Skip to content

ziyin-dl/global-anchor-method

Repository files navigation

The Global Anchor Method for Quantifying Linguistic Shifts and Domain Adaptation

This is the repo for the experiments and collected corpora in the paper `The Global Anchor Method for Quantifying Linguistic Shifts and Domain Adaptation', NeurIPS 2018.

Paper: https://papers.nips.cc/paper/8152-the-global-anchor-method-for-quantifying-linguistic-shifts-and-domain-adaptation
arXiv Category Corpora: https://gitlab.com/vinsachi/arxiv-category-corpora

@inproceedings{
  title={The Global Anchor Method for Quantifying Linguistic Shifts and Domain Adaptation},
  author={Yin, Zi and Sachidananda, Vin and  Prabhakar, Balaji},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2018}
}

The global anchor method is a powerful tool for comparing language usage between different corpora through word vectors. It can be used for

  • Transfer learning: determining whether a model trained on one corpus will transfer to another. If the corpora are very different in terms of their language usage, transfer learning may not perform well.
  • Discover linguistic shifts: one can use this method to determine the rate at which language changes with respect to time.
  • Discover domain variations: one can use this method to discover how language deviates in different domains.

In particular, we showed that the global anchor method is

Here is a short overview of what is in this directory.

Directory What's in it?
equivalence.py In the paper we showed that the alignment and global anchor methods, when viewed as metrics, are equivalent. This provides numerical verification for that claim.
jsd_loss.ipynb This is the script for computing the Jensen-Shannon divergence for the Google ngram corpus. We demonstrate the jsd method does not provide fine-grained details as our method, in particular we show it does not capture the war-effect on English language and literature.
laplacian.ipynb The script of the Laplacian method for language evolution trajectory and topic clustering.
pip_loss.ipynb The script for calculating the PIP loss for Google ngram corpus between every year.
plot.ipynb The script for the war-effect on English language evolution.
validate_equivalence.ipynb The script for empirical validation of the equivalence of the global anchor method and the alignment method.

We also provides a set of processed corpora:

Dataset name Download
Google Books Google Books Ngram Dataset (We have trained a set of word vectors for years between 1800-2008, which can be found here)
arXiv Category Corpora Repository This repo contains text corpora of academic papers separated by category from arXiv submitted between January 2007 - December 2017

Releases

No releases published

Packages

No packages published