Hugging Face's logo

Datasets:
AchrafSoltani
/
morisco-ner

Tasks:
Token Classification
Modalities:
Text
Formats:
json
Languages:
Moroccan Arabic
Size:
1K - 10K
Tags:
named-entity-recognition
darija
arabizi
moroccan-arabic
latin-script
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
tokens
listlengths
2
19
spans
listlengths
0
3
[ "lblassa", "lifiha", "db", "mafihach", "inwi" ]
[ [ 4, 4, "ORG" ] ]
[ "l3otma", "jaya", "nsafro", "ljdida" ]
[ [ 3, 3, "LOC" ] ]
[ "3andk", "lmdrassa", "nhar", "tlat?" ]
[]
[ "walakin", "3lach", "kaytsad", "nif,", "3andak", "7asasia", "mn", "chi", "7aja:", "ward,", "chjer...?" ]
[]
[ "wach", "9darti", "tamchi", "l", "argentine", "o", "tchof", "l", "atar?" ]
[ [ 4, 4, "LOC" ] ]
[ "twldat", "nhar", "l3id", "o", "taf9na", "nsamiwha", "Fetna", "kif", "f", "yawmia" ]
[ [ 6, 6, "PER" ] ]
[ "wach", "kat3raf", "chi", "maT3am", "indonisi?" ]
[ [ 4, 4, "MISC" ] ]
[ "ma", "ghadich", "nmchi", "7tel", "tlat" ]
[]
[ "ghadir", "l3rousa", "l3amariya?" ]
[]
[ "ana", "kan7tarm", "bzaf", "ra2y", "Satre" ]
[ [ 4, 4, "PER" ] ]
[ "kayn", "chi", "wahd", "8na", "kayhder", "bl", "anglais?" ]
[ [ 6, 6, "MISC" ] ]
[ "kn9raw", "kol", "nhar", "lyouma", "sebt", "o", "9arin" ]
[]
[ "mafiha", "ta", "3ib", "had", "sora", "rah", "mn", "lfirma", "dyalo" ]
[]
[ "ewa", "lah", "ya5od", "fik", "l7e9", "had", "nhar", "nta", "wyaha" ]
[]
[ "ghadi", "nghaTiw", "ga3", "lmawad", "tal", "nhar", "tlat", "jay,", "dakchi", "3lach", "bda", "l9araya", "bakri" ]
[]
[ "Lalla", "Imane,", "kifach", "dowwez", "had", "lwa9t", "kamlo?" ]
[ [ 1, 1, "PER" ] ]
[ "yamkn", "lik", "t7awl", "t9na3", "chirir", "lkbir", "bhad", "Anya", "bghat", "flos8a", "safi,", "o", "nta", "chakhS", "lwa7id", "likaybghiha", "bssa7" ]
[]
[ "kaykon", "lmostachar", "dyal", "madrasa", "kol", "tlat" ]
[]
[ "hadchi", "dmmar", "i9tiSad", "l", "almany,", "o", "sa8m", "bchkal", "ghayr", "mobachir", "f", "l7arb", "l3alamiya", "taniya" ]
[ [ 4, 4, "MISC" ], [ 11, 13, "MISC" ] ]
[ "wach", "hada", "Tobis", "lli", "tayddi", "l", "Koutoubia?" ]
[ [ 6, 6, "LOC" ] ]
[ "nssayh", "otfakri", "bli", "alalh", "makaydir", "ghir", "lifiha", "lkhiir" ]
[]
[ "Ma3ndich", "bzaf", "dyal", "SS7abat", "mn", "ghir", "Mona..." ]
[ [ 6, 6, "PER" ] ]
[ "Moha", "o", "Fatima", "3ndhom", "joj", "bnat", "o", "drri" ]
[ [ 0, 0, "PER" ], [ 2, 2, "PER" ] ]
[ "walakin", "ymken", "bant", "lia", "chi", "Tirmos", "f", "llour", "dial", "ssella." ]
[]
[ "jabt", "lik", "m3ya", "swart", "dyal", "tolnobil" ]
[]
[ "lkhmis", "sbou3", "inx2lah" ]
[]
[ "kanaDa", "kayna", "f", "2amrika", "chchamaliya" ]
[ [ 0, 0, "LOC" ], [ 3, 4, "LOC" ] ]
[ "man3ich", "7da", "mokhtalif", "m8woos", "makayfker", "ila", "f", "al3ab", "lvidio", "olmanga" ]
[]
[ "wach", "3emmrek", "Safrti", "l", "2oroppa" ]
[ [ 4, 4, "LOC" ] ]
[ "dakchi", "3lach", "3afak", "matsifthomch", "fwa9t", "m3atal", "nhar", "larba3" ]
[]
[ "ghat", "mchi", "l", "lasall", "nhar", "larb3?" ]
[]
[ "kif", "kat3raf,", "hir", "8", "safi", "nhar", "tlat" ]
[]
[ "aykhsna", "nmchiw", "ljanob", "l", "Fire", "Nation" ]
[]
[ "walakin", "3ndk", "zzher", "gddit", "nSawb", "ttamarin", "dyali", "lkhmis", "lli", "fat" ]
[]
[ "3emrek", "jerebti", "tfeddel", "sa7btek", "3la", "lwalida", "dialek" ]
[]
[ "kant", "wa7d", "lbnia", "sghira", "smitha", "Anya", "makennach", "kantkheliwha", "bou7dha" ]
[ [ 5, 5, "PER" ] ]
[ "tsabt", "blmarad", "lfaranci", "likayban", "mli", "kayb9aw", "rajlik", "itmalmlo." ]
[ [ 2, 2, "MISC" ] ]
[ "wach", "3rfti", "kifach", "noSlo", "l", "hotel", "Mansour?" ]
[ [ 5, 6, "ORG" ] ]
[ "wach", "ana", "tanhder", "m3a", "ssi", "Khalid?" ]
[ [ 5, 5, "PER" ] ]
[ "bighitini", "nkon", "James?" ]
[ [ 2, 2, "PER" ] ]
[ "blmonasaba,", "kantmnna", "tkon", "ma3rdtich", "3la", "Ihsan." ]
[ [ 5, 5, "PER" ] ]
[ "wach", "ghadi", "na7taj", "chahada", "jami3iya", "bach", "nl9a", "wadifa", "tghyir", "lbi2a?" ]
[]
[ "mora", "chi", "tlata", "dial", "simanat", "rje3t", "lfranssa", "3la", "9bel", "l5edma" ]
[ [ 6, 6, "LOC" ] ]
[ "rah", "msta9lla", "lblad", "3la", "franssa" ]
[ [ 4, 4, "LOC" ] ]
[ "ghat", "mchi", "l", "lasall", "nhar", "lkhmis?" ]
[]
[ "mazal", "n9dr", "nkhraj", "tilifoni", "onkhdm", "b", "Google" ]
[ [ 6, 6, "ORG" ] ]
[ "nafs", "lmojrimin", "f", "lbayt", "l", "2abyaD" ]
[]
[ "lmgharba", "kolxi", "f", "9atar" ]
[ [ 0, 0, "MISC" ], [ 3, 3, "LOC" ] ]
[ "Oh,", "yarbbi,", "3ndhom", "ghir", "Jack", "Daniels." ]
[ [ 4, 5, "ORG" ] ]
[ "aji", "tchofi", "howa", "li", "mchina", "lih", "w", "sawlna" ]
[]
[ "la", "akhti", "ma", "blanch", "hada", "maymknch" ]
[]
[ "la,", "hadak", "tilifon", "dyal", "Naima" ]
[ [ 4, 4, "PER" ] ]
[ "nta", "wahd", "ma", "m3ak", "hadra" ]
[]
[ "Tajine", "dial", "bTaTa", "o", "zzitoun." ]
[]
[ "dwazt", "chwiya", "dyal", "lwa9t", "f", "fransa", "o", "almanya", "o", "amrica", "ljanobiya" ]
[ [ 5, 5, "LOC" ], [ 7, 7, "LOC" ], [ 9, 10, "LOC" ] ]
[ "had", "lktab", "dyal", "Loubna." ]
[ [ 3, 3, "PER" ] ]
[ "a", "fink?", "charki", "m3a", "nas", "dakchi", "li", "katwejdi" ]
[]
[ "tamrid", "outa3lim", "makhdaminch", "fihom" ]
[]
[ "aktarya", "f", "amrica", "ljanobia", "fin", "kaynin", "7ayawnat", "fchkel", "f", "ghabat", "amazon" ]
[ [ 2, 3, "LOC" ], [ 9, 10, "LOC" ] ]
[ "khdithom", "mli", "kant", "kantfaraj", "f", "Masterchef", "f", "tlfaza." ]
[ [ 5, 5, "ORG" ] ]
[ "wach", "kat9ra", "l", "anglais?" ]
[ [ 3, 3, "MISC" ] ]
[ "ma", "ghadich", "nmchi", "7tel", "sebt" ]
[]
[ "3adi", "hit", "na3essa", "bnhar" ]
[]
[ "khaliha", "f", "yed", "lah", "o", "li", "mektaba", "like", "hiya", "hadike" ]
[]
[ "3andk", "lmdrassa", "nhar", "tnin?" ]
[]
[ "amrica", "ljanobiya", "kant", "fchkal,", "mkhtalfa", "3la", "lblad" ]
[ [ 0, 1, "LOC" ] ]
[ "wach", "msali", "ssebt", "jjay?" ]
[]
[ "anmchi", "l", "cinema", "lkhmiss", "jay,", "tji", "m3aya", "?" ]
[]
[ "wach", "rajlek", "tahoa", "mn", "Khounifra?" ]
[ [ 4, 4, "LOC" ] ]
[ "yalla8", "rje3t", "mn", "2almanya" ]
[ [ 3, 3, "LOC" ] ]
[ "kirak", "ghadi", "flmorja3a", "dyal", "imti7an", "riyadiyat?" ]
[]
[ "n9dr", "nghani", "oghniya", "mod7ika", "3la", "jotat", "maytin,", "ddoud", "kaydkhl", "o", "kaykhrj" ]
[]
[ "maghaydirouch", "l3mariya" ]
[]
[ "awili", "3la", "lbarba", "flhrira" ]
[]
[ "galt", "liya", "sahbti", "jlabatak", "mzayra" ]
[]
[ "l3rousa", "taht", "mn", "l3mariya" ]
[]
[ "ah,", "ghadi", "ntkhtabro", "f", "kolchi", "f", "lfaSl", "tani" ]
[]
[ "Lah", "ikhlf", "3lik", "assi", "Bojm3a" ]
[ [ 4, 4, "PER" ] ]
[ "ma", "ghadich", "nmchi", "7tel", "larba3" ]
[]
[ "wach", "n9adro", "n3awdo", "n7addo", "maw3id", "nhar", "l7ad?" ]
[]
[ "had", "ddar", "dyal", "Malika" ]
[ [ 3, 3, "PER" ] ]
[ "sma7", "liya", "bghit", "ngol", "safir" ]
[ [ 4, 4, "LOC" ] ]
[ "Knt", "kanDn", "ghadi", "n3red", "3la", "Assia,", "kat3rfha." ]
[ [ 5, 5, "PER" ] ]
[ "zwinin", "akhti", "ch7al", "bach", "t9amo", "3lik", "?" ]
[]
[ "kaynin", "chi", "Tyor", "wa3ra", "flbrazil" ]
[ [ 4, 4, "LOC" ] ]
[ "ghadin", "b9a", "hna", "7tel", "larba3" ]
[]
[ "idan", "hoa", "ghaybghi", "inta9l", "ltma" ]
[]
[ "meli", "mazal", "fazg,", "staghl", "lforSa", "o", "kherrej", "lkelb,", "ana", "ansekhkhen", "lik", "l7emmam", "o", "ncomandi", "lmakla", "dial", "lhnoud." ]
[ [ 16, 16, "MISC" ] ]
[ "hadchi", "mam3rofch", "bzaf", "f", "oropa" ]
[ [ 4, 4, "LOC" ] ]
[ "ghadi", "ykon", "3andna", "mti7an,", "ila", "kan", "hadchi", "likat9sad,", "simana", "jaya", "nhar", "lkhmis" ]
[]
[ "fach", "kendira", "ana", "kyji", "darbha", "zelzal" ]
[]
[ "kayt3llem", "longli", "bbou7dou" ]
[ [ 1, 1, "MISC" ] ]
[ "tsnay", "ta", "nhar", "tlat", "ghadi", "ikon", "iftitah" ]
[]
[ "imta", "ghadi", "sa3oudiya", "stenzef", "ga3", "lpetrole", "li", "3endha" ]
[ [ 2, 2, "LOC" ] ]
[ "Driss", "o", "Fatima", "mabghawch", "lmonaDa" ]
[ [ 0, 0, "PER" ], [ 2, 2, "PER" ] ]
[ "man9derch", "ydouz", "liya", "chi", "jem3a", "bla", "ksakso" ]
[]
[ "makayna", "la", "mosi9a", "faranciya" ]
[ [ 3, 3, "MISC" ] ]
[ "3adi", "hna", "kamlin", "khadmin", "fach", "makn" ]
[]
[ "wach", "darija", "logha", "wla", "lahja" ]
[ [ 1, 1, "MISC" ] ]
[ "makay3aytoch", "lik", "Einstein", "3la", "dakchi" ]
[ [ 2, 2, "PER" ] ]
End of preview. Expand in Data Studio

MoriscoNER — Moroccan Darija Arabizi NER

Named-entity recognition data for Moroccan Darija written in Arabizi (Latin script with digits such as 3, 7, 9), four entity types (PER, LOC, ORG, MISC), from the IANLP 2026 MoriscoNER paper. Code, pipelines and documentation: github.com/achrafsoltani/morisco-ner.

Contents

file content licence
train.standoff.jsonl MoriscoNER training corpus: 1,531 natively annotated YouTube comments, stand-off (comment id, token count, SHA-1 of the tokens, spans) CC BY-NC 4.0
rehydrate_youtube.py rebuilds the training tokens from the comment ids with your own YouTube Data API key, and checks each record Apache-2.0
arabizi_gold.jsonl gold test set: 400 sentences from the Darija Open Dataset (DODa), 252 entities, native-validated; {"tokens", "spans"} CC BY 4.0
arabizi_gold_ar.jsonl the test set rendered in Arabic script (token-aligned), for transliteration baselines CC BY 4.0
arabizi_draft_review.csv the test set's annotation workbook: LLM-teacher draft plus native corrections CC BY 4.0
darnercorp_gs_big_arabizi.jsonl transliterated training data: DarNERcorp gold + LLM-teacher silver, transliterated Arabic→Arabizi token by token CC BY 4.0

Spans are [start, end, TYPE] token indices, inclusive. The training corpus is stand-off because the text of YouTube comments may not be redistributed (YouTube API terms, GDPR). To rebuild it:

YT_API_KEY=... python rehydrate_youtube.py train.standoff.jsonl train.jsonl

Comments deleted or edited since collection cannot be recovered and are reported; in a spot check on 2026-10-19, 2 of 50 were no longer available.

Provenance

  • Training corpus: YouTube comments on Moroccan content, pre-labelled by an LLM teacher and corrected by two native annotators (Achraf Soltani, Soukaina Allali), Cohen's κ = 0.873 on a doubly annotated 200-comment subset.
  • Test set: DODa sentences (MIT), teacher-drafted then native-corrected; no sentence overlap with the training corpus.
  • Transliterated data: DarNERcorp (Moussa & Mourhir, Data in Brief, 2023; CC BY 4.0) plus LLM-teacher silver, kept only where the surface form is verbatim in the source.

Companion models

Citation 

@inproceedings{soltani2026morisconer,
  title     = {{MoriscoNER}: A Native {Moroccan-Darija} Latin-Script {NER} Corpus and an Encoder-Conditional Native-vs-Transliteration Study},
  author    = {Soltani, Achraf},
  booktitle = {Proceedings of IANLP 2026},
  series    = {Advances in Computer Science Applications and Research},
  publisher = {Springer},
  year      = {2026},
  note      = {in press}
}

Please also cite DarNERcorp and DODa.
Downloads last month
44
Number of rows:
1,931
Total file size:
1.53 MB

Models trained or fine-tuned on AchrafSoltani/morisco-ner