Cross-Register Projection for Headline Part of Speech Tagging

POSH: The POS-tagged HeadlIne corpus was created for the paper “Cross-Register Projection for Headline Part of Speech Tagging” published in EMNLP 2021.  This dataset contains headlines with gold annotated POS tags.

The GSCh evaluation set is here: GSCh/gsc-headline-gold.test.conllu

The smaller evaluation set of GSC headlines sampled uniformly at random (described in section 2.3) is here: gold_unconstrained_headlines/unifrand_gsc.test.conllu

The POS-tagged NYT headlines described in section 2.3 are not shared directly as this text was drawn from the New York Times Annotated Corpus (LDC2008T19), and subject to license constraints.  However, if you have access to and have untarred LDC2008T19, you can recover this evaluation set with:

    TAG_PATH="./unifrand_onlynyt.tags.json"  # mapping from NYT headline span to gold POS tag

    python build_gold_nyt_headlines.py --nyt_dir /PATH/TO/ANNOTATED/NYT/CORPUS/ --tag_path ${TAG_PATH} --num_proc 4

Increase the argument to --num_procs to process more shards from the NYT corpus in parallel and reduce build time.

Under GSCproj we also share the GSCproj folds which we used to train and validate our models.  These are not gold POS tags, and are shared purely for reproducibility sake.