iTlajtol

iTlajtol is a mobile application that can translate from Nahuatl to Spanish and vice-versa.

Requerimients

This work is based on FastAI and uses SentencePiece. To prepare an enviroment to run this models, please install the requirements.

Corpus

There are 2 main corpuses, first is Axolotl from this web. And the second one is a webscrapping result from JW using Nahuatl de la Huasteca.

Neural networks

This work also uses two NN architectures, an RNN that implements attention and a Transformer. This implementations are from FastAI and were finetunned to accomodate a low-resource language such as Nahuatl.

Models

There are several models to carry out a comparative analysis. Such models can be found on the models folder with a pkl extension and its weights as a pth file. So far only the best Transformer is available and the first RNN with attention.

Arquitecture	Corpus	Tokenization	From	To	%BLEU
Transformer	JW	SentencePiece	NCH	SPA	62.5145
RNN+Att	Axolotl	Spacy	SPA	NAH	32.2177

Data

Transformer and RNN This data contains the vocabulary and segmented pieces.

Vocab

For sentencePiece the vocabulary can be loadded with the files at vocab, such files had to be load with SentencePiece.

iOS Client

There are two clients, one written in Objective-C and the other in Swift using SwiftUI.

Cite this work

IEEEXplore

@inproceedings{9307780,
    Author = {S. K. {Bello García} and E. {S{\'a}nchez Lucero} and B. E. {Pedroza M{\'e}ndez} and J. C. {Hern{\'a}ndez Hern{\'a}ndez} and E. {Bonilla Huerta} and J. F. {Ramírez Cruz}},
    Booktitle = {2020 8th International Conference in Software Engineering Research and Innovation (CONISOFT)},
    Doi = {10.1109/CONISOFT50191.2020.00041},
    Keywords = {Spectrogram;Decoding;Tools;Mobile handsets;Internet;Engines;Speech recognition;Nahuatl;NMT;mobile;translation;attention;machine learning;CoreML;neural network;Mel spectrogram},
    Month = {Nov},
    Pages = {235-244},
    Title = {Towards the implementation of an Attention-based Neural Machine Translation with artificial pronunciation for Nahuatl as a mobile application},
    Year = {2020},
    Bdsk-Url-1 = {https://doi.org/10.1109/CONISOFT50191.2020.00041}
}

@article{springer-jcr,
    Author = {S. K. {Bello García} and E. {S{\'a}nchez Lucero} and E. {Bonilla Huerta} and J. C. {Hern{\'a}ndez Hern{\'a}ndez} and J. F. {Ramírez Cruz} and B. E. {Pedroza M{\'e}ndez}},
    Date-Added = {2021-08-22 17:05:11 -0500},
    Date-Modified = {2021-08-22 17:07:42 -0500},
    Journal = {Programming and Computer Software},
    Keywords = {machine translation; nahuatl; attention; application; transformers},
    Month = {Dec},
    Number = {8},
    Title = {Implementation of Neural Machine Translation for Nahuatl as a Web Platform: A focus on text translation},
    Volume = {47},
    Year = {2021}
}

MICAI-Springer

@InProceedings{10.1007/978-3-030-89820-5_10,
    author="Bello Garc{\'i}a, Sergio Khalil
    and S{\'a}nchez Lucero, Eduardo
    and Bonilla Huerta, Edmundo
    and Hern{\'a}ndez Hern{\'a}ndez, Jos{\'e} Crisp{\'i}n
    and Ram{\'i}rez Cruz, Jos{\'e} Federico
    and Pedroza M{\'e}ndez, Blanca Estela",
    editor="Batyrshin, Ildar
    and Gelbukh, Alexander
    and Sidorov, Grigori",
    title="Nahuatl Neural Machine Translation Using Attention Based Architectures: A Comparative Analysis for RNNs and Transformers as a Mobile Application Service",
    booktitle="Advances in Soft Computing",
    year="2021",
    publisher="Springer International Publishing",
    address="Cham",
    pages="120--139",
    isbn="978-3-030-89820-5"
}

@article{ISP-RAS,
    Author = {Sergio. Khalil. {Bello García} and Eduardo. {S{\'a}nchez Lucero} and Edmundo. {Bonilla Huerta} and Jos{\'e}. Federico. {Ram{\'\i}rez Cruz} and Jos{\'e}. Crispín. {Hern{\'a}ndez Hern{\'a}ndez} and Blanca. Estela. {Pedroza M{\'e}ndez}},
    Date-Added = {2021-08-22 17:29:22 -0500},
    Date-Modified = {2021-08-22 17:57:28 -0500},
    Journal = {Proceedings of the Institute for System Programming of the RAS},
    Title = {Pushing the Text Nahuatl Neural Machine Translation Boundaries as a Mobile Application},
    Year = {2021}
}