Dataset Viewer
Auto-converted to Parquet Duplicate
corrupted
stringlengths
1
541
clean
stringlengths
1
550
कृीम गुर्ूङ्ग
कृतिम गुरुङ्ग
साथमा नागर शैलीमा निर्मित एउटा शिवालय पनि तथा कृष्ण मन्दिर पनि छ।
साथमा नागर शैलीमा निर्मित एउटा शिवालय पनि तथा कृष्ण मन्दिर पनि छ।
ररि बनवर स्ालपोथ्री लानपूचछेोक टूँड र खुट्टा शुणतइा रतो हुनछ् भेन टाउोक काो हुुन्
शारीरिक बनावट स्यालपोथ्री लामपुच्छ्रेको ठुँड र खुट्टा सुन्तला रातो हुन्छ भने टाउको कालो हुन्छ
उनका पूसफछरु बरज्ञनों भाा प्रकाश हूणछन् भमे लोसा आहिले स्पपेनको ररजधा म्थााडर्डमा बसदे आएरु छन्
उन्का पुस्तकहरू दर्जनौं भाषामा प्रकाशित हुन्छन् भने लोसा अहिले स्पेनको राजधानी म्याडि्रडमा बस्दै आएका छन्
तीन दिवसीय समारीहका अन्तिम दिन उदयपुरबाठ ३० किमी टाढा ग्रामीण अश्वलमा गणगीरकी ठुली मैला लाग्छा
तीन दिवसीय समारोहका अन्तिम दिन उदयपुरबाट ३० किमी टाढा ग्रामीण अञ्चलमा गणगौरको ठुलो मेला लाग्छ।
सक्‍छत्‌।
सक्‍छन्।
अध्यदेशले एके यररावधमभ भरर््ना भऐाकर्मचरि कसैलाई हूनपाउने सुमश्चितता गरढिएकोो र कषबैट त्यौ अधिरुार छिनकेो न्दै कर्नचारी आन्दोलनमा छन
अध्यादेशले एकै प्रावधानमा भर्ना भएका कर्मचारी कसैलाई मुख्यसचिवसम्म हुन पाउने सुनिश्चितता गरिदिएको र कसैबाट त्यो अधिकार छिनेको भन्दै कर्मचारी आन्दोलनमा छन्
्यस्ते अरवििार्का षूर्बपर्मूख बस्न्यात पुरववसचिवहूर अधधकाीर, पम्तविर ुथ समैत ्तियारे ूद्दा दायर गरैौ धीयो
त्यस्तै अख्तियारका पूर्वप्रमुख बस्न्यात, पूर्वसचिवहरू अधिकारी, पन्तविरुद्ध समेत अख्तियारले मुद्दा दायर गरेको थियो
कानमा कुनै दबाई हाल्नमन्दा 'पहिला योग्य चिकित्सकसँग सल्लाह अवश्य लिनुहोस्‌।
कानमा कुनै दबाई हाल्नभन्दा पहिला योग्य चिकित्सकसँग सल्लाह अवश्य लिनुहोस्।
तैहहरथुममा लाथपरिवरतन अनूफुललण ऊर्जा कक्ेत्र्को वीकशफा लागि जिल्ला यवलावू धा उर्जा थजन िरनणको कामम आन्तीम चणरम पूगकेौ छ
तेह्रथुममा जलवायु परिवर्तन अनुकूलन तथा ऊर्जा क्षेत्रको विकासका लागि जिल्ला जलवायु तथा ऊर्जा योजना निर्माणको काम अन्तिम चरणमा पुगेको छ
घघटनाको जौथो दिन समितीले गुोली हान्म शंललगम सुरटको स्कैच ब्रनाएकौ र् मौटरसइकल नम्बरसमेषुइेक गरेरुो छौ
घटनाको चौथो दिन समितिले गोली हान्न संलग्न सुटरको स्केच बनाएको र मोटरसाइकल नम्बरसमेत खुलेको दाबी गरेको हो
४२८ रनफो लक्ष्य पछयाएोक अषटरेलियाा ३७ औथराम २ ३९र्नना सनेटीयो
४८२ रनको बिशाल लक्ष्य पछ्याएको अष्ट्रेलिया ३७ ओभरमा २ सय ३९ रनमा समेटियो
टंंफ कूनारी कार्
टंक कुमारी कार्किी
हरिश्जन्ढ्र शरम
हरिश्चन्द्र शर्मा
चिताा बनाउदा षिन नृतकरुो उमैरौ गणणाका अथरम दाऊर्को राखिम्छ
चिता बनाउँदा पनि मृतकको उमेरको गणनाका आधारमा दाउराको सङ्ख्या राखिन्छ
रैशम छर्मफार
रेशम चर्मकार
"हृदयाघातपछि जीवन बेकार, अनुपयोगी अनि निरर्थक हुन्छ यो घारणा अहिले चाँडो बद्ठिएको छ।"
"हृदयाघातपछि जीवन बेकार, अनुपयोगी अनि निरर्थक हुन्छ यो धारणा अहिले चाँडो बद्लिएको छ।"
ललावति केवट
ललावती केवट
सुरेस बहाुदर बी.क.
सुरेश बहादुर वी.क.
ससं १०२६ शममम) मनोबाि्जञान मन/मास्तिषफकौ वज्ञानक र्पूभाा व्यवहीक जण(ई
सं १९२० सम्म) मनोविज्ञान मन/मास्तिषकको विज्ञानक रूपमा व्यवहारीक चरण(ई
ुययुतसुको नपि विकर्णौजसतै बािचारधार र मामसकिा थीथौ
युयुत्सुको पनि विकर्णको जस्तै विचारधारा र मानसिकता थियो
भुंडी राग्ररी
भुँडी राम्ररी
मञू गजुरेलल
मञ्जु गजुरेल
नेयाम७४४ सथथान हत लागू हूदँा ससरहाा जििल्इाम साबािकका गवतीषुर, मझ रा र नवर्जपूर् गाउँँ वकास ीतिहरूलइ समेटे नवरायपरु गउँापलिकाबनाइऐको छौ
नेपालमा ७४४ स्थानीय तह लागू हुँदा सिराहा जिल्लामा साविकका भगवतीपुर, मझौरा र नवराजपुर गाउँ विकास समितिहरूलाई समेटेर नवराजपुर गाउँपालिका बनाइएको हो
महोौत्सवको
महोत्सवको
त्ोय टाऊकौ ाठिएौ हाीलाइ पनि गगटोको ूपमा आर््रुो प्ारणी बनए
त्यो टाउको काटिएको हात्तीलाई पनि गँगटोको रूपमा अर्को प्राणी बनाए
रव्बन मया
रव्वनी मिया
बीजा छमौटका मापढण्ड ५० प्रतिशत फेशबुरु मोटीङ र वामाहहा ो-रुमभीजिटबा र बँकि ० प्र्िश ुजर प्यामलकौ निर््णयबट नुै बतइएक
बिजेता छनौटका मापदण्ड ५० प्रतिशत फेसबुक भोटिङ र यामाहा शो-रुम भिजिटबाट र बाँकी ५० प्रतिशत जुरी प्यानलको निर्णयबाट हुने बताइएको छ
न्दर्भ सममग्रिहरू बाहय कीहरु सल्थान जील्ला गाऊं िककाससमतिहरु सल्थान जील्लाका ठठाऊँछर्ू गाउँहरू विकथन-४ ब्रनइएक बाा शुधाििएा लैखहरू
सन्दर्भ सामग्रीहरू बाह्य कडीहरू सल्यान जिल्ला गाउँ विकास समितिहरू सल्यान जिल्लाका ठाउँहरू गाउँहरू विकिथन-४ मा बनाइएका वा सुधारिएका लेखहरू
कौशीला दैी यौशि
कौशीला देवी जोशि
शकटास्य त्राम पति सामान्य छ।
शकटास्य नाम पनि सामान्य छ।
नैत्र्कोना बड्गुलादैशकोउत्तरी भुभगभा षर््े एक जिलला ह
नेत्रकोना () बङ्गलादेशको उत्तरी भूभागमा पर्ने एक जिल्ला हो
उमी आफ्नो कुस्ीत कारयकाललमा एज्ञै लीको नामले पर्ीचचिित छिन
उनी आफ्नो कुस्ती कार्यकालमा एजे लीको नामले परिचित छिन्
नमीया देवी धौरी
मनीया देवी चौधरी
क्षयको चर्ण इहासककययाडभरसरुो उषयोगुकौ इतिछाश सतो रहेको छ ज वविादद वैज्ञनिक प गरी र नयं-नयां खोज आनुसनघानले भरीऐको छ
क्षयको चरण इतिहास क्याडभर्सको उपयोगको इतिहास यस्तो रहेको छ जो विवाद, वैज्ञानिक प्रगति र नयाँ-नयाँ खोज अनुसन्धानले भरिएको छ
सूर्य्का प्रकाशय ऊर्जालाई अवशोषित गुेरर क्लौोरफलिको अणुउत््तैजित हूनछ्न्
सूर्यका प्रकाशीय ऊर्जालाई अवशोषित गरेर क्लोरोफिलको अणु उत्तेजित हुन्छन्
"यी सबैमन्दा एउटा अति सुन्दर एवम्‌ छरितो पक्षी सेता रडको समुद्री चिल हो, यसको दर्शन तपाइँले त्यस समय गर्न सक्रहुन्छ जब यो पानीमा आफ्रा सिकारका लागि पौडिरहन्छा।"
"यी सबैभन्दा एउटा अति सुन्दर एवम् छरितो पक्षी सेता रङको समुद्री चिल हो, यसको दर्शन तपाईँले त्यस समय गर्न सक्नुहुन्छ जब यो पानीमा आफ्ना सिकारका लागि पौडिरहन्छ।"
२५-३०
१५-३०
आईराम लआे मगर
आईराम आले मगर
पूर््व-पश्चभ ारअइिबट मचिराज्ञनार्गभा ७८ कलोभिटरफो बरी पपारर गरेयछिइला सददरमुकाम पुगिन्छ
पूर्व-पश्चिम राजमार्गको चारआलीबाट मेची राजमार्गमा ७८ किलोमिटरको दुरी पार गरेपछि इलाम सदरमुकाम पुगिन्छ
"दोस्ो जब जुनको अन्तिम सप्ताहमा अड॒ग्गुर पाक्छ तब त्यसै बेला मनसुनको झरी सुरु हुन्छ, जसद्वारा फल फुटेर कुहुन थाल्छा"
"दोस्रो जब जुनको अन्तिम सप्ताहमा अङ्गुर पाक्छ तब त्यसै बेला मनसुनको झरी सुरु हुन्छ, जसद्वारा फल फुटेर कुहुन थाल्छ।"
डेन्‍्द्रफ
डेन्ड्रफ
उषद्र्बामाचउनने; बीचाइँ गुर्नेउयदर््ाहहा
उपद्रव मचाउने; बिजाइँ गर्ने, उपद्र्याहा
पश्चिमी घाठमा
पश्‍चिमी घाटमा
सबै भन्द बढी षरचलममा र्हहेोक झरण्ड सन १९९९मा एकपारलैङ्िक महला मोनिक हैल््म्सले नबाकएा धिऐ
सबै भन्दा बढी प्रचलनमा रहेको झण्डा सन् १९९९मा एक पारलैङ्गिक महिला मोनिका हेल्म्सले बनाएका थिए
तर अनलाान पत्र्कार्िता तयो बन्धनबाट मूकत हून्
तर अनलाईन पत्रकारिता त्यो बन्धनबाट मुक्त हुन्छ
वैराग ना
वैराग- ना
सेनश इल ुमसहर्
सनेश लाल मुसहर
फ़ूम कूाार्ि राकशकौटि
फुम कुमारी राकसकोटी
ल्लललत बहाादुर गोतम
ललित बहादुर गौतम
नेपालको नेवारी तथा मै थील समूदाथमा करलोफौ ठा ँठलाई औइलाई मसला र कागुी मोलैर् सिशीमा खँेदर चा बननाइछ चसललाइ नेव रि भाशान सलच' र भैथिइ भाषाना 'गेडल' भनीनछ
नेपालको नेवारी तथा मैथिल समुदायमा कर्कलोको डाँठलाई केलाएर ओइलाई मसला र कागती मोलेर सिसीमा खाँदेर अचार बनाइन्छ जसलाई नेवारी भाषामा 'सलचा' र मैथिली भाषामा 'गेड्ली' भनिन्छ
"डा रविकान्तका अनुसार हरिया सब्जीलाई फ्रिजमा राखेर दोस्नो-तेत्नो दिन पकाएर भोजन गर्दा क्यान्सरको खतरा हुनसक्छ, जुन व्यक्तिले सर्चैंताजा सब्जीको प्रयोग गर्छ, ऊ क्यान्सरका खतराबाट प्रायः जोगिस्छ।"
"डा रविकान्तका अनुसार हरिया सब्जीलाई फ्रिजमा राखेर दोस्रो-तेस्रो दिन पकाएर भोजन गर्दा क्यान्सरको खतरा हुनसक्छ, जुन व्यक्‍तिले सधैँ ताजा सब्जीको प्रयोग गर्छ, ऊ क्यान्सरका खतराबाट प्रायः जोगिन्छ।"
यौ षनि हेर््नुुहौस नस्दरभ सागु्रीहरू ब्हय कडिहरू नैालमापाइने चराछर्ू बजजरहु
यो पनि हेर्नुहोस् सन्दर्भ सामग्रीहरू बाह्य कडीहरू नेपालमा पाइने चराहरू बाजहरू
रुरण सँह वस्नेत
किरण सिंह वस्नेत
समदरभ सामगु्ी मेपाल फ्रिरुेठ खैाडीछ रू जीबाि जीवीत नैपीलछर् सन १९०१ मा जनन्म बागलुङ जिल्लाका मानिसहरू
सन्दर्भ सामग्री नेपाली क्रिकेट खेलाडीहरू जीवनी जीवित नेपालीहरू सन् १९९१ मा जन्म बागलुङ जिल्लाका मानिसहरू
स्वास्थ्य बिस्तारे बिस्तारै खस्किन थाल्छ।
स्‍वास्थ्य बिस्‍तारै बिस्‍तारै खस्‍किन थाल्‍छ।
ूशइ भडल
सुशील मंडल
ामग बढ्दा ब्रढ्छ र् अपर्तिबढेुनसार मागु बढेन पचानउ सफैन मने जबर घट्छ
माग बढ्दा बजार बढ्छ र आपूर्ति बढेअनुसार माग बढेन, पचाउन सकेन भने बजार घट्छ
नेत्र दाब गर्ने इच्छालाई स्ैं घ्यान दि्वन्छ।
नेत्र दान गर्ने इच्छालाई सधैँ ध्यान दिइन्छ।
शन् १३० को बशकदैखि विम्नन आन्दोलन तथ सङग्ामहरुमा यस बथागको अुलमीथ ययोगबानरहैोक छ
सन् १९३० को दशकदेखि विभिन्न आन्दोलन तथा सङ्ग्रामहरूमा यस विभागको अतुलनीय योगदान रहेको छ
'यस खाल्डो ड्रलागदा खाले धेरै गोहीले मरिएको थियो।
यस खाल्डो डरलाग्दा खाले धेरै गोहीले भरिएको थियो।
अमेरिकामा यसपटक मतदानका लाग यौग्य नपालि अनेरिकमहरुौ संखया ५० छजार भनद धरै र्हैको अणुमान गरिएको छ
अमेरिकामा यसपटक मतदानका लागि योग्य नेपाली अमेरिकनहरुको संख्या पनि ५० हजार भन्दा धेरै रहेको अनुमान गरिएको छ
सबै घस्मा लुकै।
सबै घरमा लुके।
भौहसमदिम आलम
मो.हसमुदिन आलम
छेउमे देखिन्छ। ब्हेल तटका
व्हेल तटका छेउमै देखिन्छ।
ढिइयो, आपनो
दिइयो, आफ्नो
िवभिन्म ढेवीदेवताकौ यूचाअर्चनाको इगि अवस्यक य्रबनध मिल ाए
विभिन्न देवीदेवताको पूजाअर्चनाको लागि आवस्यक प्रबन्ध मिलाए
स्वसजाति
स्वरजाति
मान लाम तामाङ्
मान लामा तामाङ्
"उल्लेखनीय छ, हरितक्रान्तिका सुरुका समयमा हाप्रा भूमिमा केवल नाड़ट्रोजन र फस्फोरसको अभाव थियो, अतः उनत बीज, सिँचाड़ जल र नाइ़द्रोजत वा फस्फोरस भएका उर्वरकहरुका प्रयोगद्वारा कृषि उत्पादनमा आशातीत वृद्धि भयो।"
"उल्लेखनीय छ, हरितक्रान्तिका सुरुका समयमा हाम्रा भूमिमा केवल नाइट्रोजन र फस्फोरसको अभाव थियो, अतः उन्नत बीज, सिँचाइ जल र नाइट्रोजन वा फस्फोरस भएका उर्वरकहरूका प्रयोगद्वारा कृषि उत्पादनमा आशातीत वृद्धि भयो।"
अहिलेका व्यस्त जीवनमा सुगन्धित तेलको प्रयोग कम्ती हुँदे गइरहेको छ तथा महिलाहरु यी दिन ब्युटी पार्लरको शरण लिन थालेका छब्‌।
अहिलेका व्यस्त जीवनमा सुगन्धित तेलको प्रयोग कम्ती हुँदै गइरहेको छ तथा महिलाहरू यी दिन ब्युटी पार्लरको शरण लिन थालेका छन्।
इशश्वरि नाया षौडेल
इश्वरी माया पौडेल
सट
सिट
वैजथन्ती श्र्ेष्ड
वैजयन्ती श्रेष्ठ
शतिााम नोनियां
सिताराम नोनियाँ
वो तरिचम्दरकलैजक पिरमा परदछ
हाल यो त्रिचन्द्र कलेजको परिसरमा पर्दछ
ठििगम मगर
डिगम मगर
विनठरी गरुुङ
विन्छरी गुरुङ
जसत्ै'सोठर' नप न्रो नैपाइोक ऐउठा ोचफ रुथयक्त गुाउँदै ननाच दे आएको लकोसङङ्गु ितरुो एउटा आनर काव्यफो कहिल्वै नओइइइिनै फुल
जस्तै 'सोरठी' पनि हाम्रो नेपालको एउटा रोचक कथायुक्त गाउँदै नाच्दै आएको लोकसङ्गीतको एउटा अमर काव्यको कहिल्यै नओइलिने फूल हो
शरका रको उचच तहमा िदेसा लगनाी रर देशालई छानी छुँदैनन भन्ने छ
सरकारको उच्च तहमा विदेशमा लगानी गरेर देशलाई हानी हुँदैन भन्ने छ
रिड्कल्स
रिङ्कल्स
"९१ बीसम्र न भाारतियसेनहरुल्लै ढिवलााईई प्तुगुलक वशबठ मूकत गर्ाए।"
"१९ दिसम्बर, १९६१ मा भारतीय सेनाहरूले दिवलाई पर्तुगलीका वशबाट मुक्त गराए।"
मारगकौ अडानमाझूरुनू र स्वतन््तर राख्टरको गरमा बाई रखने सवालम अहल्लेसम्कमा प्राणमनत्र्ीहरुभन्दा ेधर्ै अघि छन ओली
भारतसँगको अडानमा नझुक्नु र स्वतन्त्र राष्ट्रको गरिमा बचाई राख्ने सवालमा अहिलेसम्मका प्रधानमन्त्रीहरुभन्दा धेरै अघि छन ओली
धधेरजसो कर्वक्रम शीक्ाआनि यानकारिसंग सभ्नधित हनु्थे।
धेरजसो कार्यक्रम शिक्षा अनि जानकारीसँग सम्बन्धित हुन्थे।
िणा फूामीर सरार
सरिना कुमारी सरदार
भनि आइसिंडिएस
अनि आइसिडिएस
थो अवस्ाथफा करण हामीले सोचैजजसो कामगर्न नसककेका हौ
यो अवस्थाका कारण हामीले सोचेजस्तो काम गर्न नसकेका हौं
पौरणिक कथनन आुसर नार्ीहरू रजशवला हुद जामैर्नाजनीनै गिएका जिि पनि पपरहु छन् तिबट मूक््ति पाऊनका ल्लागुि यससो बत बसिम्छ
पौराणिक कथन अनुसार नारीहरू रजस्वला हुँदा जानेर वा नजानीकनै गरिएका जति पनि पापहरू छन्, तिबाट मुक्ति पाउनका लागि यसको ब्रत बसिन्छ
फ्नो दाटौमाा तगारोका रूपमा देखा पर्े हाटएर मले नकौ सथानम आफ़ूूष्रति बफदार कननिष्ठ कार््यकरालाइ लयएकाा छन्
आफ्नो बाटोमा तगारोका रूपमा देखा पर्ने सबैलाई हटाएर किमले तिनको स्थानमा आफूप्रति बफादार कनिष्ठ कार्यकर्तालाइ ल्याएका छन्
शुल्लभ पप्ररशाद गौतम
सुलभ प्रसाद गौतम
यसकारण यसका गुणलाई यथावत राख्नका लागि यसलाई खुल्लामा राख्नुपर्छ अनि चाँडो प्रयोग गर्नुपर्छ।
यसकारण यसका गुणलाई यथावत राख्‍नका लागि यसलाई खुल्लामा राख्‍नुपर्छ अनि चाँडो प्रयोग गर्नुपर्छ।
डलली राम भुजले
डिल्‍ली राम भुजेल
पुल्लिंगी ज्ञीवको लंिगु हौ यशले मा मोदाइल गुवामेटसस यैद गर सलााईकौ सष्म या शु्करणु पनि भम्छन्
पुलिंगी एक जीवको लिंग हो जसले साना मोबाइल ग्यामेट्स पैदा गर्दछ जसलाईको स्पर्म या शुक्राणू पनि भन्छन्
कम्पीलेवफादरीपुरक श्र ५ मनाान र शूरकषा गर्ीअऐको’ यीयौ
कम्पनीले वफादारीपूर्वक श्री ५ को सम्मान र सुरक्षा गरिआएको’ थियो
जजटयु नमको गीद्थले सतलाइ बचान तर ररवारणसंंगकौ थुदधा उनकौ पखे ा कटािवो
जटायु नामको गिद्धले सीतालाई बचाउन खोजे तर रावणसंगको युद्धमा उनको पखेटा काटियो
वणीभनदा मन, मनभनदा सकप, षकल्पभन्ाद चितत् शरेषठ छ
वाणीभन्दा मन, मनभन्दा संकल्प, संकल्पभन्दा चित्त श्रेष्ठ छ
लेषिय र साऊढीमा आरथ्िक संकट चूलिएपनि बीदैशिन चाछनाेको षंखय अब फर्ि पुरानै अबस्थमा फररुनिे देखिऐको छ
मलेसिया र साउदीमा आर्थिक संकट चुलिएपनि विदेशिन चाहानेको संख्या अब फेरि पुरानै अवस्थामा फर्किने देखिएको छ
"यसपक्ठि क्रमशः वार्ड बोड़, नर्स अनि डाक्टरका मोबाड़ल फोनहरू सड्रूमित थिए।"
"यसपछि क्रमशः वार्ड बोइ, नर्स अनि डाक्टरका मोबाइल फोनहरू सङ्क्रमित थिए।"
चाल्नेछाँ|
चाल्नेछौँ।
लगाउनुहोस्‌|
लगाउनुहोस्।
End of preview. Expand in Data Studio

Nepali OCR Proofreading Dataset (Devanagari)

Dataset Summary

A Nepali-only (Devanagari script) text-correction dataset built for fine-tuning a small language model (target: HimalayaGPT 0.5B) as an OCR proofreader. Each example is a (corrupted, clean) pair: corrupted is Nepali text with OCR/handwriting-style errors (character confusions, missing matras, merged/split words, transposed or dropped characters), and clean is the correct text it should map to.

The set is compiled from six sources — a printed-OCR corpus, a dictionary, a news corpus, a synthetic handwriting-confusion model, a names list, and a large web-text corpus — deliberately sized and balanced rather than used in raw form, per the curation notes below.

  • Final training set: 1,000,000 pairs (train.jsonl)
  • Test set: 10,001 pairs (test.jsonl) — the untouched, official RoundTripOCR-nepali test split
  • Format: JSON Lines, two fields per row: corrupted, clean
  • Script / language: Nepali written in Devanagari (ne)

Supported Tasks

  • Text editing / grammatical-and-OCR error correction: given noisy OCR/handwriting-style Devanagari text, produce the corrected text. Intended for sequence-to-sequence or causal LM fine-tuning (corrupted → clean).

Languages

Nepali (ne), Devanagari script only. Two supplements deliberately mix in a small amount of English:

  • The code-switching rows insert English loanwords (e.g. proper nouns, institution names) into otherwise-Nepali sentences, mirroring real Nepali documents that code-switch this way.
  • All other sources were filtered to be Devanagari-dominant (see per-source notes below); no general-purpose English text is included.

Dataset Structure

Data Instances

{"corrupted": "गाइडले हामीलाई सिंहका खुद्दाको चिह देखाए।", "clean": "गाइडले हामीलाई सिंहका खुट्टाको चिह्न देखाए।"}

Data Fields

Field Type Description
corrupted string Nepali text with OCR/handwriting-style errors
clean string The correct Nepali text

Data Splits

Split Rows Source
train 1,000,000 Compiled and size-capped, see composition below
test 10,001 Official cfilt/RoundTripOCR-nepali test split, untouched

The official RoundTripOCR-nepali validation split (10,001 rows) was folded into train rather than shipped separately — only a train/test split is provided here.

Dataset Creation

Curation Rationale

The dataset was built to avoid three failure modes common in synthetic OCR-correction data: (1) source-specific artifacts (e.g. font-driven error patterns from one rendering pipeline) dominating the mix, (2) a single supplement (a names list) so large it would have swamped every other signal, and (3) an unrealistically huge final size for a 0.5B-model fine-tune. All three are addressed explicitly in the pipeline below.

Source Data

1. cfilt/RoundTripOCR-nepali (Hugging Face) — the base corpus. Printed-text OCR round-trip pairs across 49 distinct fonts.

  • Raw: 2,970,148 train rows
  • Deduplicated + min-length filtered: 2,197,411
  • Font-balanced (capped at the per-font median, 44,931): 1,965,156
  • Numeral-only/punctuation-only diffs downsampled: 1,965,156 substantive + kept-as-is minority classes
  • Length-bucket balanced (short/medium/long): 78,675 rows
  • Plus 11,801 controlled identity examples (clean == clean, for anti-over-correction / copy-bias balance) and 184,631 word-level pairs extracted from the sentence-level diffs (higher signal-to-noise)
  • Plus the official 10,001-row validation split, folded in
  • Combined "RoundTripOCR core": 284,528 rows in the final set

2. w4ashabii/Nepali_dictionary (Hugging Face) — dictionary definition sentences, used only to widen the vocabulary pool that the synthetic handwriting-confusion model draws from (not a standalone supplement).

3. ashokpant/nepali-news-dataset-large (Kaggle) — real news articles across 20 categories (Sports, Politics, Economy, Entertainment, etc.). 6,973 articles retrieved; category sizes were highly uneven (Opinion: 34,308 sentences vs. Migration: 1,239) so each category was capped at the per-category median (4,931 sentences) before use, yielding an 80,077-sentence pool. Combined with the dictionary sentences above to form the clean-text pool for handwriting-style corruption.

4. Synthetic handwriting-confusion model — a character-level confusion map (consonant look-alikes, independent vowels, matras, nasalisation marks, Devanagari digits) plus duplication/transposition/ word-drop operators, applied to the combined dictionary+news clean-text pool. 129,723 pairs in the final set.

5. nischallal/nepali-name-dataset-in-devanagari-with-gender (Kaggle, sourced from Nepal's Election Commission voters list) — person names in Devanagari. 17,998,877 raw rows deduplicated to 6,220,952 distinct names (voter rolls repeat common names for thousands of different people). Each distinct name was run through the handwriting confusion model (elevated edit rate, tuned for short tokens) to produce a corrupted/clean pair; 32,387 came out identical and were kept as identity examples. This produced 6,220,952 candidate pairs — far more than any other single source — so for the final set, a random sample of 200,000 (20% of the final total) was drawn rather than using the source in full; see Discussion of Biases below.

6. himalaya-ai/nepali-corpus-compile (Hugging Face) — a large (~31.3M document, 30.6GB) general Nepali web-text corpus, used as filler to reach the target dataset size without over-representing any other single source. Streamed (not downloaded whole) and filtered to Devanagari-character ratio ≥ 0.8 per sentence, since this corpus mixes in some English. Corrupted the same way as the handwriting supplement. 370,768 pairs in the final set, all from distinct clean sentences (no sentence corrupted more than once).

7. Code-switching supplement — synthetic: English loanwords/proper nouns inserted into otherwise-Nepali sentences, then corrupted, to cover the code-switched text that appears in real Nepali documents (institution names, NGOs, place names, etc.). 14,981 pairs.

8. Domain diversity supplement — an optional slot for user-supplied domain-specific text (e.g. gazette notices); 0 rows in this build (no source files were provided).

Final Composition

Source Rows % of train
nepali-corpus-compile (filler) 370,768 37.08%
RoundTripOCR core (curated + identity + word-level + val) 284,528 28.45%
Names (Election Commission voters list) 200,000 20.00%
Handwriting supplement (dictionary + news, synthetic corruption) 129,723 12.97%
Code-switching supplement 14,981 1.50%
Domain supplement 0 0.00%
Total 1,000,000 100%

Clean-text length by source (characters):

Source Mean Min Median Max
Code-switching 87.6 15 80 550
nepali-corpus-compile 87.4 8 81 200
Handwriting 70.1 8 63 541
Names 15.3 3 15 53
RoundTripOCR core 33.2 1 14 541

Curation and Sizing Decisions

  • Names were capped, not the raw 6.2M used. A 0.5B-parameter fine-tune doesn't benefit from millions of near-duplicate person-name corrections; a 200,000-name random sample covers the space of Devanagari name-spelling OCR confusions without one source dominating training.
  • Overall size was capped at 1,000,000 rows rather than shipping every pair collected (a full uncapped run of this pipeline would exceed 30M rows) — chosen to keep fine-tuning time and compute reasonable for a 0.5B model while remaining large enough for the task.
  • No sentence in the final set was corrupted more than once to reach volume (an earlier design that re-corrupted the same clean sentences multiple times to hit volume targets was rejected during development in favor of pulling more raw text from a larger corpus instead).
  • The English mixed into nepali-corpus-compile was filtered out by a per-sentence Devanagari-character-ratio threshold (≥ 0.8), not by the corpus's own source/language metadata (which was null/ unreliable in the raw data).

Annotations

No human annotation was performed. All clean text is either (a) the original clean side of an existing OCR-pair dataset (RoundTripOCR), (b) real prose from a dictionary/news/web corpus, or (c) a real person name from a public voter-roll dataset. All corrupted text is synthetically generated by a character-level confusion model (for sources b and c) or comes from the source dataset's own OCR pipeline (for source a).

Considerations for Using the Data

Discussion of Biases

  • Names are still 20% of the dataset by design, and all derive from a voters-list source — this means the name-spelling distribution reflects Nepal's electoral roll demographics, not a general population or naming-convention sample. A model trained on this may proofread common electoral-roll name spellings better than rarer or newer name forms.
  • The nepali-corpus-compile filler (37% of the set) determines a large share of the model's general vocabulary exposure, since it was sized specifically to fill the volume gap after other sources were pooled — its own topical/stylistic composition (a general web corpus) is inherited by the final set.
  • Corruption is synthetic, not real scanned-document OCR output (except for the RoundTripOCR-core portion, which does derive from an actual OCR pipeline across 49 fonts). The handwriting/names/corpus- compile portions use a hand-built character-confusion model, which may not fully reflect the error distribution of any specific real-world OCR engine or handwriting style.
  • Domain diversity supplement is empty (0 rows) in this build — no gazette/legal/form-style text was included, so domain-specific vocabulary in those registers is not represented.

Other Known Limitations

  • Fields beyond corrupted/clean (e.g. source font, name gender, news category) were used during curation but are not preserved in the final JSONL — only the two-field pair ships.
  • License status of upstream sources varies and was not independently re-verified per source at compile time — see Licensing Information.

Licensing Information

This compiled dataset draws on multiple upstream sources with their own, independently-set licenses:

  • cfilt/RoundTripOCR-nepali (Hugging Face)
  • w4ashabii/Nepali_dictionary (Hugging Face)
  • ashokpant/nepali-news-dataset-large (Kaggle)
  • nischallal/nepali-name-dataset-in-devanagari-with-gender (Kaggle)
  • himalaya-ai/nepali-corpus-compile (Hugging Face)

Check each source's own license/terms before redistributing this compiled dataset or using it commercially — license was not asserted here because it was not independently re-verified per source at compile time.

Citation

If you use this dataset, please cite the upstream sources listed above in addition to this compilation.

Dataset Curators

Compiled via a Jupyter notebook pipeline (01_dataset_curation_*.ipynb) that loads, filters, balances, and merges the sources described above. See the notebook for the exact, reproducible curation steps and all tunable constants (FINAL_DATASET_SIZE, NAME_TARGET_FRACTION, per-source filters and caps).

Downloads last month
-