‘Smart Translation Systems’ Feature Prominently in China’s 3-Year AI Agenda

China’s Ministry of Industry and Information Technology (MIIT) released a “3-Year Action Plan to Boost the New Generation Artificial Intelligence Industry (2018-2020)” policy paper on December 14, 2017. The plan was issued to all levels of government and administrative offices across China for implementation.

This 15-page paper, as its name says, is an action plan. Beyond the usual propaganda bylines and high level, qualitative concepts, it does set out actual targets it intends to meet by 2020. The goals are divided into eight specific commercial applications of AI, two of which relates to Natural Language Processing (NLP) – speech recognition and machine translation systems.

The other applications with targets spelled out are: autonomous vehicles, service robots, aerial drones, medical imaging diagnosis, visual identity recognition and smart home appliances.

In speech recognition, China wants to develop next generation voice interactive systems with personalization, chatbot and audio-visual integration capabilities. The systems will be applied to intelligent manufacturing and smart home systems.

By 2020, it wants to achieve 96% accuracy for Chinese speech recognition and 92% accuracy within a 5-meter range, as well as above 90% accuracy rate for understanding user conversations.

For machine translation (which the paper classified as “smart translation systems”), it wants to rely on machine learning technology to increase the accuracy and practical usability of multilingual translation systems and simultaneous interpretation.

By 2020, China wants to achieve 85% accuracy for bi-directional Chinese-English translations and distinct improvements in the translation of minority languages (within China) to Chinese.

To ensure the AI products and services meet international standards, the paper said that industry benchmarks, testing standards and safety frameworks will also be developed. However, the paper did not go into details what these standards and benchmarks will be or how exactly will the aforementioned accuracy percentages be assessed and calculated.

The paper also talked broadly about developing the basic infrastructure required for meeting these goals, including building open source datasets for model training and testing, cloud services and an intellectual property platform.

Specifically, in the areas of NLP and visual recognition, China wants to accumulate enough industry specific data for industrial production, medical treatment, finance and transportation to support startups and innovation in these verticals.

It was also mentioned that China intends to accelerate the deployment of the country’s 5G mobile network and ensure that more than 90% of the country’s broadband coverage will meet the needs of the AI industry by 2020.

Download the Slator 2019 Neural Machine Translation Report for the latest insights on the state-of-the art in neural machine translation and its deployment.