This repo contains the code for our paper entitled Multi Task Deep Morphological Analyzer : Context Aware Neural Joint Morphological Tagging and Lemma Prediction. The Web API service is accessible here.
A sample analysis:
Both the directories follow the organization:
-
preProcessing contains the code for dataset parsing. Datasets can be downloaded from the website of Universal Dependencies.
-
dataInfo contains details on data set statistics.
-
Models for all experiments:
- multiTask_with_context4.py hosts the fully BiLSTM model for a CW of 4 words.
- multiTask_with_attention.py hosts the character CNN-RNN based MT-DMA model, as reported in the paper.
- onlyFeatures.py and onlyRoots.py contain the codes for individual learning.
-
Code for MOO based GA feature selection.
-
Code for post processing, visualization, BLEU, Levenshtein and word accuracy calculation can be found in postProcessingAndVisualization.
-
Outputs on the HDTB and UDTB datasets.
-
Outputs for t-SNE plots, GA graphs, and Precision-Recall curves.
Cubic-spline interpolations for validation accuracies of population:
