The government of Macau recently commended the University of Macau’s Natural Language Processing and Portuguese-Chinese Machine Translation Lab for landing in first place in five contests at the Sixth Conference on Machine Translation.
The lab, led by Professor Derek Fai Wong, has developed numerous natural language processing (NLP) and machine translation (MT) tools to aid in communication between speakers of Portuguese and Chinese, a particularly relevant language pair for residents of Macau, which is home to a distinct dialect of Portuguese that is spoken by a small minority of Macanese citizens.
The lab is not the only research group dedicated to developing more finely tuned MT tools for Portuguese-to-Chinese and Chinese-to-Portuguese. In fact, the Macau Polytechnic Institute has hosted a yearly international Chinese-Portuguese Translation Competition since 2017. A testament to the strong demand for high-quality MT engines in this language pair, the competition hosted nearly 600 competing teams in its first four years alone.
At the Sixth Conference on Machine Translation, the lab participated alongside the Alibaba Dharma Academy in the metrics shared task, in which automatic evaluation metrics were examined. The lab’s MT engines were shown to outperform prominent companies like Google and Unbabel at the Sixth Conference on Machine Translation and received five first-place prizes, as well as two second place prizes on automatic translation quality tests.
The conference’s metrics shared task did not evaluate the Chinese to/from Portuguese language pair for which the lab is particularly specialized. As such, members of the lab teamed up with the Alibaba Dharma Academy to develop an engine called RoBLEURT, which functions in the numerous language pairs that were examined at the conference.
In addition to MT tools, the University of Macau’s Natural Language Processing and Portuguese-Chinese Machine Translation Lab has developed a bilingual Portuguese-Chinese dictionary, which includes more than 100,000 entries, as well as tools that can be used for training students in the development of their own MT systems.