Computer Netwo.rks and ISDN Systems 29 (1997) 1075- 1082
`
`Digestor: device-independent access to the World Wide Web
`
`Timothy W. Bickmore*, Bill N. Schilit 1
`
`FX Palo Alto Laboratory. 3400 Hillview Avenue. Bldg. 4. Palo Alto. CA 94304. USA
`
`Abstract
`
`Digestor is a software system which automatically re-authors arbitrary documents from the world-wide web to display
`appropriately on small screen devices such as PDAs and cellular phones, providing devic~independent access to the web.
`Digestor is implemented as an HTIP proxy which dynamically re-authors requested web pages using a heuristic planning
`algorithm and a set of structural page transformations to achieve the best looking document for a given display size. ©
`1997 Published by Elsevier Science B.V.
`
`Keywords: Device-independent access: PDA: Document re-structuring: WWW
`
`1. Introduction
`
`Access to world-wide web documents from per(cid:173)
`sonal electronic devices has been demonstrated in
`research projects [2,10,17,18), and is now becoming
`a commercial reality. General Magic's Presto!Links
`for Sony's MagicLink, AllPen's NetHopper for the
`Newton and Sharp's MI- lO (shown in Fig. l), all pro(cid:173)
`vide WWW browsers for PDA class devices, while
`the Nokia 9000 Communicator and Samsung's Duett
`provide web access capabilities from cellular phones
`(Fig. 2).
`Unfortunately, most pages on the world-wide web
`are designed for display on desktop computers with
`color monitors having at least 640x480 resolution,
`and many pages are designed with even larger moni(cid:173)
`tors in mind. This disconnect can lead to a 4-to-l (or
`greater) ratio of designed vs. available screen area,
`
`• Corresponding author, e-mail: bickmore@pal.xerox.com
`1 E-mail: schilic@pal.xerox.com
`
`Fig. I.
`
`making direct presentation of most WWW pages on
`these small devices aesthetically unpleasant, un-nav(cid:173)
`igable, and in the worst case, completely illegible.
`This presents the central problem addressed by this
`work; how to display arbitrary web pages which
`have been designed for desktop systems on personal
`electronic devices which have much more limited
`110 capabilities.
`
`0169-7552/97/$17.00 © 1997 Published by Elsevier Science B.V. All rights reserved.
`PI/ SOI69 - 7552(97 ) 00026 - 3
`
`Petitioners' Exhibit 1022, Page 1
`
`

`

`1076
`
`T. W. Bickmore, B.N. Schilit I Compurer Nerwork.s and ISDN Systems 29 ( 1997) /075- 1082
`
`. -· .
`.. -·
`. . .. ...
`
`Fig. 2.
`
`2. Approaches to device-independent access
`
`There are four general approaches to display(cid:173)
`ing WWW pages on small screen devices: device(cid:173)
`specific authoring; multiple-device authoring; client(cid:173)
`side navigation; and automatic re-authoring.
`Device-specific authoring involves authoring a set
`of WWW pages for a particular display device, for
`example a cellular phone outfitted with a display and
`communications software such as the Nokia 9000.
`The basic philosophy in this approach is that users
`of such specialty devices will only have access to a
`select set of services, and the pages for these ser(cid:173)
`vices can all be designed up-front for the device's
`particular display. Information may be provided from
`the world-wide web at large, but the desired pages
`must be pre-defined and custom information extrac(cid:173)
`tion and page formatting software must be written to
`deliver the information to the small device. This is
`the approach taken in Unwired Planet's UP.Link ser(cid:173)
`vice [16) which uses a proprietary mark-up language
`(HDML).
`ln multiple-device authoring, a range of target
`devices is identified, and mappings from a single
`source document to a set of rendered documents are
`defined to cover the devices within the range. One
`example of this is the StretchText approach [8], in
`which portions of the document (potentially down
`to the word level) can be tagged with a "level of
`abstraction" measure. Upon receiving the document,
`users can specify the level of abstraction they wish to
`view and are presented with the corresponding detail
`or lack thereof. Another example of multiple-device
`authoring is HTML cascading style sheets (CSS)
`
`[13]. In CSS, a single style sheet defines a set of
`display attributes for different structural portions of
`a document (e.g., all top-level section headings are
`to be displayed in red 18-point Times font). A series
`of style sheets may be attached to a document,
`each with a weight describing its desi r:~bi li ty to the
`document's author. The user can also specify a style
`sheet, as can the WWW browser (the "default" style
`sheet). Although the author's style sheets normally
`override the user's, the user can selectively enable or
`disable the author's, providing the ability to tailor the
`rendering of the document to their particular display.
`In client-side navigation, the user is given the
`ability to interactively navigate a single web page
`by altering the portion of it that is displayed at any
`given time. A very trivial example of this is the
`use of scroll bars on the document display area. A
`much more sophisticated approach is that taken in
`the PAD+·I system [3], in which the user is free to
`zoom and pan the device display over the document
`with infinite resolution. Active Outlining [11 ] bas
`also been implemented as a client-side navigation
`technique, in which the user can dynamically expand
`and collapse sections of the document under their
`respective section headings. Other rechniques which
`fall into this category include semi-transparent wid(cid:173)
`gets [ 12) and the Magic Lens system [4].
`Finally, automatic re-authoring involves devel(cid:173)
`oping software which can take an arbitrary web
`document designed for the desktop, along with char(cid:173)
`acteristics of the target display device, and re-author
`the document through a series of transformations so
`that it can be appropriately displayed on the device.
`This process can be performed either on the client,
`on the server, or on an intermediary HTTP proxy
`server (as in [7]) which exists solely for the purpose
`of providing these transformation services. An exam(cid:173)
`ple of this latter approach is the UC Berkeley Pythia
`proxy [9], which performs transformations on web
`page images, although the focu s of this work is on
`minimizing page retrieval time, not on producing the
`most appropriate page layout for the display device.
`There are many possible automatic re-authoring
`techniques, which can be categorized along two di(cid:173)
`mensions: syntactic vs. semantic and transforma(cid:173)
`tion vs. elision. Syntactic techniques operate on the
`structure of the page, while semantic techniques
`re ly on some understanding of the content. Elision
`
`Petitioners' Exhibit 1022, Page 2
`
`

`

`T.W Bickmore, B.N. Schilit/Computer Networks and ISDN Systems 29 (1997) 1075~ 1082
`
`1077
`
`techniques basically remove some information, leav(cid:173)
`ing everything else untouched, while transformation
`techniques involve modifying some aspect of the
`page's presentation or content. The table below illus(cid:173)
`trates these dimensions, along with examples of each
`category.
`
`Elide
`
`Transform
`
`Image reduction
`Syntactic Section outlining
`Semantic Removing "irrelevant content" Text summarization
`
`Each of the four approaches- device-specific au(cid:173)
`thoring, multiple-device authoring, client-side navi(cid:173)
`gation, and automatic re-authoring -
`has benefits
`and drawbacks. Device-specific authoring will typ(cid:173)
`ically yield the best-looking results, but limits the
`user's access to a small select set of web pages. Mul(cid:173)
`tiple-device authoring, while less total effort per doc(cid:173)
`ument than device-specific authoring, still requires
`significantly more manual design work than simply
`authoring for a single desktop platform. Client-side
`navigation holds promise if a good set of techniques
`can be developed, but the "peephole" approach taken
`in PAD++ seems very awkward to use for large doc(cid:173)
`uments, and the active outlining technique has lim(cid:173)
`ited applicability since most web pages do not use
`a strict section/sub-section organization. Automatic
`re-authoring is thus the ideal approach to provid(cid:173)
`ing broad access to the web from a wide range of
`devices, if it can be made to produce legible, naviga(cid:173)
`ble and aesthetically pleasing re-authored documents
`without loss of information.
`
`3. Understanding there-authoring process
`
`In order to gain an understanding of the pro(cid:173)
`cess required of an automated re-authoring system,
`a study was conducted to assess the characteris(cid:173)
`tics of typical web pages, and to identify candidate
`re-authoring techniques through the process of re(cid:173)
`authoring several web pages by hand.
`A relatively small sample of "typical" web pages
`-the Xerox Corporate web site [20] -was initially
`selected to focus the study. This collection of 3,188
`web pages is representative of a state-of-the-art,
`professionally-designed site. A variety of statistics
`were collected on these pages using a web crawler,
`
`to help gain an understanding of the structure and
`content of a typical page. These statistics generally
`agree with other, larger-scale studies that have been
`performed across the entire web [6,19].
`Next, a subset of the pages in the Xerox web site
`was selected for manual re-authoring. A set of pages
`from the Xerox 1995 Annual Report were selected
`and converted by hand for display on a Sharp Zaurus
`PDA with a 320 x 240 pixel screen. Detailed notes
`were kept of the design strategies and techniques
`used.
`Some of the design heuristics learned during this
`process were:
`• Keeping at least some of the original images is
`important to maintain the look and feel of the
`original document. Common techniques include
`keeping only the first, or only the first and last
`image (bookend images) and eliding the rest.
`• Section headers (Hl-H6 tags) are not often
`used correctly. They are more frequently used
`to achieve a particular font size and style (e.g.,
`bold), if they are used at all. Thus, they cannot
`be relied upon to provide a structural outline for
`most documents. Instead, documents with many
`text blocks can be reduced by replacing each text
`block with the first sentence or phrase of each
`block (first sentence elision).
`• An initial rule of thumb for images is to reduce
`them all in size by a standard percentage, dictated
`by the ratio of the display area that the document
`was authored for to the display area of the target
`device. Images which contain text or numbers can
`only be reduced by a small amount before their
`contents become illegible.
`• Semantic elision can be performed on sidebars
`which present information which is tangential to
`the main concepts presented in a page. Many of
`the Xerox pages had such sidebars which were
`simply eliminated in the reduced versions.
`• Semantic elision can also be performed on images
`which do not contribute any information to the
`page, but serve only to improve its aesthetics.
`• Pages can be categorized, and then re-authored
`based on their category. Two examples of these
`are banners and link tables. Banners primarily
`contain a set of images and a small number of
`navigation links (often only one) which serve to
`establish an aesthetic look, but contain little or
`
` 3
`
`Petitioners' Exhibit 1022, Page
`
`

`

`1078
`
`T.W. Bickmore, B.N. Schilit!Computer Networks and ISDN Systems 29 ( 1997) 1075- 1082
`
`no content. When space is at a premium, these
`can usually be omitted entirely. Link table pages
`consist primarily of a set of hypertext links to
`other pages, and very little additional content.
`These pages can usually be re-formatted into a
`more compact form which just lists the links in a
`text block.
`• Whitespace, which is taken for granted on a large
`display, is at a premium on small devices. Sev(cid:173)
`eral techniques were discovered for reducing the
`amount of whitespace in a page. Sequences of
`paragraphs (P tags) or breaks (BR tags) can be
`collapsed into one. Lists (UL, OL, and DL tags)
`take up valuable horizontal space with their in(cid:173)
`denting and bullets, and can be re-formatted into
`simple text blocks with breaks between successive
`items (also observed in [8]).
`In conclusion, to perform document re-author(cid:173)
`ing two things arc required: a set of re-authoring
`techniques (a "bag of tricks"), and a strategy for
`applying them. Of the techniques used in the manual
`re-authoring study, those most amenable to codifi(cid:173)
`cation were the syntactic elision techniques (section
`outlining, first sentence elision, image elision) and
`the syntactic transfonnation techniques (image size
`reduction, font size reduction). The design strategy
`learned during the study consisted of a ranking of the
`transfonnation techniques (i.e., try this before that)
`and a set of conditions under which each transfor(cid:173)
`mation or combination of transformations should be
`applied.
`
`4. Digestor system design
`
`Following the results of the study discussed above,
`there are two major elements to Digestor's design:
`a collection of individual re-authoring techniques
`which transform documents in various ways; and an
`automated re-authoring system which implements a
`design strategy by selecting the best combination of
`techniques for a given document/display size pair.
`
`4.1. Re-authoring techniques
`
`4.1.1. Outlining
`Section header outlining techniques provide a
`very good method for reducing the required dis(cid:173)
`play size for structured documents, such as technical
`papers and reports. The outlining process is depicted
`in Fig. 3. The contents of each section is elided from
`the document and the section header is converted
`into a hypertext link which, when selected, loads the
`elided content into the browser. When confronted
`with multiple section levels (sections, sub-sections,
`sub-sub-sections. etc.), there are two approaches to
`performing the elision. The first -
`f ull outlining -
`works by keeping only the section headers and elid(cid:173)
`ing all content, with the results looking like a table
`of contents for a book. In the second approach -
`a cutoff level in the section hier(cid:173)
`to-level outlining -
`archy is determined and all content below that level
`(including lower-level section headers) is elided, but
`all content above that level is kept.
`
`PaacTldc
`
`Scalionl
`F11f"~ U'M
`..... :lfdtdc11Ddep4Ddtot ...
`
`Scctiool
`:btl .. twobuil . . . . . ~ ~)
`b ....... ~·--=-
`
`8(~3
`:t:.il :D"'fltOiw.C~:\.. ;1ole
`._.~.....,d~b.a.M ..: .a.
`
`~-::.. ....
`
`Fig. 3.
`
`Section 1
`
`.._ 3TtJL ~'?\'ide l
`.. :;110...._,.. ,(
`... ,-_.
`
`-~u .. : .. :h~t.
`
`Petitioners' Exhibit 1022, Page 4
`
`

`

`T.W. Bicf..7rtore. B.N. Schilit!Complller Networks and ISDN Systems 29 (19971 1075-1082
`
`1079
`
`4. 1.2. First sentence elision
`Since most pages have text blocks, even when
`no section headers are present, first sentence elision
`can be a good way of reducing required screen area.
`In this technique, each text block is replaced with
`its first sentence (or phrase up to some natural break
`point), and this sentence is also made into a hypertext
`link to the original text block.
`
`4.1.3. Image reduction and elision
`Images present one of the most difficult problems
`for automatic re-authoring, because the decision of
`whether to keep, reduce, or eliminate a given image
`should be based on an understanding of the content
`and role of the image on the page. However, image
`reduction and elision can be applied without content
`understanding, as long as users are provided a mech(cid:173)
`anism by which they can retrieve the original image.
`The approach taken in Digestor is to provide a set
`of techniques which transform all images in a page
`by pre-defined scaling factors (25%, 50%, and 75%),
`and making the reduced images hypertext links back
`to the originals. In addition to image reduction, three
`syntactic elision techniques were also developed for
`images -
`elide all, first image only, and bookends
`in which all images, all but the first, and all
`-
`but the first and last are elided from t11e documt:nl,
`respectively. Elided images are replaced wilh their
`ALT text when available, or with a standard icon
`when no ALT text is available, which is also made
`into a hyperte.xt link to the original image.
`
`4.2. Autorruued re-authoring system
`
`The overall process of deciding which combi(cid:173)
`nation of techniques to apply to a given page for
`a given client display seems at first to require some
`form of human artistic ability. However, an automatic
`re-authoring algorithm has been developed which
`captures many of the heuristics used in the manual
`re-authoring exercise, and which does a fairly good
`job of producing good looking pages for a given
`display.
`The basic approach is that of a hcuri:sti~o: planner
`r 11 which searches a document transformation space
`in a best-first manner. Each state in this space rep(cid:173)
`resents a version of the document, with the initial
`J;tate reprec:enting the original "a~-authored" docu-
`
`\
`
`\
`
`Fig. 4.
`
`ment. The measure of merit (evaluation function) for
`each state is a rough estimate of the screen area
`required for the document as it exists in that state
`(Fig. 4). A state can be expanded into a successor
`:state Lhrough the application of single rransformarion
`technique. At every step in the search process the
`most promising state (that with the smallest display
`area requirements) is selected and a transformation
`applied to it, if possible. As soon as a state is cre(cid:173)
`ated containing a document version that is "good
`enough", the search is halted and that document is
`returned to the client for rendering. If the search is
`exhausted and no document version can be found
`that is good enough, lhen the best document fou nd
`during the search is returned.
`Heuristic information is used in several places in
`the planner, including: the order in which transfor(cid:173)
`mation techniques are applied to a given state; the
`pre-conditions for each transformation technique;
`and the determination of when a document version
`is "good enough" to halt the search. In general,
`transformations which make minor changes to the
`document are preferred over those which make more
`extensive changes (e.g., reducing images by 25%
`is preferable to reducing them by 75%). The pre(cid:173)
`conditions for each transformation technique specify
`
`Petitioners' Exhibit 1022, Page 5
`
`

`

`1080
`
`T.W. Bickmore, B.N. Schilit!Computer Networks and ISDN Systems 29 ( 1997) 1075-1082
`
`which other transformations they can be combined
`with (e.g., it makes no sense to apply both full outlin(cid:173)
`ing and first sentence elision to the same document),
`as well as requirements on the content and structure
`of the document that the technique is being applied
`to (e.g., only 3pply full outlining when there are at
`least three section headers). The current condition
`for "good enough" is fairly simplistic; the search is
`stopped when the area required by a document ver(cid:173)
`sion is 2.5 times the screen area of the client display
`(which assumes that the user doesn't mind scrolling
`the display a little in one direction).
`
`5. Implementation
`
`The design described above has been imple(cid:173)
`mented as an HTIP proxy server (as in [7]). The
`proxy accepts a request for an HTML document,
`retrieves the document from the specified HTTP
`server, parses the HTML and constructs an Abstract
`Syntax Tree (AST), labels each of the AST nodes
`with a unique identifier. and then retrieves any em(cid:173)
`bedded images so that their size can be determined
`(as necessary). Once this has been accomplished, the
`planner is initialized with a state containing the AST
`for the original document. During each planning
`cycle, the planner selects the state with the best doc(cid:173)
`ument version so far, then selects the best applicable
`transformation technique and applies it resulting in
`a new state and document version being generated.
`It is assumed that the convolution of transformations
`is always commutative, and several checks are used
`in the planner to ensure that redundant states are not
`constructed.
`techniques have been
`Fifteen
`transformation
`implemented and
`integrated
`into
`the system:
`FullOutline, OutlineToH I, OutlineToH2, Outline(cid:173)
`ToH3, OutlineToH4, OutlineToH5, OutlineToH6,
`FirstSentenceElision, Reducelmages25%, Reduce(cid:173)
`Images50%, Reducelmages75%, ElideAlllmages,
`FirstlmageOnly, Bookendlmages, and ReduceFont(cid:173)
`Size. Transformations manipulate the ASTin the state
`they are applied to in order to produce a new ver(cid:173)
`sion of the document (the manipulations are similar
`to those described in [5]). Whenever portions of the
`AST are elided or transformed, an HTML hypertext
`link is added into the AST referencing the node iden-
`
`tifiers of all affected AST subtrees, enabling users
`to request the original portions ot the document that
`have been modified during re-authoring.
`The system is implemented in the Java program(cid:173)
`ming language. and the HTTP proxy server software
`was based on the MBServler system (14]. In addi(cid:173)
`tion to functioning as a true proxy, this system can
`also respond to requests for certain URLs with doc(cid:173)
`uments generated by the proxy itself, which is used
`to provide the user with forms-based control over the
`proxy and there-authoring process. In the worst case
`the planner produces 80 versions of the document
`in its search space and takes about 20 seconds to
`complete on a 75 MHz Pentium (using Symantec's
`Java JIT compiler).
`The first thing that a user of Digestor will typ(cid:173)
`ically do is specify the size of display for their
`device, and indicate the size of their default browser
`font (this is needed in order to estimate the screen
`area requirements of text blocks). To do this, the
`user requests a specific control URL from the proxy,
`resulting in delivery of the form shown in Fig. 5.
`Once a user has configured the system. they can
`start retrieving documents from the web. Fig. 6 il(cid:173)
`lustrates Digestor's re-authoring capability. In this
`example, the system chose to use 25% image re(cid:173)
`duction in combination with first sentence elision to
`render the displayed version. Immediately following
`retrieval of a page, the user can request a trace of the
`re-authoring session to determine which transforma(cid:173)
`tions had been applied by requesting another control
`URL from the proxy.
`
`Fig. 5.
`
`Petitioners' Exhibit 1022, Page 6
`
`

`

`T.W Bickmore. B.N. Schilil/Compurer Networks and ISDN Systems 29 (1997) 1075- 1082
`
`1081
`
`..... Al.-~,.. •• ,.,. .......... ......... ,.(JIM"ft..... t
`
`...... fiiiii~4W.•.-t~ . . .................. 4 ....
`.... ~ ...... 4 . ..... ~" -- . . .. . . . _. ~ .,··
`-
`. . . _ .. -
`
`__ . ..,_. -. ..... .......... ..._,.,_.,.._. ~ -·-... ....
`............ - ,. .............. ., ................... ......,. ...... .
`... ,. .......
`.... ...-.--....u• .... ,.,Jt.
`,~._. ,"""""'.,.- .,.,. ... ,,...~-··1,.· ...... ....... s.!.:I.X."'' - ' ·
`"c .. o .... ••o.~~·..a::..(tt'IIMil.._w:. ......... ..,,._ ... )II:U:
`._.-.. tPe n» ... ,.,.&
`-.. -~ ,.. .... .,. • • .,._ ..... '!: ·~· ·: ·--- ............ .,. - ·
`... _
`
`lt. I•
`
`...... l,4\ IMf'Wf\ .........
`
`•WWJ w-~ 1 . .
`
`, ..... :If OI"'Otf''f.):U"f"'co<R) .. t
`
`lllA"II
`
`, ... ,..uM """1tllli;.,..""""('W'"r ~11111l .. ~~ «•·•
`• ~
`
`. ;::_ ... wt\w1t M- I ufy.e. ,_ •
`
`~'!<lSI -.~.... •• ,_,. •
`
`Cl?
`
`I'PfA A
`
`Fig. 6.
`
`6. Conclusion
`
`Digestor does a good job of automatically re(cid:173)
`authoring web pages for display on devices with
`small screens. It has been informally tested on a wide
`range of pages for a number of screen sizes, and it
`produces output which is always legible and naviga(cid:173)
`ble, although not always as aesthetically pleasing as
`one would like.
`
`6.1. Future work
`
`There is much work to be done on Digestor before it
`could be made into a widely useable product. Some of
`the general issues which need to be addressed include:
`• Improved measure of merit - The current sys(cid:173)
`tem simply adds up the space requirements of all
`images and text to arrive at an estimate of the
`screen area requirements for a document. This is
`adequate for fairly dense documents with mini(cid:173)
`mal structure, such as those in the Xerox Annual
`Report, but works poorly for documents with a lot
`of whitespace or which use advanced layout tech(cid:173)
`niques (e.g., tables). Ultimately, what is needed
`is a size estimator which performs much of the
`work performed by a browser in formatting each
`document version onto a display area. Factors
`other than required screen area may also need
`to be included, such as actual width requirements
`(users don't like to scroll horizontally), bandwidth
`requirements, and aesthetic measures.
`• More user control - Users should be able to
`adjust the various heuristics used in the planner
`
`to suit their taste. For example, they could spec(cid:173)
`ify the relative preference of the transformation
`techniques, or specify that some transforms not
`be used at all. At a higher level of abstraction,
`they could express their preferences for a set of
`trade-offs, such as "more content" vs. "larger rep(cid:173)
`resentation". In addition, the re-authoring system
`could be moved to the client and coupled with the
`browser so that the user could dynamically ap(cid:173)
`ply and undo different transformations until they
`achieved a result they liked.
`• Shared proxy - As implemented, Digestor can
`only be configured for one user at a time. Since it
`potentially has greater value as a shared public or
`workgroup service. it could be equipped to handle
`configuration information from a range of users
`-
`using a technique such as Netscape cookies
`[ 15] -
`to convey each user's preferences.
`• More transformation techniques - The area
`that Digestor could use the most improvement in is
`its set of transformation techniques. Some specific
`candidates to add include: elision of advanced page
`elements (e.g., tables, applets, shockwave plug-ins,
`etc.); folding of table rows or columns; re-format(cid:173)
`ting of lists into text blocks; folding of descriptive
`lists (as in [8]); compressing whitespace (e.g., re(cid:173)
`moving multiple P or BR tags); and text block
`summarization. A thorough study of graphic art
`techniques could be made to derive additional can(cid:173)
`didates. Transformations could also be generalized
`by the use of user-specified parameters so that,
`for example, a transformation could elide all text
`blocks containing a user-specified keyword.
`
`Petitioners' Exhibit 1022, Page 7
`
`

`

`1082
`
`T. W. Biclarwl'f!, B.N. Schilit/Compuur Networks and ISDN S,rsterns 29 ( 1997) 1075- 1082
`
`Acknowledgements
`
`Thanks to Joe Sullivan for supporting this re(cid:173)
`search.
`
`References
`
`I I] A. Barr and E. Feigenbaum, The Handbook of Artificial
`Intelligence, William Kaufmann, Inc .. Los Altos, CA, 1981,
`pp. 5!kl3.
`(2] J. Banlett, Experience with a Wireless World Wtde Web
`client, IEEE COMPCON 95, San Francisco, March 1995.
`[3) B. Bederson and J. Hollan, Pad++: A zooming graphical
`interface for exploring alternate interface physics, in: Pro(cid:173)
`ceedings of ACM UIST '94, ACM Press, 1994, ftp://ftp.es.u
`nm.edulpublpad-t-t-/pad-ujsr94.ps.gz
`[4] E. Bier, M. Stone, K. Pier, W. Buxton and T. DeRose.
`Toolglass and magic lenses: The see-through interface, SIC(cid:173)
`GRAPH '93 Confe1'!!nce Proceedings, 1993, hnp:/lwww.par
`c.xerox.comlistVprojecLs/Magic/Lenses/93Siggraph.hunl
`[5] S. Bonhomme and C. Roisin, Interactively restructuring
`HTML documents, 5th International World Wide Web Con(cid:173)
`fe1'!!nu, Paris, May 1996, http://www5conf.inria. fr/fich__bt
`mVpapers/Pl610verview.html
`[6) T. Bray, Measuring the Web, 5th lnttmational World Wide
`Web Corl{el'f!nce. Paris. May 1996. hnp://www5conf.inria.fr
`/fich_htmllpapers/P9/0verview .html
`[71 C . Brooks, M. Maz.er, S. Meeks and J. Miller, Applica(cid:173)
`tion-specific proxy servers as HTTP stream transducers, 4th
`International World Wide Web Conference, Boston, Decem(cid:173)
`ber 1995, http://www. w3.orglpub/Conferences/WWW 4/Pap
`ers/561
`[8) I. Cooper and R. Shufflebotham, PDA web browsers:
`Implementation
`issues, University of Kent at Canter(cid:173)
`bury Computing Laboratory WWW Page, November
`1995, http:/lalethea.ukc.ac. uk/Dept/Computing/Research/S
`TEP/Papers/WWWPDAI
`[91 A. Fox and E. Brewer, Reducing WWW latency and band(cid:173)
`width requirements by real-time distillation. 5th Interna(cid:173)
`tional World Wide Web Conference, Paris, May 1996, http:/
`/www5conf.inria.fr/fich_htmVpapersiP48/0verview.html
`[10) S. Gessler and A. KO[UJia, PDAs as mobile WWW
`browsers, 2nd International World Wide Web Conference,
`Chicago, October 1994), http:/fwww.ncsa.uiuc.edu/SDG/IT
`94/Proceedings/DDay/gessler/www_pda.btml
`[II) J. Hsu, W. Johnst.on and 1. McCanby, Active outlining for
`HTML doc:urncnu: An X-Mosaic implementation, 2nd ln(cid:173)
`rernaliOIIal World Wide Web Conference, Chicago, October
`1994, http://www .ncsa.uiuc.edu/SDGIIT94/ProcudingsiHC
`Ilhsu/hsu.html
`[ 12) T. Kamba, S. Elson, T. Harpold, T. Stamper and P.
`Su.lc:aviriya. Using small screen space more efficiently, CHI
`96, Vancouver, BC Canada, April 1996.
`[ 13) H. Lie and Bert Bos, Cascading style sheets, WW\V Con(cid:173)
`sortium, September 1996. hnp://www.wJ.org/pub/WWWff
`R!WD-cssl
`
`(14] Mort Bay Consulting, MBServler: Mort Bay's HTTP
`Servlet .servmg .server. http://www.oz.email.com.aulmort
`bay/MBServler/MBServler.html
`[15] Nctscape Communications Corporation, Persistent Client
`State HTTP Cookies, http:/lwww.netscape.com/newsref/std
`/cook:ie_spec.html
`(16] Uuwiu:J Ph•nct, ln.: .. UP.Link Developer's Gwde Version
`1.0, Redwood Shores. CA, July 1996, http:/lwww.uplanet.
`com
`[ 17) G. Voelker and B. Bershad, Mobisaic: An information sys(cid:173)
`tem for a mobile wireless computing environment, Work(cid:173)
`shop on Mobile Computins Systems and Application$, Santa
`Cruz, December 1994, http:/lwww.cs.washington.edu/home
`slvoelker/ mobisaic/ps/mobile94-final.ps
`[18) T. Watson, Application design for wireless computi ng. 1994
`Mobile Computing S~vsrems and Applications Workshop,
`Position Paper. Auuust 1994. http:/ljupiter.aist-nara.ac.jp/
`' keniti-n/ieee-rncsa94/wat.son.ps.gz
`[ 19] A. Woodruff, P. Aoki, E. Brewer. P. Gauthier and L. Rowe,
`An investigation of documents from the WWW. 5rh lnrer(cid:173)
`nationa/IVorld Wide Web Conft1'!!nce, Paris, May 1996, http
`:l/www5conf.ioria.fr/fich_htmVpapers/P7/0verview.html
`[20] Xerox Corporation. 1995 Annual Report. http:www.xerox.c
`om/annualreport/1995/
`
`Timothy W. Bickmore has been a
`Coosulllng Scientist at the FX Palo
`Alto Laboratory for the last two
`years, where he has been working
`on applications in mobile comput(cid:173)
`ing and synthetic characters. He is
`alsn independently working on sev(cid:173)
`enll projects for NASA in the area of
`rocket propulsion health monitoring.
`Previously, he has held research po(cid:173)
`sitions with the Lockheed Artificial
`l nte II igence Center in Palo Alto and
`Aerojet Propulsion Systems Division in Sacramento. Mr. Bick(cid:173)
`more received a M.S. degree in Computer Science from Arizona
`Slate University in 1986.
`
`8 111 N. SchUlt is a Senior Research
`Scientist at FX Palo Alto Labora(cid:173)
`tory in California. At FX PAL his
`research focuses on ubiquitous in(cid:173)
`formation access, finding the infor(cid:173)
`mation you want, anytime, anyplace.
`Previously, Bill was a Member of the
`technical staff at AT&Ts Bell Labs
`where he led the TeleWeb project.
`In 1995 Bill received a Ph.D . degree
`from Columbia Univecsity for stud-
`ies in location-dependent and con(cid:173)
`text-:lware mobile computing. His thesis research was performed
`at Xerox PARC where he worked on the Ubiquitous Computing
`initiative as a student intern and visiting scientist.
`
`Petitioners' Exhibit 1022, Page 8
`
`