This repository contains the source code to reproduce the results for our experiments, which we presented at SIGIR'18 and SIGIR'19.

The following steps apply for both papers:

• Create an anaconda environment with conda create -n unsupclir python=3.7
• Install requirements: pip install -r requirements.txt
• Set up clef-dataloaders: pip install git+https://github.com/rlitschk/clef-dataloaders.git
• Download the CLEF 2000-2003 test collection from ELRA (ELRA-E0008)

In our paper Unsupervised Cross-Lingual Information Retrieval using Monolingual Data Only, we propose a fully unsupervised framework for ad-hoc cross-lingual information retrieval (CLIR) which requires no bilingual data at all. To reproduce our results, run scripts/reproduce_sigir18.sh. This will download cross-lingual word embeddings and run all experiments.

Bibtex:

@inproceedings{litschko2018unsupervised,
  title={Unsupervised cross-lingual information retrieval using monolingual data only},
  author={Litschko, Robert and Glava{\v{s}}, Goran and Ponzetto, Simone Paolo and Vuli{\'c}, Ivan},
  booktitle={The 41st International ACM SIGIR Conference on Research \& Development in Information Retrieval},
  pages={1253--1256},
  year={2018}
}

In our paper Evaluating Resource-Lean Cross-Lingual Embedding Models in Unsupervised Retrieval, we compare different CLWE spaces in CLIR. To reproduce the results, run scripts/reproduce_sigir19_{clef,europarl}.sh.

Bibtex:

@inproceedings{litschko2019evaluating,
  title={Evaluating resource-lean cross-lingual embedding models in unsupervised retrieval},
  author={Litschko, Robert and Glava{\v{s}}, Goran and Vulic, Ivan and Dietz, Laura},
  booktitle={Proceedings of the 42nd international ACM SIGIR conference on research and development in information retrieval},
  pages={1109--1112},
  year={2019}
}