A Structured Span Selector

https://arxiv.org/pdf/2205.03977.pdf

This repository contains the open-sourced official implementation of our structured span selector paper:

A Structured Span Selector (NAACL 2022).
Tianyu Liu, Yuchen Eleanor Jiang, Ryan Cotterell, and Mrinmaya Sachan

Overall idea

For all span selection tasks (e.g. coreference resolution, semantic role labelling, question answering), we learn the latent context-free grammar of the spans of interest. The search space of spans $O(n^2)$ is reduced to the space of nonterminals $O(n)$.

Installation

First of all:

   git clone https://github.com/lyutyuh/structured-span-selector.git
   cd structured-span-selector

1. Create a virtual environment with Conda

    conda env create -f sss.yml

2. Activate the new environment

    conda activate sss

3. Install genbmm with inside-outside algorithm extension

    pip install git+https://github.com/lyutyuh/genbmm

Obtaining the CoNLL-2012 data

Please follow https://github.com/mandarjoshi90/coref and especially https://github.com/mandarjoshi90/coref/blob/master/setup_training.sh to obtain the {train, dev, test}.english.v4_gold_conll. There are 2802, 343, 348 documents in the {train, dev, test} datasets respectively.

The MD5 values are:

md5sum dev.english.v4_gold_conll
>>> bde418ea4bbec119b3a43b43933ec2ae
md5sum test.english.v4_gold_conll
>>> 6e64b649a039b4320ad32780db3abfa1
md5sum train.english.v4_gold_conll
>>> 9f92a664298dc78600fd50813246aa77

Then, run

python minimize.py ./data_dir/ ./data_dir/ false 

and get the jsonlines files.

Training

    python run.py spanbert_large 0

Evaluating

    python evaluate.py spanbert_large <checkpoint> 0

Citing

If you find this repo helpful, please cite the following version of the paper:

@inproceedings{liu-etal-2022-structured,
    title = "A Structured Span Selector",
    author = "Liu, Tianyu  and
      Jiang, Yuchen  and
      Cotterell, Ryan  and
      Sachan, Mrinmaya",
    booktitle = "Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = jul,
    year = "2022",
    address = "Seattle, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.naacl-main.189",
    pages = "2629--2641",
}