音声ブラウザご使用の方向け: SKIP NAVI GOTO NAVI

Web Posted on: May 18, 1998

DUELING SCANNERS

Peter Scialli
Shrinkwrap Computer Products
11706 Saddle Crescent Circle
Oakton, VA 22124
Phone: (703)620-4642
Fax: (703)620-4641
Email: shrink@erols.com
Website: http://www.erols.com/shrink

The first annual Dueling Scanners session was held on March 17, 1998 in conjunction with the C-SUN "Technology and People with Disabilities" conference in Los Angeles, California. Vendors of reading software, producing specialized products for blind and print-disabled individuals were invited to compete at the session.

All the products considered for participation in the session are specially developed solutions that are generally operated as "stand-alone" products for people with print related disabilities. These products convert printed reading material into synthetic speech, large print or Braille. While some individuals are able to make adaptations of commercially available optical character recognition (OCR) products in combination with alternate methods of accessing the computer screen, these approaches are varied and specific to the computer preferences and skills of the individual using such an approach.

The intent of Dueling Scanners is to present the available solutions that provide a standardized approach to accessing printed material.

The Manufacturers invited to display their products at Dueling Scanners included Arkenstone, Inc, producer of Open Book Unbound version 3.5, Kurzweil Educational Systems, producers of the Kurzweil 1000 version 3.0, Robotron, manufacturer of the Galileo reading machine and Telesensory, Inc, the manufacturer of Reading AdvantEdge version 2 software. The event was attended by all of these except Telesensory who declined the invitation to participate. Arkenstone was represented by Jim Fruchterman, President of the company and Mike May, Vice-President in charge of Sales. Kurzweil was represented by David Bradburn, Marketing Manager and Stephen Baum, Vice-President of Research and Development. Robotron was represented by John Panarese, Manager of Technology for the Visually Impaired, North American Distributor for Robotron.

The guidelines and question pool, as presented in advance to the vendors appears below. These are followed by the judge's reactions to what they witnessed. It is important to note that in examining the quality of scanned output, a scientific determination of OCR accuracy was not made. Such studies require a great deal of time and statistical resources that were not available in this setting. Since all the current scanning products for the print-handicapped use components from mainstream character recognition solutions, it was felt that data on their raw scientific accuracy could be obtained from other sources and that "viewing" documents for their general readability and

utility once scanned would be of more benefit to this audience. The anticipated audience for Dueling Scanners comprises rehabilitation professionals and other educated consumers of special technology for the blind and print-handicapped. These groups will usually have a good familiarity with adaptive computer equipment, but will not generally be experts in the intricacies of such equipment.

The following twelve guidelines were given to the participants approximately one month prior to the session:

1. Each vendor should bring to the session only one reading product. It should be the "flag ship" system--that is, a complete reading system that the vendor feels best represents the available purchasable product line. If optional equipment is available for the system, that should be brought also. It will not be necessary to bring video displays for systems with a text display option; a projection monitor will be supplied by the Center on Disability.

2. The reading system brought for display and demonstration at Dueling Scanners must be a current model that is readily available to the public that day for purchase. It may not be a prerelease of a pending product nor may it differ in any other way from an actual system that an end user could purchase on the day of the event.

3. Each vendor should bring one example of a document which it feels will exemplify the virtues of its particular reading system. While each document will be tried on every system for comparison, it is important for vendors to focus upon the virtues of their own system rather than attempting to demonstrate the perceived shortcomings of their competition.

4. The judges will supply documents which they feel will challenge the ability of each system to produce readable text. Each system will receive a rating, devised by the judges, on overall OCR accuracy, decolumnization accuracy and general readability. The judges will be free to ask specific questions regarding the nature of the work done by each system.

5. Each vendor should explain with words and/or demonstration how, using their reading system, a novice user of computer technology can best learn to scan and do basic reading of documents. Training requirements should be discussed.

6. Each vendor should explain with words and demonstration how a document may be stored for later retrieval once it has been scanned.

7. Each vendor should explain with words and/or demonstration how a scanned document can be converted into a format that is usable by a variety of word processing software.

8. Each vendor should describe and/or demonstrate how its particular reading system can be integrated into the user's software environment--assuming that the system is running on a PC where other software is run. Information should be provided as to whether or not the reading system allows the user to define different programs for document text viewing/editing.

9. The vendor should discuss the facilities in its reading system which make it possible for documents to be scanned using network-aware equipment (e.g., an HP5SI network scanner).

10. Each vendor should explain with words and/or demonstration the strategies that might be employed to convert a document, scanned by its reading system, into Grade II Braille and how their product lends itself to such a strategy.

11. Each vendor should explain with words and/or demonstration how its product addresses the needs of people with disabilities in addition to blindness. These may include but are not limited to low vision, deaf-blindness. limited hand motion, poor dexterity, etc.

12. Each vendor should explain with words and/or demonstration if it is possible for their system to be used by people who need to read text in more than one language.

The Results

The Dueling Scanners judges volunteered their participation in the event. The judges were Curtiss Chong, Director of Technology for the National Federation of the Blind, Janina Sajka, Director of Technical Services for the American Foundation for the Blind and Brian Charlson, a Rehabilitation instructor for the Carroll Center for the Blind and First Vice-President of the American Council of the Blind. Each judge is an acknowledged expert in the use of adaptive computer technology by and for blind persons.

There were two sessions of Dueling Scanners: one in the morning and one in the afternoon. This permitted two separate audiences to view the event and gave the judges two opportunities to form opinions about the products. It is important to note that Robotron participated only in the morning session. Their equipment was damaged in transit and could thus not be properly judged on an even basis with the other systems. However, it was possible to see some of the important features of the Galileo reading machine and these features will be noted where relevant. It is hoped that Robotron will be able to participate in Dueling Scanners in the future as their product clearly merits attention.

Beginning with the third guideline, the following was observed:

Kurzweil brought a document that was a very worn paperback book. It included standard type with the addition of some unusual fonts and some printing above and below the lines. Kurzweil wanted to demonstrate their two-page mode in which they can set the software to automatically divide each scanned image into two separate scanned pages. This mode should be helpful in maintaining the integrity of books with complex layouts such as those containing a mixture of text and tables. The software read the document with excellent accuracy and proper presentation of the material.

Arkenstone's Open Book had some initial trouble with this document: The system had excellent OCR quality, but did not properly decolumnize the two facing pages. It mixed some fragments of the second page in with the first page. Jim Fruchterman gave a very forthright explanation of this problem. In the case of a book where two facing pages are severely skewed relative to each other, the Arkenstone decolumnization facility may get "confused." In fact, the document in question was a paperback book with a severely damaged binding. Mr. Fruchterman was able to reposition the book on the scanner such that the two facing pages were more normally positioned relative to each other. After doing this, the result on the Arkenstone system was also excellent in terms of accuracy and presentation.

John Panarese representing Robotron next produced an advertising brochure that had a complex column layout, graphic content and multiple fonts. All three systems were able to properly read and decolumnize this document.

Finally, Arkenstone produced a page from a magazine with complex columns and a graph included on the page beneath one of the columns. All three systems slowed considerably in the processing of this document. All three properly recognized and decolumnized this document. The Kurzweil system was faster in producing a result than the Arkenstone system. The Arkenstone system attempted to read some text from within the graph while Kurzweil did not. The Robotron system was quite slow in the processing of the document, but did a good job once finished. The processing time of the damaged Robotron machine should not be a factor in these results, however.

Next, each of the judges produced a document of interest. Mr. Chong presented a photocopy of a newspaper article from the Baltimore Sun. The document was of poor quality with an inconsistent gray shaded background. All three systems properly decolumnized this document. None of the three systems read this document very well. The degraded text proved difficult to decipher. However, Arkenstone's Open Book appeared to do a better job of presenting the information on the page. The article was substantially more readable after being scanned with Open Book. During a second scan of the same document, each system, following some adjustments, improved its reading of the document. Arkenstone, however, maintained its ability to give the best result.

Ms. Sajka presented a document which was an operating manual for an electronic musical keyboard. The page had a large amount of complex text along with tables of values relevant to the operation of the keyboard. All three systems did a good job with this manual. All did a credible job of presenting the complex tables. The Kurzweil 1000 was the fastest of the three systems in processing this document.

A third document clearly gave Arkenstone the lead in demonstrating superior text reading ability: Mr. Charlson asked the vendors to read the Table of Contents page from the C-SUN Conference Program. In addition to text, the page contained the table of contents itself which was black type within a box whose background was gray speckled. Neither the Kurzweil nor Robotron systems could read the table of contents itself although they did read the other text on the page. Using its default settings, Open Book clearly and correctly read the entire table of contents. This outcome was met with applause from the audience at the morning session.

During the afternoon session, Mr. Baum from Kurzweil was able to read the table of contents after manually making contrast adjustments to the software. While Kurzweil was able to read the document following these adjustments, the judges were struck by the fact that a totally blind person would not have been able to know that something else of importance existed on the page in the typical case. Mr. Fruchterman attributed Open Book's impressive performance to Arkenstone's use of Scanfix, a unique component of Open Book which provides various forms of extra filtering for difficult text.

It is worth commenting on the stated difference in the philosophy of each vendor with respect to the intended end-user. It was found that the Kurzweil 1000 was generally faster than Open Book at completing its task of scanning, recognizing and presenting text with synthetic speech. While special modes for improving degraded text are available in Kurzweil's system, the product is shipped with these features turned off as a default condition. This results in very fast recognition speeds with good quality text. As with all the systems, poor quality text will slow the speed of processing and quite naturally decrease the accuracy of the result. Turning on special features which help decipher bad text will slow a product's processing time, but increase the accuracy of the result. Arkenstone apparently prefers to ship their product with such features turned on. This makes the product slower than it needs to be on good quality documents. However, as seen with the C-SUN program page, it can make an otherwise unreadable document impressively clear. In either the case of Kurzweil or Arkenstone, the special features may be adjusted relative to the factory defaults with only a modest knowledge of the product; such adjustments, however, may require some trial and error.

Item Five addresses the issue of the ability for a novice user of computers to get acquainted with the products. Each of the three companies discussed here provides the ability to make use of their product without vast computer knowledge. As a stand-alone reading machine, the Galileo, unlike the PC based products, does not offer the user the prospect of going beyond the major functions involved with scanning, reading and storing text. The benefit of a stand-alone system, in fact, is that the relatively few controls have dedicated functions. The audible cues provided by these controls may be tailored by the manufacturer to give very task specific information. The Galileo has a limited number of keys that are large and well separated. It appeared easy to address basic scanning functions with this unit. More advanced features such as saving text or changing operating languages require the pressing of different key combinations. Some of these might be difficult to remember without a reference card of some sort. Fortunately, the keys give good feedback and it is unlikely that any harm could come from hitting the wrong key combination.

Mr. Fruchterman briefly discussed the availability of a special version of Open Book called VERA. This stands for Very Easy Reading Appliance. VERA, while not explicitly demonstrated at this session, was presented as removing the end-user from the need to interact with the computer except at a very basic level. It uses a keypad with very few keys, some of which emulate the functions of an ordinary tape recorder. VERA, as well as Galileo, should be considered for users who have difficulty with or fear of computer use.

Arkenstone's Open Book, the system that was on display has three levels of operation: Beginner, Intermediate and Advanced. In the first of these, the system limits the number of adjustments that may be performed on the system to just the basics. All text is stored as a single long document and editing is not possible. Using a clear menu system, controlled with a computer's numeric keypad or with a dedicated and specially labeled Arkenstone keypad, a user can progress through the levels of sophistication. Open Book is about as non threatening for the beginning computer user as any application is likely to be. As will be seen later in this document, the product is capable of a great deal of sophistication if one wants it to be. Keyboard shortcuts and use in a multitasking Windows environment is quite possible and very practical.

Kurzweil 1000 may also be fully controlled from a dedicated and specially labeled keypad. The product may be a bit more challenging for the true novice in that keypad functions may change, depending upon which of three modes the keypad is in. As a user progresses through increased knowledge and comfort with computers, Kurzweil 1000 offers many sophisticated functions and a true Windows 95 standard interface which Open Book lacks. There will be more on this in the discussion of special features.

For the simple act of pressing a button and hearing scanned text being spoken, all three systems work about as well as each other. It is important to note that the Galileo lacks the ability to scan ahead in a text while imultaneously reading aloud. Kurzweil 1000 has the best implementation of this feature in that a user may press a key to scan a new page as soon as a prior page has been scanned. With the Open Book system, the user can begin listening to spoken text, but must wait for a beep from the computer before a new page may be started. In both cases, the net result is the ability to hear continuous text from a long document.

Kurzweil 1000 and Open Book both have a number of built-in "Help" features to assist the beginning user. Kurzweil provides a very useful "Help and Status" key. This key, on the computer or special keypad, once pressed will cause the function of the next key pressed to be announced without executing the function. If the Help and Status key is pressed and the subsequent key is pressed and held, the function along with a brief explanation and status will be announced. The entire text of the Kurzweil 1000 manual is also available from the keypad.

Open Book has several available means of providing help as well: In place of a Help and Status key, Arkenstone has a "Key Identifier" key. When pressed, this puts the program into a mode wherein each key pressed causes an announcement of that key's function. This is helpful, but more limited than Kurzweil's approach. Arkenstone also has a "Where Am I" key which gives a brief description of which of the product's functions is currently in use. It also provides page and position information for a document that is being read. Finally, Open Book has a Help menu from which a quick command summary or the full manual may be accessed.

There are some noteworthy differences among the products with respect to the type of documentation shipped with the products. The Galileo comes with a manual recorded on cassette, a print manual and a manual in braille upon request. Open Book ships with a printed manual, a cassette tutorial, an optional video"Getting Started" guide and a braille manual on request. Kurzweil 1000 ships with a printed manual and a Braille manual on request. The judges believe that a taped manual or tutorial is an important feature and should be part of any of these products.

Addressing guideline six, the vendors discussed the ability that each of their systems has to store documents that have already been scanned. All three systems have similar "Library" functions wherein a document may be named and stored within the product itself. Naming a document with the Galileo as with the keypad-only options of the other two systems is tedious at best. Fortunately, each system will supply a default name for a document prior to storage. Kurzweil 1000 has the unique ability to use as a default file name the first several words of a document. A user would likely want to rename the file in most cases, but it is a convenient way to quickly store a file and preserve some meaningful information in the name of the file itself. Open Book, because of its backward compatibility for people using it under the Microsoft Windows 3.1 Operating System restricts file names to eight characters and appends an extension relevant to the file format being used for storage. A default file name in Open Book, therefore, might be "DOC0001.TXT." As with Kurzweil 1000, the Open Book file name may be overwritten with a more meaningful phrase. The Galileo permits the use of the current Date and Time as a default file name. This too may be overwritten albeit in a tedious manner given that the Galileo is not useable with a computer keyboard.

All three systems can export a document to a floppy disk. This is an unusual and highly useful feature in a stand-alone reading product such as Galileo. All three systems give the operator a large choice of file formats from which to choose when storing text. In practice, most word processors and file readers will read several common file formats such as Microsoft Word or Corel Word Perfect. ASCII text is always readable. A nice feature is that the different systems can export in specialized formats such as Rich Text Format (RTF) in which font and special character information such as underlining is preserved.

The three systems can also read files that were produced by other systems, such as on a word processor. Kurzweil 1000 has the very useful feature of being able to read the Adobe Portable Document Format (PDF). The vendors all agreed that the quality of the systems' abilities to import and export text is mainly dependent on modules provided by third-party vendors. For this reason, the translations sometimes leave something to be desired in terms of their faithfulness in formatting according to the rules of a "foreign" document management application.

Questions six and seven are closely related and have already been touched upon. It is interesting to note the methods that the two PC based systems have for permitting the user to edit and otherwise work with a document that has been scanned. As has been discussed, it is possible to export documents to a number of formats so they may be used by other applications. Another use for such a function is having the ability to edit a document to make it more readable subsequent to scanning. Each system approaches this ability in a different way. Kurzweil 1000 has a built-in editor such that a document may be modified without leaving the application. While a nice feature, in a multitasking environment it shouldn't be any more trouble to launch a word processing application whose commands may already be known. Using Kurzweil's editor presumably means that the user must learn an additional set of commands. Still, Kurzweil's approach is a very practicable one. Mr. Bradburn demonstrated Kurzweil 1000's ability to rescan a page that came out poorly. The system also permits inserting a new page into the middle of a document which can be handy for including text that has been accidentally

skipped by the person doing the scanning.

Arkenstone's approach to editing documents takes advantage of multitasking in a very elegant way. Mr. Fruchterman demonstrated his ability to launch a word processing application from within Open Book. He was able to press a single keystroke, confirmed with the Enter key which launched Microsoft Word for Windows along with a Windows based screen reading application. The current document was simultaneously transferred to the word processor. While a lot had to happen, it was a single keystroke that caused it to happen and the result was impressive. It should be pointed out that for a blind person to accomplish this task, he will need to have a screen reading application available on his computer, but this is the case with a large number of users. Kurzweil's system will permit editing and printing of the text directly from the application. As with Kurzweil 1000, Open Book will permit the user to replace a poorly scanned page with a fresh attempt, but only at the end of a document. In order to insert a page into the body of the document, the Open Book user will need to scan the new page and use a word processor or text editor to insert it into a document.

Question eight deals with the use of the PC based systems in conjunction with other features of the user's operating environment. As discussed, both Arkenstone and Kurzweil have the ability to work well in the multitasking environment of MS Windows 95. Arkenstone's system will work on any computer with at least sixteen megabytes of memory and either Windows 3.1, 3.11 or 95. Mr. Fruchterman also reported that Open Book has been tested to work well under Windows 98 which is scheduled for release in June, 1998.

Kurzweil 1000 requires that the computer platform has at least twenty-four megabytes of memory and is running the Windows 95 or Windows NT 4.0 operating systems. Kurzweil 1000 is also reported to work well under Windows 98.

Kurzweil 1000 is a true Windows 95 application. This means it can take advantage of the sophistication inherent in that operating system. A notable demonstration of this came when Mr. Baum was able to show the ability to scan and process documents in the "background" while other tasks were ongoing on the computer. Such an activity is the domain of a truly advanced computer user, but it is a powerful innovation. By comparison, Open Book can remain running in the background, but cannot continue to scan or process pages unless it is the focused upon application.

Regarding question nine, neither system is specifically designed for use in a network environment. However, both systems are capable of recognizing image files produced and stored elsewhere. In a network situation, there would presumably be a centrally located scanner that would be capable of transmitting image files to a server for storage. Either system would be capable of working with the image files in this situation assuming those images could be downloaded to an individual work station.

In question 10, the judges were strongly impressed by Arkenstone's ability to demonstrate an elegant and extremely effective solution to the question of braille production. Regarding the production of hard copy braille, the Kurzweil representatives did not specifically demonstrate the task. They explained that as a Windows 95 application, a document could be exported to any number of file formats that could be recognized and translated by appropriate braille translation software. While this is certainly an accurate explanation, it was unclear that a non-technical computer user would necessarily appreciate the concept as explained.

As with the reading of the gray-shaded table of contents page, Arkenstone again clearly demonstrated an important ability: Mr. Fruchterman had a Braille embosser connected to his computer. Using Open Book's ability to launch other applications, he scanned a page and hit a single key, "Q" for Quick-Braille. After confirming his command with a press of the enter key, Open Book seamlessly launched a dos-based Braille translation program which instantly began embossing grade II text. This ability is decisively convincing, for example, in terms of its utility in a mainstream classroom setting. It is obvious from the Braille demonstration that a blind student could have in his possession a braille copy of a class handout or exam paper within seconds of his classmates receiving their printed copies.

It is worth mentioning that under Windows 3.1 and 3.11, Open Book will directly support dynamic braille displays such as the Power Braille series from Telesensory Corporation. Under Windows 95 and 98, it would be necessary for the Open Book user to use a screen access program which itself supported Braille displays, in conjunction with Open Book. Kurzweil 1000 should also be able to produce braille on a dynamic display if used in conjunction with an appropriate screen access program. This ability was not demonstrated.

Regarding Question 11, the use of each system by those with multiple handicaps, each system has a limited but potentially useful number of features. Open Book, as has been mentioned, is capable of directly supporting dynamic Braille displays and may be the only logical solution for a deaf-blind individual. Both systems are able to produce large print on a computer screen, this type being variable in terms of foreground and background color combinations. Arkenstone uses a fixed font while Kurzweil uses the font available in the original text. Open Book has the additional feature of a moving lightbar or crosshair cursor which itself can be varied in size and color. This cursor tracks text as it is being read. Either system should be useable by a person requiring a special keyboard interface. Mr. Fruchterman told of an Arkenstone customer who lacks the physical ability to scan documents, but once scanned can use a Single Switch device to read a document. A similar arrangement should be possible with Kurzweil 1000. Kurzweil 1000 has the additional feature of permitting voice commands to be given to the system via a microphone. This ability was not demonstrated so it was not possible to observe the accuracy or range of activities provided by the speech input system.

On the issue of multiple language support Arkenstone clearly has the edge. The Kurzweil representatives spoke of a German language version of the product which was not shipping on the day of the Dueling Scanners sessions. They also spoke of plans for versions of the product in several other languages. Arkenstone was able to give an impressive live demonstration of multilingual support. There are two main issues in having a scanning product for the blind provide multilingual support. First, the product must be capable of recognizing text written in a particular language. Character sets and contextual clues about text vary from language to language. The other issue is the ability of the speech synthesizer in use to use the rules of a particular language to produce speech. A third issue is the language used in a product's user interface. Generally, a foreign speaker will want the menus and prompts of a product to be in his preferred language. Open Book demonstrably addressed all of these issues.

Mr. Fruchterman switched his product from the DECtalk synthesizer it had been using to a Keynote Gold Multimedia speech synthesizer. This is a software speech synthesizer which is available in several common languages. He showed that Open Book could be set to recognize Spanish text while maintaining an English user interface. Open Book scanned and processed the text in the usual way. When reading began, the synthesizer was automatically switched into its Spanish mode and the document was read in synthetic Spanish. Mr. Fruchterman explained that Open Book is available with fourteen different language recognition modules, any combination of which may be used in the product. The user interface is available in several common languages including French, Spanish, German, Dutch and Finish.

The Galileo also has the ability to operate in several languages. It automatically changes both the interface and recognition language at the same time. The Galileo can be equipped to handle several languages at a time.

Conclusions

With respect to a laundry list of features, it is debatable and subjective as to which of these are most important. Kurzweil 1000 has an impressive list of features that were not specifically tested in these sessions. For example, Kurzweil 1000 offers a built-in spell checker and dictionary. These are very nice convenience features. Other features of significant interest include the ability to rescan a bad page in the middle of a document. Open Book does not directly support some of this functionality but does have the ability to launch third-party software with a single keystroke so is to some degree customizable with such features. When deciding which product to purchase, it is essential that the counselor and/or end-user discuss with the product dealer which features are available and what their possible utility might be.

Open Book has the ability to function on older computers running the Windows 3.1 operating system. This may help some to avoid an expensive investment in new hardware for some individuals while still providing excellent scanning and reading capabilities. Mr. Fruchterman stated that future versions of Open Book will be true Windows 95/NT compatible products. Clearly, this is where computer operating systems are migrating and Kurzweil has a technological edge with respect to having the potential to harness all the power available on state-of-the-art machines.

In terms of raw scanning and recognition speed, the Kurzweil system did noticeably better than the Open Book. However, this is offset by the Open Book's ability to read out-of-the-ordinary or severely degraded print without requiring the user to fine tune the contrast setting. This is what might be expected given the stated philosophies of each manufacturer with respect to default settings. Open Book's use of Scanfix, a feature not available in Kurzweil 1000 was striking. This was not a scientific test of OCR accuracy by any means. However, Open Book often produced a more satisfying result than the other systems in this particular venue.

Kurzweil 1000 is arguably a more powerful and sophisticated system. As a true 32 bit/Windows 95 application, it has the ability to integrate smoothly into the blind computer users operating environment. The downside of this approach is that aside from the main features of scanning, reading and storage, the Kurzweil 1000 user may be required to stretch very basic computer knowledge to a level a bit above the comfort zone. Again, this is very subjective and must be judged by the individual user. Open Book with its multiple levels of sophistication and even the option of a variant of the product that uses a special simple keypad may well be a better system for those who aren't certain of how far they want to progress in the use of computers.

Both systems permit operation in a multitasking environment using speech other than that which is built-in to the product. Kurzweil 1000 does a more convincing job of this in that it should be just as easy or difficult to use as any standard Windows application. Open Book, originally developed in Windows 3.1's infancy and at a time when blind people used only DOS based computer programs presents a very laid-back and intuitive approach. It is only in its function which permits the reading of an external file where the user "sees" anything that resembles a Windows Dialog Box. This function is limited to the "Advanced" setting of Open Book.

Among the features specifically tested at these sessions, Kurzweil 1000 was almost always faster in its processing. It also showed tolerance for the difficult condition of facing pages which were skewed relative to each other. Open Book showed more tolerance for poor quality or unusual text.

It was agreed by the judges that none of the systems was perfect. Asked to rate the two PC based systems on a scale from 1 to 100, the average subjective rating of the judges was 75 for Kurzweil 1000 and 83 for Open Book.

Again, the numbers just reported are based on subjective, non-scientific measurements of the two systems as displayed only at the March 17 sessions. It is crucial that anyone considering the purchase of such a system try each of them with material that is of importance to them. There were two main reasons for Arkenstone's relatively superior performance at Dueling Scanners: The first was the ability to read the gray-background text that the other systems did not even acknowledge. The other was Arkenstone's apparent ease of use, a product of Jim Fruchterman's excellent ability to demonstrate and explain the system. It was too often the impression that the Kurzweil representatives relied on the general principle that as a Windows 95 application, it has no real limitations within that operating system. A case in point is the ability to scan and process text while focusing on another application. This is, in fact, an impressive feature and points to some of the power available in the Kurzweil system. However, for the intended audience, who may not have the technical sophistication to appreciate some of Kurzweil's power relative to Open Book, the Arkenstone system consistently

addressed all the questions asked of it with clear and thorough demonstrations. In truth, both systems do an excellent job of reading printed text. The features beyond that are matters which are best considered on a case by case basis.

Finally, a note about the Robotron Galileo. Because the demonstration unit was damaged in transit, it was not possible to use it competitively in this setting. As a stand alone reading machine, one would expect it to be relatively limited in features. The Galileo appears to have quite an extensive list of features that could easily be of value to most people requiring a technological solution to reading print. It is hoped that Robotron will be able to participate more fully in future Dueling Scanners.

All the vendors who participated in this year's sessions indicated that their development would continue with respect to the number and quality of features in their respective systems. Dueling Scanners is but one way that the consumers of blindness related technology may stay abreast of such developments. Our sincere thanks goes to Dr. Harry Murphy and all the staff at the C-SUN Center on Disability for making this exercise possible.