This is the repository containing the code for the models in the ACL 2020 paper "Machine Reading of Historical Events".
The datasets are stored in the data folder. We encourage re-using the same splits (in the train, validation and test folders, respectively).
Each folder contains a wotd file corresponding to the Wikipedia On This Day (WOTD) dataset,
and a otd2 file corresponding to the On This Day 2 (OTD2) dataset.
The data for the On This Day (OTD) dataset is available upon request.
The datasets are stored as pandas dataframes in pickle files. Loading them should be as simple as:
In [1]: import pandas as pd
In [2]: data = pd.read_pickle("data/validation/wotd.pkl")
In [3]: data.head()
Out[3]:
YY Event Information
0 1949 Cold War: The western occupying powers approve... [However, hundreds of thousands of East German...
1 1887 Buffalo Bill Cody's Wild West Show opens in Lo... [In December 1872, Cody traveled to Chicago to...
2 1993 An election takes place in Nigeria which is la... [=\nBabangida, then a lieutenant with the 1st ...
3 1601 Long War: Austria captures Transylvania in the... [The Long Turkish War or Thirteen Years' War w...
4 1925 The Government of Turkey expels Patriarch Cons... []
The key key columns are:
YY: The year of the event.Event: The event description.Information: An array, possibly empty, of Contextual Information (CI) sentences extracted for that event.
There are two models, you can find them under the appropriate lstm and boe folders.
Instructions on how to setup each of the models and reproduce experimental results are located there.
@inproceedings{honovich2020historical,
title = "Machine Reading of Historical Events",
author = "Honovich, Or and
Torroba Hennigen, Lucas and
Abend, Omri and
Cohen, Shay B.",
booktitle = "Proceedings of the 58th Conference of the Association for Computational Linguistics",
year = "2020"
}