Serial Parsing in ACT-R With Null elements

This code implements an ACT-R based serial parser with re-analysis using a CCG inspired grammar formalism. The code can also be used to generate priming predictions for a comprehension-to-production priming paradigm.

General components of the code

In most cases, applying SPAWN to other phenomena will not require modifying the following files:

model.py

The model class with ACT-R based methods for memory retrieval and activation updates.

supertagger.py

This includes the algorithsm for parsing, supertagging + combining, and re-analysis.

ccg.py

Implementation of CCG rule application with the six CCG application rules + type raising.

Experiment-specific components of the code

The following files contain code for the specific set of experiments described in chapter 2 of this dissertation. In most cases, applying SPAWN to other phenomena will require modifying these files.

create_declmem.py

Specifying the declarative memory (i.e. "grammar" and "vocabulary").

create_training_dat.py

Templates to create data to train SPAWN models.

train.py

Code to train models on a pre-tokenized corpus where tokens are separated by space. All model hyperparameters are specified in this file.

prime.py

Code to evaluate and "adapt" or "fine-tune" already trained models.