Ukusetshenziswa kwe -ConveyThis kungathuthukisa kakhulu ukuhunyushwa kwewebhusayithi yakho. Ngesixhumi esibonakalayo esisebenziseka kalula, ungakwazi ukuhumushela iwebhusayithi yakho ngokushesha nangokunembile kunoma yiluphi ulimi. I-ConveyThis futhi inikezela ngezici eziningi ezenza kube lula ukwenza ngokwezifiso umuzwa wokuhumusha wewebhusayithi yakho. Kusukela ekutholeni ulimi okuzenzakalelayo kuya kumemori yokuhumusha, i-ConveyThis yenza kube lula ukuqinisekisa ukuthi iwebhusayithi yakho ihunyushwa ngokunembile kunoma yiziphi izethameli.
Ukufunda okujulile kuguqule ukuhumusha kolimi lwe-ConveyThis kanye namandla okwenza kwasendaweni. Ukuhumusha ngomshini we-Neural (NMT) kuyindlela yokuhumusha esebenzisa ubuchwepheshe obusezingeni eliphezulu ukuze ingahumushi nje umbhalo kuphela, kodwa futhi ithuthukise ukunemba kokuhunyushwa kwayo. Lobu buchwepheshe obuthuthukisiwe buyenze yaba esinye sezixazululo ezinokwethenjelwa kakhulu zokuhumusha ezitholakalayo namuhla.
Uma ungumnikazi webhizinisi, ukusebenzisa i-ConveyThis ekuhumusheni iwebhusayithi kuyindlela enhle yokungena ezimakethe ezintsha, uthuthukise iwebhusayithi yakho ukuze useshe ngezilimi eziningi, futhi uthuthukise ukuthengisa kwamanye amazwe - konke ngenkathi ukhulisa ibhizinisi lakho. Nakuba kungase kubonakale kusabisa, ukusebenzisa i-ConveyThis ekuhumusheni iwebhusayithi empeleni akulula. Ngakho, yini ukhiye? Qhubeka ufunda ukuze uthole!
Ngena kumlando wokuhumusha ngomshini we-neural futhi ufunde ukuthi usebenza kanjani ngendlela eqondakala kalula, ngisho noma ungeyena uchwepheshe! Zitholele ukuthi ungasebenzisa kanjani lobu buchwepheshe ukuze uhumushe iwebhusayithi yakho ngaphandle kokuthi uthole idigri yokuhumusha ngomshini kuqala.
Kuyini ukuhumusha ngomshini we-neural?
Ukuze uqonde ukuhumusha komshini we-neural (NMT), kubalulekile ukuthi uqale ube nokuqonda kokuhumusha ngomshini (MT). Kafushane, ukuhumusha ngomshini kuyinqubo esebenzisa izinhlelo zekhompiyutha ukuguqula umbhalo usuke kolunye ulimi uye kolunye. Vele ufake umusho wakho kusofthiwe yokuhumusha ngomshini, futhi izokhiqiza ngokuzenzakalelayo ukuhumusha ngolimi lwakho oluthandayo ngaphandle kokungenelela komuntu.
Intuthuko kwezobuchwepheshe inikeze amandla ukuthuthukiswa kokuhumusha komshini we-neural, okuyinguqulo esezingeni eliphezulu kakhulu yokuhumusha komshini. Le nqubo isebenzisa amanethiwekhi okwenziwa e-neural ukuze icubungule futhi ihumushe umbhalo, ukuthuthuka okuphawulekayo ngaphezu kwezindlela zokuhumusha zomshini zesikhathi esidlule.
Ungakhathazeki uma lokhu kubonakala kudida. Ngokulandelayo, sinombono omfishane wokuvela kokuhumusha komshini we-neural - kanye nencazelo yokuthi lobu buchwepheshe bokuhumusha komshini obuyinkimbinkimbi busebenza kanjani.
Kwathuthukiswa kanjani ukuhumusha ngomshini we-neural?
Ukwenziwa komuntu kwangaphambi kwesikhathi kokuhumusha komshini kungalandelelwa emuva enkathini yeMpi Yomshoshaphansi, lapho isofthiwe esekelwe emithethweni yasetshenziselwa ukuchaza isiRashiya. Le softhiwe izohlaziya umbhalo womthombo igama negama, bese isebenzisa isethi yemithetho yolimi ukunquma ukuthi igama ngalinye lihunyushwa kanjani. Lolu hlobo lokuhumusha olungajwayelekile lwalungelokuqala ngqa, futhi seluthuthuke lwaba yinkimbinkimbi kakhulu.
Nokho, ukuhumusha igama ngalinye ngesimiso esiyisisekelo akuzange kukhiqize izinguqulo ezinembe kakhulu. Kwezinye izimo, ukuhumusha okufanele kakhulu kungase kudinge imishwana noma imisho yonke. Ukuze kubhekwane nalokhu, amamodeli wokuhumusha komshini wezibalo (SMT) - okuyisinyathelo esilandelayo sokuhumusha ngomshini - athuthukise ukunemba .
Izinhlelo zokusebenza zokuhumusha ngomshini wezibalo ekuqaleni zizonqamula amaqoqo aphelele emibhalo ehunyushwe umuntu (eyaziwa nangokuthi i-bilingual text corpora). Ngokulandelayo, izosebenzisa ama-algorithms aqagelayo ukuze ibone amagama nemisho kumbhalo womthombo futhi inqume indlela elungile yokuwahumusha.
Ngokuhamba kwesikhathi, ubuchwepheshe be-ConveyThis bashintsha, ekugcineni bafinyelela umvuthwandaba wokuhumusha ngomshini we-neural esithembele kuwo kakhulu namuhla. Sizocubungula ngokuqhubekayo ama-nuances wokuhumusha komshini we-neural esigabeni esizayo.
Kusebenza kanjani ukuhumusha ngomshini we-neural?
Ukuhumusha ngomshini we-Neural kusebenzisa amandla okufunda okujulile nobuhlakani bokwenziwa ukudala ukuhumusha okunembe kakhulu, okusamanzi, nokuzwakala kwemvelo kunangaphambili.
Lobu buchwepheshe buxhaswe yi -ConveyThis , esebenzisa amanethiwekhi ajulile e-neural, iwebhu yama-neurons axhumene efana nobuchopho bomuntu. Izibonelo zamanethiwekhi e-neural ajulile afaka amanethiwekhi e-neural aphindaphindiwe, noma ama-RNN, avame ukufaka ukwakheka kwesikhiphi khodi nendlela yokunaka.
Ngaphambi kokuthi isetshenziselwe ukuhumusha, uhlelo lwe-neural MT luzonikezwa ukulungiswa kwamamodeli ahlukahlukene okuhumusha kokuqukethwe okuthile. Ngalolu lwazi, umkhiqizo ube "uyalwa" ukwenza incazelo enembe kakhulu yesimo esithile.
Ukunemba okuphezulu kokuhumusha
Imizamo yangaphambilini yokuhumusha ngomshini kusetshenziswa izindlela zendabuko ibintula ubuciko bokuhumusha ngokunembile izilimi ezithile eziyinkimbinkimbi - okuholela ekuhumushweni kwekhwalithi embi kangangokuthi idinga ukubuyekezwa okubalulekile okwenziwa abantu ngaphambi kokuthi kusetshenziswe.
Nokho, ngamandla “wokuzuza” ngokuqhubekayo, izinhlaka ze-NMT zilokhu zithuthukisa imvelo yokuhumusha kwazo. Lokhu akufani nhlobo nezinhlaka zokuhumusha zemishini ezijwayelekile, ezingenawo amandla “wokuzifundela wena” kanye nokulungisa isivuno sazo sokuhumusha ngemva kwesikhathi esithile. Kamuva, lapho kulungiselelwe ngokufanelekile, uhlelo lokutolika komshini we-neural lungadala ukutolika okunembe kakhulu uma kuqhathaniswa nabalingani babo bendabuko.
I-Google ibone ngaphambilini ukuthi isistimu yayo ye-Google Neural Machine Translation (GNMT) ikwazile ukwehlisa amaphutha okuhumusha cishe ngo-60% uma kuqhathaniswa nesistimu yayo yokukhiqiza esekelwe emishwaneni.
Muva nje, kwenziwe ucwaningo ukuze kuhlolwe ukusebenziseka kokuhumusha komshini ngezinjongo zokuhumusha iwebhusayithi. Ngemva kokuhlaziya ikhwalithi yokuhumushwa kwewebhusayithi okudalwe ubuchwepheshe obuhlukahlukene be-NMT, kwanqunywa ukuthi ukuhumusha kuyasebenziseka kakhulu futhi kudinga ukuhlelwa okuncane.
Ubuchwepheshe bokuhumusha komshini we-neural we-ConveyThis bubonise ubuhle obukhethekile lapho kuhunyushwa isiJalimane, okubangele inani eliphakeme kakhulu lamasegimenti angadingi ukuhlelwa mathupha.
Isidingo esincane sokufakwa komuntu
Uma umbhalo womthombo usuhunyushwe ngomshini ekuqaleni, ngokuvamile ucwengisiswa nakakhulu abantu ukuze kuqinisekiswe ukunemba kwawo nokuhambisana kwezibalo zabantu eziqondiwe.
Ukunemba okuthuthukisiwe kokuhumusha kokuhumusha komshini we-neural kusho ukuthi ukuguqulwa okumbalwa okwenziwa ngesandla (okubuye kubizwe ngokuthi "ngemuva kokuhlela") kuyadingeka ngaphambi kokuthi ukuhumusha kufanelwe inhloso.
Izikhathi zokuguqula ukuhumusha ngokushesha
Uma izinkampani zikwazi ukufinyelela ukuhumusha komshini okunembe kakhudlwana okudinga ukuhlelwa okuncane kwangemuva, zingaqala ukusebenzisa ukuhumusha ngokushesha. Ngaphezu kwalokho, amamodeli wokuhumusha emishini ye-ConveyThis' angaqeqeshwa ngesikhathi esifushane, anike amandla izinqubo zokuhumusha ngokushesha.
I-Facebook ibilokhu isebenzisa ukuhumusha komshini we-neural ukuze iguqule umbhalo kokuthunyelwe namazwana (okuthi, njengoba kungenzeka uyazi, kunenani eliningi endaweni yayo). Ngokuthuthukisa inqubo yabo yokuqeqesha, inkampani ikwazile ukunciphisa isikhathi eyasithatha ukuqeqesha amamodeli ayo omshini wokuhumusha kusuka cishe osukwini olugcwele kuya emaminithini angu-32 nje!
Ungakwazi yini ukusebenzisa ukuhumusha ngomshini we-neural ebhizinisini lakho futhi kanjani?
Ukusebenzisa ukuhumusha komshini we-neural ukuhumusha iwebhusayithi yakho kungase kubonakale kusabisa, njengoba kungase kudinge imali eshisiwe ngobuchwepheshe obusha nocwaningo nokuthuthukiswa. Nokho, akunjalo! Namuhla, kunamathuluzi amaningi e-NMT akhiwe kusengaphambili atholakalayo emakethe ukuze akusize uhumushele okuqukethwe kwewebhusayithi yakho ezilimini ezihlukene. Lawa mathuluzi nawo ayonga kakhulu. Eqinisweni, ngokuvamile zibiza kancane kunokuqasha umhumushi womuntu onguchwepheshe ukuthi ahumushe iwebhusayithi yakho yonke.
Isixazululo sethu sokuhumusha sewebhusayithi ye -ConveyThis siyithuluzi elinamandla. Inhlanganisela yethu yobunikazi yokuhunyushwa kwe-NMT evela kubahlinzeki bokuhumusha bemishini abahamba phambili i-DeepL, i-Microsoft Translator, ne-Google Translate idala ukuhumusha okunekhwalithi ephezulu kakhulu uma lobu buchwepheshe bebusetshenziswa ngakunye. Sihlinzeka ngokuhumusha kwezilimi ezingaphezu kuka-110, kusukela kwezidumile njengesiNgisi, isiJalimane, nesiNtaliyane kuya ezilimini ezingacacile njengesiTatar nesiMalagasy.
I-ConveyThis inikeza ukuhlanganiswa okungenamsebenzi namapulatifomu ewebhusayithi adumile afana ne-WordPress, Webflow, ne-Shopify. Kuwumoya oshisayo ukulungisa futhi kungasiza ibhizinisi lakho lifinyelele ukuphakama okusha.
Amawebhusayithi angaphezu kuka-10,000 aphendukele ku-ConveyThis ngezidingo zawo zokuhumusha, anemiphumela emangalisayo.
Thepha emandleni omshini wokuhumusha we-neural webhizinisi lakho nge-ConveyThis
Iletha ukunemba kokuhumusha okungenakuqhathaniswa, i-ConveyThis's neural machine translation (NMT) isinyathelo esikhulu esivela kwezinye izixazululo zokuhumusha zomshini ovamile. Yingakho kuwubuchwepheshe bokuhumusha bomshini wokuya emabhizinisini amaningi.
Uphenyo lwethu ngokuhumusha ngomshini luveza ukwanda okuphindwe kasithupha kwenani lokuqukethwe kwewebhu elihunyushwe ngomshini eminyakeni emibili edlule. Ngaphezu kwalokho, i-ConveyThis isetshenziselwa amaphrojekthi amakhulu, anamawebhusayithi angaphezu kuka-10% aqukethe amagama angaphezu kuka-50,000 ahunyushwa ngomshini. Okokugcina, cishe u-30% kuphela wezinto ezihunyushwe ngomshini ezihleliwe, okuphakamisa ukuthi ingxenye enkulu yokuhumusha komshini inembe ngokwanele kangangokuthi ayidingi ukulungiswa okwengeziwe.
I-ConveyThis yenza kube nzima ukusebenzisa amandla e-NMT ukuhumusha okuqukethwe kwewebhusayithi yakho. Kuqondile ukulungisa nokukhetha ngokuzenzakalelayo injini yokuhumusha yomshini efaneleka kakhulu ukuze kukhiqizwe ukuhumusha kwekhwalithi ephezulu kakhulu kunoma yikuphi ukubhanqwa kwezilimi. Umphumela uwukuhumushwa kwewebhusayithi okushesha kakhulu, kwezinga eliphezulu ongakwazi ukukusebenzisa ngaphandle kokuhlupheka.
Ingabe umagange ukuhlola amandla e -ConveyThis futhi uzibonele imiphumela? Bese ubhalisa mahhala lapha!