1.1.0 • Published 5 years ago

displacy v1.1.0

Weekly downloads
6
License
MIT
Repository
github
Last release
5 years ago

displaCy.js: An open-source NLP visualiser for the modern web

displaCy.js is a modern and service-independent visualisation library. We hope this makes it easy to compare different services, and explore your own in-house models. If you're using spaCy's syntactic parser, displaCy should be part of your regular workflow. Because spaCy's parser is statistical, it's often hard to predict how it will analyse a given sentence. Using displaCy, you can simply try and see. You can also share the page for discussion with your team, or save the SVG to use elsewhere. If you're developing your own model, you can run the service yourself — it's 100% open source.

To read more about displaCy.js, check out the blog post.

displaCy.js is written in ECMAScript 6. For full, cross-browser compatibility, make sure to use a compiler like Babel. For more info, see this compatibility table.

Using displacy.js

To use displaCy in your project, include displacy.js and initialize a new instance specifying the API and settings:

const displacy = new displaCy('http://localhost:8000', {
    container: '#displacy',
    format: 'spacy',
    distance: 300,
    offsetX: 100
});

Our service that produces the input data is open source, too. You can find it at spacy-services.

The following settings are available:

SettingDescriptionDefault
containerelement to draw displaCy in, can be any query selector#displacy
formatformat used to generate parse ('spacy' or 'google')'spacy'
defaultTexttext used if displaCy is run without text specified'Hello World.'
defaultModelmodel used if displaCy is run without model specified'en'
collapsePunctcollapse punctuationtrue
collapsePhrasecollapse phrasestrue
distancedistance between words in px300
offsetXspacing on left side of the SVG in px50
arrowSpacingspacing between arrows in px to avoid overlaps20
arrowWidthwidth of arrow head in px10
arrowStrokewidth of arc in px2
wordSpacingspacing between words and arcs in px50
fontfont face for all text'inherit'
colortext color, HEX, RGB or color names'#000000'
bgbackground color, HEX, RGB or color names'#ffffff'
onStartfunction to be executed on start of server requestfalse
onSuccesscallback function to be executed on successful server responsefalse
onErrorfunction to be executed if request failsfalse

Visualising a Parse

The parse() method renders a parse generated by spaCy as an SVG in the container.

displacy.parse('This is a sentence.', 'en', {
    collapsePunct: false,
    collapsePhrase: false,
    color: '#ffffff',
    bg: '#000000'
});

The visual settings specified here override the global settings. The available settings are collapsePunct, collapsePhrase, font, color and bg.

Rendering a Parse Manually

Alternatively, you can use render() to manually render a JSON-formatted set of arcs and words:

const parse = {
    arcs: [
        { dir: 'right', end: 1, label: 'npadvmod', start: 0 }
    ],
    words: [
        { tag: 'UH', text: 'Hello' },
        { tag: 'NNP', text: 'World.' }
    ]
};

displacy.render(parse, {
    color: '#ff0000'
});

The visual settings specified here override the global settings. The available settings are font, color and bg.

Converting output from other formats and adding your own

By default, displaCy expects spaCy's JSON output in the following style:

{
    "arcs": [
        { "dir": "left", "end": 4, "label": "nsubj", "start": 0 }
    ],

    "words": [
        { "tag": "NNS", "text": "Robots" }
    ]
}

If format is set to 'google', the API response is converted from Google's format. To add your own conversion rules, add a new case to handleConversion():

handleConversion(parse) {
    switch(this.format) {
        case 'spacy': return parse; break;
        case 'google': return({ words: ..., arcs: ... }); break;
        case 'your_format': return({ words: ..., arcs: ... }); break;
        default: return parse;
    }
}

You can now initialize displaCy with format: 'your_format'.

Changing the theme and colours

You can find the current theme settings in /assets/css/_displacy-theme.sass. All elements contained in the SVG output come with tags and data attributes and can be styled flexibly using CSS. By default, the currentColor of the element is used for colouring, meaning only need to change the color property in CSS.

The following classes are available:

Class nameDescription
.displacy-wordword text
.displacy-tagPOS tag
.displacy-tokencontainer of word and POS tag
.displacy-arcarrow arc (without label or arrow head)
.displacy-labelrelation type (arrow label)
.displacy-arrowheadarrow head
.displacy-arrowcontainer of arc, label and arrow head

Additionally, you can use these attributes as attribute selectors:

AttributeValueOn Element
data-tagPOS tag value.displacy-token, .displacy-word, .displacy-tag
data-labelrelation type value.displacy-arrow, .displacy-arc, .displacy-arrowhead, .displacy-label
data-dirdirection of arrow.displacy-arrow, .displacy-arc, .displacy-arrowhead, .displacy-label

Using a combination of those selectors and some basic CSS logic, you can create pretty powerful templates to style the elements based on their role and function in the parse. Here are a few examples:

/* Format all words in 12px Helvetica and grey */

.displacy-word {
    font: 12px Helvetica, Arial, sans-serif;
    color: grey;
}


/* Make all noun phrases (tags that start with "NN") green */

.displacy-tag[data-tag^="NN"] {
    color: green;
}


/* Make all right arrows red and hide their labels */

.displacy-arc[data-dir="right"],
.displacy-arrowhead[data-dir="right"] {
    color: red;
}

.displacy-label[data-dir="right"] {
    display: none;
}


/* Hide all tags for verbs (tags that start with "VB") that are NOT the base form ("VB") */

.displacy-tag[data-tag^="VB"]:not([data-tag="VB"]) {
    display: none;
}


/* Only display tags if word is hovered (with smooth transition) */

.displacy-tag {
    opacity: 0;
    transition: opacity 0.25s ease;
}

.displacy-word:hover + .displacy-tag {
    opacity: 1;
}

Adding custom data attributes

displaCy lets you define custom attributes via the JSON representation of the parse on both words and arcs:

"words": [
    {
        "tag": "NNS",
        "text": "Robots",
        "data": [
            [ "custom", "your value here" ],
            [ "example", "example text here" ]
        ]
    }
]

Custom attributes are added as data attributes prefixed with data-, so their names shouldn't contain spaces or special characters. If added to words, the data attributes are attached to the token (.displacy-token), if added to arcs, they're attached to the arrow (.displacy-arrow):

<text class="displacy-token" data-custom="your value here" data-example="example text here">...</text>