paraphrasing

Noun-Compound Paraphrasing

SemEval 2013 Task 4: Free Paraphrases of Noun Compounds - Given a two-word noun compound, the participating system is asked to produce an explicitly ranked list of its free-form paraphrases. The list is automatically compared and evaluated against a similarly ranked list of paraphrases proposed by human annotators.

We predict for each noun-compound the k best paraphrases, and then learn to re-rank the suggested paraphrases using the SemEval training data. Training script:

usage: semeval_2013_train.py [-h] [--k K] [--minimum_score MINIMUM_SCORE]
                             [--unrelated_threshold UNRELATED_THRESHOLD]
                             train_gold_file language_model_dir patterns_file
                             word_embeddings

positional arguments:
  train_gold_file       a tsv file with gold train paraphrases and their scores
  language_model_dir    the path to the trained language model
  patterns_file         the file with the POS patterns
  word_embeddings       word embeddings to be used for the language model

optional arguments:
  -h, --help            show this help message and exit
  --k K                 the number of paraphrases to retrieve for re-rankning,
                        default = 1000
  --minimum_score MINIMUM_SCORE
                        the minimum score to keep a paraphrase
  --unrelated_threshold UNRELATED_THRESHOLD
                        the minimal score the "is unrelated to" paraphrase has to get to be included

Test script, produces the test paraphrases and evaluates them using the task scorer:

usage: semeval_2013_test.py [-h] [--k K] [--minimum_score MINIMUM_SCORE]
                            [--unrelated_threshold UNRELATED_THRESHOLD]
                            test_gold_file language_model_dir patterns_file
                            word_embeddings reranker

positional arguments:
  test_gold_file        a tsv file with gold test paraphrases and their scores
  language_model_dir    the path to the trained language model
  patterns_file         the file with the POS patterns
  word_embeddings       word embeddings to be used for the language model
  reranker              the pkl file for the trained re-ranker

optional arguments:
  -h, --help            show this help message and exit
  --k K                 the number of paraphrases to retrieve for re-rankning,
                        default = 1000
  --minimum_score MINIMUM_SCORE
                        the minimum score to keep a paraphrase
  --unrelated_threshold UNRELATED_THRESHOLD
                        the minimal score the "is unrelated to" paraphrase has to get to be included

Note that the directory needs to include the code for the SemEval scorer, that can be installed from here.

Download pre-trained ranker from here.