Balisage Paper: Multi-channel eBook production as a function of diverse target device capabilities

Leading eBook Formats

Survey of Devices

There are many different eBook reader devices available and seemingly more every day. These devices can be divided into 2 basic classes. The first is based on an E Ink electronic paper display. These displays can currently only show shades of gray (although research is being done on color E Ink), are viewable in direct sunlight, and require no power to maintain a static image. There is no back lighting, which reduces eye strain, but makes them unreadable in the dark. The Amazon Kindle and Sony Reader are examples of this class of reader. Many within this class of reader can play MP3 and other types of audio files. The second class uses backlit LCD screens and are often more than dedicated eBook reading devices. For this project 8 devices were included in the test set. These devices come from both classes and provide a mix of different features and capabilities.

Enhanced eBooks

The term “enhanced” eBooks has been used to mean many different things. For example, David Baldacci released his book “Deliver Us from Evil” as an enhanced eBook. In this case, in addition to the story, he included behind the scenes looks at how a book is written including audio, video, photos and an alternate ending, much like the special features found on many DVDs.

Essentially an enhanced eBook is a combination of media and book content that utilizes any mixture of text, audio, still images, animation, video, and interactivity content forms. The formats used to create a literary fiction eBook can include the addition of audio-visual elements and interactive contents allowing new form of creativity. The user (e.g., reader) might have an opportunity to participate in events occurring to characters, to feel influence of a musical part of a narration and graphic part.

Text Conversion (aka Overlap Happened)

The text conversion from OSIS XML to XHTML for inclusion in the eBook file appeared to be fairly straightforward at first glance. However, further analysis revealed that there was a minimal amount of structure contained within the file. Also, the entire Bible was contained within a single file. The snippet below shows many of the markup problems:

1  <p>
2	<verse eID="Luke.20.47"/>
3	<chapter eID="Luke.20"/>
4	<chapter sID="Luke.21" osisID="Luke.21"/>
5	<verse sID="Luke.21.1" osisID="Luke.21.1"/>He looked up, and saw the rich people 
6	  who were putting their gifts into the treasury.<verse eID="Luke.21.1"/>
7	<verse sID="Luke.21.2" osisID="Luke.21.2"/>He saw a certain poor widow casting in 
8	<milestone type="x-noteStartAnchor"/>two small brass coins.
9	<note type="translation">literally, “two lepta.” 2 lepta was about 1% of a day’s wages 
10	  for an agricultural laborer.</note><verse eID="Luke.21.2"/>
11	<verse sID="Luke.21.3" osisID="Luke.21.3"/>He said, 
12 	<q sID="Luke.21.3.149" who="Jesus" n="" type="x-doNotGeneratePunctuation"/>“Truly I 
13 	  tell you, this poor widow put in more than all of them,<verse eID="Luke.21.3"/>
14	<verse sID="Luke.21.4" osisID="Luke.21.4"/>for all these put in gifts for God from their 
15	  abundance, but she, out of her poverty, put in all that she had to live on.”
16	<q eID="Luke.21.3.149" n=""/>
17 </p>

Within this <p> element, we see <verse> and <chapter> elements. However, these tags are simply empty boundary markers, denoted by the sID (e.g. line 5) and eID (e.g. line 6) attributes. Notice the verse end (line 2) and the chapter end (line 3) and beginning (line 4) at the start of the <p>. The <milestone> element (line 8) marks the beginning of the anchor text, but the end of the text is identified by the <note> start tag (line 9). In many cases, <note> elements do not have <milestone> elements to identify any text. Also notice the <q> (quote) elements. The starting indicator (line 12) occurs in one “verse” and the ending indicator (line 16) occurs within a different “verse”.

These overlapping structures presented a challenge in converting the data to XHTML. It was desired that verse numbers be inserted into the text in order to allow readers to more easily navigate the text. Since the primary use of the data is for human reading, there was not a need to specifically delineate the verse boundaries, as this could be determined programmatically later, if needed. Also paragraphs would be kept intact as much as possible.

The process of converting the OSIS markup took several steps. First of all, an XSLT stylesheet was written that attempted to convert the OSIS markup into nested XHTML structure. The stylesheet processed the beginning and ending marker elements and converted them to character sequences within the output text. The stylesheet also created the separate files for each chapter. Since output from the stylesheet needed to be well-formed in order to be processed further by other stylesheets, angle brackets were replaced with "~~" to delineate markup. Some markup was also removed including the verse ends.

<p>~~ENDchapter~~~~chapter osisID="Luke.21" sID="Luke.21"~~<verse osisID="Luke.21.1" 
sID="Luke.21.1"/>He looked up, and saw the rich people who were putting their gifts into the 
treasury.<verse osisID="Luke.21.2" sID="Luke.21.2"/>He saw a certain poor widow casting in 
<milestone type="x-noteStartAnchor"/>two small brass coins.<note type="translation" 
xid="w1aab1b7b5d709c14"> literally, “two lepta.” 2 lepta was about 1% of a day’s wages for an 
agricultural laborer.</note><verse osisID="Luke.21.3" sID="Luke.21.3"/> He said, <q n="" 
sID="Luke.21.3.149" type="x-doNotGeneratePunctuation" who="Jesus"/>“Truly I tell you, this poor 
widow put in more than all of them, <verse osisID="Luke.21.4" sID="Luke.21.4"/>for all these 
put in gifts for God from their abundance, but she, out of her poverty, put in all that she had to 
live on.”<q eID="Luke.21.3.149" n=""/></p>

The data above shows the output of the original snippet after it had been transformed by the first XSLT stylesheet.

Next a series of sed commands was run on the file to convert the character sequences into XML tags. For example the "~~ENDchapter~~" sequence was changed to </chapter>. Other sed commands were used to arrange the tags into the correct nesting order in order to be well-formed (e.g. moving </chapter> tags before <p> tags. The sed tool was used as a quick and dirty method for manipulating the overlapping markup in order to prepare the data for further XML processing.

<chapter osisID="Luke.21" sID="Luke.21">
	<p>
		<verse osisID="Luke.21.1" sID="Luke.21.1"/>He looked up, and saw the rich people 
		  who were putting their gifts into the treasury.
		<verse osisID="Luke.21.2" sID="Luke.21.2"/>He saw a certain poor widow casting 
		  in <milestone type="x-noteStartAnchor"/>two small brass coins.
		<note type="translation" xid="w1aab1b7b5d709c14">literally, “two lepta.” 2 lepta 
		  was about 1% of a day’s wages for an agricultural laborer.</note>
		<verse osisID="Luke.21.3" sID="Luke.21.3"/>He said, 
		<q n="" sID="Luke.21.3.149" type="x-doNotGeneratePunctuation" who="Jesus"/>
		  “Truly I tell you, this poor widow put in more than all of them,
		<verse osisID="Luke.21.4" sID="Luke.21.4"/>for all these put in gifts for God 
		  from their abundance, but she, out of her poverty, put in all that she had to 
                  live on.”
		<q eID="Luke.21.3.149" n=""/>
	</p>
	...
</chapter>
			

The data above show the data after it has been provcessed by the sed batch commands. The <chapter> markup now surrounds the <p> and <verse> markup now occurs within the <p>. However, the overlapping <q> markup still exists.

Finally another XSLT script took the well-formed XML source file to prepare the files needed to build the EPUB collection. Spans with class attributes were also added into the markup to allow CSS formatting of the text. For example the span with the “jesus” class (formerly <q> markup) allows for red-letter marking of the words of Jesus on eBook readers that display color. Each book, chapter and verse were also given IDs to provide anchor points for linking (e.g. footnote returns). Since milestones do not occur consistently, it was decided to drop them and simply insert a marker indicating that a footnote is present. In cases where quotes span over several paragraphs or chapters, the quotes are closed at the end of the larger structure and restarted in the preceding one. For example, the quote was closed at the end of each verse and reopened after the next verse number markup had been inserted. This prevents the verse numbers from being displayed in red. The single XML was divided into individual files, each containing one chapter.

Two different methods were considered for footnotes. In the first method, footnotes would be displayed separately from the text, but be linked via a footnote marker. The footnotes were extracted from the source XML and stored in a separate file with bi-directional links from the source to the footnotes and from the footnotes back to the referring source. The markup shown below is an example of the final product of the source text and its associated footnote.

<div class="chapter">
	<span id="Luke.21"/>
	<h2 class="chapter-title">Chapter 21</h2>
	<p>
		<span id="Luke.21.1" class="verse">1</span>
	             He looked up, and saw the rich people who were putting their gifts into the 
	             treasury.
		<span id="Luke.21.2" class="verse">2</span>     
		He saw a certain poor widow casting in two small brass coins.
		<span id="fnref-w1aab1b7b5d709c14">
			<a href="../footnotes.xml#fn-w1aab1b7b5d709c14" class="footnote" 
			title="Footnote Luke.21.2">ƒ</a>
		</span>
		<span id="Luke.21.3" class="verse">3</span> 
		  He said, <span class="jesus">“Truly I tell you, this poor widow put in more 
		  than all of them,</span>
		<span id="Luke.21.4" class="verse">4</span>
		<span class="jesus">for all these put in gifts for God from their abundance, but 
		  she, out of her poverty, put in all that she had to live on.”</span>
	</p>
</div>

<div class="footnotes">
	<h3 class="chapter-title">Footnotes</h3>
	...
           <div id="fn-w1aab1b7b5d709c14">
		<p><a href="Luke/Luke.21.xml#Luke.21.2">Luke.21.2 ⇑ </a>literally, “two lepta.” 
		2 lepta was about 1% of a day’s wages for an agricultural laborer.</p>
	</div>
	...
</div>

One problem that was encountered in using this format was the tendency of human readers to hit the "previous page" button, thinking it was a back button. This would turn the previous page in the footnotes section of the book, rather than returning back to the original text.

In the second footnote scenario, embedded footnotes were used to keep the information together and reduce the need for jumping back and forth (and confusion about the use of the "previous page" functionality. In this case the footnote appears on screen with the text, but is differentiated by a different color and spacing.

<div class="chapter">
	<span id="Luke.21"/>
	<h2 class="chapter-title">Chapter 21</h2>
	<p class="text">
		<span id="Luke.21.1" class="verse">1</span>
	             He looked up, and saw the rich people who were putting their gifts into the 
	             treasury.
		<span id="Luke.21.2" class="verse">2</span>     
		He saw a certain poor widow casting in two small brass coins.
		<span class="notes"><span class="noteverse">2</span>literally, “two lepta.” 
		2 lepta was about 1% of a day’s wages for an agricultural laborer.
		</span>
		<span id="Luke.21.3" class="verse">3</span> 
		  He said, <span class="jesus">“Truly I tell you, this poor widow put in more 
		  than all of them,</span>
		<span id="Luke.21.4" class="verse">4</span>
		<span class="jesus">for all these put in gifts for God from their abundance, but 
		  she, out of her poverty, put in all that she had to live on.”</span>
	</p>
</div>

This method worked well on color displays, but required some extra styling in order to work well on grayscale devices.

In addition to the preparation of the source text, the final stylesheet also generated several metadata files needed to construct a valid EPUB file. This includes the OSF file which contains the Dublin Core metadata, the file manifest and the spine (which lists the files in the order in which they appear). An NCX file is also generated which is used by the eBook reader to present the table of contents. Some readers are able to present collapsible TOCs, so it is important to nest the entries in order to allow that functionality when possible.

Insertion of Illustrations

Artists through the ages have used the Bible as a source for subject matter. Since the source text contains no illustrations at all, a method was developed to insert graphic images at the correct places within the text. As stated previously, most of the source material (approximately 350 images) was found on a website of the LaVista Church of Christ. These images are mostly black and white which will display well on the E Ink screens of the devices.

An XML markup scheme was developed to allow an XSLT stylesheet to insert the graphic links at appropriate places. An example of the insertion markup is shown below as is a sample of the resulting image markup in the text. The stylesheet also added the graphics to the manifest information within the OSF file.

<links>
	<image verse="Luke 16.21" file="Luke.16.21.jpg" caption="Lazarus and the Rich Man"/>
	<image verse="Luke.22.61" file="Luke.22.61.jpg" caption="Peter's Denial"/>
	<image verse="Luke.23.5" file="Luke.23.5.jpg" caption="Jesus Before Pontius Pilate"/>
	<image verse="Luke.23.23" file="Luke.23.23.jpg" caption="Jesus Beaten"/>
</links>

<br/><span>
	<img src="../images/Luke.16.21.jpg" alt="Lazarus and the Rich Man"/>
</span><br/>

Creation of Hidden Text

In most eBook readers, when searching capability is provided, it only indexes the PCDATA text, not attribute values. In order to make navigation of the Bible file easier, it was suggested that hidden text be added at each book, chapter, and verse containing alternate search strings that would allow users to go directly to any anchor point. This hidden text would also allow alternate spellings and abbreviations to be used, making navigation even more intuitive. For example, a reference to Paul's second letter to Timothy, could be entered by a user as “2 Timothy”, "2nd Timothy”, “II Timothy”, “2 Tim”, “2nd Tim”, “II Tim”, etc. These different variations are added to each anchor point in a <span> element with an class attribute set to “hidden”. The CSS stylesheet then attempts to hide the text, while making it searchable. Our Luke example with the hidden text is shown below.

<div class="chapter">
	<span id="Luke.21"/>
	<h2 class="chapter-title">Chapter 21</h2>
	<p>
		<span id="Luke.21.1" class="verse">1</span>
		<span class="hidden">Luke 21:1</span> He looked up, and saw the rich people who 
		  were putting their gifts into the treasury.
		<span id="Luke.21.2" class="verse">2</span>     
		<span class="hidden">Luke 21:2</span> He saw a certain poor widow casting in two 
		  small brass coins.<span id="fnref-w1aab1b7b5d709c14">
		<span id="fnref-w1aab1b7b5d709c14">
			<a href="../footnotes.xml#fn-w1aab1b7b5d709c14" class="footnote" 
			title="Footnote Luke.21.2">ƒ</a>
		</span>
		<span id="Luke.21.3" class="verse">3</span> 
		<span class="hidden">Luke 21:3</span> He said, <span class="jesus">“Truly I 
		  tell you, this poor widow put in more than all of them,</span>
		<span id="Luke.21.4" class="verse">4</span>
		<span class="hidden">Luke 21:4</span>
		<span class="jesus">for all these put in gifts for God from their abundance, but 
		  she, out of her poverty, put in all that she had to live on.”</span>
	</p>
</div>

This intentionally introduced redundancy met with various degrees of success. Some devices didn't support the “display:none” directive in the CSS file and displayed the text. Some didn't display the text, but also didn't index it and others performed as had been hoped.

Insertion of External Links

The ability to link to outside materials is seen as a “must-have” for enhanced eBooks. In designing this functionality, it was already known that most of the first generation readers (nook, Kindle, Sony) would not be able to perform this task. Therefore, a new facet of the project was introduced: the ability to specify which features to include in an eBook and which to skip. More on that below in the sidetrack.

Geographical names occur throughout the Bible. Many places mentioned either no longer exist or are known by different names, making it difficult for a reader to know where an even might have occurred. The Bible Geocoding project uses information from several reference sources to build KML files for use in Google Maps and Google Earth. The snippet below shows the information contained within the KML files for two locations mentioned in the book of Luke.

<Placemark>
	<name>Abilene</name>
	<description>&lt;a href="http://www.gnpcb.org/esv/search/?q=Luke+3%3A1"&gt;Luke 3:1
	  &lt;/a&gt;</description>
	<styleUrl>#obi-normal</styleUrl>
	<Point>
		<coordinates>36.091710,33.587300</coordinates>
	</Point>
</Placemark>
<Placemark>
	<name>Arimathea</name>
	<description>&lt;a href="http://www.gnpcb.org/esv/search/?q=Luke+23%3A50"&gt;Luke 23:50
	  &lt;/a&gt;</description>
	<styleUrl>#obi-normal</styleUrl>
	<Point>
		<coordinates>35.180162,31.832739</coordinates>
	</Point>
</Placemark>

An XSLT stylesheet was developed that would convert the KML into a form that would be easier for the insertion mechanism to process. This stylesheet also reversed the order of the coordinates to work in the URL call to Google Maps. The XML scheme developed for the geographic points resembles that for the image insertion.

<links>
	<geo verse="Luke.3.1" point="33.587300,36.091710" name="Abilene"/>
	<geo verse="Luke.23.50" point="31.832739,35.180162" name="Arimathea"/>
</links>

The links to Google Maps are built on the place name within the flowing text. Another XSLT stylesheet uses the verse attributes shown above to locate the starting point from where to search in the text. Recall that the only thing marking the beginning of a verse is the <span class='verse'> markup. Once the appropriate starting point is located in the tree, each text node until the following verse marker is examined to determine if it contains the value in the name attribute. If so, an HTML <a> element is wrapped around the text pointed to a coordinate in Google Maps. When the link is activated, a web browser can be opened and the page displayed.

As expected, this functionality did not work in the first generation eBook readers since they have limited browser capabilities. However, in the others, there were various degrees of success. On the iPad, the user is asked if they would like to leave the iBooks app to open the browser. On the eDGe, the map opens on the LCD screen, as expected.

Media Links

Another hallmark of enhanced eBooks is the use of multimedia resources to improve the reading experience. Many people believe that EPUB does not support media, which is simply wrong. The EPUB supports inclusion of different media types and can select the appropriate player based on the mimtype given to a particular file. The limitation is most often with the reading devices themselves. The Kindle device does not include codecs for video, while the iPad cannot run Flash. Both of these can be embedded into an EPUB file..

There is some debate in the eBook community about the best way to include media files in EPUB files. There are essentially two options: 1) include the files in the EPUB file itself; 2) reference the resources via HTML <a> tags. Each option has its benefits and drawbacks.

The inclusion of the files in the EPUB file assures that the resource is available, whether the eBook reader is online or not. It also reduces the reliance on a web browser to retrieve the information. It is also faster since there is not delay while waiting to download. The chief drawback is the impact this has on the EPUB file itself. As mentioned earlier, the EPUB file is essentially a zip file with a specified set up internal files and a “.epub” file suffix. Most audio and video formats are already compressed, so therefore they do not compress significantly inside the EPUB file. This requires memory overhead to open the files inside the readers. It also requires longer download times for the books when they are first loaded onto the eBook reader.

Referencing files produces smaller EPUB files, but requires a mechanism (usually a web browser) to retrieve and present the resources. This requires that the eBook reader be online. There will also be a lag between the time the link is selected and the resource is downloaded. This is especially noticeable with large video files.