experiment in each sub-directory in the Source Code directory.

• BERT
• RoBERTa
• mBERT
• RuBERT
• LAYERS
• Cross-lingual

---

Datasets:

Pre-processed datasets for our experiments are available in Datasets directory. If you wish to go through the process of preparing them by yourself, use the follwoing scrip files:

VNC-Tokens dataset:

• The TEST portions of this dataset is available in the Datasets directory.

Russian Datasets:

• This dataset is in the Datasets directory but if you wish to construct desired dataset, you can get the russian dataset from (https://github.com/kaharodnik/Ru_idioms). We only use texts from Russian Wikipedia and a context of up to 300 characters to the left and right of the target expression. You can achive this by running following in the Russian Dataset Prepration directory:

python Wiki.py

Datasets for our experiments are ready by this step.

---

If you have any further questions, please do not hesitate to contact me.