Prerequisites

• This package was tested on Ubuntu Mate 16.04.
• Have maven installed in your system
• Have Gurobi6.5.2 installed in your system and have the environment variables GUROBI_HOME and GRB_LICENSE_FILE setup in your path, as required by Gurobi. Gurobi would typically require adding the following lines to your .bashrc or .bash_profile, but please refer to Gurobi’s installment instructions for details.

  export GUROBI_HOME=/opt/gurobi652/linux64
  export PATH=$PATH:$GUROBI_HOME/bin
  export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$GUROBI_HOME/lib
  export GRB_LICENSE_FILE=$GUROBI_HOME/gurobi.lic
  

  Please contact the author (qning2@illinois.edu) if you cannot find the official package for version 6.5.2. Note that if you only need the system output files, you can move forward without gurobi.

Reproduce NAACL’18 Results

All the following commands should be run from the root dir of the project, i.e., TemProb-NAACL18/.

git clone git@github.com:qiangning/TemProb-NAACL18.git
cd TemProb-NAACL18
tar xf data/TemProb.txt.tar.gz -C data
sh scripts/mvn_install.sh
mvn compile

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

mkdir logs
mkdir logs/Awareness
mkdir logs/Awareness/CompareStateoftheArt_AllEventTimex
mkdir logs/Awareness/CompareStateoftheArt_PartialTBDense
mkdir logs/eval_corpus_prior
sh scripts/RunThis_All.sh > RunThis_All_log.txt

Note: Since Github is limiting the bandwidth for large files (our data/TemProb.txt is a large file), it’s very likely that you will see an error saying that TemProb.txt fails to be downloaded. In that case, please go to here and find the backup link to download it.

Again, if no errors are encountered, you should now have all the tables reported in the paper. Take a look at scripts/RunThis_All.sh and it should be rather easy to understand. For example,

• Table 3: sh scripts/eval_corpus_prior_batch_table3.sh
• Table 4: sh scripts/eval_prior_causality_table4.sh
• Table 5, line 1: mvn exec:java -Dexec.mainClass=test.global_ee_test -Dexec.args=”table5_line1” > logs/Table5_line1.txt
• …

One exception is Table 7, which was not generated automatically, but we have included the numbers in logs_refs/DONOTDELETE_Table7.txt. Another exception is the description right above Table 5 (these numbers couldn’t fit into Table 5, so we had to put them in the text). These numbers can be found in logs_refs/DONOTDELETE_Table5_improvement.

If you met with errors while evaluating the temporal awareness scores of each system, probably it’s due to python 2 vs 3 issues. Please change corresponding python commands to be python2.

Where do we find the logs?

Standard metrics (prec, rec, and F1):

• Table 3: logs/eval_corpus_prior/th0.x_sentsplit.txt where x=5~9.
• Table 4: logs/eval_corpus_prior/CausalDirection.txt.
• Table 5: logs/Table5_line*.txt where *=1~3.
• Table 6: logs/Table6_line*.txt where *=1~3.
• Table 7: logs_refs/DONOTDELETE_Table7.txt.
• Table 8: logs/Table8_line*.txt where *=1~6. In addition, logs/Table8_line3_proposed_on_partialTBDense_detail.txt contains more detailed information of System 3 in Table 6: there are performances of each of the 9 documents in the test split of TBDense, and both local and global performances. Please refer to the Sec. 2 of the paper for descriptions of local and global. However, note that the reported values in Table 6 are from global.

Temporal awareness scores:

• Table 6: logs/Awareness/table6_line*.txt where *=1~3.
• Table 8 (on partial TBDense, i.e., the top part of the table): logs/Awareness/CompareStateoftheArt_PartialTBDense.
• Table 8 (on full TBDense, i.e., the bottom part of the table): logs/Awareness/CompareStateoftheArt_AllEventTimex.

Note that sometimes the awareness evaluations are not finished by itself (due to an unknown instability in the awareness evaluation tools provided by TempEval3). You can go to the log of the standard metrics and locate a line starting with sh scripts/evaluate_general_dir.sh (usually at the bottom of each file). For example, if you see logs/CompareStateoftheArt_PartialTBDense/naacl.txt is incomplete. Since that corresponds to the 3rd line of Table 8, go to logs/Table8_line3_proposed_on_partialTBDense.txt and you will see a line I intentionally created for this situation:

sh scripts/evaluate_general_dir.sh output/Awareness/gold output/Awareness/CompareStateoftheArt_PartialTBDense/naacl naacl Awareness/CompareStateoftheArt_PartialTBDense

Run this from TemProb-NAACL18/ and you will see that logs/CompareStateoftheArt_PartialTBDense/naacl.txt is updated and complete now (may take a few seconds to complete).

I have also put the original logs I generated into logs_refs/ for your reference.

Where do we find the system outputs?

• Table 6: output/Awareness/global/table6_line*/ where *=1~3.
• Table 8 (on partial TBDense):
  • Line 1: output/Awareness/CompareStateoftheArt_PartialTBDense/caevo/
  • Line 2: output/Awareness/CompareStateoftheArt_PartialTBDense/emnlp/
  • Line 3: output/Awareness/CompareStateoftheArt_PartialTBDense/naacl/, or output/Awareness/global/table8_line3/.
• Table 8 (on full TBDense):
  • Line 4: output/CAEVO_on_TBDense
  • Line 5: output/EMNLP_on_TBDense
  • Line 6: output/EMNLPAugmentedByNAACL_on_TBDense
  • Note these three outputs are not generated by this package. Specifically, Line 4 comes from here. Line 5 comes from here. And Line 6 comes from here.

Citation

Please kindly cite the following paper: Qiang Ning, Hao Wu, Haoruo Peng, Dan Roth, “Improving Temporal Relation Extraction with a Globally Acquired Statistical Resource”, NAACL 2018 (pdf)

@inproceedings{NingWuPeRo18,
    author = {Qiang Ning and Hao Wu and Haoruo Peng and Dan Roth},
    title = {Improving Temporal Relation Extraction with a Globally Acquired Statistical Resource},
    booktitle = {NAACL},
    month = {6},
    year = {2018},
    publisher = {Association for Computational Linguistics},
    url = "http://cogcomp.org/papers/NingWuPeRo18.pdf",
}