SiameseConvNet

Performs cognate identification using Siamese Convolutional Networks

Please consider citing the following paper if you use the code:

Taraka Rama. Siamese Convolutional Networks for Cognate Identification. Proceedings of COLING 2016, Osaka, Japan, 2016. http://aclweb.org/anthology/C/C16/C16-1097.pdf

Requirements:

• Need Keras (https://keras.io/) with Tensorflow (https://www.tensorflow.org/) as a backend for running the code.

Running the program:

• Run a program as python siamese_cognates_cnn_langs_info.py 30 data/IELex-2016.tsv.asjp
• The program takes the number of concepts for training and the name of the training dataset as commandline arguments
• There are total four programs.
• The program starting with one_hot uses the 1-hot representation of a phoneme to represent a word as a matrix and then compute the similarity between the two words and then optimizes binary cross-entropy using a Adadelta optimizer
• The program starting with siamese_cognates uses articulatory features for representing a phoneme and then performs convolutional operations for identifying cognates
• The program outputs the F-scores and accuracies for a dataset to the screen
• The three datasets (Mayan, Austronesian, and Indo-European) which are used in the experiments are provided in data folder

Contact:

In case of any questions, contact taraka-rama.kasicheyanula@uni-tuebingen.de