This is the code and data for the sentence mover's similarity metrics.
The code is based on the Word Mover's Distance implementation from this repo and this paper. Make sure you follow installation info for this repo before running SMS code.
Input should be a file of tab-separated reference and hypothesis texts, one pair per line.
The file should be passed in, along with the word embedding type (glove or elmo) and the metric type (wms, sms, or s+wms).
Output will be written to the input's directory, labeled with the embedding and metric choices.
e.g., python smd.py input.tsv glove sms will calculate the SMS numbers for the file input.tsv using GloVe embeddings. The output will be written to input_glove_sms.out.