From July 18 to 20, 2023, the Second Conference on Natural Language Generation and Intelligent Writing (NLGIW 2023) was held in Urumqi, Xinjiang. Global Tone Communication Technology Co., Ltd. (GTCOM) participated as a strategic partner and demonstrated its GeWularge model technology and industrial practices. Also present at the event were high-tech companies such as Tencent AI Lab and Baidu, as well as people engaged in relevant research and practices from the academia and the industry. Participants shared research findings and practical experience, and contributed to technological innovation and research results commercialization in the field of large model and natural language generation.

The event was organized bythe Natural Language Generation and Intelligent Writing Committee (in preparation) of the Chinese InformationProcessingSociety of China and undertaken byXinjiang Normal University. It includes special presentations, workshops, and a young scholar forum, all focusedon natural language generation and big model. Specifically, topics covered at the eventincludethe basic principles of large language model, code large model, and multi-modal large model;efficient training and reasoning, supervised fine-tuning, reinforcement learning from human feedback, securityand ethics, large model application and other dimensions. This was indeed a grand gala of large model for all participants.

Atthe enterpriseforum, which was part of the conference, Dr. Chen Ziyan, vice president of GTCOM 2030 Artificial Intelligence Research Institute, was invited to give a speech entitled "GeWu Multilingual Large Model Technology and Industrial Practices". In the speech, Chen explained how GTCOM’s large model, with technological innovations, successfully addressedchallenges faced by artificial intelligence in industrial practices such as the semantic separation between languages, the connection betweendata, requirements and applications, and the difficulty in anthropomorphic social interaction generation.

GeWu is a large model system that GTCOM launched in November 2022. It includes multilingual pre-training models, multilingual machine translation super-large models, multi-modal pre-training models, etc. Dr. Chen Ziyan highlighted the new-generation multilingual generative dialogue model that the company is currently working on. Based on a high-quality parallel corpus of nearly 8 billion sentence pairs in GTCOM’s own possession and PB-level data in more than 100 languages, this model integrates over 20 million high-quality business instructions covering context-based understanding, personalization, and thought chains, and proposes a hybrid reinforcement learning technology featuring domain-specific expert alignment and social confrontation feedback, aiming to create a generative large model that is highly responsive and customizable.

The generative large model has been put into use in scenarios such as end-to-end multilingual knowledge graph generation, social media behavior generation, and field-specificreport writing. Technological updates are going on targetedat global defense and security, research data analysis, smart cities, and global strategic data research and other fields.Dr. Chen Ziyan also hadroundtable discussionswith experts fromTencent,Baiduand other companieson the difficulties and countermeasures in the application of large models.

GTCOM displays its multilingual NLP algorithm on-site the event.