Injini yokusesha yakwa-Google ingelinye lamathuluzi aziwa kakhulu futhi asetshenziswa kakhulu okuthola ulwazi olufanele kanye nokuphendula imibuzo eyahlukene. Injini yokusesha isetshenziswa emhlabeni wonke, okuvumela abasebenzisi ukuthi bathole ngokushesha ulwazi abaludingayo. Yingakho ithimba lokuthuthukisa le-Google lihlala lisebenzela ukuthuthukisa injini yalo yokusesha.
Okwamanje, usesho lwe-Google lubona umbuzo ngamunye njengesethi yamagama, abese efaniswa nemiphumela efanele. Uhlelo alusebenzi kahle ngemibuzo yezingxoxo neyinkimbinkimbi, futhi ukuqonda ulimi sekuyisikhathi eside kuyinkinga eqhubekayo.
Esikhathini esizayo esiseduze, inkampani ihlela ukusebenzisa i-algorithm entsha yokucubungula imibuzo yolimi lwemvelo, ngokusekelwe kunethiwekhi ye-neural ye-BERT (Bidirectional Encoder Representations from Transformers), eyethulwe ngonyaka odlule. I-algorithm iyakwazi ukuhlaziya umbuzo uwonke, ngaphandle kokuwuphula ube ngamagama futhi icabangele izandiso nezihlanganiso. Le ndlela izovumela umongo ogcwele wombuzo, ukuthola izimpendulo ezifanele.
Onjiniyela be-Google bachaza ukudalwa kwe-BERT neural network-based algorithm "njengempumelelo ebaluleke kakhulu yeminyaka emihlanu edlule kanye nesinye sezehlakalo ezibaluleke kakhulu emlandweni wokusesha." Okwamanje, i-algorithm entsha isetshenziselwa ukucubungula ingxenye yemibuzo yosesho lwe-Google eyenziwa ngesiNgisi. I-algorithm ekugcineni izonwetshwa kuzo zonke izilimi ezisekelwayo, kodwa kunzima ukusho ukuthi lokhu kuzokwenzeka nini.
Source: 3dnews.ru
