The standalone viewer

MAT's document component can be used, without the rest of MAT, to view, annotate, align, or compare annotated documents. You may want to embed this capability in your own Web tool or workflow. This document describes how to include the appropriate Web pages from the MAT distribution; how to generate a document which the MAT document component can understand; and how to enable hand annotation, comparison and alignment.

For instance, let's say you have a Java servlet which accesses a database in which your annotations are stored. You might use our Java library to produce a serialized document in the appropriate format, retrieve it in your Web application via AJAX, and pass this serialization to the embedded standalone document viewer. Alternatively, you could forward your annotations to your Web application in your own format, and use our JavaScript library to create the appropriate document.

An example can be found in $MAT_PKG_HOME/web/examples/standalone_viewer_example.html.

• Viewer setup
• Creating the viewer
• Creating a document
• Creating and displaying a single document
• Displaying aligned documents
• Displaying a document comparison
  
• Enabling hand annotation
  
• Details: the viewer object
• Details: the tag table

Viewer setup

The following includes are required from $MAT_PKG_HOME/web/htdocs:

• js/mat_utils.js
• js/mat_core.js
• js/core_ui.js
• js/mat_doc_display.js
• js/mat_standalone_doc_viewer.js
• css/mat_core.css

The Javascript files must be loaded in the order given. If you're running the MAT Web server with the default application prefix, you can load them in your document head as follows (where "yourserver" is the host where the MAT Web server is running):

<script type="text/javascript" src="http://yourserver:7801/MAT/js/mat_utils.js"></script>
<script type="text/javascript" src="http://yourserver:7801/MAT/js/mat_core.js"></script>
<script type="text/javascript" src="http://yourserver:7801/MAT/js/core_ui.js"></script>
<script type="text/javascript" src="http://yourserver:7801/MAT/js/mat_doc_display.js"></script>
<script type="text/javascript" src="http://yourserver:7801/MAT/js/mat_standalone_doc_viewer.js"></script>
<link rel="stylesheet" type="text/css" href="http://yourserver:7801/MAT/css/mat_core.css">

Alternatively, you can copy these six files to your own Web server.

No other files are required. Although the MAT UI relies heavily on the Yahoo! UI Toolkit, the standalone doc viewer does not, nor does it rely on any other toolkit.

Creating the viewer

In order to create the standalone viewer, you need, minimally, a tag set and a DIV where the viewer will be rendered.

<script type="text/javascript">
var Tags = [{label: "PERSON", css: "background-color: #CCFF66"},
            {label: "LOCATION", css: "background-color: #F99CC"},
            {label: "ORGANIZATION", css: "background-color: #99CCFF"}];

var docDiv = document.getElementById("docDiv");

// Set up the viewer.
var viewer = new MAT.DocDisplay.StandaloneViewer({
  tags: Tags,
  div: docDiv
});
</script>

There's a lot more you can do, which we'll return to when we look at the viewer details. For now, let's move on to creating documents.

Creating and displaying a single document

The document you provide should be a Javascript object which corresponds to the MAT JSON format, versions 1 or 2. For instance:

<script type="text/javascript">
var Document1 = {
  "signal": "David Asher is a former coordinator of North Korea policy at the State Department.",
  "metadata": {},
  "asets": [{"type": "ORGANIZATION", "attrs": [], "annots": [[65, 81]]}, 
            {"type": "LOCATION", "attrs": [], "annots": [[39, 50]]},
            {"type": "PERSON", "attrs": [], "annots": [[0, 11]]}]
};
</script>

We assume that you'd create these documents automatically somehow, by converting documents in your favorite format into MAT JSON format, or from your previous interactions with MAT. The document you pass can also be a JSON string in this format, rather than the resulting structure, or, alternatively, you can use the MAT JavaScript UI to create the appropriate document, e.g.:

<script type="text/javascript">
var Document1 = viewer.newDocument("David Asher is a former coordinator of North Korea policy at the State Department.");
Document1.createAnnotation("PERSON", 0, 11);
Document1.createAnnotation("LOCATION", 39, 50);
Document1.createAnnotation("ORGANIZATION", 65, 81);
</script>

To display this document, just call renderSingleDocument():

<script type="text/javascript">
viewer.renderSingleDocument(Document1);
</script>

The viewer will reset itself before it displays the document, so you can reuse the viewer as often as you want.

Displaying aligned documents

You can also display an alignment of multiple documents which all have the same document signal. Keep in mind that this mechanism only works for span annotations.

<script type="text/javascript">
viewer.renderDocumentAlignment([
 {doc: Document1, position: "above", initial: "1"}, 
 {doc: Document2, position: "above", initial: "2"}, 
 {doc: GoldDocument, position: "behind", initial: "G"}
]);
</script>

In this example, the argument to renderDocumentAlignment is a Javascript array of objects with the following keys and values:

The stacking order of the documents is determined by the order of the entries in the Javascript array: earlier elements are above later elements. There can be only one element with a position of "behind", obviously.

The viewer will reset itself before it displays the document alignment, so you can reuse the viewer as often as you want.

Displaying a document comparison

Unlike document alignments, a true document comparison works for both spanned and spanless annotations. These documents are created by the Python library, using the same pairing algorithm used by the scorer. So in order to generate such a document, you'll need to do some work in your own Web backend, either by capturing the output of the MATCreateComparisonDocument utility, or (if your Web backend is Python) by directly calling the libraries which that utility calls. Once you have such a document (either a JSON object or its corresponding string), you can create the comparison view as follows:

<script type="text/javascript">
viewer.renderDocumentComparison(doc);
</script>

or

<script type="text/javascript">
viewer.renderDocumentComparison(doc, {
  pivot: {
    docName: ...,
    initial: ...
  }, 
  others: [{
    position: ...,
    docName: ...,
    initial: ...
  }, ...]
});
</script>

In this second case, the second argument is a JavaScript object which describes the documents, and gives you more control over how the documents are referenced in the comparison view. The pivot is the reference document; the others are the documents which you're comparing to the reference document. The others array should be in the same order as the documents that you passed to the comparison creation document utility. The position and initial attributes have the same interpretation as they do for document alignments; the docName is a longer name by which you want the document to be described. Note that you can't change the position of the pivot; it is always behind the text.

Enabling hand annotation

If you want to do hand annotation in your document, or display annotation tables, you'll need to add a few more headers, and you'll need to provide access to the YAHOO! UI Toolkit, which we use to manage the annotation popups.

Add the following headers immediately before the load line for mat_standalone_doc_viewer.js:

<script type="text/javascript" src="http://yourserver:7801/MAT/resources/yui/build/yuiloader/yuiloader.js"></script>
<script type="text/javascript" src="http://yourserver:7801/MAT/js/workbench_ui_popup_mgr.js"></script>
<link href="http://yourserver:7801/MAT/css/workbench_ui_annotation.css" rel="stylesheet" type="text/css">

If you want to display annotation tables, you'll need one more header:

<script type="text/javascript" src="http://yourserver:7801/MAT/js/workbench_ui.js"></script>

Once you've done this, you need to do three more things.

First, we need to ensure that YUI is loaded before we create the viewer. Because of how browsers load documents, this is harder than it should be. You'll need to place the code which creates the viewer in an onload() function, and you'll also need to make use of the MAT.DocDisplay.StandaloneViewer.waitForYUI() function. Second, if you want to do hand annotation, create your viewer with the "annotatable" attribute set, and enable hand annotation using the enableHandAnnotation method. These steps are all illustrated here:

<html>
<head>
....
<script>
function onload() {
  MAT.DocDisplay.StandaloneViewer.waitForYUI(function () {
    var viewer = new MAT.DocDisplay.StandaloneViewer({
      tags: Tags,
      annotatable: true,
      div: docDiv
    });

    viewer.enableHandAnnotation();
  });
}
</script>
....
</head>
<body onload="onload()">
....
</body>
</html>

You can separate this last step, if, say, you want to control whether hand annotation is enabled via an external radio button.

Note: there is a bug in Firefox 9 where, if the first element of the document body is a block element whose default style has a top margin, such as <p>, the annotation popup will insert an invisible monitor element which erroneously blocks the collapse of the top margin of that first block element. You can get around this by enclosing the entire body of the document in a <div>.

Details: the viewer object

When you create a viewer, there are a lot of parameters you can set; by far, the most important is how you tell the viewer about the annotations you want it to recognize. There are three possible ways of doing this, with increasing levels of power and complexity:

• The "tags" parameter allows you to define simple spanned annotations. You can set attributes, but only in support of effective labels.
• The "atr" parameter allows you to define a simplified JSON annotation set descriptor, which is a list of complete annotation set descriptors, containing both spanned and spanless annotations, with as many attributes as you want, with as many types as you want. However, you're not able to specify whether to alphabetize the labels or not in the legend, or to specify a tag hierarchy for recursive menu display during annotation.
• The "taskATRFragment" parameter allows you to specify an expanded JSON annotation set descriptor, which supports everything you can specify with "atr", but also a number of display parameters which "atr" does not support.

Choose the parameter that best suits your needs, but choose only one. You can use the MATAnnotationInfoToJSON tool to generate the simplified or expanded annotation set descriptors for the "atr" or "taskATRFragment" parameters.

The viewer is instantiated with a single argument, a Javascript object with the following keys and values:

For instance, let's say that you have a tag table in the variable "Tags", as above; a div with the ID "docDiv", as above; a div with the ID "legDiv", where you want the legend rendered; and a div with the ID "updateDiv", where you want to display the annotation that's being hovered over. You'd instantiate the viewer as follows:

var viewer = new MAT.DocDisplay.StandaloneViewer({
  tags: Tags,
  taskName: "Named Entity",
  div: document.getElementByID("docDiv"),
  legendDiv: document.getElementByID("legDiv"),
  callbacks: {
    mouseOverAnnotations: function(contentObj) {
      document.getElementByID("updateDiv").innerHTML = contentObj.labels.join(", ") + (contentObj.suffix || "");
    },
    cancelMouseOverAnnotations: function() {
      document.getElementByID("updateDiv").innerHTML = "";
    }
  }
});

Viewer methods

Details: the tag table

The tag table is a list of annotations which will be recognized and displayed. It is a Javascript array of objects with the following keys and values:

If you choose to display an annotation legend (see below), the elements in the legend will appear in the same order you listed them in the tag table. Remember, the CSS will apply to every subspan of the marked text, and the text may be divided into subspans if annotations overlap, so using left and right borders is not recommended.