For this reason, we performed human side-by-side evaluations on all new models, which confirmed the gains in BLEU. For instance, several works have demonstrated how the BLEU score can be biased by translationese effects on the source side or target side, a phenomenon where translated text can sound awkward, containing attributes (like word order) from the source language. This improvement is comparable to the gain observed four years ago when transitioning from phrase-based translation to NMT.Īlthough BLEU score is a well-known approximate measure, it is known to have various pitfalls for systems that are already high-quality. With these latest updates, we see an average BLEU gain of +5 points over the previous GNMT models, with the 50 lowest-resource languages seeing an average gain of +7 BLEU. The improvements since the implementation of the new techniques over the last year are highlighted at the end of the animation.Īdvances for Both High- and Low-Resource LanguagesĪ popular metric for automatic quality evaluation of machine translation systems is the BLEU score, which is based on the similarity between a system’s translation and reference translations that were generated by people. The quality improvements, which averaged +5 BLEU score over all 100+ languages, are visualized below.īLEU score of Google Translate models since shortly after its inception in 2006. These techniques span improvements to model architecture and training, improved treatment of noise in datasets, increased multilingual transfer learning through M4 modeling, and use of monolingual data. In this post, we share some recent progress we have made in translation quality for supported languages, especially for those that are low-resource, by synthesizing and expanding a variety of recent advances, and demonstrate how they can be applied at scale to noisy, web-mined data. Many techniques have demonstrated significant gains for low-resource languages in controlled research settings (e.g., the WMT Evaluation Campaign), however these results on smaller, publicly available datasets may not easily transition to large, web-crawled datasets. And while the research community has developed techniques that are successful for high-resource languages like Spanish and German, for which there exist copious amounts of training data, performance on low-resource languages, like Yoruba or Malayalam, still leaves much to be desired. Nevertheless, state-of-the-art systems lag significantly behind human performance in all but the most specific translation tasks. “translate: English to French”.Posted by Isaac Caswell and Bowen Liang, Software Engineers, Google ResearchĪdvances in machine learning (ML) have driven improvements to automated translation, including the GNMT neural translation model introduced in Translate in 2016, that have enabled great improvements to the quality of translation for over 100 languages. The default model for the pipeline is t5-base which under the hood adds a task prefix indicating the task itself, e.g. You can use the □ Transformers library with the translation_xx_to_yy pattern where xx is the source language code and yy is the target language code. This approach might be less reliable as the chatbot will generate responses that were not defined before. You can then translate the output of the chatbot into the language of the user. You can use a Translation model in user inputs so that the chatbot can process it. Translate the input and output of the agent.This allows you to proofread responses in the target language and have better control of the chatbot's outputs. You can then train a new intent classification model with this new dataset. You can translate a dataset of intents (inputs) and responses to the target language. Translate the dataset to a new language. Translation models can be used to build conversational agents across different languages. When this happen, you can use a pretrained multilingual Translation model like mBART and further train it on your own data in a process called fine-tuning. You can find over a thousand Translation models on the Hub, but sometimes you might not find a model for the language pair you are interested in.
0 Comments
Leave a Reply. |