Home

Awesome

sentence-splitter

Split {Japanese, English} text into sentences.

What is sentence?

This library split next text into 3 sentences.

We are talking about pens.
He said "This is a pen. I like it".
I could relate to that statement.

Result is:

Sentence Image

You can check actual AST in online playground.

Second sentence includes "This is a pen. I like it", but this library can not split it into new sentence. The reason is "..." and 「...」 text is ambiguous as a sentence or a proper noun. Also, HTML does not have suitable semantics for conversation.

As a result, The second line will be one sentence, but sentence-splitter add a contexts info to the sentence node.

{
    "type": "Sentence",
    "children": [
      {
        "type": "Str",
        "value": "He said \"This is a pen. I like it\""
      },
      ...
    ],
    "contexts": [
        {
            "type": "PairMark",
            "pairMark": {
                "key": "double quote",
                "start": "\"",
                "end": "\""
            },
            "range": [
                8,
                33
            ],
            ...
        ]
    ]
}

Probably, textlint rule should handle the "..." and 「...」 text after parsing sentences by sentence-splitter.

Installation

npm install sentence-splitter

Usage

export interface SeparatorParserOptions {
    /**
     * Recognize each characters as separator
     * Example [".", "!", "?"]
     */
    separatorCharacters?: string[]
}

export interface AbbrMarkerOptions {
    language?: Language;
}

export interface splitOptions {
    /**
     * Separator & AbbrMarker options
     */
    SeparatorParser?: SeparatorParserOptions;
    AbbrMarker?: AbbrMarkerOptions;
}

/**
 * split `text` into Sentence nodes.
 * This function return array of Sentence nodes.
 */
export declare function split(text: string, options?: splitOptions): TxtParentNodeWithSentenceNode["children"];

/**
 * Convert Paragraph Node to Paragraph Node that includes Sentence Node.
 * Paragraph Node is defined in textlint's TxtAST.
 * See https://github.com/textlint/textlint/blob/master/docs/txtnode.md
 */
export declare function splitAST(paragraphNode: TxtParentNode, options?: splitOptions): TxtParentNodeWithSentenceNode;

See also TxtAST.

Example

Node

This node is based on TxtAST.

Node's type

Get these SentenceSplitterSyntax constants value from the module:

import { SentenceSplitterSyntax } from "sentence-splitter";

console.log(SentenceSplitterSyntax.Sentence);// "Sentence"

Node's interface

export type SentencePairMarkContext = {
  type: "PairMark";
  range: readonly [startIndex: number, endIndex: number];
  loc: {
    start: {
      line: number;
      column: number;
    };
    end: {
      line: number;
      column: number;
    };
  };
};
export type TxtSentenceNode = Omit<TxtParentNode, "type"> & {
    readonly type: "Sentence";
    readonly contexts?: TxtPairMarkNode[];
};
export type TxtWhiteSpaceNode = Omit<TxtTextNode, "type"> & {
    readonly type: "WhiteSpace";
};
export type TxtPunctuationNode = Omit<TxtTextNode, "type"> & {
    readonly type: "Punctuation";
};

Fore more details, Please see TxtAST.

Node layout

Node layout image.

This is 1st sentence. This is 2nd sentence.

<Sentence>
    <Str />                      |This is 1st sentence|
    <Punctuation />              |.|
</Sentence>
<WhiteSpace />                   | |
<Sentence>
    <Str />                      |This is 2nd sentence|
    <Punctuation />              |.|
</Sentence>

Note: This library will not split Str into Str and WhiteSpace(tokenize) Because, Tokenize need to implement language specific context.

For textlint rule

You can use splitAST for textlint rule. splitAST function can preserve original AST's position unlike split function.

import { splitAST, SentenceSplitterSyntax } from "sentence-splitter";

export default function(context, options = {}) {
    const { Syntax, RuleError, report, getSource } = context;
    return {
        [Syntax.Paragraph](node) {
            const parsedNode = splitAST(node);
            const sentenceNodes = parsedNode.children.filter(childNode => childNode.type === SentenceSplitterSyntax.Sentence);
            console.log(sentenceNodes); // => Sentence nodes
        }
    }
}

Examples

Reference

This library use "Golden Rule" test of pragmatic_segmenter for testing.

Related Libraries

Tests

Run tests:

npm test

Create input.json from _input.md

npm run createInputJson

Update snapshots(output.json):

npm run updateSnapshot

Adding snapshot testcase

  1. Create test/fixtures/<test-case-name>/ directory
  2. Put test/fixtures/<test-case-name>/_input.md with testing content
  3. Run npm run updateSnapshot
  4. Check the test/fixtures/<test-case-name>/output.json
  5. If it is ok, commit it

Contributing

  1. Fork it!
  2. Create your feature branch: git checkout -b my-new-feature
  3. Commit your changes: git commit -am 'Add some feature'
  4. Push to the branch: git push origin my-new-feature
  5. Submit a pull request :D

License

MIT