---
license: cc-by-sa-4.0
task_categories:
  - token-classification
task_ids:
  - multi-label-classification
language:
  - multilingual
tags:
  - language-identification
  - low-resource-languages
  - code-switching
  - span-classification
  - synthetic
size_categories:
  - 10K<n<100K
annotations_creators:
  - machine-generated
source_datasets:
  - cis-lmu/udhr-lid
  - facebook/flores
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.parquet
      - split: validation
        path: validation.parquet
      - split: test
        path: test.parquet
      - split: test_adversarial
        path: test_adversarial.parquet
      - split: test_romanized
        path: test_romanized.parquet
---

# Tessera mixed-page corpus

Synthetic multilingual web pages with **exact span boundaries and multi-label
language annotations**, for training and evaluating span-level language
identification.

## Why this is synthetic, stated up front

No public corpus annotates span boundaries and languages *inside* real multilingual
web pages at scale. Annotating one by hand across 479
languages, most of them low-resource, is not feasible and would itself be error-prone
in exactly the places that matter most.

So the gold labels here are **constructive**: each document is assembled from source
sentences of known language, and each span's offsets and language are recorded as it
is placed. The boundaries are exact by definition.

The cost is real and is not hidden: the difficulty of this corpus was chosen by its
authors. Results measured on it are labelled synthetic wherever they appear, and the
accompanying repository reports real-data checks separately.

## Contents

| split | documents | spans | languages | spans <= 20 chars | multi-label spans |
|---|---|---|---|---|---|
| train | 30,000 | 561,368 | 578 | 52.89% | 16.47% |
| validation | 3,000 | 56,117 | 534 | 55.08% | 18.25% |
| test | 4,000 | 74,200 | 549 | 55.08% | 18.31% |
| test_adversarial | 2,500 | 47,086 | 576 | 55.62% | 25.49% |
| test_romanized | 2,500 | 47,162 | 114 | 48.58% | 8.66% |

`test_adversarial` applies every degradation layer aggressively. `test_romanized`
contains only non-Latin-script body languages, romanised with probability 1.0.

## How a document is built

Pages imitate real furniture, because that furniture is *why* document-level LID
discards low-resource text: navigation, footers, advertising and copyright lines are
very often English even on a page whose body is not.

Archetypes:

- `buried_body` -- body in a mid/tail language, chrome in English. Document-level LID
  labels the page English, and the body is lost.
- `buried_minority` -- body in English with a quote or comment block in another
  language. The document survives filtering *as English*, and the minority text is
  silently absorbed into an English corpus.
- `codeswitch_heavy`, `short_span_heavy`, `monolingual` (control).

## Degradation layers

`shorten` (5-20 chars), `informalize` (repetition, emoji, case, punctuation,
abbreviation, typos -- all script-generic, no English-specific slang), `romanize`,
`codeswitch`, `mojibake` (encoding round-trip damage), `html_residue`. Each span
records which layers were applied, so results can be stratified by condition.

## Multi-label definition

Two kinds of multi-label, scored differently, and **which labels are correct is
measured rather than asserted**:

- **`any` (alternatives).** A string attested in several languages' source corpora
  gets all of them. The token `Massa` is attested in dozens of Latin-script
  languages; any one is correct, and predicting all of them earns no extra credit.
- **`all` (code-switching).** The span really contains every listed language, so each
  is separately required.
- **`script_only`.** Digits, punctuation and symbols get `und` and are scored on
  whether a system *declines* to name a language.

`primary` always records the language the text was actually sampled from, so
recall-per-language can never be inflated by permissive alternates.

## Sources and licensing

| source | licence | redistributable | share-alike |
|---|---|---|---|
| [UDHR-LID (cis-lmu/udhr-lid)](https://huggingface.co/datasets/cis-lmu/udhr-lid) | CC0-1.0 | True | False |
| [FLORES-200](https://github.com/facebookresearch/flores/tree/main/flores200) | CC-BY-SA-4.0 | True | True |

**Effective licence: CC-BY-SA-4.0.** FLORES-200 is CC-BY-SA-4.0 and its share-alike term propagates to any corpus containing its sentences.

No scraped web text is redistributed. Only material whose licence permits
redistribution is published.

## Resource tiers

Defined operationally, not editorially: `head` = in fastText lid.176's label space,
`mid` = absent from that but present in GlotLID or OpenLID, `tail` = covered by no
evaluated document-level LID. Counts: {'tail': 50, 'mid': 282, 'head': 147}.

## Known weaknesses

Please read these before using this corpus for anything consequential.

1. **Register.** UDHR is formal legal prose and FLORES is encyclopaedic. Neither is
   conversational. The `informalize` layer approximates informality; it does not
   sample it from real usage.
2. **Romanisation is systematic transliteration, not attested user romanisation.**
   Real users writing Hindi or Tigrinya in Latin script do not follow a
   transliteration standard. Anything trained or evaluated on this layer is measuring
   an approximation, and Han, Kana and Hangul are excluded entirely because
   character-reading transliteration produces text no human would write.
3. **Page structure is templated.** Five archetypes cannot cover the real diversity
   of the web.
4. **Sentence counts are very uneven.** Some languages have 40 UDHR sentences and
   others have 997 FLORES sentences; the per-language table below gives the exact
   figure, and any per-language claim should be read against it.
5. **Language identity is contested for some codes.** Where a macrolanguage and its
   varieties both appear, the labels follow the source corpus's choices, which are
   not neutral and may not match a speaker community's own view.
6. **Tail languages are only 50 of the 479
   here.** GlotLID covers 2,102 labels, so the population most at risk in this corpus
   is `mid`: languages that existing LID *knows* but that document-level application
   discards anyway.

## Language coverage

| code | script | source sentences | sources |
|---|---|---|---|
| `aar_Latn` | Latn | 57 | udhr |
| `abk_Cyrl` | Cyrl | 57 | udhr |
| `ace_Arab` | Arab | 997 | flores |
| `ace_Latn` | Latn | 1058 | flores, udhr |
| `acm_Arab` | Arab | 997 | flores |
| `acq_Arab` | Arab | 997 | flores |
| `acu_Latn` | Latn | 180 | udhr |
| `ada_Latn` | Latn | 59 | udhr |
| `ady_Cyrl` | Cyrl | 56 | udhr |
| `aeb_Arab` | Arab | 997 | flores |
| `afr_Latn` | Latn | 1057 | flores, udhr |
| `agr_Latn` | Latn | 57 | udhr |
| `aii_Syrc` | Syrc | 57 | udhr |
| `ajg_Latn` | Latn | 59 | udhr |
| `ajp_Arab` | Arab | 997 | flores |
| `aka_Latn` | Latn | 997 | flores |
| `als_Latn` | Latn | 1056 | flores, udhr |
| `alt_Cyrl` | Cyrl | 51 | udhr |
| `amc_Latn` | Latn | 56 | udhr |
| `ame_Latn` | Latn | 59 | udhr |
| `amh_Ethi` | Ethi | 1048 | flores, udhr |
| `ami_Latn` | Latn | 58 | udhr |
| `amr_Latn` | Latn | 61 | udhr |
| `apc_Arab` | Arab | 997 | flores |
| `arb_Arab` | Arab | 1055 | flores, udhr |
| `arb_Latn` | Latn | 997 | flores |
| `arl_Latn` | Latn | 60 | udhr |
| `arn_Latn` | Latn | 58 | udhr |
| `ars_Arab` | Arab | 997 | flores |
| `ary_Arab` | Arab | 997 | flores |
| `arz_Arab` | Arab | 997 | flores |
| `asm_Beng` | Beng | 997 | flores |
| `ast_Latn` | Latn | 1055 | flores, udhr |
| `auc_Latn` | Latn | 154 | udhr |
| `awa_Deva` | Deva | 997 | flores |
| `ayr_Latn` | Latn | 1056 | flores, udhr |
| `azb_Arab` | Arab | 997 | flores |
| `azj_Cyrl` | Cyrl | 56 | udhr |
| `azj_Latn` | Latn | 1055 | flores, udhr |
| `bak_Cyrl` | Cyrl | 997 | flores |
| `bam_Latn` | Latn | 1054 | flores, udhr |
| `ban_Latn` | Latn | 1057 | flores, udhr |
| `bax_Latn` | Latn | 58 | udhr |
| `bba_Latn` | Latn | 58 | udhr |
| `bci_Latn` | Latn | 58 | udhr |
| `bcl_Latn` | Latn | 59 | udhr |
| `bel_Cyrl` | Cyrl | 1055 | flores, udhr |
| `bem_Latn` | Latn | 1054 | flores, udhr |
| `ben_Beng` | Beng | 1055 | flores, udhr |
| `bfa_Latn` | Latn | 42 | udhr |
| `bho_Deva` | Deva | 1053 | flores, udhr |
| `bin_Latn` | Latn | 47 | udhr |
| `bis_Latn` | Latn | 59 | udhr |
| `bjn_Arab` | Arab | 997 | flores |
| `bjn_Latn` | Latn | 997 | flores |
| `blt_Tavt` | Tavt | 55 | udhr |
| `boa_Latn` | Latn | 61 | udhr |
| `bod_Tibt` | Tibt | 1057 | flores, udhr |
| `bos_Cyrl` | Cyrl | 59 | udhr |
| `bos_Latn` | Latn | 1056 | flores, udhr |
| `bre_Latn` | Latn | 59 | udhr |
| `buc_Latn` | Latn | 57 | udhr |
| `bug_Latn` | Latn | 1057 | flores, udhr |
| `bul_Cyrl` | Cyrl | 1055 | flores, udhr |
| `bum_Latn` | Latn | 58 | udhr |
| `cab_Latn` | Latn | 60 | udhr |
| `cak_Latn` | Latn | 57 | udhr |
| `cat_Latn` | Latn | 1055 | flores, udhr |
| `cbi_Latn` | Latn | 62 | udhr |
| `cbr_Latn` | Latn | 60 | udhr |
| `cbs_Latn` | Latn | 43 | udhr |
| `cbt_Latn` | Latn | 57 | udhr |
| `cbu_Latn` | Latn | 57 | udhr |
| `ccp_Cakm` | Cakm | 62 | udhr |
| `ceb_Latn` | Latn | 1056 | flores, udhr |
| `ces_Latn` | Latn | 1055 | flores, udhr |
| `cfm_Latn` | Latn | 57 | udhr |
| `cha_Latn` | Latn | 57 | udhr |
| `chj_Latn` | Latn | 59 | udhr |
| `chk_Latn` | Latn | 58 | udhr |
| `chr_Cher` | Cher | 115 | udhr |
| `chv_Cyrl` | Cyrl | 25 | udhr |
| `cic_Latn` | Latn | 60 | udhr |
| `cjk_Latn` | Latn | 1111 | flores, udhr |
| `cjs_Cyrl` | Cyrl | 59 | udhr |
| `cjy_Hani` | Hani | 58 | udhr |
| `ckb_Arab` | Arab | 997 | flores |
| `cmn_Hani` | Hani | 363 | udhr |
| `cnh_Latn` | Latn | 58 | udhr |
| `cni_Latn` | Latn | 52 | udhr |
| `cnr_Latn` | Latn | 59 | udhr |
| `cof_Latn` | Latn | 59 | udhr |
| `cos_Latn` | Latn | 59 | udhr |
| `cot_Latn` | Latn | 60 | udhr |
| `cpu_Latn` | Latn | 60 | udhr |
| `crh_Latn` | Latn | 1056 | flores, udhr |
| `cri_Latn` | Latn | 58 | udhr |
| `crs_Latn` | Latn | 58 | udhr |
| `csa_Latn` | Latn | 57 | udhr |
| `csw_Cans` | Cans | 43 | udhr |
| `ctd_Latn` | Latn | 60 | udhr |
| `cym_Latn` | Latn | 1057 | flores, udhr |
| `dag_Latn` | Latn | 57 | udhr |
| `dan_Latn` | Latn | 1058 | flores, udhr |
| `ddn_Latn` | Latn | 58 | udhr |
| `deu_Latn` | Latn | 1064 | flores, udhr |
| `dga_Latn` | Latn | 59 | udhr |
| `dik_Latn` | Latn | 997 | flores |
| `dip_Latn` | Latn | 57 | udhr |
| `div_Thaa` | Thaa | 57 | udhr |
| `dje_Latn` | Latn | 54 | udhr |
| `duu_Latn` | Latn | 57 | udhr |
| `dyo_Latn` | Latn | 55 | udhr |
| `dyu_Latn` | Latn | 1055 | flores, udhr |
| `dzo_Tibt` | Tibt | 1056 | flores, udhr |
| `ekk_Latn` | Latn | 60 | udhr |
| `ell_Grek` | Grek | 1113 | flores, udhr |
| `emk_Latn` | Latn | 58 | udhr |
| `eng_Latn` | Latn | 1055 | flores, udhr |
| `epo_Latn` | Latn | 1055 | flores, udhr |
| `ese_Latn` | Latn | 35 | udhr |
| `est_Latn` | Latn | 997 | flores |
| `eus_Latn` | Latn | 1056 | flores, udhr |
| `eve_Cyrl` | Cyrl | 49 | udhr |
| `evn_Cyrl` | Cyrl | 57 | udhr |
| `ewe_Latn` | Latn | 1056 | flores, udhr |
| `fao_Latn` | Latn | 1054 | flores, udhr |
| `fat_Latn` | Latn | 57 | udhr |
| `fij_Latn` | Latn | 1056 | flores, udhr |
| `fin_Latn` | Latn | 1058 | flores, udhr |
| `fkv_Latn` | Latn | 58 | udhr |
| `fon_Latn` | Latn | 1055 | flores, udhr |
| `fra_Latn` | Latn | 1055 | flores, udhr |
| `fry_Latn` | Latn | 58 | udhr |
| `fuf_Adlm` | Adlm | 58 | udhr |
| `fuf_Latn` | Latn | 58 | udhr |
| `fur_Latn` | Latn | 1056 | flores, udhr |
| `fuv_Latn` | Latn | 1105 | flores, udhr |
| `fvr_Latn` | Latn | 57 | udhr |
| `gaa_Latn` | Latn | 59 | udhr |
| `gag_Latn` | Latn | 59 | udhr |
| `gan_Hani` | Hani | 58 | udhr |
| `gaz_Latn` | Latn | 1057 | flores, udhr |
| `gej_Latn` | Latn | 58 | udhr |
| `gjn_Latn` | Latn | 60 | udhr |
| `gkp_Latn` | Latn | 56 | udhr |
| `gla_Latn` | Latn | 1059 | flores, udhr |
| `gld_Cyrl` | Cyrl | 59 | udhr |
| `gle_Latn` | Latn | 1053 | flores, udhr |
| `glg_Latn` | Latn | 1055 | flores, udhr |
| `glv_Latn` | Latn | 58 | udhr |
| `grn_Latn` | Latn | 997 | flores |
| `gsw_Latn` | Latn | 59 | udhr |
| `guc_Latn` | Latn | 58 | udhr |
| `gug_Latn` | Latn | 49 | udhr |
| `guj_Gujr` | Gujr | 1055 | flores, udhr |
| `guk_Latn` | Latn | 55 | udhr |
| `guu_Latn` | Latn | 58 | udhr |
| `gyr_Latn` | Latn | 49 | udhr |
| `hak_Hani` | Hani | 57 | udhr |
| `hat_Latn` | Latn | 1114 | flores, udhr |
| `hau_Latn` | Latn | 1152 | flores, udhr |
| `haw_Latn` | Latn | 58 | udhr |
| `hea_Latn` | Latn | 58 | udhr |
| `heb_Hebr` | Hebr | 1055 | flores, udhr |
| `hil_Latn` | Latn | 58 | udhr |
| `hin_Deva` | Deva | 1057 | flores, udhr |
| `hlt_Latn` | Latn | 55 | udhr |
| `hms_Latn` | Latn | 59 | udhr |
| `hne_Deva` | Deva | 997 | flores |
| `hni_Latn` | Latn | 57 | udhr |
| `hnj_Latn` | Latn | 58 | udhr |
| `hns_Latn` | Latn | 59 | udhr |
| `hrv_Latn` | Latn | 1055 | flores, udhr |
| `hsb_Latn` | Latn | 59 | udhr |
| `hsn_Hani` | Hani | 58 | udhr |
| `hun_Latn` | Latn | 1056 | flores, udhr |
| `hus_Latn` | Latn | 175 | udhr |
| `huu_Latn` | Latn | 58 | udhr |
| `hye_Armn` | Armn | 1068 | flores, udhr |
| `ibb_Latn` | Latn | 67 | udhr |
| `ibo_Latn` | Latn | 1055 | flores, udhr |
| `ido_Latn` | Latn | 55 | udhr |
| `idu_Latn` | Latn | 51 | udhr |
| `iii_Yiii` | Yiii | 57 | udhr |
| `ijs_Latn` | Latn | 50 | udhr |
| `ike_Cans` | Cans | 44 | udhr |
| `ilo_Latn` | Latn | 1055 | flores, udhr |
| `ina_Latn` | Latn | 58 | udhr |
| `ind_Latn` | Latn | 1057 | flores, udhr |
| `isl_Latn` | Latn | 1055 | flores, udhr |
| `ita_Latn` | Latn | 1055 | flores, udhr |
| `jav_Java` | Java | 62 | udhr |
| `jav_Latn` | Latn | 1057 | flores, udhr |
| `jiv_Latn` | Latn | 60 | udhr |
| `jpn_Jpan` | Jpan | 1140 | flores, udhr |
| `kaa_Cyrl` | Cyrl | 58 | udhr |
| `kab_Latn` | Latn | 1055 | flores, udhr |
| `kac_Latn` | Latn | 997 | flores |
| `kal_Latn` | Latn | 58 | udhr |
| `kam_Latn` | Latn | 997 | flores |
| `kan_Knda` | Knda | 1055 | flores, udhr |
| `kas_Arab` | Arab | 997 | flores |
| `kas_Deva` | Deva | 997 | flores |
| `kat_Geor` | Geor | 1055 | flores, udhr |
| `kaz_Cyrl` | Cyrl | 1055 | flores, udhr |
| `kbd_Cyrl` | Cyrl | 55 | udhr |
| `kbp_Latn` | Latn | 1056 | flores, udhr |
| `kbr_Latn` | Latn | 59 | udhr |
| `kde_Latn` | Latn | 51 | udhr |
| `kdh_Latn` | Latn | 59 | udhr |
| `kea_Latn` | Latn | 1055 | flores, udhr |
| `kek_Latn` | Latn | 57 | udhr |
| `kha_Latn` | Latn | 58 | udhr |
| `khk_Cyrl` | Cyrl | 1054 | flores, udhr |
| `khm_Khmr` | Khmr | 1056 | flores, udhr |
| `kik_Latn` | Latn | 997 | flores |
| `kin_Latn` | Latn | 1056 | flores, udhr |
| `kir_Cyrl` | Cyrl | 1055 | flores, udhr |
| `kjh_Cyrl` | Cyrl | 57 | udhr |
| `kkh_Lana` | Lana | 51 | udhr |
| `kmb_Latn` | Latn | 1113 | flores, udhr |
| `kmr_Latn` | Latn | 1055 | flores, udhr |
| `knc_Arab` | Arab | 997 | flores |
| `knc_Latn` | Latn | 1061 | flores, udhr |
| `kng_Latn` | Latn | 115 | udhr |
| `koi_Cyrl` | Cyrl | 57 | udhr |
| `kon_Latn` | Latn | 997 | flores |
| `koo_Latn` | Latn | 54 | udhr |
| `kor_Hang` | Hang | 1055 | flores, udhr |
| `kqn_Latn` | Latn | 59 | udhr |
| `kqs_Latn` | Latn | 58 | udhr |
| `kri_Latn` | Latn | 58 | udhr |
| `krl_Latn` | Latn | 183 | udhr |
| `ktu_Latn` | Latn | 118 | udhr |
| `kwi_Latn` | Latn | 145 | udhr |
| `lad_Latn` | Latn | 58 | udhr |
| `lao_Laoo` | Laoo | 1054 | flores, udhr |
| `lat_Latn` | Latn | 118 | udhr |
| `lia_Latn` | Latn | 45 | udhr |
| `lij_Latn` | Latn | 1055 | flores, udhr |
| `lim_Latn` | Latn | 997 | flores |
| `lin_Latn` | Latn | 1107 | flores, udhr |
| `lit_Latn` | Latn | 1055 | flores, udhr |
| `lld_Latn` | Latn | 59 | udhr |
| `lmo_Latn` | Latn | 997 | flores |
| `lns_Latn` | Latn | 57 | udhr |
| `lob_Latn` | Latn | 58 | udhr |
| `lot_Latn` | Latn | 140 | udhr |
| `loz_Latn` | Latn | 58 | udhr |
| `ltg_Latn` | Latn | 997 | flores |
| `ltz_Latn` | Latn | 1056 | flores, udhr |
| `lua_Latn` | Latn | 1055 | flores, udhr |
| `lue_Latn` | Latn | 58 | udhr |
| `lug_Latn` | Latn | 1046 | flores, udhr |
| `lun_Latn` | Latn | 56 | udhr |
| `luo_Latn` | Latn | 997 | flores |
| `lus_Latn` | Latn | 1055 | flores, udhr |
| `lvs_Latn` | Latn | 1112 | flores, udhr |
| `mad_Latn` | Latn | 60 | udhr |
| `mag_Deva` | Deva | 1050 | flores, udhr |
| `mah_Latn` | Latn | 58 | udhr |
| `mai_Deva` | Deva | 1055 | flores, udhr |
| `mal_Mlym` | Mlym | 1097 | flores, udhr |
| `mam_Latn` | Latn | 58 | udhr |
| `mar_Deva` | Deva | 1056 | flores, udhr |
| `maz_Latn` | Latn | 59 | udhr |
| `mcd_Latn` | Latn | 60 | udhr |
| `mcf_Latn` | Latn | 62 | udhr |
| `men_Latn` | Latn | 59 | udhr |
| `mfq_Latn` | Latn | 58 | udhr |
| `mic_Latn` | Latn | 59 | udhr |
| `min_Arab` | Arab | 997 | flores |
| `min_Latn` | Latn | 1056 | flores, udhr |
| `miq_Latn` | Latn | 51 | udhr |
| `mkd_Cyrl` | Cyrl | 1055 | flores, udhr |
| `mlt_Latn` | Latn | 1056 | flores, udhr |
| `mni_Beng` | Beng | 997 | flores |
| `mnw_Mymr` | Mymr | 56 | udhr |
| `mor_Latn` | Latn | 59 | udhr |
| `mos_Latn` | Latn | 1053 | flores, udhr |
| `mri_Latn` | Latn | 1054 | flores, udhr |
| `mto_Latn` | Latn | 51 | udhr |
| `mxi_Latn` | Latn | 58 | udhr |
| `mxv_Latn` | Latn | 58 | udhr |
| `mya_Mymr` | Mymr | 1053 | flores, udhr |
| `mzi_Latn` | Latn | 58 | udhr |
| `nan_Hani` | Hani | 58 | udhr |
| `nav_Latn` | Latn | 59 | udhr |
| `nba_Latn` | Latn | 59 | udhr |
| `nbl_Latn` | Latn | 58 | udhr |
| `ndo_Latn` | Latn | 58 | udhr |
| `nds_Latn` | Latn | 58 | udhr |
| `nhn_Latn` | Latn | 58 | udhr |
| `nio_Cyrl` | Cyrl | 51 | udhr |
| `niu_Latn` | Latn | 58 | udhr |
| `niv_Cyrl` | Cyrl | 56 | udhr |
| `njo_Latn` | Latn | 64 | udhr |
| `nku_Latn` | Latn | 52 | udhr |
| `nld_Latn` | Latn | 1055 | flores, udhr |
| `nno_Latn` | Latn | 1055 | flores, udhr |
| `nob_Latn` | Latn | 1059 | flores, udhr |
| `not_Latn` | Latn | 59 | udhr |
| `npi_Deva` | Deva | 1052 | flores, udhr |
| `nso_Latn` | Latn | 1055 | flores, udhr |
| `nus_Latn` | Latn | 997 | flores |
| `nya_Latn` | Latn | 1113 | flores, udhr |
| `nym_Latn` | Latn | 49 | udhr |
| `nyn_Latn` | Latn | 50 | udhr |
| `nzi_Latn` | Latn | 59 | udhr |
| `oaa_Cyrl` | Cyrl | 58 | udhr |
| `oci_Latn` | Latn | 1395 | flores, udhr |
| `ojb_Cans` | Cans | 44 | udhr |
| `oki_Latn` | Latn | 59 | udhr |
| `orh_Latn` | Latn | 59 | udhr |
| `ory_Orya` | Orya | 997 | flores |
| `oss_Cyrl` | Cyrl | 44 | udhr |
| `ote_Latn` | Latn | 59 | udhr |
| `pag_Latn` | Latn | 997 | flores |
| `pam_Latn` | Latn | 66 | udhr |
| `pan_Guru` | Guru | 1055 | flores, udhr |
| `pap_Latn` | Latn | 1055 | flores, udhr |
| `pau_Latn` | Latn | 57 | udhr |
| `pbb_Latn` | Latn | 56 | udhr |
| `pbt_Arab` | Arab | 997 | flores |
| `pbu_Arab` | Arab | 59 | udhr |
| `pcd_Latn` | Latn | 58 | udhr |
| `pcm_Latn` | Latn | 56 | udhr |
| `pes_Arab` | Arab | 1055 | flores, udhr |
| `pis_Latn` | Latn | 57 | udhr |
| `piu_Latn` | Latn | 57 | udhr |
| `plt_Latn` | Latn | 1050 | flores, udhr |
| `pnb_Arab` | Arab | 61 | udhr |
| `pol_Latn` | Latn | 1056 | flores, udhr |
| `pon_Latn` | Latn | 57 | udhr |
| `por_Latn` | Latn | 1112 | flores, udhr |
| `pov_Latn` | Latn | 115 | udhr |
| `ppl_Latn` | Latn | 59 | udhr |
| `prs_Arab` | Arab | 1053 | flores, udhr |
| `quc_Latn` | Latn | 58 | udhr |
| `que_Latn` | Latn | 58 | udhr |
| `qug_Latn` | Latn | 182 | udhr |
| `quh_Latn` | Latn | 57 | udhr |
| `quy_Latn` | Latn | 1057 | flores, udhr |
| `quz_Latn` | Latn | 57 | udhr |
| `qva_Latn` | Latn | 59 | udhr |
| `qvc_Latn` | Latn | 58 | udhr |
| `qvh_Latn` | Latn | 58 | udhr |
| `qvm_Latn` | Latn | 60 | udhr |
| `qvn_Latn` | Latn | 57 | udhr |
| `qwh_Latn` | Latn | 59 | udhr |
| `qxn_Latn` | Latn | 57 | udhr |
| `qxu_Latn` | Latn | 54 | udhr |
| `rar_Latn` | Latn | 59 | udhr |
| `rgn_Latn` | Latn | 58 | udhr |
| `rmn_Latn` | Latn | 116 | udhr |
| `roh_Latn` | Latn | 399 | udhr |
| `ron_Latn` | Latn | 1094 | flores, udhr |
| `run_Latn` | Latn | 1055 | flores, udhr |
| `rup_Latn` | Latn | 57 | udhr |
| `rus_Cyrl` | Cyrl | 1055 | flores, udhr |
| `sag_Latn` | Latn | 1056 | flores, udhr |
| `sah_Cyrl` | Cyrl | 57 | udhr |
| `san_Deva` | Deva | 1052 | flores, udhr |
| `san_Gran` | Gran | 51 | udhr |
| `sat_Olck` | Olck | 997 | flores |
| `scn_Latn` | Latn | 997 | flores |
| `sco_Latn` | Latn | 58 | udhr |
| `sey_Latn` | Latn | 61 | udhr |
| `shk_Latn` | Latn | 55 | udhr |
| `shn_Mymr` | Mymr | 1053 | flores, udhr |
| `shp_Latn` | Latn | 62 | udhr |
| `sid_Latn` | Latn | 57 | udhr |
| `sin_Sinh` | Sinh | 1056 | flores, udhr |
| `skr_Arab` | Arab | 63 | udhr |
| `slk_Latn` | Latn | 1053 | flores, udhr |
| `slr_Latn` | Latn | 58 | udhr |
| `slv_Latn` | Latn | 1056 | flores, udhr |
| `sme_Latn` | Latn | 58 | udhr |
| `smo_Latn` | Latn | 1060 | flores, udhr |
| `sna_Latn` | Latn | 1057 | flores, udhr |
| `snd_Arab` | Arab | 997 | flores |
| `snk_Latn` | Latn | 58 | udhr |
| `snn_Latn` | Latn | 136 | udhr |
| `som_Latn` | Latn | 1054 | flores, udhr |
| `sot_Latn` | Latn | 1054 | flores, udhr |
| `spa_Latn` | Latn | 1075 | flores, udhr |
| `src_Latn` | Latn | 59 | udhr |
| `srd_Latn` | Latn | 997 | flores |
| `srp_Cyrl` | Cyrl | 1056 | flores, udhr |
| `srp_Latn` | Latn | 59 | udhr |
| `srr_Latn` | Latn | 58 | udhr |
| `ssw_Latn` | Latn | 1057 | flores, udhr |
| `suk_Latn` | Latn | 58 | udhr |
| `sun_Latn` | Latn | 1063 | flores, udhr |
| `sus_Latn` | Latn | 58 | udhr |
| `swb_Latn` | Latn | 58 | udhr |
| `swe_Latn` | Latn | 1058 | flores, udhr |
| `swh_Latn` | Latn | 1055 | flores, udhr |
| `szl_Latn` | Latn | 997 | flores |
| `tah_Latn` | Latn | 57 | udhr |
| `tam_Taml` | Taml | 1066 | flores, udhr |
| `taq_Latn` | Latn | 997 | flores |
| `taq_Tfng` | Tfng | 997 | flores |
| `tat_Cyrl` | Cyrl | 1054 | flores, udhr |
| `tbz_Latn` | Latn | 60 | udhr |
| `tca_Latn` | Latn | 60 | udhr |
| `tdt_Latn` | Latn | 57 | udhr |
| `tel_Telu` | Telu | 1055 | flores, udhr |
| `tem_Latn` | Latn | 55 | udhr |
| `tet_Latn` | Latn | 58 | udhr |
| `tgk_Cyrl` | Cyrl | 1054 | flores, udhr |
| `tgl_Latn` | Latn | 1057 | flores, udhr |
| `tha_Thai` | Thai | 1111 | flores, udhr |
| `tir_Ethi` | Ethi | 1055 | flores, udhr |
| `tiv_Latn` | Latn | 67 | udhr |
| `tly_Latn` | Latn | 59 | udhr |
| `tob_Latn` | Latn | 59 | udhr |
| `toi_Latn` | Latn | 57 | udhr |
| `toj_Latn` | Latn | 57 | udhr |
| `ton_Latn` | Latn | 60 | udhr |
| `top_Latn` | Latn | 59 | udhr |
| `tpi_Latn` | Latn | 1055 | flores, udhr |
| `tsn_Latn` | Latn | 1055 | flores, udhr |
| `tso_Latn` | Latn | 1111 | flores, udhr |
| `tsz_Latn` | Latn | 59 | udhr |
| `tuk_Cyrl` | Cyrl | 57 | udhr |
| `tuk_Latn` | Latn | 1056 | flores, udhr |
| `tum_Latn` | Latn | 997 | flores |
| `tur_Latn` | Latn | 1056 | flores, udhr |
| `twi_Latn` | Latn | 1113 | flores, udhr |
| `tyv_Cyrl` | Cyrl | 58 | udhr |
| `tzh_Latn` | Latn | 25 | udhr |
| `tzm_Tfng` | Tfng | 998 | flores, udhr |
| `tzo_Latn` | Latn | 58 | udhr |
| `udu_Latn` | Latn | 60 | udhr |
| `uig_Arab` | Arab | 1055 | flores, udhr |
| `uig_Latn` | Latn | 59 | udhr |
| `ukr_Cyrl` | Cyrl | 1055 | flores, udhr |
| `umb_Latn` | Latn | 1104 | flores, udhr |
| `ura_Latn` | Latn | 56 | udhr |
| `urd_Arab` | Arab | 1105 | flores, udhr |
| `uzn_Cyrl` | Cyrl | 58 | udhr |
| `uzn_Latn` | Latn | 1055 | flores, udhr |
| `vai_Vaii` | Vaii | 61 | udhr |
| `vec_Latn` | Latn | 1055 | flores, udhr |
| `ven_Latn` | Latn | 89 | udhr |
| `vep_Latn` | Latn | 58 | udhr |
| `vie_Latn` | Latn | 1057 | flores, udhr |
| `vmw_Latn` | Latn | 58 | udhr |
| `war_Latn` | Latn | 1057 | flores, udhr |
| `wln_Latn` | Latn | 59 | udhr |
| `wol_Latn` | Latn | 1057 | flores, udhr |
| `wuu_Hani` | Hani | 58 | udhr |
| `wwa_Latn` | Latn | 39 | udhr |
| `xho_Latn` | Latn | 1055 | flores, udhr |
| `xsm_Latn` | Latn | 59 | udhr |
| `yad_Latn` | Latn | 55 | udhr |
| `yao_Latn` | Latn | 52 | udhr |
| `yap_Latn` | Latn | 58 | udhr |
| `ydd_Hebr` | Hebr | 1056 | flores, udhr |
| `ykg_Cyrl` | Cyrl | 51 | udhr |
| `yor_Latn` | Latn | 1054 | flores, udhr |
| `yrk_Cyrl` | Cyrl | 55 | udhr |
| `yua_Latn` | Latn | 59 | udhr |
| `yue_Hani` | Hani | 59 | udhr |
| `yue_Hant` | Hant | 997 | flores |
| `zam_Latn` | Latn | 60 | udhr |
| `zdj_Latn` | Latn | 50 | udhr |
| `zgh_Tfng` | Tfng | 61 | udhr |
| `zho_Hans` | Hans | 997 | flores |
| `zho_Hant` | Hant | 997 | flores |
| `zlm_Arab` | Arab | 24 | udhr |
| `zlm_Latn` | Latn | 58 | udhr |
| `zro_Latn` | Latn | 56 | udhr |
| `zsm_Latn` | Latn | 997 | flores |
| `ztu_Latn` | Latn | 58 | udhr |
| `zul_Latn` | Latn | 1055 | flores, udhr |
| `zyb_Latn` | Latn | 57 | udhr |

## Links

- Code: <https://github.com/NagaYu/tessera>
- Model: <https://huggingface.co/NagaYu/tessera-span-lid>
- Demo (runs in your browser): <https://huggingface.co/spaces/NagaYu/tessera>

## Citation

Built with [Tessera](https://github.com/NagaYu/tessera) -- span-level multi-label
language identification. Repository: `NagaYu/tessera-mixed-pages`.