Command Line Programs for the Blind

by Karl Dahlke

This article is available in French.

Introduction

Word processors, Internet browsers, and email clients typically present information in two dimensions, spreading text and icons across the screen. This interface is remarkably efficient thanks to the parallel processing capabilities of the retina and the visual cortex. In fact this screen to brain interface is so efficient, there is no need to count the bits as they fly by. Amateur webmasters and professional software developers routinely exploit this high band width channel, which is almost “too cheap to meter”. Web pages often present reams of irrelevant information in the form of extraneous links and rarely used buttons and icons. Carefully crafted applications are a little more organized, but they still employ toolbars and widgits that clutter the screen. This ancillary data is quickly tossed aside as your eyes focus on the item of interest. However, a blind user cannot assimilate this data at a glance, and separate the wheat from the chaff. He is forced to read every word on the page, using a voice synthesizer or braille display. These adapters are imperfect at best, as they ratchet the flow of information down to an agonizing crawl. By analogy, a high speed cable modem has been replaced with a dial-up connection, turning a rich multimedia experience into an exercise in frustration and futility. Is there a practical alternative? I believe there is, but certain critical applications must be rewritten from the ground up. To this end, I have developed a combination editor/browser called edbrowse, which is 100% text based. Output is measured and conserved like a precious commodity as it passes through the narrow channel of speech or braille. Edbrowse gives the blind user exactly what he asks for, and nothing more. Sighted users also find its unique features helpful in certain situations, which will be described below. This open source program is in the public domain, and is bundled in several Linux distributions. An early perl version (with limited functionality) runs on virtually any computer.

This has implications for software development in general. When a program or utility is “web based”, i.e. accessible through a browser, with some common sense restrictions on the underlying html and javascript, it is easily adapted to a wide range of disabilities. Each user employs the browser that caters to his particular needs and preferences. The interface is automatically tailored to the individual - no additional programming required. The semantics of the data, and its representation on the screen, or in speech, or on a braille display, have been neatly separated. Of course this approach doesn’t work for all applications, (a blind user isn’t going to play Flight Simulator), but it can be employed in many situations. Even a toolkit, such as Microsoft VB, might be enhanced to create interactive web pages, instead of using the screen and the mouse. If this project proves feasible, a wide variety of common VB applications could become accessible over night. The key is the accessible client, in this case a command line browser, combined with a suite of streaming applications that use these adapted clients as front end programs.

The benefits of this approach are not limited to the totally blind. A color-blind individual might use his browser to change the color of the background, text, headings, and hyperlinks to improve the contrast, while A user with low vision might increase the font size. Disabled users, and folks who simply prefer their text in a particular font, are hoping for a software revolution that sequesters functionality within the application, and leaves the details of the interface unspecified, to be determined by the wants and needs of the individual.

The Early Human-Machine Interface

Today’s computer professionals with laptops in hand can hardly imagine the awkward interfaces that prior generations had to endure. When I was a freshman at Michigan State University, I learned how to use a key punch machine and a card reader; that was how you “talked” to the computer. Each punch card represented one line of text, and a stack of several thousand cards might hold the source code for your master’s project. There was no file server on campus; you kept your box of cards with you at all times. And if you ever dropped your box of cards in the snow while walking from the dorm to the computer center in January, you were in a world of hurt. Even an accidental spill from table to floor was cause for consternation, as you carefully put your cards back in order. Apparently this scenario was not uncommon, prompting more than a few students to place sequence numbers on their cards in the rightmost columns - columns that were ignored by the computer. Some people left gaps in these numbers, e.g. counting by tens, so that additional lines of code could be inserted as necessary. My roommate, who did not bother to number his cards, always carried them in a sealed cardboard box. This led to more than a few jokes on my part. “Quit singing in the shower at 6 A.M., or i’ll scatter your box of cards all over the floor.”

Once a student arrived safely at the computer center with cards in hand, he might make some last minute corrections to his stack, then feed the cards, at a rate of ten per second, through the card reader. This created a “batch job”, which resulted in a printout some 20 minutes later. Wait times were highly variable, depending on load, which is why many students worked at night. Having placed the cards carefully back in his box, our weary student anxiously waits for the results of his labors. Did the program compile? Did it run? Was there an error in logic? Can he turn his printout in for a grade, or is there more work to do? If anything has to be changed, he goes back to the punch machine, hammers out new cards, slides them into position, and walks back to the card reader for another run. The smallest typo represents another hour’s work. It was not unusual to see bleary-eyed students stumbling back into the dorm at 2 or 3 in the morning, card box under one arm and a stack of printouts under the other.

Imagine my joy when the University installed interactive teletypes! These look like electric typewriters, but the keystrokes are transmitted directly to the central computer. When you hit return, the computer responds, then waits for your next command. The interface had become a dialog, which clipped along at 110 bits per second, or 300 bits per second if you glommed onto one of the newer teletypes. The paper retained a written transcript of the entire session; your commands and the computer’s responses. These machines are all but forgotten, except for the vestigial letters tty, which are an abbreviation for teletype. The software that facilitates communication between you and your computer, through the keyboard and screen, is still called a tty today. Type `tty’ into any Unix or Linux computer, and it will tell you which tty driver you are using, e.g. /dev/tty1 on console 1. A large Unix machine can have hundreds of tty drivers, supporting hundreds of simultaneous users.

The clatter of the teletype was an annoyance for most, but it was a blessing for me. I knew when the computer had responded, and the nature of that response. If a volunteer reader was not available, and the homework assignment was modest, I could log onto the system, type my program into the editor, compile the program, and run the executable, based solely on the clicks of the teletype. After my roommate read through the printout and verified the results, I tucked it away in my notebook and turned it in the next day for a grade. Although I now have a speech synthesizer at hand, I still miss the audio feedback that was an unintentional feature of the mechanical teletype. To this end, I modified the Linux tty driver to create similar sounds using the PC speaker. When the computer sends text to the screen, soft clicks accompany the nonspace characters, while a longer swoop indicates a new line, as though the print head was swinging back to the left. These modules are available from the drivers directory in the acsint project.

The chirps and clicks are subtle, and are easily ignored by those around me; yet they form an important part of my audio interface. Like the systems of yesteryear, my tty tells me when the computer has responded to my commands, and the quantity and format of that response, even before my synthesizer has spoken a word.

The acsint project also contains a linear speech adapter, the only one of its kind. Like the paper teletype, this adapter retains a log of all tty output, and allows the user to review that log, reading the entered commands and the computers responses. All other adapters, whether on MS-dos, Windows, Mac-OS, or Linux, read the words or icons on the screen. My adapter can read screen memory as well, but it typically runs in linear mode, which is optimal for the command line interface. This adapter, and various applications such as edbrowse, all work together to present a new (i.e. old) paradigm, a paper teletype inside your computer.

Over the next few years, universities around the country replaced their paper teletypes with cathode ray tube terminals, also known as CRTs. Trees everywhere heaved a sigh of relief; yet the interface was still the same. A user types a command, and the computer responds on the next line. The dialog continues.

Early Editors and Word Processors

When I began my career at AT&T Bell Labs, I learned to use the standard Unix text editor, which is simply called ed. This is a command line program, consistent with the interface of its day. If you want to see the seventh line in a file, type 7 and hit return. To find the line containing the string “xyz”, type /xyz/. Lines can be changed, deleted, moved, or copied using this program, but you only see one line at a time. Furthermore, ed is merely a text editor; it is not a word processor. Tools like troff and nroff were developed to turn text into a formatted page. For instance, .PP indicates a new paragraph, .SH indicates a section heading, and so on. Online manual pages are still written in this markup language today. Type “man ls’ into a Linux machine, and nroff is running in the background.

Although the combination of ed and nroff was primitive by today’s standards, it was perfect for me. I used ed to create documents, inserting formatting tags where appropriate, and the resulting pages were comparable to those created by my sighted colleagues who were forced to use the same text based tools.

Needless to say, this was not tolerated for long. Screen editors such as vi and emacs quickly appeared, followed by word processors such as Word Perfect and MS Word. For the first time, you could see what your document would look like before sending it to the printer. Once again, trees around the world were granted a reprieve, and everyone who touched a computer became more productive over night - everyone but me. Yes, screen readers allowed me to roam about and read the text, but I was still processing the data linearly. The benefits of a two dimensional search and scan were not available to me, and to think otherwise is to live in a state of denial. So I continued to use ed and nroff to create programs and documents for Bell Labs. I even ported ed to the IBM PC, thus giving me the same linear interface at work and at home.

Accessing the Internet

Fast forward 15 years, and the Internet consists of web pages that are text based, with embedded formatting directives that are similar to nroff. For instance, <P> tells your browser to start a new paragraph, while <H2> represents a new section heading at level 2. These are reminiscent of the .PP and .SH directives that came before. Thus, Hyper Text Markup Language (HTML), the language of the Internet, is an evolutionary descendent of nroff. Having learned most of the html codes, I was able to develop some 1,400 web pages for MathReference.com, using nothing but a text editor. If I want my web page to display Goodnight Moon in italics, then Margaret Wise Brown in a smaller font on the next line, I will enter the following commands.

<I>Goodnight Moon</I> <br><font size=-1>by Margaret Wise Brown</font>

Once again I am in the minority. Most web developers use graphical design tools such as MS Front Page or Dreamweaver, which hide the technical details of html. The interface is similar to a word processor. Arrange the text and pictures on the screen as you would like them to appear on your website, and the tool generates the appropriate html. This works well for others, but for me, the benefits of a two dimensional representation are rendered academic, as my speech adapter roams around the screen, trying to make sense of the page as a whole. It’s like looking at the world through a straw.

Although I could write web pages using ed and a few basic html commands, I was still unable to surf the net quickly and efficiently. My text editor allowed me to create a website from scratch, but there was no command line browser to help me read websites that were written by others. The closest approximation was a program called lynx, which does not employ graphical icons, and can even be run without a mouse. Indeed, many blind people still use lynx today. However, lynx remains a screen oriented application, presenting information across 25 rows and 80 columns. I was hoping for a command line interface similar to ed.

In 2001, I began writing a program called edbrowse, a combination editor browser, whose interface was fashioned after ed. This has all the features of ed, along with some new commands, such as `b’ to browse an html file, and `g’ to go to a hyperlink referenced by that web page. One can “edit” www.ibm.com as easily as one might edit a local file. Of course you cannot meaningfully change the contents of www.ibm.com, since it resides on another computer, but you can format it using the browse command, then step through the text line by line, or search for a word or phrase using the ed commands you already know. To find the next hyperlink, search for the left brace, as this indicates a link to another web page. Similarly, one can step through the fields in a fill-out form by searching for the less than sign. With practice, it is surprisingly easy to navigate through most web pages and find the information you want. Compared to other browsers, edbrowse demands more input, in the form of entered commands, and generates less output; which is precisely the paradigm for a one dimensional channel such as speech or braille.

System Administration

We recently purchased a Mac, and after I learned how to log in remotely from my Linux machine, I was able to perform all sorts of printer administration tasks, including adding and configuring a new printer, printing test pages, setting up print queues with priorities, and monitoring print jobs. This is normally accomplished through a graphical printer utility on the screen, but I found another way. The print utility is web based, so anyone with a browser can access the printers via http://localhost:631. (If you have a Mac, type this into your browser and you’ll see what I mean.) An important subsystem of the computer with a moderate level of complexity has been rendered accessible to a wide range of disabled users, thanks to the power of the Internet. Fire up your favorite browser, be it edbrowse or lynx or firefox, and manage your print jobs through an interface that has been optimized for your particular needs.

Another example of web based system administration is samba file sharing, which is accessible through http://localhost:901 on some computers. Turning to network administration, most off-the-shelf routers can now be configured through html. I hope this is the beginning of a new trend in system administration. Accounts and passwords, networking, firewalls, disk utilities, and the task manager are just a few examples of real world applications that can and should be web based. If the resulting web pages are relatively simple in their content and format, computers would become more accessible, almost over night. Most people would access these functions through the default graphical browser that is shipped with the computer, and they wouldn’t know the difference. At the same time, I would take advantage of edbrowse, which was written specifically for my needs.

Beyond this, web based administration makes it easy to configure the computer remotely. If the firewall permits, I could access the printers on your box by typing http://yourbox:631 into my browser. There is no need to log in remotely and run edbrowse on your computer, which may not be practical in any case.

Other Applications

Some applications cannot be operated effectively through a browser, and edbrowse does not claim to be the universal solution to accessibility. Consider Microsoft Excel, a well known spreadsheet application. Imagine a conversion utility that transforms an xls file to an interactive web page, ready for manipulation by your favorite browser. Each cell becomes an input field in a fill-out form, and the cells are arranged in rows and columns inside an html table. As data is modified, javascript formulas compute totals and averages, as Excel normally does. This works well for small data sets, yet some spreadsheets contain hundreds of columns and thousands of rows. Your eyes easily drift down column 79, looking for trends in the data, but edbrowse cannot do this. Long lines are tolerated, but there is no convenient way to jump to column 79. Nor can you easily locate the “payment” column for the row associated with customer “Mary Smith”. Either edbrowse must be augmented with new spreadsheet features, or a linear version of Excel must be developed, with novel commands designed to locate and modify individual cells.

This holds true for any program where eye movements (which lie outside the purview of the screen interface) must be converted into commands and responses, entirely new pathways for the linear version of the same application. Think of the application as a conversation between the program and the user. When the user ignores 95% of what the program “says”, and selects the relevant 5% by moving his eyes, that conversation must change in fundamental ways to be blind-friendly. Most screen programs implement this type of mega-output conversation; that is in fact the screen paradigm. For this reason, screen programs with high data rates must be redesigned from the ground up to run efficiently in text mode. At the same time, simpler programs can often be restructured to generate html or xml, which gives the user control over his interface through specialized clients.

In the Background

Although edbrowse was designed to make the Internet accessible to the disabled community, it has some unique features which have attracted the attention of Unix administrators and developers. Most browsers assume the presence of a human operator, to read the screen and click the mouse. In contrast, edbrowse can simply read a set of commands from a file. In addition, edbrowse includes a scripting language, with conditional logic, loops, and functions. A background cron job, run every morning by the computer, can review the contents of several web pages, and send mail to the operator if certain conditions are met. In other words, edbrowse is well suited to batch processing, without human intervention. In theory, the same tasks could be accomplished by a shell script, calling upon wget and grep and the like, but a self-contained edbrowse script is easier to write and maintain. This is why grml, a distribution aimed at system administrators, was the first to include edbrowse in its standard release. Others, such as debian, and even Free BSD, now offer edbrowse in their distributions.

Summary

Although edbrowse has become a useful tool for batch jobs on the Internet, the main focus continues to be accessibility. Pasting a screen reader on top of Chrome or Edge is a good beginning, but it hardly creates a level playing field. The disabled community deserves better. The editor, the browser, the mail client, and the spreadsheet can be made more efficient for the blind, blind-deaf, and motor impaired, if we would just take the time to redesign them with the user’s needs in mind. Once these critical applications are transformed, other utilities can be modified to leverage these clients, as illustrated by an emerging suite of web based tools and services. These efforts should be funded at the federal level, but sadly, they have received little support from government or industry to date. When everyone has access to computers, laptops, and the internet, we all benefit. Sir Bert Massie, Chairman of the Disability Rights Commission, said it best. “The overall vision must be one of a society in which everyone (disabled and non-disabled alike) can flourish and participate fully as equal citizens.”