The internet search giant submitted the patent application in December, but it was only published publicly by the China National Intellectual Property Administration on Tuesday.
The translation method works by collecting various types of data from a target animal, including “voice, body language, behavioural changes and other biological signs”, according to the application posted on the agency’s website. It then determines the “emotion” by analysing the data before converting it into a target language.
The technology utilises several subfields of AI, including machine learning, which enables computers to learn from large data sets; deep learning, which uses neural networks to analyse voice and movement characteristics; and natural language processing, which applies computational approaches to the analysis and synthesis of natural language and speech, according to Baidu.
If a voice does not match any historical emotional record, the team would manually label the voice data, update the sample and adjust the model parameters.
The publication of the patent application does not mean it has been granted yet. The process of assessing the application could take a year in the best-case scenario, according to You Yunting, a senior partner at Shanghai Debund Law Firm.