Charles Translator: A Machine Translation System Between Ukrainian and Czech to Aid Refugees and Facilitate Communication
Charles Translator is a machine translation system developed to quickly provide high-quality translation between Ukrainian and Czech in order to mitigate the language barrier faced by Ukrainian refugees in the Czech Republic following the Russian invasion of Ukraine.
Evaluating Parameter-Efficient Fine-Tuning Methods for Improving Low-Resource Language Translation
Parameter-efficient fine-tuning (PEFT) methods can effectively adapt large pre-trained language models for diverse tasks, offering a balance between adaptability and computational efficiency. This study comprehensively evaluates the performance of various PEFT architectures for improving low-resource language (LRL) neural machine translation (NMT).
Improving Zero-shot In-context Machine Translation with Anti-Language Model Decoding
Anti-Language Model (Anti-LM) decoding objective with exponential decay can significantly improve zero-shot in-context machine translation performance compared to other decoding methods.
Systematic Comparison of Retrieval Techniques for Retrieval Augmented Neural Machine Translation
The choice of retrieval technique significantly impacts the performance of retrieval-augmented neural machine translation models, with varying effects across different architectures. Optimizing for coverage and diversity of retrieved examples can yield substantial gains, especially for non-autoregressive models.
Continuous-Output Neural Machine Translation with Random Target Embeddings Outperforms Pre-Trained Embeddings
Random target embeddings can outperform pre-trained embeddings, especially on larger datasets and for rare words, in continuous-output neural machine translation.
The Future of Machine Translation Lies with Large Language Models
The emergence of Large Language Models (LLMs) like GPT-4 and ChatGPT is introducing a new phase in the Machine Translation (MT) domain, offering vast linguistic understandings and innovative methodologies that have the potential to further elevate MT.
An Empirical Evaluation of Vocabulary Trimming in Neural Machine Translation
Vocabulary trimming, a common practice in neural machine translation, fails to consistently improve model performance and can even lead to substantial degradation across a wide range of hyperparameter settings.
KazParC: A Comprehensive Parallel Corpus for Multilingual Machine Translation
KazParC is a large-scale parallel corpus designed to facilitate machine translation across Kazakh, English, Russian, and Turkish languages. The corpus was developed with the assistance of human translators and contains over 371,000 parallel sentences spanning diverse domains. The research also introduces Tilmash, a neural machine translation model that demonstrates competitive performance compared to industry-leading services.
ACT-MNMT: Auto-Constriction Turning for Multilingual Neural Machine Translation
ACT-MNMT introduces a novel mechanism for Multilingual Neural Machine Translation, addressing off-target issues and improving translation performance.
BiVert: Bidirectional Vocabulary Evaluation for Machine Translation
기계 번역의 평가를 위한 BiVert 방법론 소개
© 2024 by Linnk AI