Karibu kwenye WikiBabel

Muhtasari

Maelezo ya ubora na msanidi wa karibu ni bora zaidi kuliko habari ndogo. Kwa hiyo tulikimbia Wikipedia ya Kiingereza kupitia Google Tafsiri kwa Kiswahili, lugha inayotumiwa na watu milioni 50-100, lakini ambapo habari ndogo inapatikana kwenye mtandao. Ikiwa matumizi yanaonekana vizuri, tutaifanya kwa lugha zingine ambako maudhui yamepungua na Tafsiri ya Google hutoa matokeo mazuri. Kwa nini? Sababu tunapenda Wikipedia na tunataka watu wengine wawe nayo pia, hata kama jumuiya ya wahariri kwa lugha yao ni ndogo.

Toleo la muda mrefu

Taarifa zilizopo kwenye majukwaa ya wavuti kama Wikipedia inaonekana kutokuwa na mwisho - lakini hii sio kwa lugha nyingi. Kwa mfano, kidogo sana ya habari hii inapatikana kwenye internet ya Kiswahili, iliyoongea na watu milioni 50-100 Afrika Mashariki. Kwa mfano, Wikipedia ya Kiingereza imerasa 5.5 Milioni, wakati Swahili ina 38000.
Kutokana na kwamba zana za tafsiri za mashine kama vile Tafsiri ya Google zimepata hivi karibuni sana kutokana na maendeleo katika mitandao ya neural, na kwamba Wikipedia ya Kiingereza ni kubwa na yenye ujuzi, tumekimbia mwisho kwa njia ya zamani (kwa kipangilio cha kurasa kwa sasa), na kusababisha katika hifadhi hii.

Je, msemaji wa Kiswahili hawezi kwenda kwenye Google kutafsiri na kutafsiri ukurasa wa Wikipedia wa Kiingereza wenyewe? Hii itawahitaji kujua kwamba Wikipedia ipo, kwamba ni bora kwa lugha nyingine kuliko yao, na kwamba Google kutafsiri ipo na kutafsiri kurasa. Tunatatua tatizo la kugundua kwa kuzuia matokeo na kuifanya inapatikana kwa injini za utafutaji. Tunalenga kuwa na matokeo ya juu katika matokeo ya injini ya utafutaji kwa maneno ya kiufundi ya Kiswahili (ambapo tulipata habari chini sana kuliko kwa Kiingereza kwa mfano). Tunatakiwa kufuata mfano wa Wikipedia na kushiriki katika programu ya Msingi ya Msingi, ambapo watumiaji wanaweza kufikia maudhui haya bila kulipa kwa bandwidth.

Kwa nini Wikipedia haina kufanya hivyo? Hata marekebisho ya mashine ya kuboresha mashine ni mbali na kamilifu. Wikipedia ina lugha ya juu juu ya kurasa zao na inahitaji idhini ya kibali ili kukubali kurasa. Jamii ya lugha ya Kiswahili inayohusika katika Wikipedia bado ni ndogo sana. Pia kuna hatari ya kuangamiza kwa njia ambayo hutoa taarifa sahihi. Tafadhali heshima uamuzi wa Wikipedia na usiangalie kurasa hizi zote katika Wikipedia isipokuwa msemaji wa asili aliiangalia kwa sarufi sahihi, maana na uhalali wa maudhui. Soma zaidi hapa.

Inashiriki nyuma

Ikiwa mafanikio, maono ya muda mrefu yatakuwa na interface ya uhariri iliyowezeshwa tena na kuruhusu wasemaji wa asili kuzungumza kurasa na kuwaangalia tena katika Wikipedia ikiwa hakuna neno lililopo. Ikiwa makala ipo lakini kwa kiasi kikubwa kwa lugha ya asili, tutafanya interface ya kuunganisha. Vipengele vilivyosahihishwa vitaaminika kuongezwa kwenye mafunzo ya mafunzo ya Google kuweka na kuboresha tafsiri ya moja kwa moja katika lugha hiyo zaidi.

Jaribio

Hypothesis ni kwamba taarifa nzuri na sarufi isiyo ya kawaida bado ni muhimu. Tunatakiwa kuthibitisha au kupinga dhana hii kwa kufuatilia idadi ya watumiaji, na muhimu zaidi kwa muda gani wanaotumia kila ukurasa. Ikiwa imethibitishwa kuwa na manufaa, tunatarajia kutafsiri kurasa zaidi na kufanya mradi huu upatikanaji kwa lugha zingine ambako maudhui yamepungua, lakini tafsiri ya mashine huzalisha matokeo ya kueleweka.

Kwa habari zaidi wasiliana na olya@wikibabel.com

Welcome to WikiBabel

Summary

Quality information with mediocare grammer is better than sparse information. So we ran the English Wikipedia through Google Translate to Swahili, a language spoken by 50-100 milion people, but where little information is available on the internet. If usage looks good, we'll do it for other languages where content is sparse and Google Translate yields good results. Why? Cause we love Wikipedia and want other people to have it too, even if the editorial community for their language is small.

The long version

The information available on web platforms like Wikipedia is seemingly endless -- but this is not the case for many languages. For example, very little of this information is available on the Swahili internet, spoken by 50-100 million people in eastern Africa. As an example, the English Wikipedia has 5.5 Million pages, while the Swahili has 38000.
Given that machine translation tools such as Google Translate have gotten very good recently due to advances in neural networks, and that the English Wikipedia is vast and informative, we’ve ran the latter through former (for a subset of pages for now), resulting in this archive.

Can a Swahili speaker not go to Google translate and translate the relevant English Wikipedia page themselves? This would require them to know that Wikipedia exists, that it’s better in some other language than in theirs, and that Google translate exists and translates pages. We are solving a discoverability problem by caching the results and making them available to search engines. We aim to be fairly high in search engine results for Swahili technical keywords (where we found far less information than on the English internet for example). We intend to follow Wikipedia’s example and participate in the Free Basics program, where users can access this content without paying for bandwidth.

Why doesn’t Wikipedia do this? Even the improved machine translation algorithms are far from perfect. Wikipedia has high quality language on their pages and require community approval for acceptance of pages. The Swahili speaking community participating in Wikipedia is still quite small. There is also a risk of mistranslating in a way that gives incorrect information. Please respect Wikipedia’s decision and don’t check any of these pages into Wikipedia unless a native speaker checked it for appropriate grammar, meaning and content validity. Read more here.

Contributing back

If successful, the long term vision would be to have the editing interface re-enabled and allow for native speakers to correct the pages and check them back into Wikipedia if no article exists. If an article exists but is significantly shorter in the native language, we’ll do a merging interface. The corrected articles will hopefully be added to the corpus of Google Translate training set and improve automatic translation in that language further.

The Experiment

The hypothesis is that good information with imperfect grammar is still useful. We intend to prove or disprove this hypothesis by tracking number of users, and more importantly how long they spend on each page. If proven useful, we intend to translate more pages and make this project available for other languages where content is sparse, but machine translation yields comprehensible results.

For more information contact olya@wikibabel.com