Datasets and source code for the paper ID10M: Idiom Identification in 10 Languages.

Please consider citing our work if you use data and/or code from this repository.

Bibtex

@inproceedings{tedeschi-etal-2022-id10m,
    title = "{ID}10{M}: Idiom Identification in 10 Languages",
    author = "Tedeschi, Simone  and
      Martelli, Federico  and
      Navigli, Roberto",
    booktitle = "Findings of the Association for Computational Linguistics: NAACL 2022",
    month = jul,
    year = "2022",
    address = "Seattle, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.findings-naacl.208",
    doi = "10.18653/v1/2022.findings-naacl.208",
    pages = "2715--2726",
    abstract = "Idioms are phrases which present a figurative meaning that cannot be (completely) derived by looking at the meaning of their individual components.Identifying and understanding idioms in context is a crucial goal and a key challenge in a wide range of Natural Language Understanding tasks. Although efforts have been undertaken in this direction, the automatic identification and understanding of idioms is still a largely under-investigated area, especially when operating in a multilingual scenario. In this paper, we address such limitations and put forward several new contributions: we propose a novel multilingual Transformer-based system for the identification of idioms; we produce a high-quality automatically-created training dataset in 10 languages, along with a novel manually-curated evaluation benchmark; finally, we carry out a thorough performance analysis and release our evaluation suite at https://github.com/Babelscape/ID10M.",
}

In a nutshell, ID10M is a novel framework consisting of systems, training and validation data, and benchmarks for the identification of idioms in 10 languages.

Training and Development Data

Here you can find the automatically-created data that we used to train and evaluate our systems:

Language	Train	Dev	Sentences	Tokens	Idioms	B	I	O	Literal
Chinese	train_chinese.tsv	dev_chinese.tsv	9543	244422	1301	5272	3823	235327	3918
Dutch	train_dutch.tsv	dev_dutch.tsv	20935	548872	189	4530	10543	533799	16366
English	train_english.tsv	dev_english.tsv	37919	1199492	4568	10102	19884	1169506	27408
French	train_french.tsv	dev_french.tsv	35588	939161	188	12112	25248	901801	23238
German	train_german.tsv	dev_german.tsv	26963	722109	819	8311	11500	702298	18488
Italian	train_italian.tsv	dev_italian.tsv	29523	813445	452	8768	12353	792324	20506
Japanese	train_japanese.tsv	dev_japanese.tsv	6388	211437	165	2534	1662	207241	3852
Polish	train_polish.tsv	dev_polish.tsv	36333	862265	648	12971	14364	834930	22467
Portuguese	train_portuguese.tsv	dev_portuguese.tsv	30942	764017	559	5824	8871	749322	24816
Spanish	train_spanish.tsv	dev_spanish.tsv	28647	648776	1229	9994	13927	624855	17851

We underline that the just reported training data are automatically produced, hence they may contain errors. For further details about the produced silver data, please refer to the Section 3.1 of the paper.

Test Data

Here you can find the test sets used to evaluate our systems:

Language	Test	Sentences	Tokens	Idioms	B	I	O	Seen	Unseen	Literal
English	test_english.tsv	200	3287	142	159	373	2755	62	80	41
German	test_german.tsv	200	4529	111	181	377	3971	71	40	19
Italian	test_italian.tsv	200	5043	139	155	271	4617	87	52	48
Spanish	test_spanish.tsv	200	2240	78	133	348	1759	19	59	66

For further details about the produced test data refer to the Section 3.2 of the paper.

Pretrained Models

The pretrained models are available here:

• english.pt
• german.pt
• italian.pt
• spanish.pt

For further details about the neural architecture refer to the Section 3.3 of the paper.

How To Use

To run the code, you just need to perform the following steps:

1. Install the requirements:

   pip install -r requirements.txt
   

   The code requires python >= 3.8, hence we suggest you to create a conda environment with python 3.8.

2. To train or test the system, you just need to run the main.py file

   python src/main.py
   

   Once the program is started it asks you to specify if you want to train or test the system, the desired language, etc.

   If you train the system, model checkpoints will be saved in the src/checkpoints folder. Otherwise, if you evaluate your system, the script will load the model checkpoints stored in the src/checkpoints folder.

License

ID10M is licensed under the CC BY-SA-NC 4.0 license. The text of the license can be found here.

We underline that the source from which the raw sentences have been extracted is Wiktionary (wiktionary.org) and the BIO annotations identifying idiomatic expressions have been produced by Babelscape.

Acknowledgments

We gratefully acknowledge the support of the ERC Consolidator Grant MOUSSE No. 726487 under the European Union’s Horizon 2020 research and innovation programme (http://mousse-project.org/) and the support of the ELEXIS project No. 731015 under the European Union’s Horizon 2020 ([http://mousse-project.org/](http://mousse-project.org/)).