This article is part of a series about the changing face of Chinese tourism.

A nightmare taxi ride sparked by a malfunctioning translation app nearly ruined Chinese traveler Diao Ting’s Japanese vacation.

After arriving in the sprawling metropolis of Osaka just after midnight with suitcases in tow, Diao and a friend got lost on their way to their hotel from a subway station in April last year. The two women hopped in a cab, and to communicate their accommodation’s address to the driver, Diao whipped out her smartphone and opened the translation app Global Translator.

Even though the app claims to provide real-time two-way speech and text translation for 34 languages — including Chinese and Japanese — with an accuracy rate of 95.6 percent, it failed to work as advertised. When the confused middle-aged driver rapidly uttered a string of words into the phone in thick Kansai dialect, the app spewed out gibberish rather than the perfect Chinese the travelers expected.

“I was so disoriented that I felt sick,” Diao, a 29-year-old from Nanjing in eastern China’s Jiangsu province, told Sixth Tone. “We thought we were going to go missing the first day we arrived in Japan.”

It wasn’t until 10 minutes later that Diao recovered from the shock and realized she could use Google Maps to give directions. Finally, after driving around for nearly half an hour, the jaded travelers reached their destination. In the light of day, they discovered their hotel was only a 10-minute walk from where they had hailed the cab.

Apps that can translate from one language to another are popping up all over the world. While most tout their ability to bridge the language gap, the results often fail to meet expectations. Still, entrepreneurs and investors alike seem determined to push ahead, claiming that a perfect translation device is just around the corner.

For Chinese travelers, a quality translation app can be incredibly liberating: While younger Chinese and some of the country’s middle class are proficient in English, many people shy away from independent travel and opt instead to go with travel agencies due to linguistic and cultural concerns. According to survey results published by industry research institution China Tourism Academy, Chinese tourists rated lack of language assistance as their biggest problem in 21 out of the 24 most popular overseas destinations in 2015, the latest year for which data is available.

Today’s translation apps are the product of years of technological development. First, advancements in computer memory managed to shrink volumes of dictionaries into software that could simply translate word by word. Then, as computing power increased, translation programs started to comprehend entire phrases and simple sentences. At the same time, technologies such as voice and optical character recognition have given rise to digital translation tools that can understand language in a range of formats, including text, images, and audio.

The latest development in translation technology focuses on deciphering longer, more complex sentences with a new technique called neural machine translation (NMT). Unlike previous approaches that digest each word or phrase as a standalone unit, the artificial intelligence-powered NMT can pick up patterns within a language, significantly decreasing program training time. If fully developed, the technology could benefit international travelers around the world — particularly from China, the leading outbound tourism market with 122 million outbound travelers in 2016.

Currently, NMT systems are expensive and time-consuming to develop, and they can be easily confounded by uncommon words. Still, many Chinese and foreign companies are stepping up to the challenge. Google — whose services have long been blocked in China — made a comeback in March with a special version of its translation and mapping services for Chinese users that grants access to the app without needing to bypass the Great Firewall, the country’s system of internet filters and controls.

Baidu, on the other hand, enjoys an advantage as the leading search engine in China, giving it access to an enormous data set with which to optimize the performance of its Baidu Translate app. In September, the search engine giant unveiled a hand-held device that can translate speech in near-real time. The as-yet-unnamed translator currently supports Chinese, English, and Japanese.

Meanwhile, the domestic playing field is growing crowded with the entry of several smaller Chinese players. In addition to Global Translator — developed by Beijing Qicai Muzhi Technology Co. Ltd. — NetEase’s Youdao, Tencent’s Mr. Translator-Interpreter & Dictionary, and Sohu’s Sogou Translate all aim to crack Chinese translation. Xiaoyi, a portable device from voice recognition company iFlytek that is similar to Baidu’s new translation gadget made headlines earlier this year when the company’s chairman demonstrated the product to Premier Li Keqiang during the annual meeting of the National People’s Congress. Impressed, Premier Li — himself a fluent English speaker — described the translation technology as “the most advanced.” Xiaoyi can translate between Chinese and five languages: English, Spanish, French, Japanese, and Korean.

However, technology still struggles to translate between the notoriously complex Chinese language and members of the Indo-European language family. According to David Moser, a linguist and academic director of CET Chinese Studies at Beijing’s Capital Normal University — a study abroad program for U.S. college students — there is a strong “family resemblance” between Indo-European languages like English and German, but English and Chinese follow very different patterns. “The very concept of the sentence is different in the two languages, and such fundamental linguistic differences have an enormous impact on translation strategies,” Moser said.

Zhang Min, a computer science professor at Soochow University in Jiangsu, explained that although NMT systems can improve translation between Chinese and an Indo-European language, translation within the Indo-European language family is usually more accurate. “Chinese-to-English translation is like a high school-level quiz, whereas French-to-English translation is like a middle school-level quiz,” said Zhang. “The accuracy rates all improve thanks to NMT, but the existing gap is still there.”

The next technology revolution may well surmount these translation obstacles, but for now, globe-trotters like Diao might just have to accept that international travel will continue to push people out of their mother-tongue comfort zones.

Editor: Colum Murphy.

(Header image: People access different translation apps on their smartphones in Shanghai, Sept. 29, 2017. Wu Huiyuan/Sixth Tone)