Neural Machine Translation for the Arabic-English Language Pair

def prune_transformer(model, amount=0.3): for name, module in model.named_modules(): if isinstance(module, torch.nn.Linear): prune.l1_unstructured(module, name='weight', amount=amount) prune.remove(module, 'weight') return model # Efficient Multilingual NMT class LightTranslator(nn.Module): def __init__(self, vocab_size, d_model=512): super().__init__() self.encoder = TransformerEncoder(d_model) self.decoder = TransformerDecoder(d_model) def forward(self, src, tgt): memory = self.encoder(src) return self.decoder(tgt, memory)

Ahmed Amine Aliane March 2024 ICPAIS (International Conference on Pattern Analysis and Intelligent Systems)

Abstract

This work implements and evaluates modern neural machine translation architectures on the Arabic-English language pair, a setting where morphological complexity and limited high-quality parallel data pose challenges beyond those typically faced by well-resourced language pairs.

Method

Transformer-based sequence-to-sequence models are trained and evaluated on Arabic-English parallel corpora, with results compared across architectural and training variants.

Contribution

Provides an evaluation of neural machine translation performance for Arabic-English, establishing a baseline that later informed the model efficiency and pruning work developed for multilingual NMT.

This paper’s translation setup is the foundation for A General Framework for Efficient Multilingual Neural Machine Translation, which builds on it with model compression techniques.

Neural Machine Translation for the Arabic-English Language Pair

Abstract

Method

Contribution

Related work