I-Lilt: I-Neural Human + Machine Feedback Loop yokuhumusha nokwenza okwasendaweni

Izintambo

Izintambo wakhe i-loop yokuqala yomqondo womuntu + yomshini wokuhumusha. I-Lilt's ukuhumusha komshini we-neural Isistimu ye- (NMT) ingeyokuqala ngohlobo lwayo embonini yezobuchwepheshe bokuhumusha futhi idlulela ngale kweminikelo evela kuGoogle, Amazon, Facebook, Apple, noma iMicrosoft. Amabhizinisi afisa ukwandisa ukufinyelela kwawo emhlabeni wonke manje anendlela engcono yokuhumusha okuqukethwe kwawo ngokushesha nangokunembile.

Uma kukhulunywa ngokuhumusha, amabhizinisi abe nezinqumo ezimbili kuphela:

  1. Umusho ogcwele ukuhumusha ngomshini njenge-Google Translate.
  2. Ukuhumusha komuntu.

I-Lilt inika amandla okuhamba phambili kuwo womabili umhlaba ngokuhlanganisa ubuhlakani bokufakelwa namandla abantu ukuthola ikhwalithi yokuhumusha ehamba phambili. Uhlelo luka-Lilt lwe-NMT lusebenzisa ubuchwepheshe obufanayo be-neural obusivele busetshenziselwa ukuqhubekisela phambili ukubonwa kwenkulumo nokwakheka kwezithombe, kodwa umthelela wayo embonini yokuhumusha kusekusha futhi kuyathembisa. Ezinyangeni ezedlule, i-NMT inconywe ngochwepheshe bemboni ngekhono layo lokufanisa ikhwalithi yokuhumusha komuntu kanye nohlelo olusha lukaLilt akunjalo.

Ku-Lilt's neural feedback loop, abahumushi bathola iziphakamiso ezincike kokuqukethwe ngenkathi basebenza. Uhlelo lwe-NMT lubuka nje izintandokazi zomhumushi ukuze zivumelanise iziphakamiso zalo ngesikhathi sangempela. Lokhu kudala umjikelezo omuhle lapho abahumushi bethola iziphakamiso ezingcono kakhulu, futhi umshini uthola impendulo engcono ngokuya ngokwengeziwe. I-neural feedback loop iholela ekuhumusheni kwekhwalithi ephezulu yomuntu nomshini, okusiza amabhizinisi ukusebenzela amakhasimende amaningi, ukunciphisa izindleko, nokunciphisa isikhathi sokumaketha. I-Lilt ibiza ngama-50% ngaphansi futhi iphindwe kasikhombisa kuya ku-3.

Ipulatifomu kaLilt inikeza okulandelayo:

  • Ungalokothi uphinde uqeqeshe ama-MT Systems - Uhlelo olusebenzisekayo, oluguquguqukayo lokuhumusha umshini luvuselela imemori yalo yokuhumusha kanye nohlelo lwe-MT esikhathini esingaphansi komzuzwana isikhathi ngasinye lapho umhumushi eqinisekisa ingxenye ethile.
  • Ukuxhuma Okungenamthungo Kwabantu Nemishini - Hlanganisa ukuhumusha komuntu nomshini nezinye izinhlelo zebhizinisi nge-API esekwe emazingeni. Noma sebenzisa uhla olukhulayo lukaLilt lwezixhumi zangokwezifiso.
  • Ukuphathwa kwe-Agile Project - Ideshibhodi yeProjekthi yeKhanban ikuvumela ukuthi ubone ngeso lengqondo isimo samanje samaphrojekthi weqembu lakho nomsebenzi wokuhumusha.

Ideshibhodi yePhrojekthi ye-Lilt

Esifundweni sokuqhathanisa esiyimpumputhe esenziwe yiZendesk, abahumushi bacelwa ukuthi bakhethe phakathi kokuhumusha okusha kwe-NMT kukaLilt nohlelo lwangaphambilini lweLilt lokuhumusha umshini (MT). Abasebenzisi bakhethe ukuthi i-NMT ibe nekhwalithi efanayo noma ephezulu kunezinguqulo zangaphambilini ezingama-71% zesikhathi.

Siyakuthanda ukuxhumana phakathi komhumushi ongumuntu kanye nekhono lakhe lokuqeqesha izinjini zethu ze-MT. Kwasho ukuthi lapho senza utshalomali ekuhumusheni kwabantu, kuzophinde kube nomthelela kwikhwalithi yezinjini zethu ze-MT. UMelissa Burch, umphathi wezokuxhaswa kwe-inthanethi eZendesk

Abasunguli be-Lilt uJohn DeNero noSpence Green bahlangana ngenkathi besebenza ku-Google Translate ngo-2011, futhi baqala i-Lilt ekuqaleni kuka-2015 ukuletha ubuchwepheshe kumabhizinisi anamuhla nabahumushi. I-Lilt inikeza izixazululo zebhizinisi kanye nokuhumusha kwe-ecommerce futhi.

UCABANGANI?

Le sayithi isebenzisa i-Akismet ukunciphisa ugaxekile. Funda ukuthi idatha yakho yokuphawula isetshenziswa kanjani.