SADIe: Structural-Semantics for Accessibility and Device Independence

Currently, the World Wide Web is visual-centric, with web sites often being designed only with the presentation of data in mind. A consequence of this design perspective is that information contained within the data is only accessible implicitly through the layout of the web page, rather than explicitly through the data itself. For example, if you are a sighted user, you will be aware that on the left-hand-side of this webpage, there is a menu. You know this because there is a box, within which there is a column of text. The short phrases of text within this column change colour when the mouse is moved over them, indicating that the text links to other areas of the website. However, there is nothing stating that this is a menu. The knowledge telling you that it is a menu is implicitly tied to how the links are presented on screen. While this implicit knowledge is relatively easy to access for sighted users, it is often difficult to access for visually impaired computer users.

Transcoding is the adaptation of content from one format into another. There have been many previous attempts at transcoding web pages so that implicit information is more accessible to visually impaired users. These attempts can be roughly segregated into two groups:

1. Rule Based Transcoding: Systems that use rule based transcoding attempt to apply patterns or known statistics to web pages in order to extract the implicit information. For example, it may be assumed that if there is a column of links on the left-hand-side of the page, then that column of links is a menu. Whilst this rule works for most websites (indeed, it would work for the SADIe website which you are now reading), it is not necessarily always true. However, rule based transcoding systems do have the advantage that they can work with any website. One set of rules fits all websites and so you can attempt to transcode anything in the web.
2. Semantic Transcoding: Systems that use semantic transcoding attempt to capture the meaning of the elements of a webpage. This is usually achieved by annotating all the elements that occur within the page. For example, if there is a menu contained within the webpage, then a comment is added to the source code that states “this is a menu”. The transcoding system then has to look for the annotations and manipulate the content as it deems appropriate. The benefit of such systems is that they are incredible accurate. A menu is clearly labelled as a menu, so therefore it cannot be anything else (unless the annotator made a mistake). However, the big disadvantage of this is that it is time consuming, tedious and sometimes impossible to annotate every webpage on every website in order for it to be transcoded.

The SADIe project is an investigation into a hybrid solution to transcoding web pages in order to aid visually impaired users. Returning to the menu example above, we can surmise that a menu is a menu because it looks like a menu. Assuming that it is the presentation of the links that creates a menu, we must ask ourselves the question “what tells the browser to make a collection of links look like a menu?” One of the answers to this question is that the Cascading Style Sheet (CSS) associated with the webpage tells the browser to render the links as a menu. Therefore, we can assume that the rendering information encoded within the CSS of the webpage provides the implicit information that tells the user where the menu is on the webpage. Therefore, if we know what the elements of the CSS are rendering on screen, we can capture the knowledge and make it more explicit. This gives us the ability to accurately transcode a webpage in the same manner that annotations allow semantic transcoding to accurately transcode a webpage, but without the need for annotating the page itself. Instead, we just annotate the CSS file. However, it tends to be the case that a single CSS file describes the rendering of an entire website. Websites can span thousands of web pages, so by accurately describing a single file, we can potentially transcode thousands of web pages on the website.

SADIe therefore gives us the accuracy of semantic transcoding, without the time consuming tediousness of annotating every web page. Yet we also get the ability to transcode thousands of webpage in a similar fashion to rule-based transcoding, but with a higher degree of accuracy.