This package tokenizes (splits) words, sentences and graphemes, based on Unicode text segmentation (UAX #29), for Unicode version 15.0.0. Details and usage are in the respective packages:

uax29/words

uax29/sentences

uax29/graphemes

uax29/phrases

Why tokenize?

Any time our code operates on individual words, we are tokenizing. Often, we do it ad hoc, such as splitting on spaces, which gives inconsistent results. The Unicode standard is better: it is multi-lingual, and handles punctuation, special characters, etc.

Uses

The uax29 module has 4 tokenizers. In decreasing granularity: sentences → phrases → words → graphemes. Words is the most common use.

You might use this for inverted indexes, full-text search, TF-IDF, BM25, embeddings, etc. Anything that needs word boundaries.

If you're doing embeddings, the definition of “meaningful unit” will depend on your application. You might choose sentences, phrases, words, or a combination.

Conformance

We use the official Unicode test suites. Status:

Quick start

go get "github.com/clipperhouse/uax29/words"

import "github.com/clipperhouse/uax29/words"

text := []byte("Hello, 世界. Nice dog! 👍🐶")

segments := words.NewSegmenter(text)            // A segmenter is an iterator over the words

for segments.Next() {                           // Next() returns true until end of data or error
	fmt.Printf("%q\n", segments.Bytes())        // Do something with the current token
}

if segments.Err() != nil {                      // Check the error
	log.Fatal(segments.Err())
}

See also

jargon, a text pipelines package for CLI and Go, which consumes this package.

Prior art

blevesearch/segment

rivo/uniseg

Other language implementations

C# (also by me)

JavaScript

Rust

Java

Python