Code for a seminar work on the subject of learning the English past tense using RNNs.
This is a PyTorch implementation of the Annotated Encoder Decoder, an RNN described in Bahdanau et al. (2015), built using the tutorial at https://bastings.github.io/annotated_encoder_decoder/, for exploring Morris Halle's (1978) hypothesis regarding the encoding of phonemes as feature vectors by a learner of a natural language.
For running the experiment, clone the repository locally and run 304366891_english_past_tense_rnn.py using python 3.6+ (64-bit), after installing the required libraries.
Tested on Windows 10 and Linux, with python 3.6+
Required libraries: (install using python -m pip install <lib_name>)
- pytorch
- numpy
- matplotlib
- scipy
I've compiled installation scripts for Windows and Linux (should also work on a Mac), which can be found in this repository. Notice that the Torch library is quite large (ca. 700 MB). Please be patient :)
Author: Michael Glukhman 304366891 (c) 2019