This repo provides the code for the experiments in our paper:
Jakub Dotlacil, Puck de Haan - Parsing model and a rational theory of memory (2021)
All required packages can be found in requirements.txt. The pyactr package can be installed with pip together with the other requirements, or directly from the source files in the pyactr folder. Instructions to do so can be found at https://github.com/jakdot/pyactr.
The folder /examples_and_garden_paths includes code to recreate the results of the parser on the garden path sentences as depicted in the paper. The parser can be run by executing the file run_parser_act.py. All necessary data is present in the folder. Code for creating the plots can be found in plotting.rnw.
The folder /model_nsc_reading_data includes code to recreate the experiments done with the NSC. To parse a sentence with the ACT-R based parsing model, one can first run the script run_parser.py -- which builds parse trees and generates activations. Then, it is possible to run the script run_parser_act.py, which generates reading times given the data. Parameter estimation can be done by running the script parallel_estimation_smaller.py. Estimation is done in a parallelized manner to speed up the process, we recommend the user to make use of this feature, since it decreases the runtime significantly. The subfolder /generate_plots contains the script to plot the results: plotting_csv.rnw and the resulting figures used in the paper. The other subfolder, /parses, contains the PTB parses, parse trees and sentence that have been excluded from the experiments. It should be noted that the content of the actions.csv and blind_actions.csv files are based only on a small subset of the Penn Treebank (PTB) that is publicly available in the nltk package. If you require the entire PTB dataset, you can contact us.
The folder /train_for_nsc includes code to train the parsing model on PTB data.It should be noted that training is optional, since all data needed to recreate the experiments is already present in the /model_nsc_reading_data folder.