Dr. Ming Zhou

Founder/CEO, Langboat Ltd

Ming Zhou#

Dr. Zhou Ming, founder and CEO of Beijing Langboat Technology Co., Ltd.

Zhou Ming obtained his doctoral degree from the Computer Science Department of Harbin Institute of Technology in 1991 and developed one of the China's earliest Chinese English machine translation system CEMT-I. From 1991 to 1999, he engaged in postdoctoral research and teaching at Tsinghua University. He led the research and development of the famous Sino-Japanese machine translation product J-Beijing in Kodensha, Japan from 1993 to 1999. He joined Microsoft Research Asia in 1999 and led NLP research for more than 20 years.

He is a top NLP expert in the world, ACL fellow, fellow of The Chinese Computer Federation, and fellow of The Chinese Artificial Intelligence Society, as well doctoral supervisor at five famous universities in China. He is one of the leaders in the field of NLP in China and internationally, having served as the president of ACL and the Vice Chairman of the Chinese Computer Federation. He ranks among the top in the world in terms of the number of papers published at the NLP highest conference such as ACL.

He served as the vice president of Microsoft Research Asia, leading the R&D in the NLP field of Microsoft, covering technologies such as large language models, machine translation, search, chat and dialogue systems, all becoming the core technology of Microsoft's important products such as Windows, Office, Azure, Cognitive Service, and Xiaobing. The world-famous large language model UniLM led by him combines the advantages of BERT and GPT and won the 2019 Wuzhen Internet Leading Technology Award.

At the end of 2020, he left Microsoft Research Asia which he has worked for 21 years, and started to incubate a large language model company with the support of Innovation Works. Langboat Technology was officially established in June 2021. Langboat has successively received investments from Innovation Works, Scientists Fund, Lenovo Venture Capital, Eight Road Capital, and Zhongguancun Science City.

The company has developed the famous "Mengzi" series of large language model technologies and applied them to important fields such as finance. At present, the company is one of the leading LLM companies in China especially in the financial field.

Langboat won first prize in the 2021 China HICOOL Global Entrepreneurship Competition and first place in the AI and finance tracks. It receives the Technology Pioneers of Davos in 2023, and it was listed as one of the five companies with the potential to become OpenAI in China by The Information in 2023.

Achievements in R&D#

Ming Zhou has made significant and innovative contributions to NLP in the areas of machine translation, language learning, text generation, chatbot and pre-trained models. According to Google Scholar, his papers have been cited almost 33,000 times and his H-index is 101. He was the author with the most papers published in ML&NLP area across the period of 2012-2020, according to https://www.marekrei.com/blog/ml-and-nlp-publications-in-2020/.

Ming has greatly contributed to machine translation research. He was one of the China’s pioneers on Chinese-English machine translation research. Back to 1991, as his PhD thesis research, he created China’s first Chinese-English machine translation system (CEMT-I) with a hierarchical phrase structure rule system he designed for Chinese parsing, Chinese-English syntactical transfer and English generation. In 1998, he created a Chinese-Japanese machine translation products(J-Beijing) with a dependency structure based rule system. J-Beijing later became the market leader in Japan. After he joined MSRA in 1999, his MSRA-NLP group conducted many innovative research on statistical machine translation such as web mining of massive parallel data, alignment at document, sentence and word levels, lexicon extraction, collocation extraction, collocation translation, and effective decoding methods strengthened with syntactical structures. In 2008, his group, along with MSR Redmond NLP Group and other research partners, received the top Chinese-English MT quality ranking in Open MT evaluation made by the National Institute of Standards and Technology(NIST). After 2012, his group started experimenting deep learning methods to machine translation, moving from adding deep learning-based features into statistical machine translation to successful end-to-end training of a neural machine translation. His group contributed a statistical machine translation engine enhanced by deep learning features to a speech-speech English-Chinese translation system developed by MSRA Speech Group, which was successfully demonstrated at Microsoft Research Asia’s 21th Century Computing Conference in October, 2012 by then MS Chief Research Officer Rick Rashid (Ref: Microsoft Research shows a promising new breakthrough in speech translation technology - The AI Blog). This demonstration inspired worldwide research interests on speech-speech translation with deep learning approaches. In 2018, a new type of neural machine translation system created by his team along with other MSR teams is able to translate sentences of news articles from Chinese to English with the same quality and accuracy as a person(Ref: Neural Machine Translation Enabling Human Parity Innovations In the Cloud - Microsoft Translator Blog). The innovative methods out of above-mentioned research had been continually integrated into Microsoft Translator, a multilingual machine translation cloud service.

Ming has also greatly contributed to language learning research. In 2005, his group created English Writing Assistant system(EWA) which prompts relevant example sentences for user’s input word sequence to help the user write more authentic sentences. This system was integrated into MS Office for English as second language users. In 2010, his group created Engkoo (means “English vault”), a groundbreaking piece of software that takes advantage of NLP and speech technologies to build massive sets of bilingual terms and sentences to create a new kind of language-assistance technology for Chinese users to enable them to ultimately master English as a native speaker might. In his design, the system unifies human translation mined from the web, machine translation, and a language-learning experience into one user-friendly search-and-explore interface. This language assistance system received Asia Innovation Award issued by WSJ. It was released by Bing in 2012 to provide English learning services to large number of Chinese users. (Ref: Software Aids Language Learners - Microsoft Research)

Ming has greatly contributed to text generation. In 2001, his team created statistical language model based IME for MS windows, supporting both Chinese and Japanese. In 2005, he invented MS Chinese Couplet System. Chinese couplet is a unique cultural heritage of China. One person challenges the other person with a sentence. The other person relies with a sentence equal in length and word segmentation, in a way that corresponding words in the two sentences match each other, obeying a series of constraints on semantic, syntactic, and lexical relatedness. At that time, this task is viewed as a difficult problem in AI and has not been well explored in the research community. For the first time in NLP history, he presents a phrase-based SMT approach to generate the second sentences and uses a set of filters to remove candidates violating linguistic constraints (Red: Generating Chinese Couplets using a Statistical MT Approach - ACL Anthology). A popular system based on his algorithm was created allowing users to automatically complete their Chinese couplets and Chinese character riddles. It has been widely viewed as the first ML-generated poetry system in China(Ref: Chinese Tradition Inspires Machine Learning Advancements, Product Contributions - Microsoft Research).

Ming has also greatly contributed to chatbot. His team was the research power for chatbot engines behind XiaoIce, Microsoft’s social chatbot which was first released in China in 2015, then its Japanese version and English version were released in Japan and in the US respectively. His group developed the retrieval-based engine for response generation, increasing the natural conversation rounds from 4.3 turns to 21 turns. After that, his group developed a novel deep learning engine for response generation, further improving the chatbot conversation capability. A hybrid engine of both methods leads to more diversified responses, making XiaoIce the state-of-the-art chatbot in the world before the pre-trained model approach becomes available in 2019. With accumulated 200 millions users in China over the years, XiaoIce was viewed as Microsoft’s flagship AI products in the period of 2015-2019(Ref: Like a phone call: XiaoIce, Microsoft’s social chatbot in China, makes breakthrough in natural conversation - The AI Blog).

Ming has also significantly contributed to pre-trained models in recent 5 years. His team created various pre-trained models with a large-scale self-supervised pre-training across tasks, languages and modalities. Among these models, a new unified pre-trained language model (UniLM) can be fine-tuned for bother language understanding and generation tasks. The model is pre-trained using three types of tasks: unidirectional, bidirectional and sequence-to-sequence prediction. The unified modeling is achieved by employing a shared Transformer network and utilizing specific self-attention masks to control what context the prediction conditions on. UniLM compares favorably with BERT on the GLUE benchmark, and the SQuAD 2.0 and CoQA question answering tasks. UniLM achieves new state-of-the-art results on five natural language generation datasets, including improving the CNN/DailyMail abstractive summarization, the Gigaword abstractive summarization, the CoQA generative question answering and other tasks. UniLM was then extended to cover multiple modalities including vision, speech and layout (Ref: GitHub - microsoft/unilm: Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities). UniLM incorporating the language understanding and generation into one pre-trained model is very influential in research community. In 2019, UniLM received Internet Leading Technology Award, China’s prestigious Prize on Internet Innovation.

In addition to the above-mentioned innovative research, Ming’s group’s research on social media analysis is also very influential. His paper “Learning Sentiment-Specific Word Embedding for Twitter Seam understanding and generation. CodeXGLUE includes a collection of 10 tasks across 14 datasets and a platform for model evaluation and comparison. CodeXGLUE also features three baseline systems, including the BERT-style, GPT-style, and Encoder-Decoder models, to make it easy for researchers to use the platform. The availability of such data and baselines can help the development and validation of new methods that can be applied to various program understanding and generation problems.

In 2021, he founded Langboat company (https://langboat.com) which develops advanced NLP technologies to serve the enterprise customers with SaaS and on-premise solution. With advanced technologies on lightweight pre-trained models, machine translation and text generation, Langobat is growing into one of the leading companies in large language models and NLP field in China.

Service excellence in NLP community#

Extended service to the ACL or other scientific organizations whose mission aligns with that of the ACL:

Ming has done a significant amount of service to the ACL community. He started serving on the ACL executive committee in 2017 and became the President of ACL in 2019. He then served on the nominating committee from 2020 to 2022. He is the major contributor to the launch of ACL Asia-Pacific Chapter(AACL), which has significantly promoted ACL and NLP in Asia-Pacific region.

In 2020, Ming was elected vice-president of CCF, ACM counterpart in China, with 100,000+ members. Prior to this, he was the head of CCF-NLP, the major NLP association in China during 2015-2019.

He served on the editorial board of the Computational Linguistics during 2009-2013. He also served many conferences in different roles. He was the Area Chair of ACL 2000 and ACL 2003, PC chair of AIRS 2004, Chair of Interactive Demo/Poster of IJCNLP 2004, Area chair of IJCNLP 2005, Area chair of EMNLP/HLT 2005, Area chair of NAACL/HLT 2006, Area chair of COLING-ACL2006, Area chair of IJCAI 2007, Workshop chair of ACL 2008, Publicity chair of SIGIR 2009, PC member of CIKM 2004 and CIKM 2005, MT Summit 2004 and MT Summit 2005, AMTA 2002, SIGHAN 2005, SIGHAN 2006, AIRS 2005, AIRS 2006. Area chair of COLING 2010, Area chair of SIGIR 2012 and SIGIR 2013. PC chair of CCF NLP&CC 2012 and general chair of NLP&CC 2013, PC Chair of CNCC-2019 (China’s most influential conference on computer research and industry).

Educational and outreach activities#

Ming was a teacher at Tsinghua for 8 years prior to joining MSRA in 1999. He is PhD supervisor at 5 universities of China including Harbin Institute of Technology, Tianjin University, Nankai University, Beihang University and University of Science and Technology of China. Over the past 20 years, he has supervised 20 PhD students and his group has trained 500 interns from more than 30 universities in China, Japan, Singapore, South Korea, US, Canada, Australia.

He was the director of the MS-Tsinghua University Joint Lab on Media and Network and MS-Harbin Institute of Technology Joint Lab on NLP for over ten years, overseeing the research directions and joint projects. He made an important contribution to the establishment of NLPCC which is called China’s ACL conference with highly selective acceptance and using English as official language to bridge the NLP research community of China with the world. He also significantly contributed to the launch of an annual China-Japan NLP workshop (CJNLP) to promote the collaboration between China and Japan as well as other Asian countries.

As vice-president of China Computer Federation(CCF), he has devoted significant efforts to advance the collaboration between universities and industrial companies via a series of programs such as CTO Club and Technical Frontier technical lectures. He established the CCF-Startups Summit at CNCC (China’s National Computer Conference) to promote the culture of entrepreneurship and strengthen the collaboration between fundamental research and real-world applications.

ACL Fellow#

With outstanding academic research, industrial contributions, outstanding NLP community service, and important contributions to NLP education and cooperation in China and Asia, Dr. Zhou Ming was honored to be elected as a member of the 2023 ACL.

"For significant contributions to machine translation, language learning, text generation as well as the growth of NLP community in China and Asia”.