Kayan Aikin Tantance Ingancin Fassara da Hanyoyin Aiki dangane da Kayan Aikin CAT

1. Gabatarwa

Babu fassara guda ɗaya da ta dace da wani rubutu, amma akwai fassarori daban-daban da za a iya yin su, kowannensu yana biyan buƙatu daban-daban a fagage daban-daban. Bukatun fassarar shari'a, alal misali, sun bambanta sosai da na talla ko littafin mai amfani dangane da daidaito da bin ƙa'idodin yanki. Kayan Aikin Fassara da Kwamfuta (CAT) sun zama muhimmi don sarrafa rubutun da aka daidaita, masu maimaitawa kamar kwangiloli da takaddun fasaha. A cikin shekaru ashirin da suka gabata, amfani da su ya canza hanyoyin aiki da fahimtar sarrafa fassara gaba ɗaya.

Kayan aikin CAT suna taimaka wa masu fassara ta hanyar ingantawa da sarrafa ayyukan fassara, suna ba da fasalulluka kamar sarrafa nau'ikan takaddu da yawa ba tare da canzawa ba. Haɗa Fassarar Injin (MT), musamman Fassarar Injin ta Jijiyoyi (NMT), ta hanyar ƙari sun ƙara canza fagen, wanda ya haifar da raguwar lokacin isarwa da kasafin kuɗi sosai. Waɗannan canje-canje sun yi tasiri kai tsaye ga saurin da hanyar tantance fassara. A tarihi, tantance inganci wani tsari ne na ɗan adam, yana gabatar da babban "abu na ɗan adam" na zahiri (Zehnalová, 2013). Kayan Aikin Tabbatar da Ingantaccen (QA) na zamani suna wakiltar ƙoƙarin ƙarshe na shawo kan waɗannan iyakoki ta hanyar gano kurakurai na rubutu, rashin daidaituwa, da rashin daidaito cikin sauri.

Wannan takarda ta mai da hankali kan kayan aikin QA masu zaman kansu, waɗanda, a lokacin rubutawa, suna cikin waɗanda aka fi amfani da su saboda sassaucinsu wajen aiki da nau'ikan fayiloli daban-daban, sabanin na cikin gida ko na gajimare waɗanda za su iya iyakance nau'in.

2. Kayan Aikin CAT da Kayan Aikin Taimakon su

Babban abubuwan taimako a cikin yanayin kayan aikin CAT sune Ƙwaƙwalwar Fassara (TMs) da Tushen Kalmomi (Term Bases). Na ƙarshe yana da mahimmanci musamman don gudanar da tantance ingancin fassara.

Ƙwaƙwalwar Fassara (TM) an bayyana ta a matsayin "... cibiyar bayanai na fassarorin da suka gabata, yawanci bisa ga jumla-jumla, neman duk wani abu mai kama da jumlar da ake fassarwa" (Somers, 2003). Wannan aikin ya sa kayan aikin CAT su zama masu tasiri musamman ga rubutun da aka daidaita tare da tsarin maimaitawa.

Tushen Kalmomi yana tabbatar da daidaito a cikin amfani da takamaiman kalmomi a cikin aikin fassara, wanda shine tushen inganci, musamman a fagen fasaha, shari'a, ko likitanci.

3. Ma'auni na Ƙasashen Duniya da Tsarin Ingantaccen

Amfani da ma'auni na ƙasa da ƙasa, kamar ISO 17100 (Ayyukan Fassara) da ISO 18587 (Gyaran Bayan Fassarar Injin), sun kafa tsarin tushe don ayyana "inganci" a cikin ayyukan fassara. Waɗannan ma'auni suna zayyana buƙatu don hanyoyin aiki, albarkatu, da iyawa, suna motsa masana'antar zuwa ƙarin ma'auni na inganci na zahiri da ma'auni. Suna ba da tushen da za a iya saita kayan aikin QA da kuma tantance sakamakonsu.

4. Kayan Aikin QA Masu Zaman Kansu: Halaye da Kwatanta

Ganin cewa ba za a iya haɓaka kayan aikin QA na duniya da suka dace da kowane nau'in rubutu da buƙatun inganci ba, kayan aikin masu zaman kansu da ake da su suna da sifa ɗaya: babban matakin daidaitawa. Masu amfani za su iya ayyana da daidaita ma'auni da ƙa'idodi masu yawa don daidaita tsarin QA zuwa takamaiman buƙatun aikin, buƙatun abokin ciniki, ko nau'ikan rubutu.

4.1 Siffofi na Gama-gari da Daidaitawa

Binciken yau da kullun da kayan aikin QA masu zaman kansu ke yi sun haɗa da:

• Tabbatar da rubutu da nahawu.
• Daidaiton kalmomi dangane da takamaiman tushen kalmomi.
• Daidaiton tsarin lambobi da kwanan wata.
• Ingantaccen alamar (tabbatar da cewa alamun tsari daga tushen an sanya su daidai a cikin maƙasudi).
• Binciken canza raka'a ma'auni.
• Gano sassan da ba a fassara ba.
• Binciken bin ƙwaƙwalwar fassara da aka ƙayyade.

Ƙarfin daidaita hankalin waɗannan bincike da ƙirƙirar ƙa'idodi na al'ada shine babban abin banbance tsakanin kayan aikin.

4.2 Binciken Sakamako na Aiki

Takardar ta haɗa da binciken kwatancen rahotannin sakamako daga kayan aikin QA masu zaman kansu guda biyu da suka shahara (ana nufin takamaiman sunaye amma ba a bayyana su a cikin abin da aka fitar ba). Binciken ya nuna yadda kowane kayan aiki ke aiki lokacin sarrafa fassarar rubutu ɗaya, yana nuna bambance-bambance a cikin rarraba kurakurai, salon rahoto, da nau'ikan batutuwan da aka yi alama (misali, gaskiya mara kyau da kurakurai na gaske). Wannan tabbatarwa ta aiki tana da mahimmanci don fahimtar amincin kayan aikin a cikin yanayin duniya.

5. Ayyukan Masana'antu da Sakamakon Zaɓe (Bayanin Shekaru 12)

Binciken ya haɗa binciken da aka gudanar a cikin shekaru 12 a cikin masana'antar fassara. Waɗannan zaɓe sun bayyana ayyukan da masu fassara, masu bita, manajoji ayyuka, da LSPs (Masu Ba da Sabis na Harshe) suka ɗauka don tabbatar da ingancin fassara. Manyan abubuwan da ke faruwa sun haɗa da ƙara haɗa kayan aikin QA cikin daidaitattun hanyoyin aiki, canjin rawar gyaran ɗan adam tare da MT, da ƙaruwar muhimmancin bin daidaitattun hanyoyin aiki. Bayanin mahalarta yana ba da haske mai inganci game da "dalilin" waɗannan ayyuka, yana haɗawa da bayanan ƙididdiga daga binciken kayan aiki.

6. Fahimta ta Asali & Ra'ayin Mai Bincike

Fahimta ta Asali: Takardar ta gano daidai cewa kayan aikin QA na zamani ba su ne maganin gaba ɗaya na zahiri ba, amma masu tacewa masu daidaitawa. Ƙimarsu ba ta cire hukuncin ɗan adam ba, amma a tsara da ba da fifiko ga bayanan da aka yi hukunci a kai. Canjin gaske shine daga bita na zahiri, gabaɗaya zuwa gyara na tushen bayanai, na tushen batutuwa.

Tsarin Ma'ana: Hujjar Petrova tana bin hanya mai ban sha'awa: 1) Amincewa da zahirin zahiri da bambance-bambance a cikin fassara. 2) Nuna yadda kayan aikin CAT/MT suka ƙera tsarin, suna haifar da sabbin buƙatun sauri da daidaito. 3) Sanya kayan aikin QA a matsayin matakin binciken da ake buƙata don wannan sakamakon masana'antu. 4) Mafi mahimmanci, haskaka daidaitawa a matsayin babban sifa, tare da yarda da rashin yiwuwar maganin da ya dace da kowa—kashi na gaskiya da yawanci ba ya ɓacewa daga tallan kayan aiki.

Ƙarfi & Kurakurai: Ƙarfinsa shine ra'ayinsa na aiki, na ƙasa yana kwatanta sakamakon kayan aiki—wannan shine inda robar ta haɗu da hanya. Bayanan zaɓe na shekaru 12 kyakkyawan ruwan tabarau ne na tsawon lokaci. Duk da haka, babban aibi shine rashin ƙaƙƙarfan tsari, mai ma'auni don tantance masu tantancewa. Ta yaya muke auna daidaito da tunawa da kayan aikin QA wajen gano kurakuran fassara na gaskiya da samar da hayaniya? Takardar ta taɓa kwatanta sakamako amma ba ta daure ta a cikin ma'auni na yau da kullun kamar maki F1 ($F_1 = 2 \cdot \frac{precision \cdot recall}{precision + recall}$). Idan ba tare da wannan ba, da'awar game da "amincin" sun kasance na labari. Bugu da ƙari, yana rage nauyin fahimi na daidaita waɗannan kayan aikin yadda ya kamata—mummunan daidaitawa na iya zama mafi muni fiye da babu kayan aiki kwata-kwata, yana haifar da tunanin aminci na ƙarya.

Fahimta Mai Aiki: Ga LSPs: Ku ɗauki zaɓin kayan aikin QA a matsayin tsarin taswirar daidaitawarsa zuwa mafi yawan bayanan kurakuran ku da buƙatun abokin ciniki. Haɓaka ma'auni na ciki. Ga Masu Fassara: Kada ku kalli tutocin QA a matsayin umarni, amma a matsayin faɗakarwa. Mai yanke hukunci na ƙarshe dole ne ya kasance hankalin ɗan adam mai iyawa wanda ya san mahallin, batun da aka jaddada a cikin ayyukan fasahar fassara kamar na Pym "Exploring Translation Theories". Ga Masu Haɓaka Kayan Aiki: Gaba gaba ba ƙarin bincike ba ne, amma bincike mai hankali. Yi amfani da NMT ba kawai don fassara ba, amma don hasashen kuskure—kamar yadda AI na Grammarly ya samo asali fiye da binciken ƙa'ida mai sauƙi. Haɗa ƙa'idodin AI masu bayyanawa (XAI) don gaya wa mai amfani *dalilin* da ya sa wani abu zai iya zama kuskure, ba kawai cewa yana ɗaya ba.

7. Cikakkun Bayanai na Fasaha & Tsarin Lissafi

Duk da yake takardar ba ta da lissafi sosai, ƙa'idar tushen binciken QA za a iya tsara ta ta ƙididdiga. Babban ra'ayi shine ciniki tsakanin Daidaito da Tunawa.

• Daidaito ($P$): Matsakaicin batutuwan da aka yi alama waɗanda suke kurakurai na gaske. $P = \frac{Gaskiya Mai Kyau}{Gaskiya Mai Kyau + Gaskiya Mara Kyau}$
• Tunawa ($Hankali$): Matsakaicin kurakurai na gaske waɗanda aka yi alama da nasara. $R = \frac{Gaskiya Mai Kyau}{Gaskiya Mai Kyau + Gaskiya Mara Kyau}$

Haɓaka kayan aikin QA ya haɗa da daidaita wannan ciniki, wanda sau da yawa aka taƙaita shi ta maki F1: $F_1 = 2 \cdot \frac{P \cdot R}{P + R}$. Kayan aiki mai babban daidaito amma ƙarancin tunawa yana rasa kurakurai da yawa. Kayan aiki mai babban tunawa amma ƙarancin daidaito yana cika mai amfani da ƙararrawa na ƙarya. "Iri-iri na saituna" da aka ambata a cikin takardar a zahiri yana ba masu amfani damar daidaita ƙofar yanke shawara don fifita daidaito ko tunawa bisa ga buƙatun aikin (misali, babban tunawa don takaddun shari'a, mafi girman daidaito don abubuwan talla).

8. Sakamakon Gwaji & Bayanin Ginshiƙi

Binciken kwatancen sakamakon kayan aikin QA guda biyu na takardar za a iya fassara shi a cikin ginshiƙi:

Ginshiƙi: Kwatancen Sakamakon Kayan Aikin QA na Ƙirƙira don Samfurin Rubutun Fasaha
(Ginshiƙi mai kwatanta Kayan Aiki A da Kayan Aiki B a cikin nau'ikan da yawa.)

• X-axis: Rukunin Kurakurai (misali, Rashin Daidaituwar Kalmomi, Tsarin Lamba, Rubutu, Rashin Daidaituwar Alama, Alamar Rubutu).
• Y-axis: Adadin Batutuwan da aka Yi Alama.
• Ginshiƙai: Ginshiƙai masu launi biyu a kowane rukuni, ɗaya don Kayan Aiki A, ɗaya don Kayan Aiki B.
• Lura: Ginshiƙin zai iya nuna cewa Kayan Aiki A yana alamar ƙarin yuwuwar "Alamar Rubutu" da batutuwan "Salo", yayin da Kayan Aiki B ya fi tsanani akan "Rashin Daidaituwar Alama" da "Kalmomi". Wannan a zahiri yana nuna cewa kayan aiki daban-daban suna da hankali na tsoho daban-daban da saitin ƙa'idodi, suna haifar da rahotanni daban-daban daga tushen kayan. Za a iya nuna jadawalin layi na biyu da aka lulluɓe yana nuna ƙimar gaskiya mara kyau (wanda aka tabbatar da hannu), yana nuna cewa ƙimar alama mafi girma ba ta daidaita da mafi girman daidaito ba.

9. Tsarin Bincike: Nazarin Lamarin da ba na Lamba ba

Yanayi: Wani LSP yana fassara jerin igiyoyin UI na software don na'urar likita daga Turanci zuwa Jamusanci.

Aikace-aikacen Tsarin:

1. Ayyana Ma'auni na Ingantaccen: Dangane da ISO 18587 da buƙatun abokin ciniki, ayyana ma'auni masu mahimmanci: 1) Rashin haƙuri ga kurakuran kalmomi daga tushen kalmar likita da aka amince. 2) Tsananin daidaito don saƙonnin gargaɗi. 3) Tsarin lamba/kwanan wata bisa ga ma'aunin DIN. 4) Ƙuntataccen tsawon UI (babu ambaliya).
2. Daidaituwar Kayan Aiki:
   • Loda tushen kalmar likita na musamman na abokin ciniki kuma saita binciken kalmomi zuwa "kuskure."
   • Ƙirƙiri ƙa'idar QA na al'ada don alamar duk wata jumla da ta wuce haruffa 50 don yuwuwar ambaliyar UI.
   • Saita binciken tsarin lamba zuwa yankin Jamusanci (misali, 1.000,00 don dubunnan).
   • Kashe binciken zahiri kamar "salo" ko "jimla mai ban tsoro" don wannan abun ciki na fasaha.
3. Haɗa Tsarin: Kunna kayan aikin QA bayan daftarin fassara na farko kuma sake yin shi bayan gyaran bayan. Yi amfani da rahoton farko don jagorantar edita, na biyu a matsayin ƙofar yarda ta ƙarshe kafin isarwa.
4. Bincike: Kwatanta ƙididdigar kurakurai tsakanin daftari da na ƙarshe. Tsari mai nasara yana nuna raguwa mai tsanani a cikin kurakurai masu mahimmanci (kalmomi, lambobi) yayin da ƙananan tutoci na iya ci gaba. Wannan yana haifar da bambancin inganci mai ma'auni don rahoton abokin ciniki.

10. Aikace-aikacen Gaba & Hanyoyin Ci gaba

1. Bincike Mai Ƙarfin AI, Mai Fahimtar Mahalli: Matsawa bayan ƙa'idodi masu tsayi, kayan aikin gaba za su yi amfani da NMT da Manyan Samfuran Harshe (LLMs) don fahimtar mahalli. Misali, maimakon kawai alamar rashin daidaituwar kalma, kayan aikin na iya ba da shawarar kalmar da ta dace dangane da yankin rubutun da ke kewaye, kamar yadda samfuran GPT na OpenAI ke yi a cikin koyo na ciki.
2. Hasashen Maki Ingantaccen: Haɗa siffofi daga kayan aiki kamar TAUS DQF ko samfuran ƙididdige ingancin fassara (kamar yadda cibiyoyi kamar Jami'ar Edinburgh suka bincika) don hasashen makin inganci don sassan ko gabaɗayan ayyuka dangane da amincin MT, tarihin mai fassara, da tarihin tutocin QA.
3. Haɗin Kai na Tsarin Aiki & Haɗin Kai: Ci gaba zuwa daidaitattun APIs (kamar waɗanda ƙungiyar GALA ke haɓakawa) suna ba da damar kayan aikin QA su shiga cikin kowane yanayin CAT ko TMS (Tsarin Gudanar da Fassara), tare da bincike na ainihi, mai mu'amala maimakon sarrafa tara.
4. Mayar da hankali kan Kurakuran Aiki da Al'adu: Bincike mai ci gaba don gazawar aiki (misali, matakin da bai dace ba na yau da kullun ga al'adar manufa) da mahallin gani (don multimedia/ƙaddamarwa), yin amfani da hangen nesa na kwamfuta don bincikin fassarar rubutu-a-cikin-hoto.
5. Mataimakan AI Na Musamman: Ci gaba daga kayan aikin alamar kuskure zuwa mataimakan haɗin gwiwa masu himma waɗanda ke koyon takamaiman salon mai fassara da tsarin kurakurai na gama-gari, suna ba da shawarwari na riga-kafi yayin aikin fassara kanta.

11. Nassoshi

1. Petrova, V. (2019). Kayan Aikin Tantance Ingancin Fassara da Hanyoyin Aiki dangane da Kayan Aikin CAT. A cikin Proceedings of the 2nd Workshop on Human-Informed Translation and Interpreting Technology (HiT-IT 2019) (shafi na 89–97).
2. Somers, H. (Ed.). (2003). Kwamfutoci da Fassara: Jagorar mai fassara. John Benjamins Publishing.
3. Zehnalová, J. (2013). Subjektivita a objektivita v hodnocení kvality překladu. Časopis pro moderní filologii, 95(2), 195-207.
4. International Organization for Standardization. (2015). ISO 17100:2015 Ayyukan fassara — Bukatun ayyukan fassara.
5. International Organization for Standardization. (2017). ISO 18587:2017 Ayyukan fassara — Gyaran bayan fassarar injin — Bukatun.
6. Pym, A. (2014). Bincika ka'idodin fassara (bugu na 2). Routledge.
7. Specia, L., Shah, K., de Souza, J. G., & Cohn, T. (2013). QuEst - Tsarin ƙididdige ingancin fassara. A cikin Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics: System Demonstrations (shafi na 79-84).
8. TAUS. (2020). Tsarin Ingantaccen Ingantaccen. An dawo daga https://www.taus.net/dqf