While mainland China is developing into a powerhouse in natural language processing (NLP) technology, Hong Kong is also looking to crash the party with several start-ups working on chatbots for specific applications, including ones that can communicate in Cantonese. Chatbots, computer programmes which conduct conversations via auditory or textual methods, have become substantially smarter in recent years, growing in tandem with the rapid rise of artificial intelligence. Inside the AI revolution that’s reshaping Chinese society Hong Kong chatbot companies including Clare.AI, Mindlayer and Rocketbots are all expecting to roll out major projects with clients in coming months, allowing users to have conversations with artificial intelligence on platforms such as Facebook Messenger and WeChat. Gerardo Salandra, chief executive of Rocketbots, said chatbot technology will be used to enhance customer service, with 80 per cent of the world’s customer service work taken over by machines over the next three years. “In English the technology is ready. It’s just the adoption that is lacking,” he said. Beijing to set out artificial intelligence development plan up to 2030 He said world-class NLP engines, like those developed by tech heavyweights such as Google and IBM, can correctly understand as much as 95 per cent of questions asked in English. But he added that, when it comes to understanding Cantonese, the accuracy is still very low, hovering around an average of 30 per cent. Salandra said the technology developed by Rocketbots leverages the best NLP engines available and applies an extra layer of artificial intelligence that for sentiment analysis and topic modelling. “I can provide you with 55 per cent of accuracy [in Cantonese]. This is the highest in the market.” Why robots won’t be stealing your job or leaving you bankrupt Separately, Clare.AI said its independently developed Cantonese-language NLP engine is dedicated to understanding conversations about finance and achieves an accuracy of over 70 per cent. “We focus on one single vertical. We take proprietary data from banks and take data such as call logs and emails and feed them to our bots. So they are very good at understanding conversations on finance,” said Clare.AI co-founder Bianca Ho, adding that such a dedicated speciality gives them a competitive edge over general-purpose NLP engines. We take proprietary data from banks... and feed them to our bots. So they are very good at understanding conversations on finance Bianca Ho, Clare.AI co-founder Ken Yeung, another co-founder of Clare.AI, said an independent NLP engine enables the start-up to better ensure data privacy and security. “We build the technology based on research and open sourcing,” he said, “We are not relying on Google or Microsoft api or other cloud services so we can deploy our technology securely. That resolves a lot of the security and privacy issues.” Yeung said a great deal of customer service and human resources work will be handled by artificial intelligence, especially in the banking industry which spends about US$150 million a year on customer service. Ho said that unlike chatbots in Mandarin and English, which have long proliferated on various messaging platforms, the market for Cantonese-language chatbots remains uncharted territory, which is now being tapped by local AI start-ups. Salandra said Rocketbots subscribes to third-party NLP engines because he thinks the chatbot business should focus more on being the intermediary between users and NLP engine providers. “It’s too late to build your own NLP from scratch,” he said, “Big companies like Google, IBM and Amazon all have their NLPs and they are hogging most of the data. Who do you think will win out?” AlphaGo's China showdown: why it's time to embrace artificial intelligence According to Salandra, external NLP engines charge chatbot companies US$0.001 to 0.006 for each message processed. Unlike in English where words are separated by spaces, sentences in Chinese are just a chunk of characters Mew Kin Li, founder of Mindlayer Limited However, he agrees there is still demand for homegrown NLP engines, especially in Hong Kong as Hong Kong Monetary Authority requires banks to keep finance-sensitive data in their own servers rather than sharing them with global tech companies. As a result, a dedicated focus on building chatbots for banks would be a good strategy in the current landscape, he added. Mew Kin Li, founder of Mindlayer Limited, another Hong Kong start-up which built its own NLP engine from scratch, said Cantonese language is particularly challenging because it is a mixed language. “As a Chinese dialect, it’s already hard to find boundaries to tokenise each word,” he said, “Unlike in English where words are separated by spaces, sentences in Chinese are just a chunk of characters.” Li added that the lack of local research papers and talent in the field also inhibits development of Cantonese language NLP engines. “Academic papers primarily focus on English,” he said. “When the whole setting is different, results are also very different.” Li said that top engineers in the industry and academia often end up leaving Hong Kong to work on the mainland or in the US.