Friends-and-Enemies

This repository contains:

• R scripts for all analyses
• .csv files of all analysis inputs/outputs
• a folder with all the Python scripts used to run the VOISeR simulations
• a folder with scripts associated with VOISeR validation tests (testing for consistency/regularity effects)

VOISeR model

VOISeR is a simple computational reading model to support the friends-and-enemies research.

Requirement

tensorflow >= 1.13

If you are using TF 2.0, see the repository below:
https://github.com/CODEJIN/VOISeR_TF20

Structure

• The using of 'Orthography → Hidden' and 'Hidden → Hidden' is selectable.
• The model reported in the paper has both connections (O->H and H->H).

Dataset

Data were obtained from 'The English Lexicon Project':

Balota, D. A., Yap, M. J., Hutchison, K. A., Cortese, M. J., Kessler, B., Loftis, B., ... & Treiman, R. (2007). The English lexicon project. Behavior research methods, 39(3), 445-459.

The "ELP_groupData.csv" file was used to train VOISeR.

Run

Command

python VOISeR.py [parameters]

Parameters

• -dir <path>

  • Determines the type of hidden layer. You can enter either LSTM, GRU, SCRN, or BPTT.
  • This parameter is required.

• -ht B|H|O

  • Determines which layers' activation is used for the hidden activation calculation.
    • B: Using both of previous hidden and output
    • H: Using previous hidden
    • O: Using previous output
  • This parameter is required.

• -hu <int>

  • Determines the size of the hidden layer. You can enter a positive integer.
  • This parameter is required.

• -lr <float>

  • Determine the size of learning rate. You can enter a positive float.
  • This parameter is required.

• -e <int>

  • Determine the model's maximum training epoch.
  • This parameter is required.

• -tt <int>

  • Determine the frequency of the test during learning. You can enter a positive integer.
  • This parameter is required.

• -fre

  • If you enter this parameter, model will use the frequency information of words in the training.

• -emb <int>

  • If you enter this parameter with integer value, model use the embedding about the orthographic input.
  • The inserted integer value become the size of embedding.
  • The default value is None.

• -dstr <path>

  • If you enter this parameter, the target pattern become the distributed pattern.
  • If you don't enter, the target pattern become one-hot structure.
  • Please see the example: 'phonetic_feature_definitions_18_features.csv'

• -try <int>

  • Attach an tag to each result directory.
  • This value does not affect the performance of the model.

Analysis

Command

python Result_Analysis.py -f <path>

Parameter

• -f <path>
  • Results directory to run the analysis on.
  • This parameter is required.
  • Ex. VOISeR_Results/HT_B.HU_300.LR_0005.E_10000.TT_1000.DSTR_True