---
license: cc
configs:
- config_name: afr
  data_files:
  - split: train
    path:
    - afr.json.gz
- config_name: ain
  data_files:
  - split: train
    path:
    - ain.json.gz
- config_name: alt
  data_files:
  - split: train
    path:
    - alt.json.gz
- config_name: amh
  data_files:
  - split: train
    path:
    - amh.json.gz
- config_name: ang
  data_files:
  - split: train
    path:
    - ang.json.gz
- config_name: ara
  data_files:
  - split: train
    path:
    - ara.json.gz
- config_name: arn
  data_files:
  - split: train
    path:
    - arn.json.gz
- config_name: arq
  data_files:
  - split: train
    path:
    - arq.json.gz
- config_name: arz
  data_files:
  - split: train
    path:
    - arz.json.gz
- config_name: asm
  data_files:
  - split: train
    path:
    - asm.json.gz
- config_name: ast
  data_files:
  - split: train
    path:
    - ast.json.gz
- config_name: avk
  data_files:
  - split: train
    path:
    - avk.json.gz
- config_name: awa
  data_files:
  - split: train
    path:
    - awa.json.gz
- config_name: aze
  data_files:
  - split: train
    path:
    - aze.json.gz
- config_name: bak
  data_files:
  - split: train
    path:
    - bak.json.gz
- config_name: bar
  data_files:
  - split: train
    path:
    - bar.json.gz
- config_name: bel
  data_files:
  - split: train
    path:
    - bel.json.gz
- config_name: ben
  data_files:
  - split: train
    path:
    - ben.json.gz
- config_name: ber
  data_files:
  - split: train
    path:
    - ber.json.gz
- config_name: bos
  data_files:
  - split: train
    path:
    - bos.json.gz
- config_name: bre
  data_files:
  - split: train
    path:
    - bre.json.gz
- config_name: bua
  data_files:
  - split: train
    path:
    - bua.json.gz
- config_name: bul
  data_files:
  - split: train
    path:
    - bul.json.gz
- config_name: bzt
  data_files:
  - split: train
    path:
    - bzt.json.gz
- config_name: cat
  data_files:
  - split: train
    path:
    - cat.json.gz
- config_name: cbk
  data_files:
  - split: train
    path:
    - cbk.json.gz
- config_name: ceb
  data_files:
  - split: train
    path:
    - ceb.json.gz
- config_name: ces
  data_files:
  - split: train
    path:
    - ces.json.gz
- config_name: cha
  data_files:
  - split: train
    path:
    - cha.json.gz
- config_name: chv
  data_files:
  - split: train
    path:
    - chv.json.gz
- config_name: ckb
  data_files:
  - split: train
    path:
    - ckb.json.gz
- config_name: cmn
  data_files:
  - split: train
    path:
    - cmn.json.gz
- config_name: cor
  data_files:
  - split: train
    path:
    - cor.json.gz
- config_name: csb
  data_files:
  - split: train
    path:
    - csb.json.gz
- config_name: cym
  data_files:
  - split: train
    path:
    - cym.json.gz
- config_name: dan
  data_files:
  - split: train
    path:
    - dan.json.gz
- config_name: deu
  data_files:
  - split: train
    path:
    - deu.json.gz
- config_name: dsb
  data_files:
  - split: train
    path:
    - dsb.json.gz
- config_name: dtp
  data_files:
  - split: train
    path:
    - dtp.json.gz
- config_name: ell
  data_files:
  - split: train
    path:
    - ell.json.gz
- config_name: eng
  default: true
  data_files:
  - split: train
    path:
    - eng.json.gz
- config_name: enm
  data_files:
  - split: train
    path:
    - enm.json.gz
- config_name: epo
  data_files:
  - split: train
    path:
    - epo.json.gz
- config_name: est
  data_files:
  - split: train
    path:
    - est.json.gz
- config_name: eus
  data_files:
  - split: train
    path:
    - eus.json.gz
- config_name: ext
  data_files:
  - split: train
    path:
    - ext.json.gz
- config_name: fao
  data_files:
  - split: train
    path:
    - fao.json.gz
- config_name: fin
  data_files:
  - split: train
    path:
    - fin.json.gz
- config_name: fkv
  data_files:
  - split: train
    path:
    - fkv.json.gz
- config_name: fra
  data_files:
  - split: train
    path:
    - fra.json.gz
- config_name: frr
  data_files:
  - split: train
    path:
    - frr.json.gz
- config_name: fry
  data_files:
  - split: train
    path:
    - fry.json.gz
- config_name: gcf
  data_files:
  - split: train
    path:
    - gcf.json.gz
- config_name: gla
  data_files:
  - split: train
    path:
    - gla.json.gz
- config_name: gle
  data_files:
  - split: train
    path:
    - gle.json.gz
- config_name: glg
  data_files:
  - split: train
    path:
    - glg.json.gz
- config_name: gos
  data_files:
  - split: train
    path:
    - gos.json.gz
- config_name: grc
  data_files:
  - split: train
    path:
    - grc.json.gz
- config_name: grn
  data_files:
  - split: train
    path:
    - grn.json.gz
- config_name: guc
  data_files:
  - split: train
    path:
    - guc.json.gz
- config_name: hau
  data_files:
  - split: train
    path:
    - hau.json.gz
- config_name: hbo
  data_files:
  - split: train
    path:
    - hbo.json.gz
- config_name: heb
  data_files:
  - split: train
    path:
    - heb.json.gz
- config_name: hin
  data_files:
  - split: train
    path:
    - hin.json.gz
- config_name: hoc
  data_files:
  - split: train
    path:
    - hoc.json.gz
- config_name: hrv
  data_files:
  - split: train
    path:
    - hrv.json.gz
- config_name: hrx
  data_files:
  - split: train
    path:
    - hrx.json.gz
- config_name: hsb
  data_files:
  - split: train
    path:
    - hsb.json.gz
- config_name: hun
  data_files:
  - split: train
    path:
    - hun.json.gz
- config_name: hye
  data_files:
  - split: train
    path:
    - hye.json.gz
- config_name: ido
  data_files:
  - split: train
    path:
    - ido.json.gz
- config_name: ile
  data_files:
  - split: train
    path:
    - ile.json.gz
- config_name: ilo
  data_files:
  - split: train
    path:
    - ilo.json.gz
- config_name: ina
  data_files:
  - split: train
    path:
    - ina.json.gz
- config_name: ind
  data_files:
  - split: train
    path:
    - ind.json.gz
- config_name: isl
  data_files:
  - split: train
    path:
    - isl.json.gz
- config_name: ita
  data_files:
  - split: train
    path:
    - ita.json.gz
- config_name: izh
  data_files:
  - split: train
    path:
    - izh.json.gz
- config_name: jav
  data_files:
  - split: train
    path:
    - jav.json.gz
- config_name: jbo
  data_files:
  - split: train
    path:
    - jbo.json.gz
- config_name: jpn
  data_files:
  - split: train
    path:
    - jpn.json.gz
- config_name: kab
  data_files:
  - split: train
    path:
    - kab.json.gz
- config_name: kas
  data_files:
  - split: train
    path:
    - kas.json.gz
- config_name: kat
  data_files:
  - split: train
    path:
    - kat.json.gz
- config_name: kaz
  data_files:
  - split: train
    path:
    - kaz.json.gz
- config_name: kbd
  data_files:
  - split: train
    path:
    - kbd.json.gz
- config_name: kha
  data_files:
  - split: train
    path:
    - kha.json.gz
- config_name: khm
  data_files:
  - split: train
    path:
    - khm.json.gz
- config_name: kir
  data_files:
  - split: train
    path:
    - kir.json.gz
- config_name: kmr
  data_files:
  - split: train
    path:
    - kmr.json.gz
- config_name: knc
  data_files:
  - split: train
    path:
    - knc.json.gz
- config_name: kor
  data_files:
  - split: train
    path:
    - kor.json.gz
- config_name: krc
  data_files:
  - split: train
    path:
    - krc.json.gz
- config_name: kzj
  data_files:
  - split: train
    path:
    - kzj.json.gz
- config_name: lad
  data_files:
  - split: train
    path:
    - lad.json.gz
- config_name: lat
  data_files:
  - split: train
    path:
    - lat.json.gz
- config_name: ldn
  data_files:
  - split: train
    path:
    - ldn.json.gz
- config_name: lfn
  data_files:
  - split: train
    path:
    - lfn.json.gz
- config_name: lij
  data_files:
  - split: train
    path:
    - lij.json.gz
- config_name: lin
  data_files:
  - split: train
    path:
    - lin.json.gz
- config_name: lit
  data_files:
  - split: train
    path:
    - lit.json.gz
- config_name: ltz
  data_files:
  - split: train
    path:
    - ltz.json.gz
- config_name: lvs
  data_files:
  - split: train
    path:
    - lvs.json.gz
- config_name: lzh
  data_files:
  - split: train
    path:
    - lzh.json.gz
- config_name: lzz
  data_files:
  - split: train
    path:
    - lzz.json.gz
- config_name: mal
  data_files:
  - split: train
    path:
    - mal.json.gz
- config_name: mar
  data_files:
  - split: train
    path:
    - mar.json.gz
- config_name: max
  data_files:
  - split: train
    path:
    - max.json.gz
- config_name: mfa
  data_files:
  - split: train
    path:
    - mfa.json.gz
- config_name: mhr
  data_files:
  - split: train
    path:
    - mhr.json.gz
- config_name: mkd
  data_files:
  - split: train
    path:
    - mkd.json.gz
- config_name: mlt
  data_files:
  - split: train
    path:
    - mlt.json.gz
- config_name: mon
  data_files:
  - split: train
    path:
    - mon.json.gz
- config_name: mri
  data_files:
  - split: train
    path:
    - mri.json.gz
- config_name: mus
  data_files:
  - split: train
    path:
    - mus.json.gz
- config_name: mya
  data_files:
  - split: train
    path:
    - mya.json.gz
- config_name: nds
  data_files:
  - split: train
    path:
    - nds.json.gz
- config_name: nld
  data_files:
  - split: train
    path:
    - nld.json.gz
- config_name: nnb
  data_files:
  - split: train
    path:
    - nnb.json.gz
- config_name: nno
  data_files:
  - split: train
    path:
    - nno.json.gz
- config_name: nob
  data_files:
  - split: train
    path:
    - nob.json.gz
- config_name: nog
  data_files:
  - split: train
    path:
    - nog.json.gz
- config_name: nov
  data_files:
  - split: train
    path:
    - nov.json.gz
- config_name: npi
  data_files:
  - split: train
    path:
    - npi.json.gz
- config_name: nst
  data_files:
  - split: train
    path:
    - nst.json.gz
- config_name: nus
  data_files:
  - split: train
    path:
    - nus.json.gz
- config_name: oci
  data_files:
  - split: train
    path:
    - oci.json.gz
- config_name: orv
  data_files:
  - split: train
    path:
    - orv.json.gz
- config_name: oss
  data_files:
  - split: train
    path:
    - oss.json.gz
- config_name: ota
  data_files:
  - split: train
    path:
    - ota.json.gz
- config_name: pam
  data_files:
  - split: train
    path:
    - pam.json.gz
- config_name: pcd
  data_files:
  - split: train
    path:
    - pcd.json.gz
- config_name: pes
  data_files:
  - split: train
    path:
    - pes.json.gz
- config_name: pms
  data_files:
  - split: train
    path:
    - pms.json.gz
- config_name: pol
  data_files:
  - split: train
    path:
    - pol.json.gz
- config_name: por
  data_files:
  - split: train
    path:
    - por.json.gz
- config_name: prg
  data_files:
  - split: train
    path:
    - prg.json.gz
- config_name: que
  data_files:
  - split: train
    path:
    - que.json.gz
- config_name: rhg
  data_files:
  - split: train
    path:
    - rhg.json.gz
- config_name: rif
  data_files:
  - split: train
    path:
    - rif.json.gz
- config_name: rom
  data_files:
  - split: train
    path:
    - rom.json.gz
- config_name: ron
  data_files:
  - split: train
    path:
    - ron.json.gz
- config_name: run
  data_files:
  - split: train
    path:
    - run.json.gz
- config_name: rus
  data_files:
  - split: train
    path:
    - rus.json.gz
- config_name: sah
  data_files:
  - split: train
    path:
    - sah.json.gz
- config_name: san
  data_files:
  - split: train
    path:
    - san.json.gz
- config_name: sat
  data_files:
  - split: train
    path:
    - sat.json.gz
- config_name: sdh
  data_files:
  - split: train
    path:
    - sdh.json.gz
- config_name: shi
  data_files:
  - split: train
    path:
    - shi.json.gz
- config_name: sjn
  data_files:
  - split: train
    path:
    - sjn.json.gz
- config_name: skr
  data_files:
  - split: train
    path:
    - skr.json.gz
- config_name: slk
  data_files:
  - split: train
    path:
    - slk.json.gz
- config_name: slv
  data_files:
  - split: train
    path:
    - slv.json.gz
- config_name: sme
  data_files:
  - split: train
    path:
    - sme.json.gz
- config_name: spa
  data_files:
  - split: train
    path:
    - spa.json.gz
- config_name: sqi
  data_files:
  - split: train
    path:
    - sqi.json.gz
- config_name: srp
  data_files:
  - split: train
    path:
    - srp.json.gz
- config_name: sux
  data_files:
  - split: train
    path:
    - sux.json.gz
- config_name: swc
  data_files:
  - split: train
    path:
    - swc.json.gz
- config_name: swe
  data_files:
  - split: train
    path:
    - swe.json.gz
- config_name: swg
  data_files:
  - split: train
    path:
    - swg.json.gz
- config_name: swh
  data_files:
  - split: train
    path:
    - swh.json.gz
- config_name: tam
  data_files:
  - split: train
    path:
    - tam.json.gz
- config_name: tat
  data_files:
  - split: train
    path:
    - tat.json.gz
- config_name: tet
  data_files:
  - split: train
    path:
    - tet.json.gz
- config_name: tgl
  data_files:
  - split: train
    path:
    - tgl.json.gz
- config_name: tha
  data_files:
  - split: train
    path:
    - tha.json.gz
- config_name: tig
  data_files:
  - split: train
    path:
    - tig.json.gz
- config_name: tir
  data_files:
  - split: train
    path:
    - tir.json.gz
- config_name: tlh
  data_files:
  - split: train
    path:
    - tlh.json.gz
- config_name: tmr
  data_files:
  - split: train
    path:
    - tmr.json.gz
- config_name: tok
  data_files:
  - split: train
    path:
    - tok.json.gz
- config_name: tpw
  data_files:
  - split: train
    path:
    - tpw.json.gz
- config_name: tuk
  data_files:
  - split: train
    path:
    - tuk.json.gz
- config_name: tur
  data_files:
  - split: train
    path:
    - tur.json.gz
- config_name: tzl
  data_files:
  - split: train
    path:
    - tzl.json.gz
- config_name: udm
  data_files:
  - split: train
    path:
    - udm.json.gz
- config_name: uig
  data_files:
  - split: train
    path:
    - uig.json.gz
- config_name: ukr
  data_files:
  - split: train
    path:
    - ukr.json.gz
- config_name: urd
  data_files:
  - split: train
    path:
    - urd.json.gz
- config_name: uzb
  data_files:
  - split: train
    path:
    - uzb.json.gz
- config_name: vec
  data_files:
  - split: train
    path:
    - vec.json.gz
- config_name: vep
  data_files:
  - split: train
    path:
    - vep.json.gz
- config_name: vie
  data_files:
  - split: train
    path:
    - vie.json.gz
- config_name: vol
  data_files:
  - split: train
    path:
    - vol.json.gz
- config_name: war
  data_files:
  - split: train
    path:
    - war.json.gz
- config_name: wuu
  data_files:
  - split: train
    path:
    - wuu.json.gz
- config_name: xmf
  data_files:
  - split: train
    path:
    - xmf.json.gz
- config_name: yid
  data_files:
  - split: train
    path:
    - yid.json.gz
- config_name: yue
  data_files:
  - split: train
    path:
    - yue.json.gz
- config_name: zgh
  data_files:
  - split: train
    path:
    - zgh.json.gz
- config_name: zlm
  data_files:
  - split: train
    path:
    - zlm.json.gz
- config_name: zsm
  data_files:
  - split: train
    path:
    - zsm.json.gz
- config_name: zza
  data_files:
  - split: train
    path:
    - zza.json.gz
---
# Paraphrases of Sentences from [Tatoeba](https://tatoeba.org/en)

Downloaded: 2024-07-08

**Description:**
- Large multilingual paraphrase dataset
- If two sentences translate to the same sentence in a foreign language, then those two sentences could be paraphrases.

**Creation Process:**
1. Translation pairs of sentences are first linked in a large undirected graph.
2. The connected components of the graph contain sentences with similar meanings.
3. Sentences in the same language in the same component are taken to be paraphrases.
4. Only clusters of 2-20 sentences are counted. Any larger cluster is likely to come from irrelevant sentences linked by bad translation.
5. Only languages with at least 10 clusters are included here.

**Format:**
- GZipped JSON files
- Each line contains one cluster of paraphrase sentences in JSON format

**Dataset Configs:**
- [ISO 639-3 language codes](https://en.wikipedia.org/wiki/List_of_ISO_639-3_codes) as listed on Tatoeba

**Sample entry:**

```json
{
  "id": 5,
  "sentences": [
    "Today is June 18th and it is Muiriel's birthday!",
    "It's 18th June today, and also Muiriel's birthday!",
    "Today is June 18th, the birthday of Muiriel!",
    "Today is June 18 and it is Muriel's birthday!"
  ]
}
```

`id`: The cluster ID of the sentences. Clusters with the same ID in other languages mutually translate to this set of sentences.

`sentences`: List of sentences belonging to this cluster. That is, those sentences are paraphrases of each other.

# Limitations

Many important things are lost in translation. Even if two sentences translate to the same thing in a foreign language, they might not be good paraphrases.

- Context and connotations ("What's up, Mike?", "What's wrong, Mike?")
- Gendered common nouns ("I need to get rid of it.", "I need to get rid of her.", "I need to get rid of him.")
- Ambiguous cases that could be the same or different.
  - "I'll shoot it down.", "I'll kill it with a gun."
  - "What kind of people do you prefer?", "What's your type?", "What kind of guys do you like?"

Also some languages have important features such as:

- [Familiar vs. formal forms](https://en.wikipedia.org/wiki/T%E2%80%93V_distinction)
- Verb tenses, modality, and moods (so a sentence pair can be paraphrases in one language but not in others)
- Vocabulary, noun distinctions, and cultural concepts (so foreign translations will lose the original meaning of the sentence)
- Idioms, dialects, slang, jargon