nlp-corpus

lots of weird english sentences

npm install nlp-corpus

_{by
Spencer Kelly}

see french, german, and spanish translations

nlp-corpus is a proud series of weird texts from a delicious smattering of sources - aimed at getting cosmopolitan flavours of english - _highbrow, lowbrow and unibrow_ - dialects, typos, shakespeare, unicode, 19th century, aggressive emoji, and epic nsfw slurs into your training data.

it is 50,000 sentences, or 5mb, split into 50 files of randomized sentences.

it's role is mainly to kick the tires a bit, as creatively as possible, for fuzzy linguistic parsing.

- suggestive American rock lyrics
- campy _Friends_ tv-show transcripts
- vulnerable drug-trip reports from Erowid
- singaporean SMS messages
- State of the union logorrhea
- generally-offensive 90's rap
- Legal descriptions in NAFTA
- 20th century romantic fiction
- pedantic arguments on reddit
- arcane and dense jeopardy questions

Note that some of this text is nsfw, or containing offensive content, badly-formatted unicode, weird indentation, ascii art, antiquated shorthands, etc.

These texts were found just clicking around on the internet. Running them blindly through your parser should be considered fair-use, but please don't commercially republish them, or anything like that.

ok go.

``bash npm install nlp-corpus`

running this library server-side loads a subset of the documents - abt 3mb total

`javascript import corpus from 'nlp-corpus'

// all 10k sentences, in an array let arr = corpus.all()

// or load just a few: arr = corpus.some(400)

//random sentence let str = corpus.random() //random 5 sentences let arr = corpus.some(5) //n can only be <= 1,500`

or on the client-side, there's a one-liner that fetches the docs:

`html``

ok go.

``bash npm install nlp-corpus`

running this library server-side loads a subset of the documents - abt 3mb total

`javascript import corpus from 'nlp-corpus'

// all 10k sentences, in an array let arr = corpus.all()

// or load just a few: arr = corpus.some(400)

//random sentence let str = corpus.random() //random 5 sentences let arr = corpus.some(5) //n can only be <= 1,500`

or on the client-side, there's a one-liner that fetches the docs:

`html``

nlp-corpus

ok go.

Contents:

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

nlp-corpus

ok go.

Contents:

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3

$3