Datasets:
text
stringlengths 1
63k
⌀ | language
class label 201
classes | dataset_source
stringclasses 19
values |
---|---|---|
Dia hoy ny mpanapaka tamin'i Baroka: Mandehana, ka miere ianao sy Jeremia, ary aoka tsy hisy olona hahita izay itoeranareo. | 0plt_Latn
| mt560 |
Lamun urang soméah bari ngucapkeun salam, manéhna bisa boga pandangan nu alus ngeunaan urang jeung Allah nu disembah ku urang. | 1sun_Latn
| mt560 |
Натомість ті райони, які мають найнижчі рівні імміграції, найбільш ворожі й нетерпимі щодо іммігрантів. | 2ukr_Cyrl
| mt560 |
¿Serás exaltada hasta el cielo? | 3spa_Latn
| lti |
O tom da pergunta do senhor deputado surpreende um pouco a Comissão. | 4por_Latn
| mt560 |
ဒါ မှ မဟုတ ် ကျွန ် ုပ ် တို ့ ရဲ ့ အပြုအမူ ၊ | 5mya_Mymr
| mt560 |
(Разни извори -- 03 / 02 / 09-10 / 02 / 09) | 6mkd_Cyrl
| mt560 |
Syempre, an paghimo hin sugad kadaku nga desisyon nagkikinahanglan hin maopay nga pagplano. | 7war_Latn
| mt560 |
Eupša go be go sa felele moo, gobane Jehofa o be a sa kgahlišwe ke seo. | 8nso_Latn
| mt560 |
Ca kanamu gàngune ma amoon na fa juróom-ñaari làmp yu yànj, di juróom-ñaari Xeli Yàlla yi. | 9wol_Latn
| mt560 |
Ndethĩwa aathĩnĩka takeka nĩwaanangĩie kĩveti kĩu mũtwaano wakyo, ndaĩa yakyo, na ngwatanĩo yakyo na Yeova. | 10kam_Latn
| mt560 |
ഇത്, സംസാരിക്കാൻ സമയമുണ്ടാക്കേണ്ടത് ആവശ്യമാക്കിത്തീർക്കുന്നു. "- വീക്ഷാഗോപുരം, 1993 ആഗസ്റ്റ് 1, പേജ് 4. | 11mal_Mlym
| mt560 |
forbairt modheolaíochta i gcás suirbhé deonach maidir le foréigin bunaithe ar inscne, i gcomhar leis na hinstitiúidí, comhlachtaí, oifigí agus gníomhaireachtaí Eorpacha a ghníomhaíonn sa réimse seo; | 12gle_Latn
| mt560 |
Jesusan jiwäwip amtañatakejj Biblia liyisaw wakichtʼassna. | 13ayr_Latn
| mt560 |
Участники попросили дать разъяснения по классификации расходов на наркотические средства, услуги проституток, безмоторные средства, используемые в транспортных, а не в рекреационных целях, услуги городского и пригородного транспорта, осуществляемые с использованием более одного вида транспорта. | 14rus_Cyrl
| mt560 |
چې کله روبين واپس کُوهى ته راغلو اؤ وې ليدل چې يُوسف هلته نۀ وو ، نو هغۀ د خفګان په وجه خپلې جامې وشلولې. | 15pbt_Arab
| mt560 |
Nen inmunan siglo, saray Kristiano so mabetbet a puntirya na panagpauges, a sikara so ipapasen a managgaway - kaugsan. | 16pag_Latn
| mt560 |
(AП, АФП, Радио Б92, Европска комисија - 30 / 03 / 04) | 6mkd_Cyrl
| mt560 |
Osii apini sɛ: "Ma ɛda a wɔwoo me mu no nyera... | 17twi_Latn
| mt560 |
Ne gjetem qe semundja e Lindes nuk dukej si semundja tipike e meshkujve. | 18als_Latn
| mt560 |
Kiekviena narė turi vieną nedalijamą balsą, o jai atstovaujama, kai posėdyje pagal bendrojo susirinkimo darbo tvarkos taisykles asmeniškai arba surengus telekonferenciją dalyvauja bent vienas atstovas. | 19lit_Latn
| mt560 |
ታማኙ ባሪያ ከቅቡዓን ክርስቲያኖች የተውጣጣ ነው ብሎ ማሰቡ ተገቢ የሆነው ለምንድን ነው? | 20amh_Ethi
| mt560 |
Fakat kimse prensesi yıldıramadı. | 21tur_Latn
| mt560 |
దానికి ఆధారం, ఇక్కడ చూపించబడిన ఉత్తరంలో కనిపిస్తుంది. | 22tel_Telu
| mt560 |
Ła łinea dei albari ła se cata in genare sui monti, marcando, par ogni łatitudine e condision climatica, l'altitudine masima de espansion dei albari e, par quanto riguarda łe rejoni połari, ła łatitudine più ełevà parmesa par i boschi. | 23vec_Latn
| wili2018 |
Sesetengah bertujuan untuk memberitahu dan memberikan ukuran untuk mencegah dan juga secara berkesan menghentikan siberbuli dan siber-gangguan berterusan. | 24zsm_Latn
| leipzig |
ھونەرمەند لە سەردەمی نازییەکان ڕوبەڕوی زۆر ئەزیەت و داپلۆسین ھات تەنیا لەبەر ئەوەی بۆچونەکانی لەگەڵ بیر و باوەڕی نازییەکان ناکۆک بوو ناوییان نابوو (ھونەرمەندە ھەڵوەشاوەکھ) لەو کاتە ھەر بە دزییەوە لە ژوورێکی بچوکی ماڵەکەی خۆیدا کاری ھونەری دەکرد پێوانەی تابلۆکانی لە پانایی دەستێک تێپەڕی نەدەکرد،بەڵام لە کۆتایی شەردا نۆلدە بە قەبارەی گەورە زیاترلە ١.٣٠٠ تابلۆی بە ئەنجام گەیاند کە ئێستا لە ماڵەکەی خۆی نیشان دەدرێت. | 25ckb_Arab
| lti |
«Кимиё» ном китобе бо хати зебои ӯ рӯнавис шудааст, ки аз 303 варақ иборат буда, таҳти рақами 1561 дар Ганҷинаи дастнависҳои шарқии Институти шарқшиносии АИ Ҷумҳурии Ӯзбекистон маҳфуз аст. | 26tgk_Cyrl
| lti |
เรา ได ้ รับ โอกาส ที ่ พระเจ ้ า จะ นับ เรา เป ็ น มิตร ของ พระองค ์. | 27tha_Thai
| mt560 |
Նա կենաց ծառ է նորանց համար որ բռնել են նորան, եւ նա իրան պինդ բռնողին երջանիկ կ'անէ "(Առակաց 3 ։ | 28hye_Armn
| mt560 |
Da sprach die ältere zu der jüngeren: Unser Vater ist alt, und es giebt niemand mehr auf Erden, der Umgang mit uns haben könnte, wie es aller Welt Brauch ist. | 29deu_Latn
| lti |
II гасырыннан башлап Бөек Бабыл әсирлегендә булган? | 30tat_Cyrl
| mt560 |
Baada ya kuhitimu timu ya Olimpiki ya 2008, kIpruto alishinda medali ya dhahabu katika mbio hiyo. | 31swh_Latn
| lti |
Masha law law gaw, nse nsa sumla n yu ai sha nga lu na matu grai shakut ra ai. | 32kac_Latn
| mt560 |
Gowy dost edinmek üçin myhmansöýer bolmaly. | 33tuk_Latn
| mt560 |
regulas I pielikumu groza šādi: | 34lvs_Latn
| mt560 |
Cosnóidh an Coimisiún agus na Ballstáit faisnéis rúnda gnó. | 12gle_Latn
| mt560 |
U ri: "Laha ku nga ni loyi a funengetaka rivengo ku ni milomu ya mavunwa." | 35tso_Latn
| mt560 |
അദ്ദേഹം പറയുന്നു: "തീർച്ച യാ യും! | 11mal_Mlym
| mt560 |
Grein um fiskiørn sædd á Hálsvøtnum á Sandoynni; www.Portal.fo | 36fao_Latn
| lti |
Olsem wanem pasin isi i stap ples klia long samting God i mekim long Kein? | 37tpi_Latn
| mt560 |
(b) Nye ovanyãli vafendeli va Suku va ka linga? | 38umb_Latn
| mt560 |
Eger siz muňa kanagatlanarly jogap berseňiz, begenersiňiz. | 33tuk_Latn
| mt560 |
Τηλ: + 357 22 871600 Latvija sanofi- aventis Latvia SIA Tel: + 371 67 33 24 51 | 39mlt_Latn
| mt560 |
Dangosa gariad hyd yn oed pan fydd hi'n anodd | 40cym_Latn
| mt560 |
কমিশন | 41ben_Beng
| mt560 |
Ahab te gen yon palè nan Samari, e palè sa a te yon konstriksyon estrawòdinè! | 42hat_Latn
| mt560 |
Cu siguranță există o teamă că roboții le vor lua joburile oamenilor, ceea ce e adevărat în anumite domenii. | 43ron_Latn
| mt560 |
ብልኡም መንፈስ ኢዮም ዝዛረቡ: ንዘሎኒ ጥርጣረ ድማ ብሕያውነት ይሰምዑኒ ኢዮም ። " | 44tir_Ethi
| mt560 |
Il-Kummissjoni kkonkludiet li l-ksur tal-Artikolu 28 ma kienx iġġustifikat mill-Artikolu 31 tal-Ftehim TRIPS. | 39mlt_Latn
| mt560 |
Yonatan se egɔme be David ye Mawu tia be wòanye fia ɖe ye fofo teƒe, eye edze eŋu be yeakpe ɖe ye xɔlɔ ̃ a ŋu. - 1 Samuel 19: 1, 2; 20: 30 - 33; 23: 14 - 18. | 45ewe_Latn
| mt560 |
Yusuf sangat menghargai keloyalan; tapi istri Potifar tidak. | 46ind_Latn
| mt560 |
چرچ هنن ڏند ڪٿائن جي سدائين نندا ڪئي ۽ انکي جنسي بي راھہ روي جي تبليغ تي ٻڌل قرار اڏنو. اهو ئي ڪارڻ آهي تہ هن سال بہ عيسائي پادرين هن ڏينهن جي نندا ۾ سخت بيان ڏنا۔ بينڪاڪ ۾ تہ هڪ عيسائي پادريءَ ڪجھہ ماڻهن کي ساڻ وٺي هڪ اهڙي دوڪان کي نذرآتش ڪري ڇڏيو. جنهن تي ويلنٽائن ڪارڊ وڪرو ٿي رهيا هئا۔ | 47snd_Arab
| lti |
Ontwikkeling van de particuliere sector | 48nld_Latn
| mt560 |
มี การ ดำเนิน ตาม แบบ แผน คล ้ าย ๆ กัน นี ้ ใน ปัจจุบัน ด ้ วย. | 27tha_Thai
| mt560 |
Yehowa gaw lam shagu hta hkum tsup ai majaw tinang myit hpe hkang lu ai lam hta mung hkum tsup ai. | 32kac_Latn
| mt560 |
پھر وہ میرے بلانے سے اوربھی زيادہ بھاگتے رہے | 49urd_Arab
| mt560 |
Chúa Giê - su trả lời bằng một minh họa so sánh người không thuộc dân Do Thái với "chó con". | 50vie_Latn
| mt560 |
Tinklą sudaro Europos Sąjungos valstybių narių valdžios institucijos ir agentūros, kurioms pavesta užduotis užkirsti kelią korupcijai ar kovoti su ja. | 19lit_Latn
| mt560 |
एका तहात सम्राट फर्डीनंड तिसरा याने आणि स्वीडनने सह्या केल्या आणि दुसऱ्या तहात, सम्राटाने व फ्रान्सने सह्या केल्या. | 51mar_Deva
| mt560 |
La mondialisation a accru l 'espérance de vie et l 'amélioration du niveau de vie dans de nombreux pays pauvres. | 52fra_Latn
| mt560 |
Naye ekituufu kiri nti Yakuwa bulijjo ayagala okukuuma abantu be obutatuukibwako kabi. | 53lug_Latn
| mt560 |
Od czasu traktatu z Maastricht art. 157 traktatu WE określa inicjatywy w ramach polityki przemysłowej, za pomocą których Komisja może koordynować działania państw członkowskich. | 54pol_Latn
| mt560 |
Boni inggih punika sarwa woh-wohan sané rupannyané bunter-bunter alit. | 55ban_Latn
| lti |
Euroopa Parlamendi ja nõukogu direktiiv 2009 / 66 / EÜ, 13. juuli 2009, põllu- või metsamajanduslike ratastraktorite roolisüsteemi kohta | 56est_Latn
| mt560 |
На граници са марзом Лори налази се планина Ачкасар (у склопу Џавахетских планина), највиши врх у марзу са надморском висином од 3.196 метара. | 57srp_Cyrl
| lti |
Bibiliya igira iti "icyakora no muri ubwo burwayi bwe ntiyigeze ashaka Yehova, ahubwo yirukiye mu bavuzi." | 58kin_Latn
| mt560 |
(ข) เหตุ ใด คุณ จึง ตั ้ งใจ แน ่ ว แน ่ ว ่ า จะ ดำเนิน ใน ทาง แห ่ ง ความ รัก? | 27tha_Thai
| mt560 |
Namnet dukka først opp på eit kart basert på landmålingane til Discovery Investigations i 1926-30. | 59nno_Latn
| lti |
Se i miràcuj | 60fur_Latn
| lti |
Pian epektibo so panagbangat mo, nepeg mon ipapuso so Salita na Dios (Nengnengen so parapo 17) | 16pag_Latn
| mt560 |
Niha ew bêtir serbilind in ku ew aîdî teşkîlata Xwedê ne. " | 61kmr_Latn
| mt560 |
हालाँकि, दुनों अमेरिका महादीप पनामा थलजोड़ से जुड़ल हवें आ पनामा नहर से अलग होखे लें, ई दक्खिनी महादीप परंपरागत रूप से अलगा से एगो महादीप मानल जाला आ कभी-काल्ह के इनहन के एक्के सुपरमहादीप के रूप में भी देखल जाला आ अंगरेजी में अमेरिकाज़ कहि के बोलावल जाला। राजनीतिक सीमा के आधार पर पनामा–कोलंबिया के बाडर के एह दुनों महादीप सभ के बीचा के बिभाजक मानल जाला। | 62bho_Deva
| lti |
Gezien het feit dat Japan een belangrijke handelspartner is - de bilaterale handel met de EU is jaarlijks goed voor 120 miljard euro en de EU is voor Japan de op twee na grootste handelspartner en de op een na grootste investeerder - moeten we beseffen dat een bilateraal akkoord gepaard gaat met het risico dat multilaterale akkoorden met de rest van de wereld, vooral de ontwikkelingslanden, worden ondermijnd. | 48nld_Latn
| lti |
Paraiškos pateikimas ir sutarties sudarymas. | 19lit_Latn
| mt560 |
Аллаһы Патшалыгы җирдә яшәгән кешеләр өчен нәрсә эшләячәк? | 30tat_Cyrl
| mt560 |
Kolmanneksi Starbucksin rooli ei ollut Alki LP:n roolia vähäisempi missään näistä liiketoimista. | 63fin_Latn
| mt560 |
Is féidir teistiméireachtaí a fháil ó fhostóirí a bhí agat roimhe seo (ainm an fhostóra, ainm na cuideachta, an post a bhí agat ann, uimhir fostaí). | 12gle_Latn
| mt560 |
Erangi te haere o te kamera ra te kowhao o te ngira he mea takoto noa, he whakauaua rawa ia te haere o te tangata taonga ki roto ki te rangatiratanga o te Atua | 64mri_Latn
| mt560 |
Maiyarig ni Jehova iti sarikedked ken nalagda a torre iti panangsaluad ken panangsaranayna iti ilina. | 65ilo_Latn
| mt560 |
Io, e bibi ena vakawati na dau dina. | 66fij_Latn
| mt560 |
तर इतर काहींनी, इंटरनेटचा मुळीच वापर न करण्याचे ठरवले आहे. | 51mar_Deva
| mt560 |
Vývozné náhrady na sirupy a niektoré iné výrobky z cukru vyvezené bez ďalšieho spracovania uplatniteľné od 28. septembra 2007 [1] Poznámka: Miesta určenia sú definované takto: | 67slk_Latn
| mt560 |
Yo llamo a esto la ecología material. | 3spa_Latn
| mt560 |
غَشِبْتٗ مِيَ كٗلَ سِيَ دٗنْ مَدٗ، سُوَنْزٗلَنْ دَلِلْ دَ سَنْدِ كُرِ-كُرِ مِ سَلْس حَدَأَ كُرُ نَ فَلْيِنْ كٗنٗنْغَ نزَ سَجدِنْ كَمَدَ رَكْسَ كٗمْبُ كٗلَنْزَأَ سَدِنْبَمَ، ُرُ رَكْسَ سَسَمْبِنْ بَ، تِيِ سَْ دَ ندُبْتَنَ مَنْ غٗنْيِأَ | 68knc_Arab
| nllbSeed |
Thật là một cuộc tàn sát khủng khiếp! | 50vie_Latn
| mt560 |
ગુજરાતના રાજકારણમાં એક વખતે સૌથી શક્તિશાળી ગણાતાં મહિલા નેતા આનંદીબહેન પટેલ આવનારી વિધાનસભા ચૂંટણીમાં કેટલું મહત્વ ધરાવે છે? | 69guj_Gujr
| xlsum |
ئەگەر ئەم چەکانەی ڕوسیا لە کوبا جێگیرکرابان بەڕاستی ناتۆ دەکەوتە خەتەرەوە چونکە سیستەمی پارێزگاری ئەوان لە خاکی ئەمریکا شکستی | 25ckb_Arab
| lti |
De hecho, ha limitado su presencia en el extranjero al negarse a contribuir a un rescate de la eurozona, intervenir en Siria o usar la fuerza para contener el desarrollo nuclear de Irán (a pesar del decidido apoyo israelí). | 3spa_Latn
| mt560 |
과거에하느님의충실한종들도하느님의뜻을행하기위해힘써노력하지않으면안되었습니다. | 70kor_Hang
| mt560 |
Ka baka leo, Jošua o itše: "Xe e le nna le ba lapa la - ka, rena re dirêla Morêna." | 8nso_Latn
| mt560 |
▪ Kasi vikenda wuli kuti Mariya waŵe nyina wa Yesu, ndipo vikacitika wuli kuti Yesu waŵe na awiske ŵaŵiri? | 71tum_Latn
| mt560 |
Ma yella ulac daɣen laman deg-wen ɣef wayen i wen-iwekkel walebɛad ̣, amek ara wen-d-yefk Sidi R ̣ ebbi ayen i wen-ihegga i kunwi? | 72kab_Latn
| mt560 |
Vor der Industriellen Revolution gab es kein nennbares Wirtschaftswachstum. | 29deu_Latn
| mt560 |
Ongelykheid in lewenstandaarde en in geleenthede vir gesondheidsorg en skoolopleiding is net 'n paar van hulle. | 73afr_Latn
| mt560 |
On Sunday, the US-backed and Kurdish-led Syrian Democratic Forces (SDF) insisted that foreign fighters would not be tried "on our land... they will be tried by their own countries". | 74eng_Latn
| xlsum |
ادخل ، لو سمحت. | 75acq_Arab
| madar |
Qalbigiinnu yuusan murugoon. | 76som_Latn
| lti |
May pagmamalaking masasabi niya: "Ako'y isang kuta, at ang aking dibdib ay parang mga moog." | 77tgl_Latn
| mt560 |
لایا گیا | 49urd_Arab
| mt560 |
Dataset Card for "open-lid-dataset"
Dataset Summary
The OpenLID dataset covers 201 languages and is designed for training language identification models. The majority of the source datasets were derived from news sites, Wikipedia, or religious text, though some come from other domains (e.g. transcribed conversations, literature, or social media). A sample of each language in each source was manually audited to check it was in the attested language (see the paper) for full details.
Supported tasks
This dataset is intended for training high-coverage language identification models (e.g. OpenLID). It is compatible with the FLORES-200 evaluation benchmark.
Languages
There are 201 languages included in the dataset with varying amounts of data: the largest class (English) contains 7.5 million lines of data, and the smallest (South Azerbaijani) contains 532 lines of data. The mean number of lines per language is 602,812. A full breakdown of lines of data per language is available on the repo.
Dataset Structure
Data Instances
Each entry in the dataset consists of a line of data, a language label included script information, and a tag indicating the source.
{
"text": "¿Serás exaltada hasta el cielo?",
"language": "spa_Latn",
"dataset_source": "lti"
}
Data Splits
Only a train split is provided. The dataset is designed to be compatible with the FLORES-200 evaluation benchmark.
Dataset Creation
Curation Rationale
Recent work has found that existing language identification algorithms perform poorly in practice compared to test performance. The problem is particularly acute for low-resource languages: Kreutzer et al. (2022) found a positive Spearman rank correlation between quality of data and size of language for all of the \ac{lid}-filtered multilingual datasets they studied. In addition, for a significant fraction of the language corpora they studied, less than half of the sentences were in the correct language. They point out that such low-quality data not only leads to poor performance in downstream tasks, but that it also contributes to `representation washing', where the community is given a false view of the actual progress of low-resource natural language processing.
There are several open language identification models offering quick classification and high language coverage (e.g. CLD3, No Language Left Behind). However, to the best of our knowledge, none of the commonly-used scalable language identificaiton systems make their training data public.
This dataset aims to address that gap by curating and combining sources of open training data for language identification and by auditing a sample of all languages in each source to check reliability.
Source Data
The majority of the source datasets were derived from news sites, Wikipedia, or religious text, though some come from other domains (e.g. transcribed conversations, literature, or social media). We provide a full list at the end of this model card along with the licensing information for each source.
Initial Data Collection and Normalisation
Our initial aim was to cover the same languages present in the FLORES-200 Evaluation Benchmark so that we could use this dataset for evaluation. However, during the curation process, we decided to exclude three languages. Firstly, though Akan and Twi are both included as separate languages in FLORES-200, Akan is actually a macrolanguage covering a language continuum which includes Twi. Given the other languages in FLORES-200 are individual languages, we decided to exclude Akan. Secondly, FLORES-200 includes Modern Standard Arabic (MSA) written in Latin script. It is true that Arabic dialects are often written in Latin characters in informal situations (e.g. social media). However, MSA is a form of standardised Arabic which is not usually used in informal situations. Since we could not any find naturally-occurring training data, we excluded MSA from the dataset. Finally, we excluded Minangkabau in Arabic script because it is now rarely written this way, making it difficult to find useful training data.
The first step in our manual audit was to check and standardise language labels, as these are often inconsistent or idiosyncratic. We chose to copy the language codes in FLORES-200 and reassign macrolanguage or ambiguous language codes in the data sources we found to the dominant individual language. Whilst this resulted in more useful data for some languages, for other languages we had to be more conservative. For example, we originally reassigned text labelled as the macrolanguage Malay (msa_Latn) to Standard Malay, but this led to a large drop in performance as the former covers a very diverse set of languages.
Two of the authors then carried out a manual audit of a random sample of all data sources and languages: one a native Bulgarian speaker (able to read Cyrillic and Latin scripts and Chinese characters), and the other a native English speaker (able to read Latin, Arabic and Hebrew scripts). For languages we knew, we checked the language was what we expected. For unfamiliar languages in a script we could read, we compared the sample to the Universal Declaration of Human Rights or failing that, to a sample of text on Wikipedia. We compared features of the text which are common in previous language identification algorithms and could be identified easily by humans: similar diacritics, word lengths, common words, loan words matching the right cultural background, similar suffixes and prefixes, and vowel/consonant patterns. For scripts we could not read, we checked that all lines of the sample matched the script in the Universal Declaration of Human Rights.
We kept preprocessing minimal so that the process was as language agnostic as possible. We used the scripts provided with Moses to remove non-printing characters and detokenise the data where necessary. We then filtered the data so that each line contained at least one character in the expected script (as defined by Perl) to allow for borrowings. Finally, we sampled proportionally to $ p_l^{0.3} $, where $ p_l $ is the fraction of lines in the dataset which are in language $ l $. This aims to ameliorate class skew issues.
Considerations for Using the Data
Social Impact of Dataset
This dataset covers a number of low-resourced languages. This makes it a potentially useful resource, but due to the limited amount of data and domains, care must be taken not to overclaim performance or coverage.
Discussion of Biases
Our work aims to broaden natural language processing coverage by allowing practitioners to identify relevant data in more languages. However, we note that language identification is inherently a normative activity that risks excluding minority dialects, scripts, or entire microlanguages from a macrolanguage. Choosing which languages to cover may reinforce power imbalances, as only some groups gain access to language processing technologies.
In addition, errors in language identification can have a significant impact on downstream performance, particularly (as is often the case) when a system is used as a `black box'. The performance of our classifier is not equal across languages which could lead to worse downstream performance for particular groups. We mitigate this by providing metrics by class.
Additional information
The dataset was curated from the sources listed below by Laurie Burchell and Nikolay Bogoychev.
Licensing Information
License considerations for each source are given below. Open use for non-commercial purposes is covered by all licences.
If you view any part of this dataset as a violation of intellectual property rights, please let us know and we will remove it.
Source | Description | License |
---|---|---|
Arabic Dialects Dataset | Dataset of Arabic dialects for Gulf, Egyptian, Levantine, and Tunisian Arabic dialects plus MSA | No explicit license; website describes data as "some free and useful Arabic corpora that I have created for researchers working on Arabic Natural Language Processing, Corpus and Computational Linguistics." |
BLTR | Monolingual Bhojpuri corpus | CC BY-NC-SA 4.0 |
Global Voices | A parallel corpus of news stories from the web site Global Voices | The website for Global Voices is licensed as Creative Commons Attribution 3.0. There is no explicit additional license accompanying the dataset. |
Guaraní Parallel Set | Parallel Guaraní-Spanish news corpus sourced from Paraguyan websites | No explicit license |
HKCanCor | Transcribed conversations in Hong Kong Cantonese | CC BY 4.0 |
IADD | Arabic dialect identification dataset covering 5 regions (Maghrebi, Levantine, Egypt, Iraq, and Gulf) and 9 countries (Algeria, Morocco, Tunisia, Palestine, Jordan, Syria, Lebanon, Egypt and Iraq). It is created from five corpora: DART, SHAMI, TSAC, PADIC, and AOC. | Multiple licenses: Apache License 2.0 (SHAMI); GNU Lesser General Public License v3.0 (TSAC); GNU General Public License v3 (PADIC). DART and AOC had no explicit license. |
Leipzig Corpora Collection | A collection of corpora in different languages with an identical format. | The Terms of Usage states "Permission for use is granted free of charge solely for non-commercial personal and scientific purposes licensed under the Creative Commons License CC BY-NC." |
LTI | Training data for language identification | From the README: "With the exception of the contents of the Europarl/, ProjectGutenberg/, and PublicDomain/ directories, all code and text in this corpus are copyrighted. However, they may be redistributed under the terms of various Creative Commons licenses and the GNU GPL. Copying the unmodified archive noncommercially is permitted by all of the licenses. For commercial redistribution or redistribution of modified versions, please consult the individual licenses." |
MADAR Shared Task 2019, subtask 1 | Dialectal Arabic in the travel domain | The MADAR Corpus has a custom license, the text of which can be found in this repo. |
EM corpus | Parallel Manipuri-English sentences crawled from The Sangai Express | CC BY-NC 4.0 |
MIZAN | Parallel Persian-English corpus from literature domain | CC BY 4.0 |
MT560 v1 | A machine translation dataset for over 500 languages to English. We have filtered out data from OPUS-100, Europarl, Open Subtitles, Paracrawl, Wikimedia, Wikimatrix, Wikititles, and Common Crawl due to issues with the fidelity of the language labels. | Apache License 2.0 |
NLLB Seed | Around 6000 sentences in 39 languages sampled from Wikipedia, intended to cover languages lacking training data. | CC BY-SA 4.0 |
SETIMES | A parallel corpus of news articles in the Balkan languages | CC-BY-SA 3.0 |
Tatoeba | Collaborative sentence translations | CC BY 2.0 FR |
Tehran English-Persian parallel corpus (TEP) | Parallel Persian-English sentences sourced from subtitles | GNU General Public License |
Turkic Interlingua (TIL) Corpus | A large-scale parallel corpus combining most of the public datasets for 22 Turkic languages | CC BY-NC-SA 4.0 |
WiLI-2018 | Wikipedia language identification benchmark containing 235K paragraphs of 235 languages | Open Data Commons Open Database License (ODbL) v1.0 |
XL-Sum | Summarisation dataset covering 44 languages, sourced from BBC News | CC BY-NC-SA 4.0 |
Citation Information
If you use this dataset, please cite all the authors in the citation file who compiled the source datasets, plus the OpenLID paper:
@inproceedings{burchell-etal-2023-open,
title = "An Open Dataset and Model for Language Identification",
author = "Burchell, Laurie and
Birch, Alexandra and
Bogoychev, Nikolay and
Heafield, Kenneth",
editor = "Rogers, Anna and
Boyd-Graber, Jordan and
Okazaki, Naoaki",
booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.acl-short.75",
doi = "10.18653/v1/2023.acl-short.75",
pages = "865--879",
abstract = "Language identification (LID) is a fundamental step in many natural language processing pipelines. However, current LID systems are far from perfect, particularly on lower-resource languages. We present a LID model which achieves a macro-average F1 score of 0.93 and a false positive rate of 0.033{\%} across 201 languages, outperforming previous work. We achieve this by training on a curated dataset of monolingual data, which we audit manually to ensure reliability. We make both the model and the dataset available to the research community. Finally, we carry out detailed analysis into our model{'}s performance, both in comparison to existing open models and by language class.",
}
Contributions
Thanks to @hac541309 and @davanstrien for adding this dataset.
- Downloads last month
- 657