Public repository with code for a computational investigation of the teaching method "inventive spelling". This is a teaching method used in reading and writing skill acquisition in primary schools. In germanophone countries, this method is known as the Reichen-method or simply as Lesen durch Schreiben.

We propose a computational model based on two recurrent neural networks that allows to imitiate certain aspects of the inventive spelling.

Overview of computational model for reading and writing acquisition based on inventive spelling.

Set up a new conda env:

conda create --name invspell python=3.6

Install the requirements and the package in editable mode:

pip install -r requirements.txt
pip install -e .

For example, to train the model on the childlex data, run:

python scripts/run.py --data_dir ${PWD}/data --epochs 250 --print_step 1 --task childlex \
--batch_size 2500 --save_model 125 --test_size 0.05 --learn_type normal --reading True \
--optimization Adam --dropout 0.5