kadingir

This is an open source implementation of

Oshikiri, T., Fukui, K., Shimodaira, H. (2016). Cross-Lingual Word Representations via Spectral Graph Embeddings. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics. (To appear)

Contents

• src/ : Source codes (C++ & Rcpp version)
• cpp/ : Source codes (C++ only version)
• experiments/ : Code used in the experiments
• tools/

Implemented methods

• CL-LSI [Littman+ 1998]
• Eigenwords [Dhillon+ 2012] [Dhillon+ 2015]
  • One-Step CCA (OSCCA)
  • Two-Step CCA (TSCCA)
• Eigendocs
• Cross-Lingual Eigenwords (CL-Eigenwords) [Oshikiri+ 2016]

Required datasets for experiments

• Reuters Corpora (RCV1) http://trec.nist.gov/data/reuters/reuters.html
• Europarl corpus http://www.statmt.org/europarl/
• questions-words.txt https://code.google.com/archive/p/word2vec/
• The WordSimilarity-353 Test Collection http://www.cs.technion.ac.il/~gabr/resources/data/wordsim353/
• text8 corpus http://mattmahoney.net/dc/text8.zip

Submodules

• ntessore/redsvd-h - GitHub
• docopt/docopt.cpp - GitHub
• word2vec - Google Codes

References

• Oshikiri, T., Fukui, K., Shimodaira, H. (2016). Cross-Lingual Word Representations via Spectral Graph Embeddings. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics. (To appear)
• Dhillon, P., Rodu, J., Foster, D., and Ungar, L. (2012). Two step cca: A new spectral method for estimating vector models of words. In Langford, J. and Pineau, J., editors, Proceedings of the 29th International Conference on Machine Learning (ICML-12), ICML ’12, pages 1551–1558, New York, NY, USA. Omnipress.
• Dhillon, P. S., Foster, D. P., and Ungar, L. H. (2015). Eigenwords: Spectral word embeddings. Journal of Machine Learning Research, 16:3035–3078.
  • Source code : paramveerdhillon/swell - GitHub
  • Output : Codes | Paramveer Dhillon
• Littman, M. L., Dumais, S. T., & Landauer, T. K. (1998). Automatic cross-language information retrieval using latent semantic indexing. Cross-Language Information Retrieval, 51–62.

License

GPL v3