Riba:
- Ability don rarrabe tsakanin haske da cikakkun matakan gyare-gyare da zabar matakin da ya dace bisa manufa da ganuwa na rubutu
- Ƙarfin aiwatar da ingantaccen kwararar gyare-gyaren da ke gyara abin da ake buƙata kawai, yana guje wa duka gyare-gyare da gyare-gyare.
- Ikon gudanar da zagaye na daban don lamba, ƙin yarda, da lokacin da kuma auna ingancin gyara bayan lokaci tare da gyara nisa
Hanyar da aka fi sani da ƙwararrun fassarar aiki a yau ba fassarar daga karce ba ce, amma MTPE (Machine Translation Post-Editing): na'urar tana samar da ɗan gajeren daftarin aiki, mai fassarar ɗan adam ya gyara shi kuma ya shirya shi don bugawa. A cikin wannan rukunin, zaku koyi matakai guda biyu na gyarawa na yau da kullun (haske da cikakke), yadda ake amfani da wanne rubutu, ingantaccen kwarara bayan gyara, da kuma tarkon "over-editing". Manufar ita ce zama daidaitaccen editan bayan da ke tabbatar da inganci yayin kiyaye lokacin da injin ya ba ku.
Matakai biyu: haske da cikakken gyare-gyare
An bayyana tsarin bayan masana'antu a matakan daidaitattun matakai guda biyu (ISO 18587 shine ma'auni na duniya don wannan kasuwancin):
Haske bayan gyarawa: Manufar ita ce a bayyana rubutun a sarari kuma daidai - ba cikakke ba. Kuna gyara kurakurai masu karkatar da ma'ana, bayanan da ba daidai ba, kurakurai na nahawu; Amma ba ku yin wani salo mai salo don "zai fi kyau". Ya dace da: amfani na ciki, bayani mai sauri, rubutu marar gani, fassarori masu mahimmanci.
Cikakkun gyare-gyare: Manufar ita ce ba za a iya bambanta abin da ake fitarwa da fassarar ɗan adam ba. Ma'ana, kalmomi, sautin, kwarara, sautin - duk dole ne ya kasance na ingancin shirye-shiryen watsa shirye-shirye. Ya dace da: duk rubutu tare da ganuwa iri waɗanda za a buga, isar da su ga abokin ciniki.
Haɗa matakan biyu yana haifar da matsaloli: idan kun yi amfani da gyare-gyaren haske zuwa rubutun da za a buga, ingancin zai zama mara kyau; Idan kun yi cikakken gyare-gyare a kan bayanin ciki, za ku rasa lokaci da kuɗi. Bayyana tare da abokin ciniki matakin da ake sa ran kafin fara aiki.
Tukwici: Lokacin da abokin ciniki ya ce "fassarar injin, arha", galibi suna son "haske" amma suna magana game da rubutun da za a "buga" - tattauna wannan sabani a gaba. Idan tsammanin bai bayyana ba, lokacin da aka gama aikin za ku tambayi "me yasa wannan ba cikakke ba ne?" ko "me yasa wannan ya dauki lokaci mai tsawo haka?" Ana gardama.
Karfin gyarawa
Babban maƙiyin MTPE mafi ƙaƙƙarfan aiki shine gyare-gyare: sake rubuta jumlar da injin ya ba ku daidai amma ba zaɓinku ba, kawai saboda "hakan ne zan rubuta shi." Wannan yana lalata duk fa'idar MTPE. Ƙa'idar a bayyane take: Idan fitarwar injin daidai ne, abin fahimta kuma ya dace, kar a taɓa shi - kawai gyara shi idan ba daidai ba. Canje-canje na ɗanɗano yana iyakance a cikin cikakken gyare-gyare kuma kusan an haramta shi a cikin haske bayan gyarawa.
Akasin haka kuma yana da haɗari: ƙarƙashin gyara - ƙaddamar da mashin ɗin daidai amma jumla mara kyau kamar "ya yi kyau." Ma'auni shine horo na "gyara abin da ya kamata, bar abin da ba dole ba."
Tsanaki: Gyaran baya sau da yawa yana faruwa a cikin novice post-editors, yayin da rashin gyara yana faruwa saboda gajiya da matsin lokaci. Don nisantar ko dai matsananci, "wannan canjin yana inganta ma'ana / daidaito / sautin, ko kuwa dandano na ne kawai?" tambaya.
Ingantaccen kwarara bayan gyarawa
Gudun tafiya zuwa mataki-mataki mai amfani:
- Duba farko, gyara daga baya. Karanta ta cikin rubutun sau ɗaya; duba duka, sautin, da kurakurai masu maimaitawa.
- Tuta manyan abubuwan haɗari. Lambobi, kwanan wata, rashin yarda, sunaye masu dacewa, raka'a - ba waɗannan yawon shakatawa daban.
- Daidaita kurakurai na ma'ana. Duk abin da ke karya amincin tushen ya zo a farko.
- Daidaita lokaci da daidaito. Match Termbase da fassarorin da suka gabata.
- Gyara kwarara da sautin (a cikin cikakkiyar gyarawa). Sanya jimloli masu kamshi kamar fassarar.
- Karatun ƙarshe. Ba tare da kallon tushen ba, kawai karanta rubutun da aka yi niyya kamar mai karatu; Idan kun makale, gyara shi.
Kuna iya amfani da LLM a matsayin "ido na biyu" a cikin wannan kwarara: ta hanyar ciyar da na'urar fitarwa zuwa LLM kuma gaya masa "alama kawai wuraren haɗari," kuna rage makafi - amma yanke shawara na ƙarshe shine naku.
uku mini lokuta
Case 1 - Matsayin da ya dace ya adana lokaci. Wani kamfani ya sami fassarar tushen ilimin cikin kalmomi 40,000 don ma'aikata su fahimce su. Mai fassarar ya yi amfani da haske bayan gyarawa: ya gyara kurakurai 60 waɗanda suka gurbata ma'anar, ba tare da taɓa wani salo ba. An gama aikin 55% cikin sauri fiye da cikakken gyarawa kuma an cimma manufar.
Case na 2 - Biyan Kuɗi na Gyara. Wani sabon editan post-edita ya sake rubuta madaidaitan jimloli don sanya su "fitowa cikin kwanciyar hankali." Ma'aunin ya nuna cewa kashi 70% na jimlolin da aka gyara sun riga sun karɓa; Ƙarin lokacin da aka kashe ya tsawaita aikin kwanaki 2 kafin jadawalin kuma ya shafe ribar. Kungiyar ta rubuta dokar "gyara kawai idan ba daidai ba".
Case 3 - Rashin isasshen gyara kama. A cikin littafin jagorar na'urar likitanci, na'urar ta fassara kalmar "kada ku nutse" a hankali amma ba ta cika ba, tana cewa kawai "tsabta" maimakon "kada ku nutse." Babban abu mai haɗari zagaye ya kama wannan kwaro na tsaro; Wannan ya nuna yadda rashin ingantaccen gyara zai iya rikidewa zuwa hadari.
Samfura huɗu masu kwafi
1) Ƙaddara matakin gyara bayan gyara:
Matsayinku: Mashawarcin aikin MTPE. Zan bayyana makasudi da ganuwa na rubutu. Faɗa mini ko haske ko cikakken gyara bayan ya dace da wannan aikin, dalili, da abin da za a gyara ko barin a kowane matakin.Manufar rubutu/gani: [bayyana]
2) Duban ido na biyu (zuwa LLM):
A ƙasa akwai fassarar tushe da inji. Tuta kurakurai HIGH RISK waɗanda editan baya zai iya rasa: shubuha, tsallake/ƙara bayani, lamba mara kyau/kwanaki/suna, kuskuren ƙin yarda, lokaci mara daidaituwa. Yin shawarwarin salo.Madogararsa: [...] | Fassara: [...]
3) Haske bayan edita umarni:
DAN KYAU bayan gyara fassarar injin ɗin a ƙasa: gyara kowane sassa masu ruɗani, kuskure ko maras tabbas kawai. Kada ku canza salo ko dandano. Bar kowace jumla wadda take daidai kuma mai fahimta kamar yadda take. A taƙaice jera abin da kuka canza kuma me yasa. Fassarar: [...] | Source: [...]
4) Ikon gyarawa (zuwa gyaran ku):
A ƙasa akwai jumlar injin da gyara na. Ga kowane gyara, ka ce: shin wannan canjin (a) daidaitaccen ma'ana/daidaitacce, (b) yana buƙatar kalmomi/daidaitacce, ko (c) zaɓin salo ne kawai? (c) alama wadanda a matsayin "na iya zama ba dole ba".Machine: [...] | Gyaran kaina: [...]
Rauni mai ƙarfi / Ƙarfi mai ƙarfi
Rauni: "Gyara wannan fassarar." (Ba a san matakin ba; LLM ya sake rubuta komai, ya yi gyara.)
Ƙarfafa: "Cikakken gyara wannan fassarar na'ura: sanya ma'anar, ƙayyadaddun kalmomi da ƙwarewa a shirye don bugawa, amma ku kasance da gaskiya ga tushen, kada ku ƙara. Kada ku canza madaidaicin jimloli ba dole ba. Yi lakabin dalilin kowane canji (ma'ana / lokaci / gudana) tare da kalma."
Bambance-bambance: hanzari mai ƙarfi yana ba da matakin, iyaka da buƙatun gaskatawa; yana hana duka gyare-gyare da gyare-gyare.
Jadawalin kwatanta matakin
Girman
Haske bayan gyarawa
Cikakkun gyare-gyare
manufa
Daidai da fahimta
Ingancin fassarar ɗan adam
kurakurai na ilimin harshe
an gyara
an gyara
Daidaiton lokaci
matakin asali
cika
sautin / salo
m
Shirye don bugawa
Ingantaccen salo
ba a yi ba
Iyakance, idan ya cancanta
Rubutun da ya dace
na ciki, wucin gadi, gist
buga, alama
Lokaci / farashi
ƙananan
babba
Kuskuren gama gari
- Ba bayyana matakin tare da abokin ciniki ba. Tsammanin "ya zama mai arha" da "ya zama cikakke" rikici.
- Gyaran baya. Sake rubuta madaidaicin jumla don jin daɗi da lalata fa'idar lokaci.
- Rashin isasshen gyara. Ƙaddamar da jumla mai kyau amma ba daidai ba; Mafi haɗari sune kurakuran aminci/ ƙidaya.
- Ba yin zagaye daban don abubuwan haɗari masu girma. An yi watsi da kurakurai na lamba da ƙima a cikin kwarara.
- Tsallake karatun ƙarshe. Karanta rubutun da aka yi niyya ba tare da duba tushen ba yana nuna duk wani rashin ƙarfi da ya rage.
Ƙimar aunawa: gyara nisa da tsawon lokaci
Yi amfani da sauƙaƙan alamomi guda biyu don auna yadda “darajar” ɗin bayan gyara yake. Na farko shine lokaci: bambanci tsakanin fassarar juzu'i ɗaya daga karce da bayan-gyare-gyare; Shin MTPE a zahiri yana saurin sauri, ko kuma an cika ku ta hanyar gyarawa? Na biyu shi ne edit nesa: shi ne rabo na nawa fitarwa na inji da ka canza. Matsakaicin nisa na gyarawa (kusan babu taɓawa) alama ce ta ƙarancin gyarawa; Idan tazarar gyara ta yi tsayi da yawa (kusan ana sake rubutawa), ko dai rubutun bai dace da na'ura ba ko kuma kuna gyarawa. Idan kun bi waɗannan alamomi guda biyu a cikin ayyuka da yawa, za ku ga zahiri a cikin wane nau'in rubutu na MTPE ke kawo riba, wanda shine mafi hikimar fassara daga karce. Ba za ku iya inganta tsarin da ba ku auna ba; Dubi waɗannan lambobi biyu maimakon hankali.
a takaice
MTPE shine babban hanyar fassarar aiki a yau kuma yana da matakai guda biyu: haske bayan gyarawa, wanda ya sa rubutun ya zama daidai kuma mai fahimta, da kuma cikakken gyarawa, wanda ya kawo shi ga ingancin ɗan adam. Babban editan post-edita ya fayyace matakin kafin fara aiki; yana gyara abin da ya dace kawai kuma yana guje wa gyare-gyare da gyare-gyare; ya yi wani zagaye na daban na duba lamba, rashin amincewa, da lokaci; Yana amfani da LLM azaman saitin idanu na biyu amma yana yanke shawara ta ƙarshe da kansa.
Aikin aikace-aikace
Samu fassarar inji na rubutu na kalmomi 150-200. Aiwatar da haske bayan gyarawa da farko kuma auna lokacin; sannan a yi amfani da cikakken gyare-gyare zuwa rubutu guda (akan kwafi mai tsabta) kuma kwatanta lokaci da adadin canje-canje. A ƙarshe, yi amfani da samfurin “ikon gyaran gyare-gyare” don tantance yawan canje-canjen da kuke yi a zahiri dole ne kuma nawa ne masu salo.
jerin abubuwan dubawa
- [ ] Na fayyace haske/cikakken matakin tare da abokin ciniki.
- [ ] kawai na gyara shi idan kuskure ne; Ban canza madaidaitan jimloli ba tare da larura ba.
- [ ] Na yi wani zagaye na daban na tantance lamba, kwanan wata, suna da rashin kyau.
- [ ] Na tabbatar da jumlar tsaro/mafi mahimmanci tare da tushen akan haɗarin rashin isasshen gyara.
- [ ] Na karanta ba tare da duba tushen ba.