MCPredictor

Experiment code for:

Long Bai, Saiping Guan, Jiafeng Guo, Zixuan Li, Xiaolong Jin, and Xueqi Cheng. "Integrating Deep Event-Level and Script-Level Information for Script Event Prediction", EMNLP 2021

1. Corpus

Corpus can be found in LDC: https://catalog.ldc.upenn.edu/LDC2011T07 , since this dataset use documents from year 1994 to 2004, please use at least the second edition.

2. MCNC dataset

MCNC dataset processing code can be found here: https://mark.granroth-wilding.co.uk/papers/what_happens_next/ .

Please use python2.7 environment to run this code.

Please follow README.md and bin/event_pipeline/rich_docs/gigaword.txt to construct the dataset. bin/entity_narrative/eval/experiments/generate_sample.sh is used to generate dev/test dataset.

2.1 Modification of Granroth-Wilding's code

Please let me know if I forget any changes.

2.1.1 modify bin/run

Since some computer run in other languages, which may raise error when using JMNL, please set system language to english:

java -classpath $BUILD_DIR:$DIR/../src/main/java:$DIR/../lib/* \
    -DWNSEARCHDIR=$DIR/../models/wordnet-dict \
    -Duser.language=en \
    $*

2.1.2 modify bin/event_pipeline/1-parse/preprocess/gigaword/gigaword_split.py

It is recommended to use absolute directory #!<code-dir>/bin/run_py instead of #!../run_py

It is recommended to use lxml engine in BeautifulSoup: soup = BeautifulSoup(xml_data, "lxml")

2.1.3 modify directories

Data directories in following files should be changed to user's data directory:

• bin/event_pipeline/config/gigaword-nyt
• bin/event_pipeline/rich_docs/gigaword.txt
• bin/entity_narrative/eval/experiments/generate_sample.sh

2.1.4 modify bin/event_pipeline/1-parse/candc/parse_dir.sh

Change to : ../../../run_py ../../../../lib/python/whim_common/candc/parsedir.py $*

2.1.5 modify unavailable URLs in lib/

C&C tool: https://github.com/chbrown/candc

OpenNLP: http://archive.apache.org/dist/opennlp/opennlp-1.5.3/apache-opennlp-1.5.3-bin.tar.gz

Stanford-postagger: https://nlp.stanford.edu/software/stanford-postagger-full-2014-01-04.zip

2.1.6 extract tokenized documents

Since original texts are needed, <data_dir>/gigaword-nyt/tokenized.tar.gz should be decompressed into the same directory. Replace <data_dir> with the place you want to store the extracted data.

2.1.7 build java files

Change directory to the root of this code, then:

mkdir build
javac -classpath <code_root>/build:<code_root>/src/main/java:<code_root>/lib/* -d build/ src/main/java/cam/whim/opennlp/Tokenize.java
javac -classpath <code_root>/build:<code_root>/src/main/java:<code_root>/lib/* -d build/ src/main/java/cam/whim/opennlp/Parse.java
javac -classpath <code_root>/build:<code_root>/src/main/java:<code_root>/lib/* -d build/ src/main/java/cam/whim/opennlp/StreamEntitiesExtractor.java
javac -classpath <code_root>/build:<code_root>/src/main/java:<code_root>/lib/* -d build/ src/main/java/cam/whim/opennlp/Coreference.java
javac -classpath <code_root>/build:<code_root>/src/main/java:<code_root>/lib/* -d build/ src/main/java/cam/whim/opennlp/Tokenize.java

Replace <code_root> with the absolute path of the root of this code.

Notice: if you want to run PMI (i.e., Chambers and Jurafsky, 2008), please also build java files in src/main/java/cam/whim/narrative/chambersJurafsky.

3. Installation

Use command pip install -e . in project root directory.

Use command pip install -r requirements.txt to install dependencies.

Environment: python>=3.6.

4. Preprocess

Use command python experiments/preprocess.py --data_dir <data_dir> --work_dir <work_dir> to preprocess data. Following arguments should be specified:

• --data_dir: the directory of MCNC dataset
• --work_dir: the directory of temp data and results

On my working platform, It takes about 7 hours to generate the single chain train set, and takes about 10 hours to generate the multi chain train set. Please make sure that the process will not be interrupted.

5. Training

train mcpredictor:

python experiments/train.py --work_dir <work_dir> --model_config config/mcpredictor-sent.json --device cuda:0 --multi

train scpredictor:

python experiments/train.py --work_dir <work_dir> --model_config config/scpredictor-sent.json --device cuda:0

6. Testing

test mcpredictor:

python experiments/test.py --work_dir <work_dir> --model_config config/mcpredictor-sent.json --device cuda:0 --multi

test scpredictor:

python experiments/test.py --work_dir <work_dir> --model_config config/scpredictor-sent.json --device cuda:0

7. Citation

If you find the resource in this repository helpful, please cite

@inproceedings{bai-etal-2021-integrating,
    title = "Integrating Deep Event-Level and Script-Level Information for Script Event Prediction",
    author = "Bai, Long  and Guan, Saiping  and Guo, Jiafeng  and Li, Zixuan  and Jin, Xiaolong  and Cheng, Xueqi",
    booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2021",
    address = "Online and Punta Cana, Dominican Republic",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.emnlp-main.777",
    pages = "9869--9878",
}