Alipi Project Summary

Contact: dinesh at servelots dot com Link: alipi.janastu.org This work was supported by isif.asia

The problem addressed by this project, is the issue of web-accessibility for the printimpaired in the Web 2.0 era. By print-impaired, we mean a large class of people who cannot read the web content due to illiteracy, partial literacy or language issues. It is especially important to address this segment now, since they are likely to attain physical access to the Internet via mobile devices in the near future. Our proposal was initially to look at W3C1 WAI-ARIA2 recommendations and look into recommendations on similar lines for print-impaired web-application users. As ARIA investigations started, we noticed that the print-impaired did not fit into the existing WAI categories of disabled people as they are not disabled physically. This lead to possibility of the development of a web-framework that would serve as a re-narration web. After basic prototyping work, we have developed a specification; a supporting tool set and several samples of demonstrative content that would indicate, for example, that various government web-pages can have the potential of reaching millions of people who are printimpaired. The initial prototyping work and investigations on WAI-ARIA took us 4 months. The next 10 months have been the process of experimentation, tool-set development and specification drafting. Towards the completion of this project time-line we have begun the process of a dialog with organizations working with domestic workers (mostly print-impaired), local Kannada language web-portals and the W3C to help evaluate and adapt Alipi tools and guidelines.
1. Project

objectives

The objective of the proposal is an investigation into approaches that help the printimpaired access the web-pages that are often dominated by text. In our initial project proposal, we had mentioned the following as our objectives: Understand the scope and extent of WAI-ARIA standards in the context where communities configure their own Web application. 2. Provide an open source accessibility standards-conforming Web 2.0 development platform for content and data management on the Web that reflects on experience of both visually handicapped and the rural semi-literate. 3. Simultaneously content provision -both for visually handicapped and illiterate who demand a visually rich navigation- during both the development of their application
1.
1

World Wide Web Consortium, www.w3c.org

2 Web Accessibility Initiative (WAI) - Accessible Rich Internet Applications (ARIA) http://www.w3.org/WAI/

intro/aria.php

and its deployment for others.

Project implementation Statement of Alipi

Alipi work as outlined here has three sections that describe the model and a section on implementation related work.

A1 Accessibility guidelines for the print-impaired

The W3C WAI recommendations discuss accessibility issues and address them by providing accessible design solutions such as a text equivalent for images and other multimedia content to make it accessible for visually-impaired users; or a non-text equivalent for text targeting a deaf audience for instance. However, the need of Alipi in addressing accessibility is different. Print-impaired users are people able to use their vision and their hearing capabilities, but have difficulties accessing written text. A non-disabled user navigates within the web page and understands its structure instantly by relying on image connotations or paragraph titles for example. It is however frustrating for a print-impaired user to use assisstive technologies such as content readers in order to understand the page structure: using an auditory description is not adapted to their needs since they can see and would rather rely on their vision than their hearing. Another barrier is the language. In fact, if the spoken/written language is not familiar to the user, it would not help them understand what is going on. Thus, the idea of Alipi accessibility guidelines is to allow a lay out a web page's content in a certain way that allows print-impaired users to understand its structure by indicating connectedness of fragments in a page.

A1.1 Associations
As mentioned above, Alipi indicates the structure of a web page and explicitly defines the relationship between fragments of its content elements. The screen-shot below, from the fire safety page (c.f http://a11y.in/a11ypi/idea), represents one example of how to make obvious these relationships. Here, as a user points to the fire image, a fragment of the page text that is associated with this image is highlighted. This association can be utilized by, for example, a text to speech tool such as a screen reader to read the text associated to the image. In the next section, we will see other ways in which these associations can further assist the case of print-impaired.

These relationships can be established by several HTML techniques. Here we indicate a simple technique called the Id-class method that we use to mark-up these associations.

Id-Class Method
The Id-Class (or HTML tagging) method consists of using simple HTML attributes to add some sort of meta-information to the content of a web page. It is a simple solution to relate objects in a page inspired from the CSS classes. A same style can be applied to various elements, possibly completely unrelated, by adding the same class attribute to these elements. And this is very similar to the projects goal. The only idea missing is to relate some elements to another element like an image or a title. To this end, the id attribute also natively present in HTML can be used. Following this method, a group of elements with the same class name can be related in a semantic way to a unique element in the page3. The following mark-up (see box below) shows that image of fire brigade that is identified by image1 id is also effectively labelled as asso0. Now notice that there is a fragment of text that is labelled as belonging to class called as asso0. This utilization of HTML id and class effectively indicates the associations that we need.

<body> <div id=''asso0'' class=''header''>

3

See Johanna Royers report on HTML tagging on http://alipi.janastu.org/

<img class=''fire_brigade'' id=''image1'' src=''images/images3.jpeg'' alt=''Firebrigade''/> </div> <div id=''div1'' class=''header asso0''> A fire department ... thereof. </div> </body> This association only says that elements are related. It does not, by itself, give any information about how the elements should be rendered in the browser. Thus, using the Id-Class, several items can relate to one object by using its Id. Withing the same page, this relationship is straightforward. To link resources on different pages, namespaces are to be utilized. In other words, the Id-Class association can extend beyond a page::
<div xmlns:ex="http://www.example.com/index.html" > <div class="ex:id_value"> text image audio or another </div>

</div> where id_value is the id of an element placed on the page referred to by the namespace. We note here that the URL protocol already accommodates pointing to fragments of a page using the #id notation. For example http://a11y.in/a11yPi/idea/firesafety.html#div0 refers to the element at http://a11y.in/a11yPi/idea/firesafety.html that can be identified by its id attribute having the value: div0. The associations also make it easier to deliver a web-page as a slide show or effectively as an audio/video rendition, on smaller screens such as those of mobile phones and tablets. For instance, the fire safety page mentioned above can be rendered by serializing the associated information fragments on a mobile device. A snap-shot of which can look like this:

A2 Re-Narration Web for Contextualization, Localization or Translation

Another aspect that is significant to print-impaired is the process, and the possibility, of the webcontent that can be localized by a community that is interested in narrating some web-page fragments to a locality of interest. For example, people from Karnakata in India, have only a few documents available in Kannada, their native language; and cannot access most web pages that are in English unless they are very comfortable with English. The solution to this issue provided by Alipi is re-narration: a given document can be rendered to address a certain community either by translating the whole document in another language or by making its content more understandable (e.g explaining what a certain law is about), or simply by rendering some of its components (e.g providing an audio description for a text or adding a connotation to a video). Alipi provides renarration tools and documentation at alipi.janastu.org. This document gives guidance on how to make a valid re-narration and explains how re-narrations would evolve within the Alipi community. Take the Fire Safety web-page example above. The following is a blog about this page:

In this post on Blog of Aravind, two fragments about fire safety are written about. However, in addition to typical blogging of content, Blog of Aravinds post intends to indicate certain semantic relationship to the Fire Safety page discussed above. In particular, that the image of the fire engine in this post is a recommended replacement image for people from Karnataka who visit the original site; and that the paragraph written in Kannada is a recommended replacement for people who prefer to read Kannada - rather than English. Alipi recommendations suggest that these intentions in Blog of Aravinds post be accommodated in a way that the browsers of tomorrow can process such meta-information. This meta-information, that some fragments of this post are intended as replacement suggestions for other fragments in other pages - and recommended for a target group, can be provided via HTML tag attributes (or Microdata, etc) . The HTML source of this page can look like:

Note the use of foruri attribute to point to the image in the original Fire Safety page and the rec attribute to indicate a recommendation or a target group. Narration Choice Now assume that there are other such posts out there on the Web that are possible localization of

fragments of this page for different contexts. These can be pulled up by a browser, possibly through a search engine service, and a relevant choice of alternative narrations can be offered to the person browsing the Fire Safety site. For example, when a person from India visits the Fire Safety site, a notification can look like:

The top bar in yellow is indicating that there are two choices of alternative narratives. One in Hindi language and another in Kannada language. If the user chooses the Kannada option, the page can render by using the recommended narratives for Kannada, which could look like:

Note that fire engine image and one of the paragraph in the page are substituted with the respective content from Blog of Aravinds post that was recommended for Kannada people. When there is alternative content in audio, the user can opt to listen to the audio or choose the default audio via a text-to-speech tool. Again, now a mobile device can render this on small screens:

Another accessibility issue might reside within the page's content itself. For instance, let's

consider the Ayodhya Verdict4, which is an 8132 page work only available in English. See
http://elegalix.allahabadhighcourt.in/elegalix/DisplayAyodhyaBenchLandingPage.do

This work is inaccessible because it is a technical document, very lengthy and hard to understand An effective solution to this accessibility deficiency would be a derivative to this document, created by a competent community in a way that it is understandable by an audience less skilled in jurisprudence. As an example, one possible re-narration of the Ayodhya Verdict would be a short summary of the verdict in a regional language made by a lawyer. This latter document would be more accessible because it is less technical, is shorter and is easier to understand. Example: http://www.scribd.com/doc/38464799/Justice-S-U-Khan and a news article, http://
www.ndtv.com/article/india/ayodhya-verdict-allahabad-high-court-says-divide-land-in-3-ways56063, on a popular news media web-site that one may subscribe to.

This real-world analogy of using media-intermediaries is the basis of Alipi's Re-narration concept. Also see filter described in the next section to see how subscriptions are modelled. These attributes are: Attribute target required Description indicates properties about the audience the re-narration document is targeting. The target attribute is either a single or several white space separated property:value couples. Some of the valid properties can be: lang: indicates the language of the targeted community. geoloc: indicates its location. occupation: indicates the profession occupied by the community; example: its a narration for doctors Indicates the link to the re-narrated element. about attribute in HTML can be used instead of a new foruri Indicates the type of the re-narration. It represents the L-relatedness between the original document and its renarration. Its default value is re-narration, which can be overridden to be more specific. Some possible values are: annotation, summary, abstract...etc. Indicates the version of the re-narration.

foruri

required

ren-type

optional

version

required

4http://en.wikipedia.org/wiki/Ayodhya_dispute

Please note that the actual version of the document is obtained by combining the foruri attribute with the version attribute. Indeed, two documents can carry the same version when they are in fact re-narrations for different elements. ren-date ren-author optional optional Indicates the date of the re-narration. Specifies the author of the re-narration.

The following example is a re-narration of an image designated by foruri, targeting the community living in India and speaking Kannada, that uses various attributes of re-narration. <img id="img1" src="images/fire_engine.jpeg" idea/firesafety.html#image0" target="lang:kn geoloc:India" ren-author=Alipi version=1.0 ren-date=21/2/2011> foruri="http://www.a11y.in/a11ypi/

Caveat: Alipi processes re-narrations by picking up elements via their Ids and re-narrating them. However, some pages are not well authored. A possible example would be an article carrying a single Id element, and therefore, selecting a particular paragraph to re-narrate means picking up the whole article. To address this deficiency, Alipi can use XPath5 when there is no straightforward way to select a single element. However XPath is likely to be less consistent compared to an inscribed Id.

A3 Mediated Filter Services

Filters are a way for a web page to declare certain re-narrators as authorized or recommended or favorites. A filter is an XML file that lists URLs for re-narrators along with meta-data about each URL (the re-narrator's identification, active subject categories, relative ranking, or FoaF details) so that each page would have its favorite re-narrators. The Alipi's Filter consists of XML tags. All URLs must follow the RFC-3986 standard for URIs, the RFC-3987 standard for IRIs, and the XML standard. A Filter: Begins with an opening <urlset> tag and end with a closing </urlset> tag. Specifies the namespace (protocol standard) within the <urlset> tag. Include an <author> entry for each URL, as a parent XML tag. Include a <url> child entry for each <author> parent tag.
5http://www.w3.org/TR/xpath/

All other tags are optional. A sample Filter schema: Attribute <urlset> required Description Encapsulates the file and references the current protocol standard. Tag associated with each URL entry. The remaining tags are children of this tag. URL of the page. The name of the author or the community of renarrators The priority of this re-narrator relative to other renarrators. Valid values range from 0.0 to 1.0. The default priority of an author is 0.5. Indicates the profession of the author (e.g sociologist, lawyer...etc)

<author>

required

<url> <name>

required optional

<priority>

optional

<profession>

<optional>

The following example shows a Filter in XML format: <?xml version="1.0" encoding="UTF-8"?> <urlset xmlns="http://alipi.janastu.com/Filters"> <author> <url>http://www.example.com/</url> <name>Alipi</name> <profession>Engineer</profession> <priority>0.8</priority> </author> <author> <url>http://www.another_example.com/</url> <name>Janastu</name> <profession>Teacher</profession> </author> </urlset>

Using Filters A page can provide multiple Filters. In order to do so, the ren-filter attribute embedded in the <header> tag is to be used. It specifies one or more white space separated filters for the page. The order (from left to right ) in which filters are provided indicates their rank. If no ren-filter value is provided, the Alipi's default filter is used. Example of a ren-filter attribute specifying recommended re-narrators: <header ren-filter= http://example.com/community1.xml http://another_example.com/community2.xml >...</header> In the example above, it is indicated that the authors belonging to community1 and community2 are recommended. Order in the listing can be indicative of a preference. Here, community1 is rather to be solicited than community2. A goal of Alipi is to enable localization and contextualization of laws and policy documents that concern the citizens of a country, such as India, so that these documents become available on the mobile phones of the many print-impaired people. Towards this, we have authoring guidelines that document authors can use. Then the re-narration model so an effective process can be initiated via the communities of interest or through those who have a mandate towards such activity. Filters help identify such communities of interest in certain context. For example, in the case of government documents that are put online, it may be natural for the authorities to announce the authorized re-narrators filter on their web-sites. This can become a directive to the Alipis narration recommendation algorithm using which only the official translations or localization/contextualization are provided as choices to a user. See the page titled ReNarrationAct on alipi.janastu.org web-site to follow our case study of re-narrating a document regarding the law related to minimum wages and using an Android based phone to demonstrate that the print-impaired community of domestic workers can now browse this document using the Alipi toolbar on the Firefox browser. This notion of Filters can also help bring to the Web, a parallel of print and news media organizations. For example, a bangalore.healthren.org can announce a Filter with list of favorites who they recommend as good re-narrators for health related web-pages for the locality of Bangalore, Karnataka. A user can subscribe to this Filter and choose the narratives from this list over others. Eventually, it can remain as an end-user choice in spite of these suggested or authorized Filters as user can prefer the narrations from a list of friends over that of the authorized or that of the subscriptions, esp., in certain cases.

Positioning of Alipi Accessibility

Let G be a directed graph where the nodes are documents that exist on the web. There is an edge from d1 to d2 with a label L, if d2 is related to d1 in the sense described by label L. Strictly

speaking, d2 and d1 could reference the same URI-accessible document, but d2 could be a transformation of d1. For example, d2 could be a re-rendering of d1 where d2 is WAI-accessible to someone with color-blindness, or d2 could be accessible to vision-impaired people. WAI concerns itself with generating relatedness, not with identifying relatedness, i.e. the standard effectively makes it possible to generate d2 given d1. This kind of relatedness is primarily presentational (and thus, implicitly semantically related in a somewhat obvious way). Alipi concerns itself with more generic semantic relatedness of documents, and also concerns itself with identifying relatedness as well as generating relatedness. i.e. given a document d1, it is interested in finding (either by identifying an existing one, or by generating one) a d2 that is related to d1 in the sense of L. This is a really hard problem to solve efficiently for different notions of L-relatedness. Given a document d1, how will the set of L-related documents be discovered? Will they be generated (ex: machine translation across languages)? Or will they be fetched based on existing semantic markup on d1? Or, will they be fetched based on existing semantic markup on d2's? Or, will a document repository (e.g.: the web) be crawled to identify the set of L-related documents? If so, given a candidate document d2, what metrics will be used to determine if d1 and d2 are sufficiently closely L-related? Clearly, different domains and applications will require different standards of L-relationship between d1 and d2. In light of the previous discussion, to avoid getting lost in an overgeneralized problem, Alipi focuses on a set of different projects in specific sub-domains where L is well-defined, and specifies a set of attributes which enable either the identification or generation of L-related documents.
Alipi is therefore a web-accessibility project with a difference: it allows users to re-narrate the Web, explaining web-content in ways that are more relevant and accessible for print-impaired and others who are lost in translation.

A4 An Index of On-going Work

Alipi model development was supported by prototype work that gave us the comfort of its feasibility over time. It also alerted us to various possibilities and hardships due to Web security issues in turning this prototype into a publicly available toolbar as a Firefox addon. The Alipi toolbar is supported by a web-service6 that simulates a web-search index that responds to a re-narrations query for re-narrations available for a web-site (an url). The alipi toolbar for Firefox and its avatar for an Android based mobile device are available as git-hub repositories from alipi.janastu.org A number of face-lifting, scaling and alternative implementations using intermediate services are being considered now so that the next phase of activity can be supported
6

By Venkatesh Choppella, through the virtual-labs server hardware.

by communities of interest who are enthusiastic about the possibilities of Alipi for their domain of work.

All work is available through http://alipi.janastu.org And other reports, experience notes and announcements will be updated on this site.

People who actively participated:

Arvind Khadri, Shalini, Ajay Kulkarni, Anand Bala, Megha, Prashanth, Navya, Johanna Royer, Shantanu Choudhary, Subbu Sastry, Sky Croeser, Yassine El Ghali, Suzan Uskudarli, Venkatesh Choppella, Rakendu, Amrapalli, Anuja Mirchandani, T B Dinesh, Anon and group of students, Maraa and Janastu support.