Samfura da za ku iya kwafa (fork)
Wannan babi yana ɗauke da samfuran da za a iya sake amfani da su waɗanda sauran sassan AfriPlaybook ke nuni zuwa gare su (katinan rumbun bayanai (dataset cards), rubutun ƙididdigewa (evaluation scripts), fom ɗin amincewa (consent forms), da tsarin aiki (project charters)) a matsayin ainihin wuraren farawa da za ku iya kwafa, canza suna, da kuma daidaitawa.
Ƙa'idar abin da ya dace ya kasance a nan ita ce: kowane samfuri fayil ɗaya ne, wanda za a iya kwafa (forkable), tare da ra'ayin edita a haɗe, kuma an haɗa shi da babin da ya kawo dalilin yin sa. Idan samfuri ba shi da babi a bayansa da ke bayanin dalilin yin sa, bai dace ya kasance a nan ba tukunna.
Sauke a matsayin PDF
Kowane samfuri a wannan shafin yana da maɓallin "Download as PDF" a kusurwar dama ta sama (ko dama ta ƙasa a kan ƙananan fuskokin na'ura). Yana buɗe akwatin bugawa (print dialog) na manhajar bincikenku (browser) tare da tsarin bugawa na AfriPlaybook (layin launi mai jan ƙasa, kanun labarai masu koren daji, lissafin shafi, da tubalan lamba (code blocks) masu launin rawayar yashi), ta yadda samfurin da aka buga zai kasance mai tambari kuma mai sauƙin karantawa, ba kawai zubar da bayanan Docusaurus ba. Zaɓi "Save as PDF" a matsayin inda za a ajiye a cikin akwatin bugawar. Yana aiki a kan Chrome, Firefox, Safari, da Edge a kan kwamfuta; manhajojin bincike na wayar hannu suna ba da irin wannan zaɓin ta hanyar allon raba bayanai (share sheet) na manhajar.
Akwai su a yanzu
- Samfurin kundin binciken rumbun bayanai (Dataset search log template): kundin cike bayanai na inda kuka bincika rumbun bayanai da ke akwai, yadda kowane ya samu maki a kan jerin abubuwan dubawa na Kafin Ku Fara (Before You Start), da kuma abin da kuka yanke shawara akai. Yi amfani da shi kafin ku tsara sabon rumbun bayanai.
- Samfurin katin rumbun bayanai (Dataset card template): katin da aka samo daga Datasheets-for-Datasets wanda aka faɗaɗa da tambayoyin yanayin Afirka waɗanda AfriPlaybook ke ganin ba za a iya sasantawa a kansu ba (amincewar al'umma, ƙa'idojin rubutu, ɗaukar yaruka, sauya harshe (code-switching), da ƙuntatawa kan sake amfani da shi a gaba). Samfurin kwafa-da-cikewa (fork-and-fill) wanda ya kamata a fitar tare da kowane rumbun bayanai (corpus). An samo asali daga babin rubuta bayanai (documentation) da kuma babin shari'a, amincewa, da haƙƙin mallakar al'umma (legal, consent, and community IP).
- Tushen rubutun ƙididdigewa (Evaluation script skeleton): Lambar farawa ta Python wacce ke tabbatar da manufofin edita na AfriPlaybook: bayar da rahoto ga kowane harshe, bayar da rahoto ga kowane aji, fifita chrF don fassara, fifita CER don magana mai ɗauke da tsarin kalmomi (morphology-rich speech), da kuma tilas ɗin samfurin ƙididdigewar ɗan adam (human-eval sampling hook). Kwafa (fork), saka ma'aunin da kuka zaɓa, sannan ku fitar da bin ƙa'idar kyauta. An samo asali daga ƙa'idojin asali (core principles) da kuma ƙa'idojin edita na babin Kafin Ku Fara (Before You Start).
- Samfurin fom ɗin amincewa (Consent form template): fom ɗin rubuce + tsarin magana da baki + ƙarin amincewar al'umma + tsarin aikin mai shiga tsakani (ombudsperson) waɗanda idan aka haɗa su suke samar da kyakkyawan tsarin amincewa don tattara bayanan NLP na harsunan Afirka tare da masu bayar da gudummawa waɗanda ba su iya karatu da rubutu ba da kuma rumbun bayanai da aka samo daga al'umma. An samo asali daga babin shari'a, amincewa, da haƙƙin mallakar al'umma (legal, consent, and community IP) da kuma Mataki na 0 na babin shigar da harsunan da ba a cika amfani da su ba (long-tail language onboarding).
- Samfurin ƙa'idojin lakabi (Annotation guidelines template): wurin farawa wanda bai dogara da wani takamaiman aiki ba (task-agnostic) wanda aka tsara shi bisa tsarin ƙa'idojin MasakhaNER 2 da AfriSenti. Yana ɗauke da ma'anoni, tsarin yanke shawara tare da lokuta masu sarƙaƙiya (corner cases), tsarin lakabi (label set), tsarin rubutu da kula da alamomin wasali (diacritic), burin IAA da tsarinsa, yanke hukunci (adjudication), da kuma kundin canje-canje (change log) wanda ke nuna sauyin ƙa'idoji a fili. An samo asali daga babin Tsarin Lakabi (Annotation Design) da kuma kowane Nazarin Hali (Case Study) inda sauyin ƙa'idoji ya haifar da matsalolin inganci.
- Samfurin tsarin aiki (Project charter template): yarjejeniyar al'umma ta Mataki na 0 wacce ta ƙunshi manufa, ɓangarorin da abin ya shafa, harshe da tsarin rubutu, iyakar bayanai, haƙƙin mallakar al'umma da lasisi, ƙungiya + lokaci + kasafin kuɗi, gudanarwa, abubuwan da za a samar (deliverables), da ɗa'a. Takardar da ke fitowa daga tuntuɓar amincewar al'umma kuma wacce jagoran aikin, mai ɗaukar nauyi na hukuma, masu kula da al'umma, da mai shiga tsakani (ombudsperson) ke sanyawa hannu. An samo asali daga shigar da harsunan da ba a cika amfani da su ba Mataki na 0 (long-tail language onboarding Step 0).
- Samfurin katin ƙirar fasaha (Model card template): katin da aka samo daga Model-Cards wanda aka faɗaɗa da sassan gaskiyar ƙaddamarwa (deployment-realism) waɗanda AfriPlaybook ke ganin ba zaɓi ba ne (non-optional): jinkirin matakin da ake hari (p50 + p95 a kan matakin wayar da za a yi amfani da ita), raguwar inganci bayan rage girma (quantised quality drop), ƙididdigewar sauya harshe (code-switched evaluation), ɗaukar nau'ikan rubutu (script-variant coverage), ƙwarewar mai amfani wajen sauke ƙirar fasaha ba tare da intanet ba (offline model-download UX), da haɗarin kwaikwayon murya (voice-cloning risk) don TTS. An samo asali daga babin ƙaddamarwa (deployment).
- Samfurin waiwayen nazarin hali (Case-study retrospective template): tsarin waiwaye mai tambayoyi goma sha bakwai da aka yi amfani da shi a duk faɗin babin Nazarin Hali (Case Studies). An jera shi a nan don sauƙin ganowa.
Samfura bakwai a layi ɗaya kowanne
Idan kuna gina sabon aikin NLP na harshen Afirka daga Mataki na 0 zuwa ƙaddamarwa, za ku kwafi (fork) waɗannan samfuran kusan a wannan tsarin:
- Tsarin aiki (Project charter): sanya hannu a kansa tare da al'umma KAFIN tattara bayanai.
- Fom ɗin amincewa (Consent form): tsarin aiki don amincewar mutum ɗaya + al'umma + waɗanda ba su iya karatu da rubutu ba.
- Ƙa'idojin lakabi (Annotation guidelines): masu sarrafa sigogi (version-controlled), tare da kundin canje-canje.
- Rubutun ƙididdigewa (Evaluation script): yana tabbatar da tsarin kowane harshe + kowane aji + fifita chrF/CER kai tsaye.
- Katin rumbun bayanai (Dataset card): ana fitar da shi tare da fitar da rumbun bayanai (corpus).
- Katin ƙirar fasaha (Model card): ana fitar da shi tare da fitar da ƙirar fasaha (model), tare da ƙarin gaskiyar ƙaddamarwa.
- Waiwayen nazarin hali (Case-study retrospective): bayan an fitar da aikin, bayar da gudummawar amsoshi goma sha bakwai ɗin zuwa ga babin Nazarin Hali (Case Studies).
Ƙa'idojin edita don samfura
- Fayil ɗaya ga kowane samfuri. Babu ƙananan samfura a ciki. Idan wurin farawa yana da fiye da ɗaya, raba shi.
- Gajeren kanun bayani da ke bayanin abin da za a canza da kuma dalili. Kowane samfuri yana zuwa da bayanan yadda za a yi amfani da shi a sama; fayil ɗin da ba shi da bayanin mahallinsa (context) tarin hanyoyin haɗi ne kawai, ba samfuri ba.
- Tsoffin tsare-tsare masu ɗauke da ra'ayi (Opinionated defaults). Samfurin da ya bar kowane gurbi fanko ba samfuri ba ne, fom ne. Cika tsoffin tsare-tsaren da abubuwan da editan AfriPlaybook ya fi so (chrF, bayar da rahoto ga kowane harshe, CC BY-NC 4.0, tsarin aiki na haɗin gwiwa) sannan ku bar aikin da ke kwafa (forking) ya canza inda mahallinsa ya bambanta.
- An haɗa shi da babin da ya kawo dalilin yin sa. Duk wani mai karatu da ya fara samun samfuri kuma bai san dalilin da ya sa guraben suke yadda suke ba, ya kamata ya kasance danna ɗaya kacal daga samun dalilin.
- Ana dubawa kowane wata shida. Samfura suna lalacewa da sauri fiye da rubutu; ɗakunan karatu (libraries) suna canzawa, lasisi suna sabuntawa, samfuran bayanai suna tsufa. Kowane samfuri yana zuwa da kwanan watan "last reviewed" (lokacin da aka yi bita na ƙarshe).
Mummunan tsari guda ɗaya (The one anti-pattern)
Kada ku ɗauki samfuri a matsayin cikakken iko kawai don yana cikin AfriPlaybook. Kowane aiki yana da mahallin da samfurin bai sani ba. Samfurin wuri ne na farawa wanda ke kiyaye muku rubuta bayanan da ake maimaitawa (boilerplate); guraben naku ne don ku yi musu tambayoyi. Gurbin da ba za ku iya amsawa da gaskiya ba, gurbi ne da samfurin ya yi kuskure a yanayinku.



