Gartner: Alibaba voice semantic technology second in the world

July 7, 2012 - Gartner, an international authority on research, recently announced the Cloud AI Developer Services Key Capabilities Report, in which Ali Language AI technology received the highest scores for four projects and ranked second in the world in terms of total score. This is the best result achieved by Chinese companies in this field since the birth of the report.

It is reported that the Gartner report evaluates and ranks global cloud AI service vendors such as Google, Amazon AWS, Microsoft, and Alibaba around three dimensions: language AI, computer vision, and machine learning platforms.

Ultimately, Ali won first place in four scores: speech recognition, natural language generation/speech synthesis, language understanding/processing, and text analysis, and came in second overall.

Gartner believes that by 2025, 70% of new applications will integrate AI models, and cloud AI services can lower the development threshold of AI applications.

Language AI, as a fundamental technology in AI, has had many technical breakthroughs in the past few years, especially the pre-trained model structure represented by Bert and Transformer has pushed the related research to a climax, accelerating the process of moving from perceptual intelligence to cognitive intelligence in AI.

As one of the earliest teams to invest in the research of pre-trained language models, Dharma Institute has developed a deep language large model system AliceMind on top of Ali cloud's infrastructure, including general pre-training model StructBERT, multi-lingual pre-training model VECO, large Chinese pre-training model PLUG, etc., with reading, writing, translation, Q&A, search, summary generation, dialogue The technology system has won 35 championships and is open source for global developers.

In intelligent speech, the system built by Dharma Academy based on self-researched SAN-M network structure and SCAMA streaming end-to-end speech recognition framework can meet the requirements of different scenarios in terms of latency and accuracy, and reduce the error rate of speech recognition in difficult scenarios by nearly 30% while improving computational efficiency and reducing service costs. SAMBERT can generate vivid and expressive speech with optimized modeling in multiple dimensions, such as emotion, rhythm and sound quality.

At present, Damo Institute has more than 300 papers in the field of language AI included in the International Top Society, and the research results are applied to medical, e-commerce, electricity, law and other fields.

Take cross-border e-commerce as an example, the first thing to do is to cross the language barrier when domestic goods go abroad. 70% of buyers in Alibaba's international station communicate in English and 30% use small languages, while 96% of domestic sellers are powerless in small languages, Damo Institute's machine translation system can provide mutual translation services in 214 languages and translate hundreds of millions of words for 2 million small and medium-sized domestic businesses every day, helping domestic goods go to the world.

It is reported that IDC "2021H2 China AI cloud services market research report", Ali won the first of two public cloud market for speech and semantics, which means that Ali language AI is at the forefront of the industry in terms of product technical strength and market share.