1. Gabatarwa

Haɗa manyan chatbots, musamman ChatGPT, cikin koyon harshe yana wakiltar sauyi mai girma a fasahar ilimi. Wannan binciken yana bincika takamaiman aikace-aikacen tsara umarni (prompt engineering) don amfani da Manyan Harsunan AI (LLMs) wajen koyar da Sinanci a matsayin harshe na biyu (L2). Binciken ya dogara ne akan Tsarin Ma'auni na Turai don Harsuna (CEFR) da aikin Ma'auni na Sinanci na Turai (EBCL), yana mai da hankali kan matakan farko A1, A1+, da A2. Babban hasashe shi ne cewa umarni da aka tsara da kyau za su iya takaita abubuwan da LLMs ke fitarwa don su yi daidai da ƙayyadaddun ƙamus da tsarin haruffa, ta haka ne suke ƙirƙirar tsari mai tsari, muhallin koyo wanda ya dace da matakin.

2. Nazarin Adabi & Bayanan Baya

2.1 Juyin Halitta na Chatbots a cikin Koyon Harshe

Tafiya daga tsarin da ke da ƙa'ida kamar ELIZA (1966) da ALICE (1995) zuwa AI na zamani na ƙirƙira yana nuna sauyi daga hulɗar da aka rubuta zuwa tattaunawa mai ƙarfi, mai sane da mahallin. Tsarin farko suna aiki akan daidaita tsari da bishiyoyin yanke shawara, yayin da LLMs na zamani kamar ChatGPT suna amfani da gine-ginen koyo mai zurfi, kamar samfurin Transformer, wanda ke ba da damar fahimtar harshe na halitta da ƙirƙira wanda ba a taɓa ganin irinsa ba.

2.2 Tsarin CEFR da EBCL

CEFR yana ba da ma'auni daidaitaccen ma'auni don ƙwarewar harshe. Aikin EBCL ya daidaita wannan tsarin musamman don Sinanci, yana ayyana daidaitattun jerin haruffa da ƙamus na kowane mataki. Wannan binciken yana amfani da jerin EBCL A1/A1+/A2 a matsayin ma'auni na zinariya don kimanta bin abubuwan da LLM ke fitarwa.

2.3 Kalubalen Sinanci a matsayin Harshe na Rubutu na Alama (Logographic)

Sinanci yana gabatar da matsalolin koyarwa na musamman saboda tsarin rubutunsa wanda ba na haruffa ba, na alama (logographic). Ƙwarewa yana buƙatar haɓaka gane haruffa, tsarin rubutu (stroke order), lafazi (Pinyin), da wayar da kan sautin lokaci guda ɗaya. Dole ne a shiryar da LLMs don ƙarfafa waɗannan ƙwarewoyi masu alaƙa ba tare da ɗaukar nauyin ɗalibin farko ba.

3. Hanyoyin Bincike & Tsarin Gwaji

3.1 Dabarun Tsara Umarni (Prompt Engineering)

Hanyar binciken ta ta'allaka ne akan tsarin tsara umarni. An tsara umarni don ba da umarni a sarari ga ChatGPT don:

  • Yin amfani da haruffa kawai daga ƙayyadaddun jerin matakin EBCL (misali, A1).
  • Haɗa ƙamus mai yawan amfani wanda ya dace da matakin.
  • Samar da tattaunawa, darussa, ko bayani waɗanda suka haɗa abubuwan baki (Pinyin/sautuna) da rubutu (haruffa).
  • Yin aiki a matsayin malami mai haƙuri, yana ba da gyare-gyare da bayani masu sauƙi.

3.2 Sarrafa Haruffa da Ƙamus

Babban ƙalubalen fasaha shi ne tilasta ƙayyadaddun ƙamus. Binciken ya yi amfani da hanyar biyu: 1) Bayani a sarari a cikin umarni, da 2) Binciken bayan samarwa don auna kashi na haruffa/ƙamus da suka faɗi a waje da jerin EBCL da aka yi niyya.

3.3 Ma'auni na Kimantawa

An auna bin umarni ta amfani da:

  • Ƙimar Bin Tsarin Haruffa (CSAR): $CSAR = (\frac{N_{valid}}{N_{total}}) \times 100\%$, inda $N_{valid}$ shine adadin haruffa daga jerin matakin EBCL da aka yi niyya kuma $N_{total}$ shine jimillar haruffan da aka samar.
  • Bincike mai inganci na dacewar koyarwa da dabi'ar hulɗa.

4. Sakamako & Bincike

4.1 Bin Tsarin Haruffa na EBCL

Gwaje-gwajen sun nuna cewa umarni da suka ambaci jerin haruffan EBCL A1/A1+ a sarari sun inganta bin umarni sosai. Abubuwan da aka samar tare da waɗannan umarnin da aka takaita sun nuna CSAR sama da 95% ga matakan da aka yi niyya, idan aka kwatanta da tushe na kusan 60-70% don umarnin "Sinanci na farko" na gaba ɗaya.

4.2 Tasiri akan Haɗa Ƙwarewar Magana da Rubutu

Tattaunawar da aka umarta ta yi nasarar haɗa bayanan Pinyin da alamomin sauti tare da haruffa, yana ba da ƙwarewar koyo mai yawa. LLM na iya samar da darussan mahallin da ke neman ɗalibai su daidaita haruffa da Pinyin ko gano sautuna, suna ketare shingen "maimaita ƙamus da rubutun Sinanci (sinographic recurrence)".

4.3 Muhimmancin Ƙididdiga na Binciken

Jerin gwajin t-tests sun tabbatar da cewa bambanci a cikin CSAR tsakanin umarnin da aka sanar da EBCL da umarnin sarrafawa yana da mahimmanci a ƙididdiga ($p < 0.01$), yana tabbatar da ingancin hanyar tsara umarni.

Babban Sakamakon Gwaji

Bin Umarnin EBCL: >95% bin haruffa ga matakan A1/A1+.

Bin Umarnin Tushe: ~65% bin haruffa.

Muhimmancin Ƙididdiga: $p < 0.01$.

5. Tattaunawa

5.1 Manyan Harsunan AI (LLMs) a matsayin Malamai na Musamman

Binciken ya tabbatar da yuwuwar LLMs da aka umarta da kyau su yi aiki a matsayin "chatbots na musamman". Suna iya samar da kayan aiki marasa iyaka, masu bambancin mahalli waɗanda aka keɓance ga takamaiman matakin ɗalibi, suna magance babban iyaka na litattafai masu tsayi ko ƙa'idodin app ɗin harshe da aka tsara a baya.

5.2 Iyakoki da Kalubale

Iyaka sun haɗa da: 1) "ƙirƙira" na LLM na lokaci-lokaci wajen gabatar da ƙamus maras niyya, yana buƙatar ƙirar umarni mai ƙarfi. 2) Rashin tsarin tsarin karatun da aka gina a ciki—alhakin ɗalibi ko malami ne ya jera umarni yadda ya kamata. 3) Bukatar kimantawa na mutum a cikin madauki don tantance ingancin koyarwar abubuwan da aka samar fiye da bin ƙamus kawai.

6. Ƙarshe & Ayyukan Gaba

Wannan binciken yana ba da tabbacin ra'ayi cewa dabarun umarni na iya daidaita abubuwan da AI ke ƙirƙira tare da ƙirƙirar tsarin ƙwarewar harshe kamar CEFR/EBCL. Yana ba da hanyar da za a iya maimaitawa don amfani da LLMs a cikin koyo na L2 mai tsari, musamman ga harsunan rubutu na alama (logographic) kamar Sinanci. Ayyukan gaba yakamata su mayar da hankali kan haɓaka tsarin inganta umarni ta atomatik da nazarin dogon lokaci wanda ke auna sakamakon koyo.

7. Bincike na Asali & Sharhin Kwararru

Babban Fahimta

Wannan takarda ba kawai game da amfani da ChatGPT don koyon harshe ba ne; yana da darasi mai zurfi a cikin takaita ƙirƙirar AI don daidaitaccen koyarwa. Marubutan sun gano daidai cewa ƙarfin danye, maras iyaka na LLM abin alhaki ne a cikin ilimin farko. Nasarar su ita ce kula da umarni ba a matsayin tambaya mai sauƙi ba, amma a matsayin takaddar ƙayyadaddun bayani wacce ke ɗaure samfurin zuwa ƙayyadaddun iyakokin tsarin EBCL. Wannan ya wuce gama-gari na "hira da mai magana na asali" kuma ya shiga cikin fagen ƙirar tsarin karatu ta kwamfuta (computational curriculum design).

Tsarin Ma'ana

Hujja ta ci gaba da ma'ana kamar tiyata: 1) Amincewa da matsalar (fitar da ƙamus maras sarrafawa). 2) Shigo da mafita daga ilimin harshe da ake amfani da shi (ma'auni na CEFR/EBCL). 3) Aiwatar da mafita ta fasaha (tsara umarni a matsayin matsalar gamsar da iyakoki). 4) Tabbatar da hujja ta hanyar gwaji (auna ƙimar bin umarni). Wannan yayi daidai da hanyoyin bincike a cikin koyo na inji inda aka ƙirƙiri sabon aikin asara (a nan, umarni) don inganta takamaiman ma'auni (bin EBCL), kamar yadda masu bincike suka tsara ayyukan asara na al'ada a cikin CycleGAN don cimma takamaiman ayyukan fassarar hoto zuwa hoto (Zhu et al., 2017).

Ƙarfi & Kurakurai

Ƙarfi: Mayar da hankali kan Sinanci yana da hikima—harshe ne mai wahala, mai buƙatu sosai inda ake buƙatar mafita masu yawa na koyarwa. Tabbatar da hujja ta hanyar gwajin ƙididdiga yana ba binciken amincewa da yawanci ba a cikin takardun AI-a-cikin ilimi. Kuskure Mai Muhimmanci: Binciken yana aiki a cikin sararin samaniya na bayanan sakamakon ɗalibi. Ƙimar bin haruffa 95% abin burgewa ne, amma shin yana fassara zuwa saurin samun haruffa ko ƙwaƙƙwaran tunawa da sauti? Kamar yadda aka lura a cikin nazari kamar Wang (2024), tasiri mai kyau na chatbots akan aikin koyo a bayyane yake, amma hanyoyin ba haka ba. Wannan binciken yayi magana da kyau game da "ingancin shigarwa" amma ya bar "shan" da "fitarwa" (Swain, 1985) na tsarin koyo ba a auna su ba.

Fahimta Mai Aiki

Ga malamai da masu haɓaka fasahar ilimi: Daina amfani da umarni na gaba ɗaya. Samfurin yana nan—ku kafa hulɗar ku ta AI a cikin ƙirƙirar tsarin koyarwa. Mataki na gaba shine gina ɗakunan karatu na umarni ko tsaka-tsaki wanda ke amfani da waɗannan ƙayyadaddun EBCL/CEFR bisa ga binciken matakin ɗalibi. Bugu da ƙari, binciken ya jaddada buƙatar "APIs na koyarwa"—daidaitattun hanyoyin sadarwa waɗanda ke ba da damar ma'auni na abun ciki na ilimi su sanar da gina tambayoyin LLM kai tsaye, ra'ayin da ake bincika ta hanyar ƙaddamarwa kamar IMS Global Learning Consortium. Gaba ba malamai na AI suna maye gurbin malamai ba ne; malamai na AI ne da aka ƙera su da kyau don aiwatar da iyakar tsarin karatu da jerin da manyan malamai suka ayyana.

8. Cikakkun Bayanai na Fasaha & Tsarin Lissafi

Babban ma'aunin kimantawa ya dogara ne akan ma'auni na bin umarni na yau da kullun. Bari $C_{EBCL}$ ya zama jerin haruffa a cikin jerin matakin EBCL da aka yi niyya. Bari $S = \{c_1, c_2, ..., c_n\}$ ya zama jerin haruffan da LLM ke samarwa don umarni da aka bayar.

An ayyana Ƙimar Bin Tsarin Haruffa (CSAR) kamar haka: $$CSAR(S, C_{EBCL}) = \frac{|\{c_i \in S : c_i \in C_{EBCL}\}|}{|S|} \times 100\%$$

Tsarin umarni yana nufin haɓaka CSAR da ake tsammani a cikin rarraba amsoshin da aka samar $R$ don umarni $p$: $$\underset{p}{\text{maximize}} \, \mathbb{E}_{S \sim R(p)}[CSAR(S, C_{EBCL})]$$ Wannan yana tsara inganta umarni a matsayin matsalar ingantawa ta bazuwar (stochastic optimization).

9. Sakamakon Gwaji & Bayanin Chati

Chati: Ƙimar Bin Haruffa ta Nau'in Umarni da Matakin CEFR
Chati na sanduna zai nuna babban binciken. Axis na x zai wakilci yanayi uku: 1) Umarni na "Na Farko" na Gabaɗaya, 2) Umarni da aka Sanar da EBCL-A1, 3) Umarni da aka Sanar da EBCL-A1+. Axis na y zai nuna Ƙimar Bin Tsarin Haruffa (CSAR) daga 0% zuwa 100%. Sanduna biyu a kowane yanayi za su wakilci sakamako don kimanta matakin A1 da A1+ bi da bi. Za mu lura:

  • Umarni na Gabaɗaya: Sanduna a ~65% ga duka kimanta A1 da A1+.
  • Umarnin EBCL-A1: Sanduna mai tsayi sosai (~97%) don kimanta A1, da sanduna mai matsakaicin tsayi (~80%) don kimanta A1+ (saboda yana ɗauke da wasu haruffan A1+).
  • Umarnin EBCL-A1+: Sanduna mai tsayi (~90%) don kimanta A1+, da sanduna kaɗan kaɗan (~85%) don kimanta A1 (saboda ya zama babban jerin A1).
Wannan chati zai nuna a sarari ribar takamaiman da aka samu ta hanyar umarni da aka yi niyya ga matakin.

10. Tsarin Bincike: Misalin Lamari

Yanayi: Malami yana son ChatGPT ya samar da tattaunawa mai sauƙi ga ɗalibin A1 wanda ke yin gaisuwa da gabatar da kansa.

Umarni Mai Rauni: "Rubuta tattaunawa mai sauƙi a cikin Sinanci don masu farawa."
Sakamako: Yana iya haɗa haruffa kamar 您 (nín - kai, na yau da kullun) ko 贵姓 (guìxìng - sunanka), waɗanda ba ƙamus na A1 na yau da kullun ba ne.

Umarni da aka Ƙera (Dangane da Hanyar Bincike):
"Kai malami ne na Sinanci ga masu farawa gaba ɗaya a matakin CEFR A1. Yin amfani da HARUFFA KAWAI daga jerin haruffan EBCL A1 (misali, 你, 好, 我, 叫, 吗, 呢, 很, 高, 兴), samar da ɗan gajeren tattaunawa tsakanin mutane biyu suna haduwa a karon farko. Haɗa Pinyin da alamomin sauti ga duk haruffa. Ka kiyaye jimloli zuwa matsakaicin haruffa 5 kowanne. Bayan tattaunawar, ka ba da tambayoyin fahimta guda biyu ta amfani da ƙayyadaddun haruffa iri ɗaya."

Sakamakon da ake tsammani: Tattaunawa mai ƙarfi da aka sarrafa ta amfani da kalmomin A1 masu yawan amfani, tare da daidaitaccen Pinyin, yana aiki a matsayin kayan aikin koyarwa wanda ya dace da matakin.

11. Aikace-aikace na Gaba & Jagorori

  • Tsarin Umarni Mai Daidaitawa: Haɓaka tsaka-tsakin AI wanda ke daidaita ƙayyadaddun umarni bisa ga kimanta aikin ɗalibi na ainihin lokaci, ƙirƙirar hanyar koyo mai daidaitawa da gaske.
  • Haɗa Hanyoyi Daban-daban (Multimodal): Haɗa umarni na tushen rubutu tare da gane magana da haɗawa don ƙirƙirar cikakkun kayan aikin aiki na magana/jin wanda kuma ya bi ƙayyadaddun sauti da sauti.
  • Gabaɗaya Tsarin Tsarin: Yin amfani da hanyar guda ɗaya zuwa wasu tsarin ƙwarewa (misali, ACTFL don mahallin Amurka, HSK don gwajin Sinanci na musamman) da sauran harsuna masu ƙaƙƙarfan rubutu (misali, Jafananci, Larabci).
  • Albarkatun Ilimi na Buɗe Idanu: Ƙirƙirar ɗakunan karatu na buɗe ido na umarni da aka tabbatar, na takamaiman mataki don harsuna da ƙwarewa daban-daban, kama da ra'ayin "Promptbook" da ke fitowa a cikin al'ummomin AI.
  • Kayan Aikin Taimakon Malami: Gina kayan aikin da ke ba malamai damar samar da kayan aiki na musamman, daidaitattun takardun aiki, da kimantawa cikin sauri, rage lokacin shirya.

12. Nassoshi

  1. Adamopoulou, E., & Moussiades, L. (2020). An overview of chatbot technology. Artificial Intelligence Applications and Innovations, 373-383.
  2. Council of Europe. (2001). Common European Framework of Reference for Languages: Learning, teaching, assessment. Cambridge University Press.
  3. Glazer, K. (2023). AI in the language classroom: Ethical and practical considerations. CALICO Journal, 40(1), 1-20.
  4. Huang, W., Hew, K. F., & Fryer, L. K. (2022). Chatbots for language learning—Are they really useful? A systematic review of chatbot-supported language learning. Journal of Computer Assisted Learning, 38(1), 237-257.
  5. Imran, M. (2023). The role of generative AI in personalized language education. International Journal of Emerging Technologies in Learning, 18(5).
  6. Li, J., Zhang, Y., & Wang, X. (2024). Evaluating ChatGPT's potential for educational discourse. Computers & Education, 210, 104960.
  7. Swain, M. (1985). Communicative competence: Some roles of comprehensible input and comprehensible output in its development. Input in second language acquisition, 235-253.
  8. Wallace, R. S. (2009). The anatomy of A.L.I.C.E. In Parsing the Turing Test (pp. 181-210). Springer.
  9. Wang, Y. (2024). A meta-analysis of the effectiveness of chatbots on language learning performance. System, 121, 103241.
  10. Weizenbaum, J. (1966). ELIZA—a computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36-45.
  11. Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired image-to-image translation using cycle-consistent adversarial networks. Proceedings of the IEEE international conference on computer vision (pp. 2223-2232).
  12. European Benchmarking Chinese Language (EBCL) Project. (n.d.). An samo daga ma'ajiyar aikin EU da ta dace.
  13. IMS Global Learning Consortium. (n.d.). An samo daga https://www.imsglobal.org/