Wong Fai, professor in the Natural Language Processing & Portuguese-Chinese Machine Translation Laboratory (NLP2CT) at the University of Macau (UM) and his team presented several papers at the 61st Annual Meeting of the Association for Computational Linguistics (ACL 2023) and attracted considerable attention. UM’s advancements in related fields will promote technological cooperation between Macao and Portuguese-speaking countries and help Macao play its role as a platform between China and Portuguese-speaking countries.
The ACL 2023 conference received 4,559 submissions from around the world, of which 945 were accepted, accounting for 20.7 per cent of the total submissions. Five papers contributed by UM were accepted as main conference papers and two were accepted as findings papers. These papers covered a diverse array of subjects, including machine translation (MT), grammar error correction, and natural language generation.
During the conference, PhD students Liu Shudong and Zhan Runzhe presented two papers on machine translation training and evaluation. The first paper introduces a training and extraction method that can effectively transfer knowledge from a high-resource MT model to a child model, thus enabling the translation of low-resource language pairs. Compared to the previous transfer learning approaches that only focus on initialising model parameters, the new approach facilitates knowledge transfer in the training and inference stages. The second paper focuses on the automatic evaluation of MT and proposes a self-supervised approach to dynamically adapt the model to deal with out-of-distribution issues in cross-domain evaluation, thereby enhancing prediction performance. Both methodologies have been submitted for a patent application.
Natural language processing (NLP) research plays a crucial role in modern artificial intelligence (AI), driving the development of AI products such as the revolutionary ChatGPT. Over the past five years, UM has presented over 20 papers at the ACL conference, demonstrating the university’s leading expertise in NLP. Under the leadership of Prof Wong, the research team has been dedicated to NLP research for two decades and published a number of research results, demonstrating its world-leading research capacity. The team’s participation in the conference was supported by the Science and Technology Development Fund of the Macao SAR and UM.
The ACL conference is widely regarded as the top academic conference in the field of NLP. According to Google Scholar’s metrics, its conference proceedings are ranked first in the H5 Index of NLP publications. It is also listed as a Class A conference in the List of Academic Conferences and Periodicals Recommended by the China Computer Federation. Over the years, the conference has received sponsorship from major technology giants such as Google, Meta, Microsoft, and Apple, providing a platform for researchers from both academia and industry around the world to conduct academic exchange, and advancing the field with concerted efforts.
澳門大學自然語言處理與中葡機器翻譯實驗室(NLP2CT)教授黃輝及其研究團隊的多篇論文獲第61屆國際計算語言學學會年會(ACL 2023)錄用並於會上發表,引起學術界廣泛關注。澳大在相關領域的先進成果將推動澳門與葡語系國家的科技合作,助力澳門發揮中葡平台的作用。
今年,ACL收到來自全球共4559份投稿,當中945份獲得錄用,佔投稿總數20.7%。澳大共有五篇論文被接受為主會議論文,兩篇論文被接受為研究成果論文(Findings)。相關論文涵蓋機器翻譯、語法錯誤糾正、自然語言生成等多個領域。
會議上,澳大博士生劉樹東和詹潤哲發表了兩篇關於機器翻譯訓練和評估的論文。第一篇論文提到的訓練與檢索方法,能夠高效地將知識從高資源機器翻譯模型遷移到子模型,從而實現低資源語言對的翻譯。與以往僅關注模型參數初始化的遷移學習方法相比,該新方法進一步促進訓練和推理階段的知識遷移。第二篇論文主要研究機器翻譯的自動評估,提出一種自監督的動態調整方法,以解決跨領域評估中的分佈外預測問題,從而提高預測性能。目前,兩個研究方法已提交專利申請。
自然語言處理研究是現代人工智能(AI)技術中不可或缺的一部分,推動了人工智能產品的發展,如革命性智能對話系統ChatGPT等。過去五年,澳大在ACL上發表論文超過20篇,反映了澳大在自然語言處理研究的領先地位。在黃輝的帶領下,研究團隊在20年的研究生涯發表了多項研究成果,研究水平亦逐步邁向世界前沿。是次團隊參會獲得澳門特別行政區科學技術發展基金和澳門大學的資助。
ACL是權威的自然語言處理領域頂級學術會議。根據谷歌學術數據統計,ACL出版物在相關領域的學術出版物H5指數中位列第一,在《中國計算機學會推薦國際學術會議和期刊目錄》認定其為A類國際學術會議。ACL歷年來受谷歌、Meta、微軟、蘋果等大型科技企業贊助,為來自世界各地的學術界與工業界人員提供優秀的研究交流平台,促進學術觀點的碰撞和創新,從而推動相關領域的發展。