We provide an implementation of $K$-Grassmeans for word sense induction, disambiguation, and representation. Details please refer to our paper:

• Jiaqi Mu, Suma Bhat, and Pramod Viswanath. "Geometry of Polysemy." arXiv preprint arXiv:1610.07569 (2016).

In src/, we provide three scripts:

• induction.py: to get intersections for word sense induction directly from corpus.
• induction-from-file.py: to get intersections for word sense induction from a given set of sentences.
• representation.py: to generate a labeled corpus.

We provide three demos to try:

• demo-induction.sh
• demo-induction-from-file.sh
• demo-representation.sh

When dealing with a training corpus, please chunk a large corpus into smaller files to avoid memory overflow.

You will need to setup following parameters:

• funcWordFile: a list of function words (an example is provided in data/)
• polyListFile: a list of target polysemous words (an example is provided in data/)
• directory: a base directory
• vocabInputFile: an input vocabulary file (an example is provided in data/)
• vecInputFile: a binary vector file (an example is provided in data/)
• corpusPath: a directory to chunked corpus (an example is provided in data/)
• algoPath: an output directory