Find­ing the right word

China Daily (Hong Kong) - - FOCUS - On the brink of bank­ruptcy Grow­ing up Con­tact the writer at [email protected]­nadai­

China now has tech­nol­ogy that can tran­scribe Man­darin, Can­tonese, and English sound bites into text and an­a­lyze their mean­ings. It’s a ma­jor break­through for busi­ness, though some are not pleased. Some di­alect-speak­ers say its lim­ited ap­pli­ca­tion threat­ens to marginal­ize their di­alects. Miles Wen Haofu, founder and CEO of Fano Labs, is not wor­ried. He’s fix­ing the prob­lem.

The ad­van­tage of this new tech­nol­ogy is that con­sumer com­ments on com­pany hot­lines can be recorded, and tran­scribed into text. Cor­po­rate re­search staff can re­view the mes­sages and gauge con­sumer con­cerns ef­fi­ciently. Yet, the sys­tem can­not tran­scribe many deriva­tives of Man­darin. Some peo­ple speak in di­alects that are marked by strong ac­cents, are not widely used and can­not be in­cor­po­rated into the sys­tem in any prac­ti­cal way.

Wen’s AI-pow­ered sys­tem has been up­graded to em­brace South­west­ern Man­darin, the branch of Man­darin spo­ken by 260 mil­lion peo­ple mostly in cen­tral and south­west­ern China. It also rec­og­nizes Chi­nese spo­ken by those who have spent time over­seas and of­ten in­flect their speech with for­eign lan­guages. The sys­tem can iden­tify English words thrown into Can­tonese or Man­darin.

The sys­tem does voice-to-text trans­la­tion us­ing nat­u­ral lan­guage pro­cess­ing tech­nol­ogy, a branch of ar­ti­fi­cial in­tel­li­gence. It teaches com­put­ers the nat­u­ral way hu­mans talk and turns the in­for­ma­tion into data for sys­tem train­ing.

The sys­tem works much in the same way that chil­dren learn lan­guages. Chil­dren hear peo­ple around them us­ing words that even­tu­ally be­come fa­mil­iar. Though not able to re­spond to the un­fa­mil­iar words at first, they ab­sorb ver­bal cues, which form pat­terns in the brain, and lan­guage learn­ing pro­ceeds.

Speech recog­ni­tion tech­nol­ogy works in much the same way. Com­put­ers store huge files of sound sam­ples linked to cor­re­spond­ing text. The com­put­ers are trained to iden­tify the speech pat­terns and make pre­dic­tions as to the most likely trans­la­tion, Wen ex­plained.

Fano Labs can­not match the ad­vances made by tech gi­ants like Baidu at the mo­ment. While Baidu claims an ac­cu­racy rate of 97 per­cent for lan­guage tran­scrip­tion, Fano Labs still has a long way to go in im­prov­ing its ac­cu­racy rate, which cur­rently stands at 80 per­cent.

Though Fano Labs lags be­hind in ac­cu­racy, its abil­ity to trans­late South­west­ern Man­darin in China is unique, Wen said.

Ba­sic recog­ni­tion of sounds is not enough for the AI to carry out ef­fec­tive tran­scrip­tion. It needs to go fur­ther. Speech recog­ni­tion sys­tems have to be able to dis­tin­guish pho­nemes and homonyms to a much greater ex­tent in Chi­nese than in English.

There are some spo­ken words in Chi­nese di­alects that peo­ple can read but can’t write. These words are quite com­mon but still beyond the reach of cur­rent tech­nol­ogy.

“We co­op­er­ate with lin­guists to use the word with the clos­est mean­ing or sound to pre­dict the caller’s mean­ing,” Wen said. The sub­sti­tute words help the sys­tem place the words in con­text and achieve a clearer un­der­stand­ing of the caller’s in­tent.

Bound by a con­tract not to dis­close the names of his cor­po­rate clients in Hong Kong, Wen said the Fano Labs sys­tem is al­ready in use in the call cen­ters of mo­bile net­work providers and prop­erty com­pa­nies.

In Sichuan prov­ince — the main re­gion that speaks the South­west­ern Man­darin di­alect — the com­pany’s cor­po­rate clients in­clude Sichuan Ra­dio and TV Net­work, China Mer­chants Bank and China Tele­com Global.

Fano Labs also in­cludes staff train­ing and eval­u­a­tion as part of its ser­vice to back up voice-to-text trans­la­tion. Taken as a whole, com­pa­nies are able to an­a­lyze the tran­scrip­tions and de­ter­mine the needs of call­ers, Wen said. The startup, a lit­tle more than 2 years old, has al­ready at­tracted ven­ture in­vest­ments from Hong Kong and the main­land.

Last year, the com­pany re­ceived in­vest­ments worth mil­lions of US dol­lars from Hori­zons Ven­tures, the ven­ture cap­i­tal firm which man­ages the pri­vate in­vest­ment of Hong Kong ty­coon Li Ka-shing. It was the first time it in­vested in a Hong Kong tech startup.

The startup to­day has more than 30 em­ploy­ees, and gen­er­ated mil­lions of Hong Kong dol­lars in­come last year. De­spite all that, Wen told China Daily, Fano Labs is still in the red.

Even the most suc­cess­ful en­trepreneurs have tales of woe about their early days. Wen, a con­fi­dent straight-A grad­u­ate of the Univer­sity of Hong Kong, was no ex­cep­tion.

Wen came from Jilin prov­ince in north­east­ern China. He was among the top 2 per­cent in the na­tional col­lege en­trance exam in 2007, and was ad­mit­ted to HKU on a full schol­ar­ship, study­ing in the Depart­ment of Elec­tri­cal and Elec­tronic En­gi­neer­ing. He grad­u­ated with the high­est GPA in his ma­jor.

Wen’s tu­tor Vic­tor Li On-kwok, chair of in­for­ma­tion en­gi­neer­ing at HKU, sug­gested he study for a doc­tor­ate de­gree, to “know some­thing bet­ter than any­one else on the planet”. Wen fol­lowed the ad­vice and got his PhD at HKU.

When he started the busi­ness in 2015, Li and a school­mate came on board. “AI may be a hot topic now but, five years ago, if some­one said his or her PhD was in ar­ti­fi­cial in­tel­li­gence, peo­ple would be skep­ti­cal,” said Wen.

Peo­ple didn’t be­lieve AI had much po­ten­tial. In the first eight months into busi­ness, the en­trepreneurs ap­plied their tech­nol­ogy to seven or eight pro­jects. But none was sold.

By the end of May 2016, things got bet­ter and then much worse over a 24-hour pe­riod. On May 30, when in Chongqing for a Hong Kong Science Park event, Wen got a prom­ise for his first con­tract. Vin­cent Lo Hong-shui, chair­man of Shui On Land, was fas­ci­nated by his ideas.

Head­ing a com­pany that builds shop­ping malls and res­i­den­tial prop­er­ties in China, Lo showed in­ter­est when he heard Wen’s idea that cus­tomers can use AI to find places they were look­ing for in shop­ping malls — all they had to do was ask a ro­bot, and a com­puter mon­i­tor would show the way.

Lo thought it was a great idea, and of­fered to col­lab­o­rate with Wen. It was a ma­jor boost for Wen’s morale at a time when he was get­ting fa­tigued by the seem­ingly in­evitable cold shoul­der to his ideas wher­ever he went.

Wen was ex­cited to share the good news with his part­ners, only to get shocked. The school­mate, one of Wen’s part­ners, took the oc­ca­sion to de­clare he wanted out and was leav­ing the part­ner­ship to take an of­fer from a big com­pany that pro­vides him with a solid in­come.

The part­ner had done all the pro­gram­ming up to that point. Wen’s tu­tor, Li, worked part-time and didn’t have time to write the code. Wen had to do it on his own — a mas­sive project.

The day after break­ing up with his part­ner, Wen drunk him­self into stu­por and woke up at 4 am the next morn­ing, feel­ing driven.

From June 1 that year, Wen met peo­ple from Shui On Land to dis­cuss big plans. He worked dur­ing the day and coded at night, sleep­ing less than two hours a day. That went on for six months.

Wen was nearly broke. He had only HK$1,000 in the bank and could barely keep up with his per­sonal ex­penses.

Wen had spent al­most all his money buy­ing out his former part­ner, and had to bor­row 30,000 yuan ($4,400) from his par­ents. “It was the first time I had asked for money from them since I came to Hong Kong,” Wen noted.

His part­ner’s hasty de­par­ture had put him on the brink of bank­ruptcy.

Hav­ing learnt the les­son from his ex­pe­ri­ence with his ex-part­ner, Wen now makes it a point — al­ways to have a con­tin­gency plan.

After six months’ work, the deal with Shui On Land fell through. Wen was able to carry on. Later, he took an or­der from China Light and Power, a ma­jor elec­tric com­pany in Hong Kong, while the Shui On project was still in devel­op­ment.

CLP wanted Wen to de­velop a chat­bot for its call cen­ter, an­swer­ing fre­quently asked ques­tions. This time, ev­ery­thing came to­gether and Wen’s rep­u­ta­tion started grow­ing. He had proven he was able to tran­scribe spo­ken Can­tonese to text, even when it was mixed with English. Other com­pa­nies with call cen­ters in Hong Kong came knock­ing, look­ing for Fano Labs to build sim­i­lar pro­grams.

Wen saw end­less po­ten­tial in the di­alect tran­scrip­tion mar­ket. While hold­ing on to his edge in Can­tonese, he wanted to ex­plore a Chi­nese di­alect spo­ken by many more peo­ple.

“Nearly 100 mil­lion peo­ple speak Can­tonese around the world, while some 260 mil­lion speak South­west­ern Man­darin in China, where I see a huge mar­ket,” said Wen.

Last March, Fano Labs set up a sub­sidiary in Chengdu, Sichuan prov­ince in south­west­ern China. The sub­sidiary com­pany is de­vel­op­ing an in­ter­ac­tive de­vice for the tourism bu­reau in Meis­han, a city near Chengdu. “If a tourist wants to know the his­tory of an an­cient stone, he can ask, and our de­vice will tell him the story,” Wen said.

He plans fur­ther ex­plo­ration in other south­ern di­alects, such as South­ern Min spo­ken mostly in Fu­jian prov­ince and Hakka.

Li said this will hone their mixed lan­guage pro­cess­ing skills and bring him closer to his dream of dom­i­nat­ing the South­east-Asian mar­ket and, from there, the whole Asia-Pa­cific re­gion. Wen hopes to grow Fano Labs into a uni­corn com­pany.

That goal is writ­ten in the com­pany’s Chi­nese name You Guang Ke Ji — “you guang” means “hav­ing light”, and “ke ji” means tech­nol­ogy. Wen said the aim is to use tech­nol­ogy to il­lu­mi­nate blind spots in in­for­ma­tion tech­nol­ogy.

As Wen looks for­ward to the fu­ture, he also looks back. He has hung on to the note he wrote in that mem­o­rable morn­ing at 4 am, set­ting his own rules of en­gage­ment and driv­ing him to keep on. As he pro­cesses more lan­guages and di­alects, he re­mem­bers the drive, that in­sis­tent voice, which keeps telling him to never give up on suc­cess.

Newspapers in English

Newspapers from China

© PressReader. All rights reserved.