CogCompTime: A Java repository from CogComp

Description

CogCompTime is a state-of-the-art package for understanding time in natural language text. It can automatically (1) extract time expressions (e.g., yesterday and next Monday), (2) convert time expressions to computer-understandable format (e.g., 2018-05-31), and (3) extract verb events on the main-axis of stories and the temporal relations between them (e.g., event1 is before/after event2). This is the backend system of the online demo here.

Prerequisites

This package was tested on Ubuntu Mate 16.04.
maven
Gurobi . The environment variables GUROBI_HOME and GRB_LICENSE_FILE setup in path, as required by Gurobi. Gurobi would typically require adding the following lines to .bashrc or .bash_profile (please refer to Gurobi's installment instructions for details.)
```
export GUROBI_HOME=/opt/gurobi652/linux64 # "652" is the version number 6.5.2; we also tested with 8.1.0
export PATH=$PATH:$GUROBI_HOME/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$GUROBI_HOME/lib
export GRB_LICENSE_FILE=$GUROBI_HOME/gurobi.lic
```
Note that if you can also run it without Gurobi, so that the transitivity constraints are not enforced during inference (which may be problematic in some applications)

Compile

All the following commands should be run from the root dir of the project, i.e., CogCompTime/.

git clone git@github.com:CogComp/CogCompTime.git
cd CogCompTime
sh scripts/install.sh # to install gurobi.jar into the local .m2 repo
mvn compile

Resources

Please also download WordNet-3.0 and TemProb.txt from my another repo TemProb-NAACL18 and put them into CogCompTime/data/ before you move forward.

Example Usage

If no error messages pop up, you're can move forward by

sh scripts/example.sh

Again, if no errors are encountered, you should be able to see the output in data/output/GeorgeLowe-long.html (open it via any internet browser and you will see the temporal graph you see in our online demo) and data/output/GeorgeLowe-long.txt (the timeline you see in our online demo). Compare them with the reference output data/output/REF-GeorgeLowe-long to see if everything is the same.

Trouble shooting

The first time you run the code, one component of CogCompTime will automatically download some resource files from CogCompGroup servers (specifically, that component is illinois-nlp-pipeline. Those resource files will be downloaded and saved into your $HOME/.cogcomp-datastore. If you have trouble downloading them, you can try this back-up link (300MB) and unzip it into $HOME/.cogcomp-datastore.
There's a known issue in illinois-nlp-pipeline described here. That is, when you use illinois-nlp-pipeline to process semantic role labeling (SRL), it's possible that it requires more than 32GB of memory. CogCompTime indeed relies on that SRL component. So if you do see this issue, you can try CogCompTime on a machine with more than 32GB in memory.

Citation

Please kindly cite the following paper: Qiang Ning, Ben Zhou, Zhili Feng, Haoruo Peng and Dan Roth, CogCompTime: A Tool for Understanding Time in Natural Language EMNLP (Demo Track) (2018) (pdf)

@inproceedings{NZFPR18,
    author = {Qiang Ning and Ben Zhou and Zhili Feng and Haoruo Peng and Dan Roth},
    title = {CogCompTime: A Tool for Understanding Time in Natural Language},
    booktitle = {EMNLP (Demo Track)},
    month = {11},
    year = {2018},
    address = {Brussels, Belgium},
    publisher = {Association for Computational Linguistics},
    url = "http://cogcomp.org/papers/NZFPR18.pdf",
}