This script ranks lexical substitution candidates based on the AddCos metric (Melamud et al. A simple word embedding model for lexical substitution, 2015). The substitution candidates are taken to be the PPDB paraphrases for the target word.
This script is specifically for test sentences/instances that have been POS-tagged, with
each token in the format word_POS. Accordingly the word embedding model used for
lexical substitution should have its vocabulary in the same format, word_POS.
Test sentences should be stored in a text file with the following (tab-separated) format,
where idx indicates the index (starting from 0) of the target word to be substituted
within in the sentence, and tgt.P indicates the target with a simple part of speech
label, i.e. bug.N:
<tgt.P> <sentence_id> <idx> <sent_NN toks_NNS ,_, tagged_VBD ...>
python lexsub_addcos_ppdb.py <LEXSUBFILE> <PPDBFILE> <EMBEDDINGFILE> <OUTFILE> [<MINSCORE>]
where LEXSUBFILE is the tab-separated file containing (pos-tagged) test instances in
the above format; PPDBFILE is a PPDB file (i.e. XL or XXL) as available for download
from the PPDB site, EMBEDDINGFILE is a file
containing word embeddings in gensim word2vec format, and OUTFILE is the name of the
destination file to write the results.
Optionally specify a MINSCORE which is the minimum PPDB score for a paraphrase to be
considered a potential substitute.
This code provides the implementation of the AddCos metric as used in the paper:
Marianna Apidianaki, Guillaume Wisniewski, Anne Cocos, and Chris Callison-Burch. 2018. Automated Paraphrase Lattice Creation for HyTER Machine Translation Evaluation. In Proceedings of NAACL 2018 (Short Papers). New Orleans, LA.
If you are looking for a full implementation of the hytera pipeline from that paper (in which this code is just part of the pipeline), you can find it here.