Alibaba DAMO Academy wins SemEval's best paper to help AI understand human language better

News on July 19, SemEval-2022, the world's largest semantic evaluation competition, recently announced that this year's only "Best System Paper Award" will be awarded to researchers from Alibaba Damo Academy and other institutions. They designed a named entity recognition (NER) system that incorporates knowledge for 11 languages, including Chinese and English. It can accurately identify key entity information such as people's names, place names, institutions, works, etc., which effectively improves AI's understanding of human language. Ability.

SemEval (Semantic Evaluation) is an authoritative international competition in the field of natural language processing. It has a history of more than 20 years and is hosted by the Lexicon and Semantics Group of the International Association for Computer Linguistics (ACL). It aims to make AI To analyze and understand the meaning contained in human language.

SemEval has two best paper awards: Best Task Paper Award and Best System Paper Award. Popular understanding is that one is to ask questions and the other is to solve problems. The joint research team of Alibaba DAMO Academy, Shanghai University of Science and Technology, Zhejiang University, and Singapore University of Technology and Design won this year's Best System Paper Award. The article that stood out from 221 candidate papers is called "DAMO-NLP at SemEval- 2022 Task 11: A Knowledge-based System for Multilingual Named Entity Recognition》.

SemEval-2022 Best System Paper

The winning team participated One of the 12 tasks of SemEval-2022: Multilingual Complex Named Entity Recognition (Multilingual Complex Named Entity Recognition). Named entity recognition (NER) is a basic work in the field of natural language processing. It refers to the identification of entity words (Entities) with specific meanings in text, mainly including names of people, place names, organization names, proper nouns, etc.

The task requires researchers to design a system that can identify entities in 11 languages ​​including Chinese and English, including sentence patterns that are mixed with multiple languages, including "stalks", abbreviations, and colloquialisms. , achieve accurate identification. For example: "In 2016, she guest-starred in the HBO TV series Game of Thrones." The AI ​​needs to recognize and understand the abbreviated organization name "HBO" and the title of the work "Game of Thrones."

Dharma Academy System won the first place in the overall score

The paper proposed a new set of multi-language named entities that incorporate knowledge The recognition system won 10 first places among the 13 sub-items of the competition tasks, ranking first in total score, which greatly improved the industry level.

Generally speaking, because words have ambiguous meanings, we can only accurately understand words based on context, and the same is true for AI. The power of the new system is that it allows AI to understand complex entity words even if there is no context. According to the researchers, the system introduces additional external knowledge to build a multi-lingual general knowledge base, which expands the contextual information of the text through interactive retrieval to eliminate ambiguity; coupled with multi-stage fine-tuning, it can accurately identify entity information .

Dharma Academy System Principle

According to reports, this award-winning research has been widely used in translation, search, human-computer dialogue, etc. The field has broad application prospects. Currently, DAMO Academy's machine translation system can provide translation services in 214 languages, translating hundreds of millions of words for 2 million domestic small and medium-sized businesses every day, helping domestic products to go global. The latest report "Cloud AI Developer Service Key Capabilities Report" by Gartner, an authoritative international research organization, points out that Alibaba Language AI has ranked second in the world, the best result in the history of Chinese companies.

The above is the detailed content of Alibaba DAMO Academy wins SemEval's best paper to help AI understand human language better. For more information, please follow other related articles on the PHP Chinese website!