CA2277209C - A system and method for the dynamic presentation of the contents of a plurality of documents for rapid skimming - Google Patents

Abstract

A method and system for the dynamic presentation of the contents of a plurality of documents on a display is disclosed. The method and system comprises receiving a plurality of documents and providing a plurality of topically rich capsule overviews corresponding to the plurality of documents.
The method and system also includes dynamically delivering document content encapsulated in the plurality of capsule overviews. In so doing, a system and method in accordance with the present invention can present thematic capsule overviews of the documents to users. A capsule overview is derived for the entire document, which will depict the core content of an average length article in a more accurate and representative manner thaw utilizing conventional techniques. The capsule overviews, delivered in a variety of dynamic presentation modes, allow the users to quickly get a sense of what a document is about, and decide whether they want to read it in more detail. If so, the system and method greatly facilitate the process of focused navigation into the parts of the document which may be of particular interest to the user. In a preferred embodiment, the capsule overviews include a containment hierarchy which relates the different information levels in a document together, and which includes a collection of highly salient topic stamps embedded in layers of progressively richer and more informative contextualized text fragments. The novel presentation metaphors which the invention utilizes are based on notions of temporal typography, in particular for exploiting the interactions between form and content.

Description

~0 /-bb 1 5 : Lb YIrUM : 5HW YtK t1 HSSUI. 1 H 1tS 1 U : b5i0 4b.i ~154b t'Hl.at b/ .5~
WO 99/Z61 T2 PCTIUS98n43$4 A SYS'fl~M AND ME'I'HpD FOR THE DYNAMIC p)E~ESENTATiON pF TR,E
CpIVTENTS OF A Pl<,,URALl't7l OF DOCUMENTS FpR RAPID SK~MING
FIELD OF THE INVEhITION
The present invention relates generally to a system and method for reviewing documents. More particularly, the present invention relates to presentation of documents in a manner that allows the ustr to quickly ascertain their contents.
~ACKCROUHD OF THE ;NVEN'x'ION
Documents obtained via an eIecnronic medium (l.c., the Internet or on-line services, such as AfJL, Compuserve or other services) are often provided in such voi~nne that it is important to be able to sumrnarizc them. Oftentimes, it is desired to be~able to to quickly obtain a brief (l.c., a few sentences or a paragraph length) summary of the document rather than reading it in its completeness. Most typically, such documents span several paragraphs to several pages in length. The present inventions concerns itself with this fond of document, hereinafter referred to as average I~gtia document.
Present day summarization technologies fall short of delivering fully informative 1 s summaries of documents_ To some extent, tb~is is so because of shortcomings of the state-of the-art in natural language processing; in general, the issue of how to customize a summarization procedure for a specific information seeking task is still an open one.
however, given the rapidly growing volume of document-based information on-line, the need for auy kind of document abstraction mechanism is so great that summarization 2o technologies are beginning to get deployed in real world situations.
The majority of techniques for "stunmarization", as applied to average-length documents, fail within two broad categories. A class of techniques mice a document for certain pre-specified pieces of information, typically defined a priori, on the basis of fixing the most characteristic features of a known domain of interest. ether approaches 2s rely, in effect, on 'te-using' certain fragmenss of the original text;
these have been identified, typically by some similarity metric, as closest in meaning to the whole document. This categorization is not a rigid one- a number of approaches (as exhibited, for inGStaace, in a rent worlrshop on Association jot Co~npu~ationat Linguistics, "Proceedings of a Workshop on Intelligent, Scalable, Text Summarization,"
Madrid, 07-99 15:29 FROM:SAWYER & ASSOCIATES IU:650 493 4549 PAGE 6/35 WO 99126172 PCL/us9~384

2 Spain, 1997) use song z~tions of topicality ($. Bogizraev and C. Kennedy, "~~e-based content characxerization of text documents," in Proceedings of ACL 97 Workshop on Intelligent, Scalable Text Summarization, Madrid, Spain, 1997), (E. Hovy and C.Y_ Lin, "Automated text sununarization in SUMMA~ST," in proceedings ofACL 97 Workshop on Intelligent, ,Scalable Text Summarization, Madrid, Spain, 1997), lexical chains (It. Barzilay and M_ Elhadad, "Using lexical chains for text sumnxarization," in proceedings ofACL 97 Workshop on Intelligent. Scalable Text Summarization, Madrid, Spain, I997), and discourse structure (D. Mareu, "From discourse s~ucteues to text summaries", in Proceedings of ACL 97 Workrhop on fnteltigerrt, Scalable Text IO Summarization, Madrid, Spain, 1997}, (1~T. Fiahn and M. Strobe, "Centered segmentation:
seating up the centering model to global discourse structure," in Proceedings ofACL-EACLl97, 35th Annual Meeting of the Association for Computational ,~,inguistra acrd 8th Conference of the European Chapter of the Association for Computational Linguistics, Madrid, Spain, I 99'7), thus laying claim to newer $ets of methods.
~ 5 Still, at a certain level of abstraction, all approaches share a fundamental similarity: summarization methods today rely, in essence, on substantial data reduction over the, original document source. Such a position leads to several usability questions.
Given the extracted fragments which any particular method has identified as worth preserving, what is axe optimal way of encapsulating these into a coherent whole, 2o for presenting to the user? Acknowledging that differextt information management tasks may require different kinds of sugary, even from the same document, how should the data discarded by the redexction process be retained, in case a reference is necessary to a part of the document not originally included in the s~.m~mary? What are the trade-offs in fixing the granularity of analysis: far instance, are sentences better than paragraphs as ?5 information-bearing passages, or are phrases even better? Of particular importance to this invention is the question of 'user involvement " 1~rom the end-user's point of view, making judgements, on the basis of a summary, concerning what a document is about and whether to pay it closer attention would engage the user in a sequence of actions:
look at the stmnmary, absorb its semantic impact, infer what tire document might be 3o about, decide whether to consult tk~e source, somehow call up the full docurneat, and navigate to the points} of interest. Given that this in~aduces a serious amount of cognitive and operational overhead, what are the implications for the user when they are va/-~~ 15:~~ rttuM:SHWYtII es H55U(..lHlt,S lU-b510 4U.i 4b4J rH~,~
wo ~mn Pcr~us faced with a,large, and growing, number of dents to deal with on a daily basis?
These are only some of the questions concerning the acceptability of summazization technology by end users. There is particular urgency, even the cunxatly evolving notion of'~nfonnation push", where content arriving unsolicited, and in large quantities, at individual workstations threatens users with real and imunediate information overload. To the extent that broad coverage summarization techniques are beginning to get deployed in real world situations, it is still the case that thesse techniques are based primarily on sentence extraction methods. In such a context, the above questions take on more specific nnterpretations_ 'fhtzs, is it appropriate to concatenate t o together the sentcnets extracted as representative-especially when they come from disjoint parts of the source document? What could be done, within a sentence extraction framework, to ensure that all 'themes' in a document get repres~e~atcd by the set of sentences identified by the technology? How can the3arring effect of'daagling' (and unresolved) references in the selection-without nay obvious means of identifying the t 5 referents in the original text-be overcome? What mechani~ns could be developed for offering the user additional information from the document, for more focused attention to detail? What is~ the value of the sentenrx, as a basic information-bearing unit, as a window into a mufti-docurrtent space?
To illustrate some of these issues, consider several examples from an operational z0 news tracking site: the News Channel page of Excite, an information vendor and a popular search engine host for the World Wide Web, which is available via the "ongoing Coverage" section of the news tiaclong page, (httpJ/nt.excite.com). Under the heading of Articles about IRS Abuses Alleged, some entries read:
zs itEHO ON SundaylReform Taxes ~..
The problem, of course, is that the enemies of the present system are aft grinding drtfereri!
axes. How true, how tsue. and dfttv for most of the people who sit on the Fir~t~ce Comrt~ittes. (E'srsrt found: I8 4ct 199 Scheduled IRS Layoffs Far 500 Arm..
The Agency~s origf»af plan called for elinti~latirtg as many as 5,000 jobs in field offices and 07-99 15:30 FROM: SAWYER 8~ ASSOCIATES ID:650 493 4549 PAGE A/35 wo ~r~in Pcrms~sru3sa at the Wa~ltir~gton tee~adq'?ha way his has harned out, it works to the ager~ty~s adv2ntage, the emplpyees~ ~e arid the wrongs advantage." (First found: t7 October 1997.) both examples present summaries as sentences which alinost seamlessly follow one another. While this may account for acceptable readability, it is at best misleading, as in the original documents these sentences are several paragraphs apart.
This makes it hard to know that the references to 'Flow bue, how true ; in the f rst exarnpIe, and 'Tlse way this has turned out ; in the second, are not whatever might be mentioned in the to preceding sumrxzary sentences, but are, in fact, hidden somewhere in the original text of the docuxnents_ Opening references to 'The problem ; and the agency', are hard to resolve. The thrust of the second article---namely that there is a reversal of an anticipated situation-is zxot at all captured: it turns out that the missing paragraphs between the summary sentencxs discuss how the planned 5,000 layoffs have been ~5 reduced to °4,000, then 1,400 and finally settled at about 500", and that °now, even those 500 workers will not be cut°. As it turns out, some ir~dicaxion to this effect might have been surmised from the full title of the article, Scheduled IRS Layaj~s For 500 Are Canceled; unforitmately, this has been ~ by a data reduction sh~ategy which is insensitive to notions of linguistic phrases, auxiliary verb constructions, mood, and so zo forrh.
In the extreme case, such summaries can range from under-informative {as illustrated by the first example above to misleading (the second example), to pL~inly devoid of any useful information. Another example from the same site reads:
zs Technology News from Wired Neiws Ttlis is more than 500 times thinner than a human hair.
"Don't expect one in a present under your Christmas tree this ya~ar.' 3o Accordingly, a particular problem that must be addressed is how to 'fill in the gaps° which the data reduction process necessarily intiroduces as a summary is con~ructed by choosing ceatain fzagn~ents hom the original source. Presently, lmown ways fvr filling such gaps, assuming of course these are cvea perceived, is by the a~ciive user involvement of requesting the entire document.

Id7-99 1 S : 30 FROM = SAWYER 2i~ ASSOC ( A'TES ( U : b510 493 4 b4'J- rAht ~/,sb WO 99I2b17Z PC'~YUS98~3g4 Currently, there is a relatively rigid mechanism, typically sensitive to a mouse click, or some similar interactive command, with the simple semantics of "bring up the entire document, possibly with the point of view focused on the particular sentence of the summary whickt received the click, presented in its natural document context, and maybe highlighted". Clearly, having a richer data structure would facilitate greater flexibility in intenactiox~s with what would be, in effect, a whole range of dynamically reconfigured sunlmaties at different IeveI of granularity and detail.
There is sill one problem, however: the prncGSS of filling in the gaps requires active user involvement. In principle them is nothing wrong with this_ In practice, real to information management environments involve working with a large nurztber of documents. It is far ;&om clear that users will have the energy, bandwidth, dedication, and concentration required to assns, absorb, and act upon summaries for each one of these documents, by clicking their way thmugh each member of a long static list.
Accordingly, what is needed is a system and meZhOd for prusentirtg a plurality of t s documents to a user in a more expeditious fashion than when utiliaag conventional techniques. In a preferred embodiment, the system and method should be able to analyze documents with multiple topics. The analysis would typically be used to produce summary-Iike abswa~ctions of the documents at varying levels of granularity and detail.
The system and method should be easy to implement and cost~.effective_ Furthermore, 20 the document presentation should contain; relevant information from throughout the document, not just a selection of sentences that may miss significant topics.
The system and method should allow the presentation to be sensitive to multilayer analysis, should be able to present salient and contexZUalixed highlights of a document and should make the document available to the user seamlessly, by an active user inte~Ce_ Finally, the 2s presentation should be adaptable such that a user decides whether helshe desires to be actively involved in the presentation. The present invention addresses these needs_ SUMMARY OF T'HE INVENTiUN
A method and system for the dynamic presentation of the contents of a plurality 30 of documents on a display is disclosed. The method and system comprises receiving a plurality of documents and providing a pltuality of topically rich capsule overviews corresponding 107-U9 1 S : 30 F'l2uM : SAWYI=:k ~ ASSOI.: t A TES t U : b5d 4U3 4549 rH~a, 1 ~di .s5 wo ~nsi7z Pcr~rs~z~, to the plurality of documents, The method and system also includes dynamically delivering document content dated in the plurality of capsule overviews.
In so doing, a system and zuethod in accordance with the present invention can present thematic capsule overviews of the documents to users. A capsule overiew is derived for the entire document, which will depict the core content of an average length article in a more accurate and representative manner than utilizing conventional techniques. The capsule overviews, delivered in a variety of dynamic presentation modes, allow the user to duickly get a sense of what a document is about, and decide 1 o whether they want to read it in more detail. If so, the system and method greatly facilitate the process of focused navigation into the parts of the document which may be of particular interest to the user.
In a preferred embodiment, the capsule overviews nnclude a containment hierarchy which relates the different information IeveLs in a document together, and l s which includes a collection of highly salient topic stamps embedded in layers of progressively richer and more iaformarive contextualiud text frag~ne~nts.
The novel presentation metaphors which the invention utilizes are based on notions of temporal typography, in particular for exploiting the iuteraetio~
between form and content.
1~R~EF DESCRON OF TH>E DR.AWrIVGS
Figure 1 is a block diagram illustratsag a canventiozlal ramputer system that serves as one type of operating enviromnent for the present inveutian.
Figure 2 is a simple flow chart illustrating a method for the dynamic presentation zs of a plurality of documents in accordance with the present invention.
Figure 3 is a flow chart of a system and method for characterizing the content of a doetmleat in accordance with the present invd'ttion.
Figure 4 is an example of an article and its segmentation into topically separate sections.
3o Figure 5 is a depiction of a TopicsTicker in accordance with the prrsent invention.
Figure 6 is a flow chart of the basic operation of a RSVP viewer in accordance 07-99 15:31 FROM: SAWYER & ASSOCIATES ID-650 493 4549 PAGE 11/35 wo ~rmn pc~rnrs with the present invention.
Figure 7A-7F are depictions of a RAPIB SERIAL VISUALIZATION
PR,IwSENTATIC?N (RSVP) views in accordance with the present invention.
Figures 8A-$C are depietions of a ViewTool Viewer in accordance with the s present invention.
DFSC)E~1~'IpN OF THE INVENTION
The present invention relates to the rapid presentation of the content of an average length document. The following description is presented to enable one of t0 ordinary skill in the art to make and use the invention and is pr~ovi~ded in the context of a patent application and its requirements. 'Various modiscations to the preferred embodiment will be readily apparent to those skilled in the art and the generic principles herein may be applied to other embodimtnts_ 'fllus, the present invention is not intended to be Limited to the embodiment shown hut is to be accorded the widest scope consistent l S with the principles arid fees desenbed herein.
A system and method in accordance with the preseat invention would typically be util~ed 1I1 a COnvenhbnal ComplttCr System.
20 'l. Overview 2 GynarreiG Pr,~sentation of Doa.~ment Content 2A. Capsule Overviews 2B. Salience-Based Content Characterization 2C. Anaphora Resolution and f~ca1 salience 25 2D. Discoerrse Salience and Capsule Overview

3. Capsule Overviews as Document Abstractions

4. Filling in ttye Csaps: User lnvoivement

5. Document Characterization by Topics 30 SA. Capsule OveNiew I~xampte

6. Temporal Typography for Dynamic Docarment Delivery

7. lTnualaation of Document Cornent

8. Dynamic Docunsent Viewers 3s so TopicsTicker viewer 8B. Rapid Serial Visual Presentation RSVP) Viewer 8C. ViewTool Viewer 8D. Viewer Summary 1. Overview A system and method in accordance with the present invention would typically be utilized in a conventional computer system.
Figure 1 is a block diagram illustrating a conventional computer system 100, which serves as one type of operating environment for the present invention.
The 1o computer system 100 includes a display 110, a keyboard 120, a pointing device 130, a storage medium 140, a processor 150, and a memory 160, which are all connected by a bus 170. The processor 150 operates in conjunction with the memory 160, which are all connected by a bus 170. The processor 150 operates in conjunction with the memory 160 to execute a rendering system 180 that enables the processor 150 to provide, and 15 present, the content characterization from text files stored in some form of computer-readable medium, such as a CD-ROM, or from a network. One of ordinary skill in the art should also recognize that the present invention could be utilized in a variety of data processing systems, and in particular, display devices, and its use would be within the spirit and scope of the present invention. For example, the present invention could be 2o utilized in Network Computers (NC) and their use would be within the spirit of the present invention. In another example, the present invention could be implemented by a server utilizing a technique in accordance with the invention to provide content characterization and a client could provide the resultant display.
The present invention provides a method and system for utilizing novel 25 presentation metaphors of documents that enables users to rapidly skim the documents in order to get the "gist" of their contents. This is accomplished through the dynamic presentation of topically-rich "capsule overviews" of documents. The concept of capsule overviews is described fully in U. S. Patent No. 6,185,592, entitled "A
System and Method for Characterizing Content of Text Documents" filed on even date 3o herewith, and assigned to the assignee of the present application .
By utilizing the capsule overviews of documents derived by a system and method rJ /-Sb 1 S : ,j 1 tIGUM : SHw YtrC d HSSUI. I H It,S 1 U : b510 4b.i 4 S4S
t'Hl.,t l .j/.iS
~~5~
in accordance with the teachings of the above identified application, a system and method in accordance with the present invention can oi~'er an entirely novel approach to the information overload problem. Using topically-rich capsule overviews, a system and method in accordance with the present invention can present thematic outlines of the documents to users. These overviews allow the user to quickly get a sense of what a document is about, daeide whether they want w read it in mere detail, and quickly navigate to the poirn(s) of document of particular intemsi to them. The following discussion will describe with particularity the dynanuc presentation of document content for average length documents.
2. I~va is Presentation yf Documen c'nnt~t Figure 2 is a simple flow chart illustrating a method for the dynamic presentation of a plurality of documents in accordance with the present invention. As is seen, first a plurality of documents are received by a data processing system or the like, via step 2p2_ t5 Next; a plurality of topically rich capsules overviews which corresponds to the plurality of documents, via step 204. Finally a plurality of doctuxients are presented, via step 206.
As before mentioned, by utili~ng the above-.described capsule overviews of documents, a system and method in accordance with the present invention can offer an effective solution to the information overload problem.
2o The solution in accordance with the present invention to the problem of effectively communicating to the end user the 'gist' of an on-Iine document, or of a collection of on-line documents, is based on the idea of relating form and content, by means of dynamic visual treatment of written language, or temporal typography.
Qnly recently has the possibility of escaping the static and rigid constrajnts of writing on paper 25 been fully appreciated. 'along; in Temporal Typography, Characterization of Tirne-Varying Typographic Forms (Master's thesis, MIT Media Lab, 1995), has stated.
"Imagine looking at a small area on a computer screen. Words appear and disappear on the screen one by one. As they appear, meaning is expressed as forms change .
dynamically over time_ The combined effect of the message, form and rhythm express a 3o tone of voice, emotion or personality as if you hear a person speak_ Although the two mediums, spoken and written words, are vastly different, the analogy may give you a sense of the expressive potential of temporal typography."

YJ ! -JJ ~ J - JG (-ICVt't - Jf-1w lGIC aY f-IJJVI. ~!-t ~LJ L U - OJYJ ~tJJ
'tJ'fJ C'rIVL a-f~ J:7 WO 99I~6172 PCI7LJS9$nd3$4 The notion, essentially, is to relate highlights of the core meaning of a message to ways of visually enhancing their impact, or at least mimicking (some ofj their semantic load. In the immediate context of this disclosure, this translates to questions of what might be appropriate visual metaphors for representing semantic objects like topical Phrases, shifts in discourse structure, or contextualization of information-bearing phrasal units.
There are Several appealing aspects to dynamically presenting abstractions of document content. The user need not be actively involved_ as documents arrive at the desktop, they can be analyzed and the resulting content ions can be displayed 1o autonomously. Should the user have the time or inclination to focus on a particular document, interactive controls will be at their disposal; alternatively, each new arrival can be presented under its own schedule, followed by another, and so on. The presentation cycle can be customized to make use of arbitrary combinations of granularity of expressiveness. Notions like semantic highlights and demarcation of ~s context are easily mapped onto visual metaphors, and thus natuFally support the expression of content by means of variations of form. CogAitively, short phrases with high semantic load are amenable to punctuated display following a natural rhythm of visual perception.
In s~mamaty, delivering document content abstractions dynamically makes it 2o possible to fully exploit a variable depth analysis of docents (which will be discussed in detail below), maintains synchronicity with the continrbous flow of information into one's personal workspace, and allows for smooth integration of passive absorption of the analyses by the end-user with active participation in more focr~d document perusal.
The following discussion in accordance with the present invention highlights a 25 document analysis teclenoIogy which seeks to derive docema~ent content characterizations designed to exhibit the semantic properties described above:
- collections of highly salient topical phrases, - embedded in layers of progressively richer and more informative contextualized text fragrnents, 30 - with contexts calculated as meaningful fiagsneats defined by a containment hierarchy of information-bearing phrasal units, and - organized as capsule overviews which track the occurrence of topical phrases and other discourse referents across the document discourse.

b J-59 1 5 : ;j~ tEtUM : SNWY~;12 t5. ASSO(! ! AiES ! 11 : bbb 493 454a t~H~.t t 5i .sb WO 9911.61T2 PCTNS9&1A384 Next, some essential features of temporal typography are dcscn'bed, as it relates to dynamic delivery of document content. These lead into some conclusions about interfaces for content visualization and, accordingly, a range of viewers designed for the purposes of rapid skimming of on-line documents in order to get the 'gist' of their contents is presented. First, however, the concept of topically-rich capsule overviews is discussed in some detail.
2A. 1 t0 A capsule overview is not a true s~mary, in that it does not attempt to convey document content as a sequence of sentences. Instead, it is a semi-formal and normalized representation ofthe document, derived after a process of data reduction over the original text.
Through capsule overviews, a document's content is characterized in a way that is t s representative of the foil flow of the document. 'this is in contrast to passage extraction techniques, whuch typically highlight only certain fragxner~ts. Also, capsule overviews are derived by carrying out linguistically int~sive analysis of the text in a doe~unent, which seeks tic prominence of linguistics expressions, rather than just occurrence of certain pre-specified, or highly frequent, words and phrases-thus the system and 20 method described here can be applied to any document, independent of domain, style or aenle.
A capsule overview is not an instantiated template. A primary consideration of the content eharacteri~on system and method desen'bed here is that they should not be specific to any docmnent source or type_ A rap~zle ov~riew is a coherently presented 35 list of linguistic expressions which refer to the most prominent objects mentioned in the doeumtnt, i.e., its topic starrlps, and furthermore provide richer specil'zcatiort of the relational contexts (e.g., verb phrase, minimal clauses) in which these expressions appear.
To fiuther illustrate the coneepts associated with a capsule overview, refer now to 30 the following news article shown in Table 1. (Maridng certain phrase units within single quotes is an annotation device, for subsequent references to the text from within this disclosure document; these annotations were not part of the original article) d/-99 15:3 rreUM:SAWY~1~ 1~ HS50GIATES 1U=bSd 493 4549 rYAGE 16/;35 W4 99r161?2 PCI'/t1S98n4384 Priest Is Charged with Pope Attack 'A Spanish t~riest~ was charged here today with attearpting to murder the Pope_ ',loan Fernandex Krohn', aged 32, was arrested after 'a man armed with a bayonet' apAroached the Pope while he was 5~ying prayers at Feti~na on Wednesday night.
According to the ppli~e, 'Fernandez' told the investigators toddy that 'he' trained for the past six rr,anths for the assault 'He' was alleged to have claimed the Pope looked furious" on hearing 'the priest's' a~dsm of his handling of the chctrch's affairs. If found guilty, 'the Spaniard' faces a prison sante~xe of 15-20 years"
IO
There are a number of reasons why the title, "priest Is Charged with Pope Attack°, is a highly reprcscntative abstraction of the content of the passage. It is encapsulates the essence ofwhat the story is about: there are two actors, identified by their most prominent characteristics; one of them has been attacked by the other, the perpetrator ha_s been charged; there is an implication of malice to the act The title brings the complete set of salient facts together, is a thoughtfully composed statement, designed to be brief yet informative. Whether a pres~ntt day nahixal language analysis program can zo derive--without being primed of a domain and genre-the information required to generate such a summary is arguable. ('This is assuming, of course, that natlnaI language generation techniques could, in their own right, do the pIa~ing and delivery of such a concise and information packed message) However, part of the task of delivering accurate content characterization is being able to identify the components of this 25 abstraction (e.g., 'priest', 'pope attack', 'cliarged with'). It is from these components that, eventually, a true sununary of this document would begin to be constructed.
It is also precisely these components, viewed as phrasal units with certain discourse properties, that a capsule overview should present as a characterization of the context of the document Accordingly, in the present invention, the most salient and 3o therefore most representative phrasal units, as well as the relational expressions they are associated with. are identified to provide the core content of the documeat.
To describe the generation of capsnlc overviews in accordance with the present invention in more detail refer now to Figure 3 and the accompanying text.
Figure 3 is a 07-99 15=33 FROM: SAWYER 8 ASSOCIATES ID:650 493 4549 PAGE 17/35 WO 99126172 PC't'~US98I?A3g4 flow chart of a system and muhod for characterizing the content of a documeat in accordance with the present invention. As is seen m the figure, first a list of discourse referents are provided within the docwment via step 302. Then, the document is divided into separate segments based upon changes in topic, via step 304. Thereafter, the s discourse referents are linked together into co-reference classes, via step 306_ Next, the salience for each afthe discow~e referents is rxIculated, via step 30$_ After those calculations are performed, then it is determined which discourse referents have the highest values within a segment, via step 3I0.
The core information unit that the invention concerns itself with is the set of to discourse referents in a document. Discoiuse referents are typically realized as aoun phrases. In essence, these are the entities-actors and objects--around which a story unfolds_ In order to determine, and maintain, an accurate model of what a docenn~t is about, it is >:lecessary to be able to identify the ways in which the same entity is referred to in the text, as well as to establish co-referentiaIxty among different hnentions' in the t 5 text of the same entity. The sample document in Table 1 provides examples of the same entity being referred to in different ways is the text {°priest', "a Spanish priest', "Fernandez", and "he", is the second paragraph, all refer to the sauce person), as well as of different entities being referred to by the same text string {"he' in the firrsl: paragraph refers to the Pope, while "he' in the second paragraph refers io the priest).
zo Thereafter, discourse referents with the highest salience values are labeled as topic stamps, via step 312. The local contexts arowr><d each of the topic stamrps are identified, via step 3I4. Finally, from this information a capsule overview of the document is constructed with, a given degree of gxanularity via step 316. A
key concept associated with generation of the capsule overviews is the calculation of salience values z5 for the discourse referents, which are then used for deter~aining topic stamps in the document. ?he following will discuss salience Eased calculations in more detail.
2B_ Saiien~~$ased Gonten~~'~haracte~~:~t:..., Salience is a measure of the r~eiative prominence of objects in discourse:
objects 3o with high salience are the focus of attention; those with low salience are at the periphery.
In an effort to resolve the problems facing a term-based approach to content characterization, as discussed in the background of the application, a procedure in YJ ~ - U .7 l J - J J C CC V 1' 1 - .7 H. W 1 G .C Ow H J J V 1. a H 1 G J L U
- O J VJ y ~ J '3 J Y J C' H La G l Cl / J J
wo 9grW n pcrms~rz0.~8.a accordance with the present invention has been developed which uses a salience feature as the basis for a "ranking by tmpOrc~" of an uttstrucitued referent set;
ultimately, this facilitates topic stamp iclentif~o~ By determining the salience of the members of a referent set, an ordering can be imposed which, in conutection with an appropriate choice of tlu~eshold value, permits the reduction of the entire referent set to duly those exprcssiozts that identify the most pronvnent participants in the discourse.
'This reduced set of terms, in combination with information about local context at various levels of ~anulanty (verb phrase, minimal clause, sentence, etc.) offers an accurate and detailed characterization of a documents content. This may rhea be folded into an appropriate t0 presentation metaphor such as that will be described 1>exeinafter.
Crucially, such an analysis satisfies some important requirements of usability of document content abstractions: it is concise, it is coherent, and it does not izttroduce cognitive overload. In a more general sense, this method utilizes a strategy for scaling up the phrasal analysis techniques utilized by standard term identification and template instantiation 15 technologies, which bas at its core the utilization of a crucial feature of discourse structure: the pronninence, over some segment of text, of particular refereuts.rsomething that's missing from the traditional ter~ology for 'bare' terrui~ology identification.
2C_ Anaphora Reselutier~ and .oval Saiie~cP
20 For the purposes of determining how discourse referents relate to objects in the world of the document, a simplif~ring assumption is made that every noun phrase identified by extended phrasal analysis constitute a -mention" of a participant in the discourse. In order to determine which expressions constitute mentions of the same referent, the method described here crucially relies upon being able to carry out anaphora 2s resolution and ca-referent ideutit~c~tion. T.~guistic expressions that are identified as coreferential arc grouped into equivalence classes, and each equivalence class is taken to represent a unique referent in the discourse. ?he set of such equivalence classes constitutes the full referent set from which, ultimately, topic stamps will be derived_ A distinctive feature of the anaphora resolution algorithm is that it has been 3o specially adapted to work from a shallow syntactic base: Specifically, ii does not require foil syntactic analysis of the text. This makes the method applicable to any text document, irtive of its domain, style, or genre. This type of anaphora resolution t1 I - J J a J - v v a at v a'a - .~ r. w W . vC Jw-v J .J U ~. a H 1 G J a U -U C.7 ELI -t ,7 .~ -a ~ ~s .r wy. a .r l v v algorithm is described, in foil detail, in the paper "Anap~re for Everyone:
Pronominal Anaphora Resolution Wirhout a Parser," by C. Kennedy and B. Boguraev, which was presented at the 16th International Conference on Corr~purational Linguistics, Copenhagen, 9Jenmark, August S-9, t 996.
The immediate result of anaphora resolution is to reduce the extended phrase set of all mentions of objects in the discotuse; the larger consequence is that it provides the basis for the identification of topic stamps, as ii introduces both a worming definition of salience and a formal mechanism for determining the salience of particular linguistic expressions_ This connection between anaphora resolution, co-reference identifrcaiion, s 4 discourse salience, and semantic prominence is described in fuller detail in "Anaphora for Everyone: Pronominal Anaphora Resolution Without a Parser," (C. Kennedy and B.
Baguraev, in Proceedings of COLIlVG 96 (16th International Conference on Compurarional LingwLrrics), Copenhagen,17K, August S-9, l 996) and "Anaphora in a Wider Context: T~ac~ing Discourse Referents" (C. Kennedy and B. Boguraev, in W.
l s WahLster, Editor, Proceedings of ECAl 96 (12th European Conference on .4rn'ftcral Intelligence), Budapest, Hungary, August I1-16, I99f. 3ohn Wiley and Sons, Ltd., ~,oadonlNew York)_ Roughly speaking, the anaphora resolution procedure locates an antexedent for an anaphoric expression by first cIiminating all impossible candidate antecedents, then 2o ranking the remaining candidates according to a salience measure and selecting the most salient candidate as the antecedent. This measume, which is referred to as 'local salience', is a function of how a candidate antecedent expression satisfies a set of grammatical, syntactic, and contextual parameters. These constraints are typically referred to as "salience factors". Individual softener factors are associated with numerical values, as 25 shown below.
°sent": tt?0 iff the expression is In the ccare~nt sentence.
"cntx": 50 ifif the expression is in the current discourse segments "subj-: 80 iff the expression is a subject"
30 "eost~: TO iff the exgrogsi4n is in an existential cor~ruCtivn.
-puss": 65 of the expresstoms a possesswe.
"ace': 5~ df the express~oms a direct object.
"dot': 40 iff the expression is an indirect object.

d ~-r~'.~ 15: s4 rttuM:SHwYtre 6 H5.7UL 1H1G5 ~U:0510 ~l~J "1J'tJ CHW G LYJ/
JJ
wo 99r~6 a ~2 Pcrrt~s~~sa ~oblq~: 30 iff the expression is the complement of a preposition.
'head: 80 iff the expression is not contained in another phrase.
"arg': 50 iff the exprg,~.sion is not contained u1 an adjunct.
The local salience of a candidate is the sum of the values of the salience factors that are satisfied by sorrxe member of the equivalence class to which the candidate belongs; values may be satisfied at most once by each member of the class. The most important aspect of these numerical values for our concerns is that they impose a i0 relational structure on the salience factors, which in turn provides the basis for ordering referents according to their relative prominence is the discourse (in other words, what is important is not so much the values themselves but the fact chat they denote that, for instance, 'subj ~ factor indicates higher prominence than "ace°, itself more prominent than °oblg", and so forth).
IS
2Y). Dl n s snle An important feature of local salience is that it is variable: the salience of a referent decreases and increases according to the frequency with which it is mentioned (taking into account subsequent anaphoric expressions). When an anaphoric link is 20 established, the anaphor is added to the equiwale~nce class to which its antecedent belongs, and the salience of the class is boosted accordingly. If a refcxent ceases to be mentioned in the text, however, its local salience is incrementally decreased;
this reflects decay in its prominence. This approach works well for the purpose of anaphora resolution, because it provides a realistic representation of the antecedent space for an 25 anaphor by ensuring that only those zeferents that have mentions within a local domain have increased prominence. However, the ultimate goal of salience-based content characterization differs from that of anaphora resolution in an important rGSpect_ In order to determine which linguistic expressions should be presented as broadly representative of the content of a document, it is necessary to generate a picture of the prominence of 3o referents across the entire discourse, not just within a local domain.
For illustz-ation of the intuition underlying this idea, consider the news article discussed in Table 1. intuitively, the reason why "priest' is at the focus of the title is that there are no less than eight references to the same actor in the body of the story (markod 07-99 15:34 FROM: SAWYER 8 ASSOCIATES IO:650 493 4549 PAGE 21/35 wo v9r~ c ~a Pcrnls by single quotes in the exaruple); tnoreover, these references occur in prominent syntactic positions: five are subjects of main clauses, two are subjects of embedded clauses, and one is a possessive- (This example also illustrates the rationale behind the above-described salience factors.) Similarly, the reason why °Pope attack" is the secondary object of the title is that a constituent of tl~e compound, 'pope", also receives multiple mentions (five), although these references tend to occur in less prominent positions (twa are direct objects).
In order to ge~tera~te the broader picture of discourse structure needed to inform the selection of certain expressions as most salient, and therefore most representative of la content, an elaboration is introduced of the local salience computation descn~bed above that uses the same conditions to calculate a non-decreasing, global salience value for every referent is the text. This non-decreasing salience measure, which is referred to as 'discourse salience', reflects the distributional properties of a referent as the text story unfolds- In conjunction with the "tracking" of referents made available by anaphora I5 resolution-as discussed at some length in "Anaphora in a wider context:
Tracking discourse referents" (C. Kennedy and B_ Boguraev, in W. Wahlster, editor, Proceedings of ECAI 96 (12th European Conference ore .4rtif~cial .Intehfgence), Budapest, Hungary, Aug. I I-16, 1996. 3ohn Wiley and Sons, Ltd, London/New'York~-discourse salsenee provides the basis for a coherent representation of discourse structure that indicates the z0 topical promizzence of individual mentions of referents in isolated segments of text.
Most importantly, discourse salience provides exactly the information that is needed to impose the type of importance-based ranIdng ofreferents which is required for the identification of topic stamps- Specifically, by associating every referent with a discourse salience value, the topic stamps can be identified for a segment of text $ as the 25 n highest racked referents in S, where n is a scalable value.
The notion "segment of text" plays an extremely important mole in the content characterization task, as it provides tree basic information-structuring units around which a capsule overview for a document is constructed.. Again, the example from Table 1 gives a useful illustration of the important issues- Tlle reason that the tide of this passage 3o works as an overview of its content is because the text itself is fairly short. As a text increases in length, the "completeness' of a short description as a characterizaxion of content deteriorates. If the intention is to use concise descriptions consisting of one or d'!-99 1 S = :34 tRUM : SAWYER 8 ASSOC 1 ATES 1 U : bSd 493 454U YHt~t Lti,sS
WD 99lZ6I72 t'CT/US9'81x4384 two salient phrases--i.e., topic stamps-along with infatmation about the local context in which they appear as the primary information-beazing emits for a capsule overviev~r, then it follows that texts longer than a few paragraphs must be broken down into smaller units or °segments".
In order to solve this problem, a document is recast as a set of "discourse segments", which correspond to topically caheretlt, contigetous sections of text. One approach to segmentation which works well for the p>aposes of this method implements a similarity-based algorithm along the lines of that described by Hearst, in her paper entitled °MuIti-Paragraph Segmentation of Expository Text." {M. ~earst, in 32nd .4~snual t o Meeting of the flssociation for Computationarl Linguistics, Las Cruces, New Mexico, 1994), which identifies discourse segments in text using a lexical similarity measure. By calculating the discourse salie~ace of referents with respect to the results of discourse segmentation, each segment can be associated with a listing of those expressions that are most salient within the segment, l.c., each segrntent can be assigned a set of topic stamps_ 1 S The result of these calculations, namely the set of segment-topic stamp pairs, ordered according to linear sequencing of the segments in the text, can then be resumed as the capsule overview far the entire document. In this way, the problem of content characterization of a large tent is reduced to the problem of finding topic stamps for each discourse segment.
3. ~ys 1e Overviews as Dr~pn~~p~ Abstractions Striving to balance the conflicting requirements of depth and accuracy of a surnmacy with those of domain- and genre-independence, the notion of a capsule overviews has been developed as content abstraction for text documents, explicitly 2s designed to capture "aboutz~ss"_ Qne of the problems of information management, when presented with a growing surplus of text documents, is getting some appreciation--rapidly, compactly, and yet with a usable degree of depth and representativeness--of the info~on contained in a doclmment Informally, this is usually referred to as the "aboutness" of a document, and is represented as a set of highly salienx, and by that token most representative, phrases in the dacumtnt By viewing topicality in its stricter;
linguistic, sense, the previous section defined topic stamps to be the most prominent of these phrases, introduced into, a~td then elaborated upon, the document body.
4n the ~0/-bb 1S=.55 teGUM:SHWYt.t< 6 HSbU(_.lHlt.S IU:bSlO 493 4S4b YHI.;~. G..i/aS
wo ~armn pcrms9sna~a basis of this definition, the above-identified computational, algorithmic, procedure has been developed for generating a set of abstz~ions for the core meaning in the document, ultimately resulting in a capsule overview of the docuanent based upon suitable presentation of the most representative, and most contentfuI, expressions in the text.
These abstractions comprise layered and inter-related phrasal units at different levels of granularity and depth of document analysis. To further describe this concept of granularity refer now to the following discussion.
Gzanularity is closely tied to context. In general, the information in a given sentence is best expanded by being able to position this sentence in its paragraph context;
t0 likewise, the theme and topics) in a paragraph can be further elaborated by relating the paragraph to the segment of discourse which encompasses the theme in its entirety. This is a natural containment hierarchy; relating the different information levels in a document together. Such a hierarchy can also be extended in sub-sentential direction-phrasal units indicative of topicality axe clearly wholly contained in sentences;
15 fiuthcrmore, a phrasat containment hierarchy could also be utilized to provide contextualized information concerning the topical phrases themselves.
Imagine that in the second example above (Example 2, page 5) some mechanism has determined that the phrase ~°Scheduled rRS Layoffs °is topically indicative- Assuming some focused mining in the vicinity of such an 'anchor' by a phrasal grammar of a certain zo type, this topic phase could be further eontextualized to "Scheduled IRS
LayoJj''s For 500 Are Canceled ". This is an example of phrasal containment of information-bearing phrasal units- Similar expansion of topic in context might yield, for the initial discourse segment of the document, progressively larger and more informative fiagmcnts fioim it.
25 Egam Ip~-4 TOPICAL PHRASE: -Scheduled IRS t.ay~s TOPIC IN RELATIONAL CONTEXT: -there wltl be no Iayroffs' TOPICAL SENTENCE: ~ Yesterday, the IRS said ~,~w~lt be no w fl SENTENCE tN PARAGRAPH CONTEXT: (ore than a year ago, The lnteernal Reverwe Service pla»r~ed widespread job nuts.
Y

1!J /-~U 1 5 = J5 tKUl"1 . aHW YCtG d H55U1.. t H LGS 1 a : O.W 1(J y~"f 't J'-1.J C-H~aG G't/ JJ
wo mzsm ~c: rms9sr2t3s~s PARAGRAPH WITHIN TOPICALLY
COHERENT AISCOURSE THE111aE: ~Iona than a year ago ~j~
lalrot~. ' Confrw~ed wfth ~Oniorral criticism acrd calls for reform in light of some I~dy pub~ed reports of abuse acts toward taxpa~, as well as stau»d~r union opposition td the cress, the IRS said 10 er»ploy~ees at risk of losir~g their jobs would be r~igrred to impwe t~tome~' service, leelp taxpaY'~rs rasphre peps and irrcrgase compJiartce with tax laws.
is This example illustrates the notion of granularity ofdocument analysis, and is especially indicative of how a contaimnent hierarchy of Iayerod information~&om very compact and representative topical phrases aiI the way to full and rich discourse segments--can be utilized to represent and zrzaintain strong notion of cont~xtualization in a document abstraction.
2o The example also shows the value of being able to identify phzasal units s~nalier than sentences, arrange them in layers corresponding to the informational containment hierarchy, and perform certain semantic operations over them. These operations fall largely in the area of reference identification, co-rcfcrentiality, and topic tracking;
consider, for example, the processes of relating 'layoffs °to 'scheduled IRS layoffs ;
z5 identifying ~ntemal Revenue Service "and X125"as referring to the same abject, resolving anaphora in general, and so forth.
4_ F'rliing.io the Ga~~- U r ~nvoWe~ent rt is clear that granularity of analysis and contaiztment hiecarclxy of inforrization-3o bearing phrasal units with different (yet complementary) discourse properties and function could be utilized very effectively to implement a "zooming' fimction into andlor out of a given doctimcnt. In this way funding ont more~of what is behind a document 'summary' is, in effect, filling in the gaps in such a stmzrnary in a controlled fashion, guided by incrementally revealing pmgrcssiveiy larger and more informative contexts.
3s Conceptually, this is not dissitniIar to the notion of 'percentage of shrink factor", typically utilized by sentence-based sumursarizes, where a user can specify that a document should be condensed to N percent of its foil extent. There is, however, a crucial difference here. When re-casting a document from, say, I O% to 20~/o shrink 10/-~~ lb:,j5 tttUM:SHWYthC t1 H55Ul.1HLtb tU-b510 4~.i ~15~1~ YHlat Lb~.fS
w4 99rW n pc~rr~s~243~
z~
factor, there is no way to specify ahead of time, nor to kaow after the event, how the additional sentences relate to she original IO%. In contrast, when a document is re-cast in terms of infornnation-bearing units a Ievel higher than what its current representation uses--for instance, as a set of relational contexts immediately surrounding its topic stamps--there is a guarantee that the user's sense of what the document is about is Incrementally and monotonically tnriched.
This makes it possible to use the capsule overview technology in accordance with the present invention to enable a user to get an mediate and accurate imprarion of what a particular document is about~ As a capsule ovc~view is a small window into the 1o core content of a document, it is a useful abstraction for compact representation of content Once engaged, however, the user can still use this window to drill, arbitrarily deeply, into the underlying information layers. Before discussing the approach to visualization of document content and presentation metaphors for using capsule overviews as mediators, and facilitators, of dynamic document content delivery, the basic t s notions of topically rich capsule overviews as layered abstraetiotss of document content are exemplified hereinbelow.
5. m r 20 5A. Capsule Overview ~xatnple The following discussion describes an example of an article the analysis of which utilizes the present invention. As descrl'bed in sections 2 and 3 above, the operational components of salience-based content characterization fall in the following categories:
discourse segmentation; phrasal azralysis (of nominal expressions and their relational 2s contexts); anaphora resalutian and generation of a referent set;
calculation of discourse salience and identification of topic stamps; and enriching topic stamps with information about relational context(s). Some of the fimctionaliry follows directly from technology developed for the purposes of phrasal identification, suitably augmented with mechanisms for maintaining phrase containment; in particular, both relation 3o identification and extended phrasal analysis arc carried out by nmning a phrasal grammar over a stream of text tokens taggod for lexical, morphological, and syntactic information, and for grammatical function; this is in addition to a gz-ammar minixag for terms and, generally, referents.

d'!-99 1 S : ;~e rRUM : SAWYER ~ ASSOC: l A TES l D : 650 493 4 S49 PAGE 16/

WO 991Z6h2 2.2 P~'T1IJS98~24~84 In a preferred embodiment the base level linguistic analysis is provided by the LINGSOFT supertagger, see F. KarLsson, A. Voutilainen, I. Heikldla, and A.
Antilla, "Constraint Grammar; ,q ~~ge ~'~~ Stem for Parsing Free TexP, iViouton de Gruyter, 1995. The later, more semantically-intensive a}gorithms are dtsa-ibed in s detail in "Anaphora for Everyone: Pronominal Anaphora Resolution Without a Parser"
{C_ Kennedy and $_ Bogutaev, is pr-oceedings of COLfNG-96 (16th International Conference orr Cornputarional Linguistics), Copenhagen, DK, 1996) and 'Anaphora in a Wider Context: Traeiang Discourse Referents' (C. Kennedy and B. Bogutaev, in W.
W~lster> editor, Proceedings of EC.4X 96 (12th European Conference on Artlfu:ial l a Intelligence), Budapest, Hungary, I 996. John Whey and Sans, Ltd, LondonlNew York).
The procedure is illustrated by highlighting certain aspects of a ~Ie overview of an article 400 shown in 1r figure 4. 'I'l~e doc~une~nt is of medium-to-large size (approximately fow pages in print), and focuses on the strategy of Gilbert Amelio (former CEO
of Apple Computer) concerning a new operating system for the Macintosh. Tvo long to quote is here in full, the followizxg passage from the berg of the article contains the first, second and third segments (shown at 402, 404, and 406 in Figure 4), as identified by the discourse segmentation component. (In the figure, segment boundaries are marked by extra vertical space, this markup is far illus~abion purposes only, aad indicates the result of running the discourse segnea~tation algorithm. No such demarcation exists in the 2o source of the article itselfj.
The capsule overview was automatically generated by a fully implemented, and oP~o~, system, which incorporates all of the processing components identified above_ The relevant sections of the overview of the article 400 (for the three segments of the passage quoted) are shown in Tables 3, 4 and 5 below_ '-5 The topic stand for the three segments ~t02, 404, and 406 corxstitute the core data out of which a capsule overview is constructed; these are shown underlined immediately following the segment sequence identifiers (in square brackets}_ The capsule overview itself displays the topic stamps (highlighted in single quotes) in their relational contents.

YJI-'JJ aJ-.~O a /tV1'1-JHWIGIC als H.~JU1..~H~GJ aa-UJYJ '-3JJ '-tJ'-tJ
rHl.aC. L// JJ
WO 99I26171 PCTI(JS98~l1~iB4 f~I : _ 'Apple' would swoop in and take Wicrosoft's~ customers?
'Apple' lost 1816 million;
'~i~aoft' made 32.2 billion.
'MicrasoR' has a market valeae thirty times that of /Apple' it makes sense for'Apple' 'Apple' is in a position 'Apple' needs something dramatic [2j ~ machines; o~h'ne sy~~n Today's 'desktop machines', he [Gilbert Arnelio~ says Tomornow~s 'machines' must accommodate rivets of data Time to scrap your'opesating system' and start over The 'operating system' is the software that controls to go with the 'reengineered operating syst,~' Tably4 zo f33 Gitbert Ameho; new ot~tina ss~~
'Amelio', 53, brings a lot of credibility to this task 'leis' [Gilbert Ametioj resume includes where is 'Arrwlio' going to get this 'new operating system'?
radical redesign in 'operating synsms' that 'Amelio' is taltting about 2S Table 5 'fhe division of this passage into segments, and the segment-based assignment of topic stamps, exemplifies a capsule overviear~s ~actdng" of the underlying coherence of 3o a story. The discourse segmentation component recognizes shifts in topic-in this example, the shift from discussing the relation betr~reen Apple and Microso~
to same remarks on the future of desktop computing to a summary of Amelio's background and plans for Apple's operating system. Layered on top of segmentation are the topic stamps themselves, in their relational contexts, at a phrasal level of granularity.
35 The first segment (T'able 3) sets up the discussion by positioning Apple opposite Microsoft in the marketplace and focusing on their major products., the opa~ating rl ~-bb 15: Jb tICV!'1--JHWYLIC Q HJJUl.IHIG.J lU:bSd c~b~y ybyb t'HhL LC//,35 WO 99r1617Z PGTN99&Z<384 systans. 'ibe topic stamps identified for this segment, 'apple' and "microsoft", together with their local contexts, are both indicative of the introductory character of the opening paragraphs and highly representative of the gist of the first segment. Note that the apparent uninfvrmativeness of some relational contexts, for example, '...
'Apple' is in a position ...', does not pace a serious problem_ An adjustment of the granularity -at capsule overview presentation time (see below)--reveals the larger context in which the topic stamp occurs (e_g., a sentence), which in turn inherits the high topiealiry ranking of its anchor. "'Apple' is in a position where standing pat almost certainly means slow death"
14 For the second segment (Table 4) of the sample, °operating system"
and adesktop macb~ines" have been identified as representative. The set of topic stamps and contexts illustrated provides au encapsulated snapshot of the segmeat, which introduces Amelio's views on coming challenges for desktop machines and the general concept of a~n operating system. Again, even if some of these arc somewhat under-specified, mare is detail is easily available by a change ~ g~nularity, which reveals the definitional nature of the even larger context "The 'operating system' is the software that controls how your computer's parts_.."
The third segment (Table S) of the passage exemplified above is associated with the stamps aGilbert AmeIio' and 'new operating system". The reasons, and linguistic 2o rationale, for the selection of these particular noun phrases as topical are essentially identical to the intuition behind "priest" and "pope attack" being the eeutral topics of the example in Table 1 _ The computational justification for the choices lies in the extremely hipJh values of salience, resulting from taking into account a number of factors: co-refea~entiality between 'amelio' and °Gilbert Ameiio", co-refereatiality between "amelio' 25 and "his°, syntactic prominence of "amelio' (as a subject) promoting topical status higher tlaar~ for instance 'Apple' (which appears in adjunct positions), high overall frequency (four, countixzg the anaphor, as opposed to three for "Apple"--even if the two get the same number of text occurrences ins the segment), and boost in global salience measu~, due to "priming' ef~'ects of both referents for ~Gilbert Ameiio' and "operating system" in the 3o prior discourse of the two preceding segments. Compared to a single phrase sumumary in the form of, say, 'Amelio seeks a new operating system", the overview for the closing i0'7-99 1 S = :i'7 r'RUM : SAWYE.t~ ti ASSU~: f A f'ES l U : tiSld 4~..i ~i b~l~ r-H(.,a L~/ "y5 w0 ~nbin a segment comes close; arguably, it is even better than any single phrase sunu~y.
As the discussion of this example illustrates, a capsule overview is derived by a process which facilitates partial understanding of the text by the user. The final set of topic stamps is designed to be representative of the core of the document content. rt is compact, as it is a significantly cut-down version of the full list of identified terms, It is highly informative, as the terms included in it are the most prominent ones in the document. It is repr~rentasive of the whop document, as a separate topic tracking module effectively maintains a record of where and how referents occur in the entire span of the text. As the topics are, by definition, the primary eonte~nt-bearing entities in a t 0 document, they offer accurate approximation of what that document is about.
6. Temporal Ty~;~uhv for D~~am~ic Docament Deljverw bynamic content delivery is based on ideas of temporal typography developed by Wong ('1''_~'_ 'along, ~'emporal typography, characterization of time-varying typographic is forms, Master's thesis, MIT Media Lab,1995). This work develops a synergy of psychological studies of reading, gxaphic desagn, and temporal presentation of text.
Graphic design history is rich with examples of exp~menting with visual treatment of written language. Designers have begun to explore temporal prcstntation of text in television and film media. Studies of reading, which to a large extent form the basis of z4 Wongs work, have explored dynamic presentation of content, related to the interactions between meaning and intent of a text-based message. However, Wong's studies of the dynamzc relationship between meaning and delivery formats assume that the annotations for meaning in her experiments have been done by hand. ~a eontxast, this invention is concerned with leveraging an automatic document content analysis technology, capable z5 of delivering meaning analyses and content abstractions precisely of the kind which can be effectively coupled with dyztamic content delivery.
'7. 'Visual' lion o Docntnent .otiteot Previously, the predoz~ninaut current mechanism for mediating the spectrum between a summary of a document dad a corn~plete version of the same document was briefly discussed. In addition to a dizect hyperteact rendering of extracted sentences, in their full document contexts, two variations on this approach are the VFSPA
slider and ID/-~~ 15_..i/ tt2UM=SHWYtPC t1 HSbV1_.lHlt.S 1U-b51~ 4~.i ~ib~i~ rHlrt .3i0/.35 WO 99IZ6172 PGTIUS9l324,384 HYPERGEN. VESPA is an experimantal interface to Apple's sentence-based stunmarizer (Advanced Technologies group, Appte Computes, Cupertino, CA, Apple Information Access Toolkit: Developer Notes and AP~s, 199'7), whose main feature is a slider which dynamically readjusts the shrink factor of a document summary.
HYPERGEN exploits notions of phrasal contaimxnent within sentence units, in an attempt to elaborate a notion similar to that of granularity of aaatysis and context introduced earlier in this document: in a process called sentence simplification, Mahesh (K. Mahesh, Hypertext summary e~ctraction for fast document browsing, in Proeeedirrgs ofAAAI
Spring Symposium on Natural Lm~gur~uage Processing for the World Aide Web, pages 95-t4 I 04, Stanford, CA, 19975) uses phrases as °senteace surrogates", which are theca straightforwardly as hypertext links to the sentezxces themselves. .
As part of ~ ongoing investigation of ving large information spaces, researchers at Xerox PAltC have looked at a variety of structured data types (such as Izierarchically structured data, calendars, and bibliographic databases)_ Some general t 5 principles derived horn that work have been applied to documents: the DOCUMENT LENS is a teclanzqne for viewing 2~?5 information, designed far component presentations of multi-page documents_ Vtrithout going into detail, what is of particular relevance here is the strung notion offocus plrar context which drives the design. The visualization, b~owever, does little in terms of using any kind of document 2o summary or other abstraction, and is of a predominantly static nature (even though it is eotmmely respopszve to user interaction, as it attempts to combine a 'bird's eye view' of the entire document with a page browsing metaphor). More recently experimental prototypes have bees developed for interfaces which treat term sets {in the information retrieval sense, i.e. flat lists of index terms) as document surrogates: the focus of such 25 designs is visually on presenting notions like distribution of terms aa~oss the document, and on mediating access to local context for a given term (R. Rao, J.G.
Pedersen, M.A-Hearst, J.D. Macinlay, S.K. Card, L.Masinter, p.-K. Halvotsen, and G.G.
Robertson, "Rich interaction in the digital library", Communication of the ACM. 3$(4}:29-39, 1995;
M.A_ Hearst, "Tilebars: Visuaiizatxon of team distribution information in full text 3o information access," in ACM S,~C"rCHl Con~'ereace on Ifuman Factors In Computing Systems, Denver, CO, 1995). Ultimately, however, these interfaces still offer only a direct link between two states, the document surrogate and its full form.

d/-Sb Lb:,s/ tt~CUM:SHWYt,tt i5~ Hb'-.~Ul.iNlt.S lU:tiSld x]9;3 4b4S (-'HC:6 .!1/:3'~
WO 99I261~ PCTNS9iilZ4384 With the migration of news delivery over the World Wide Web and the growth of information 'push' vendors, some new methods are begimung to emerge for presaitation of news stories which use notions of dynamic delivery of content. Most of these are variations on the same theme: news delivery using a ticker metaphor. Thus both ABC's s news site (httpJlwww_abc_com) and Pointeast (httpl/www.pointeast.eom) employ a traditional horizontal ticker, CNN Interactive (httpJlwww.enn.com) arrange their ticker vertically, while CBS {httpJlwww.uttm.com) combine a ticker with photos from a major story.
The important insight here is that tickers aze dynamic objects, which can be ~ 0 programmed to continuously update themselves from a news feed and to cycle in a pre-defined regime, therefore not requiring user intervention. Furthermore, they can be dispatched to an area of the workspace {moruitor screw) where constant, yet inobtrusive, news delivery can take place in the periphery of the user's main activity-_ thus a choice exists between proactive cngageznent with the news source, and passive {and almost ~ 5 subliminal) monitoring of news data.
None of the examples above, however, combines a ticker with an automatic summarization engine. To a large extent this is because sentences--especially ineonseeutive ones, in the absence o~ visual markers for discontinuity-~do not Iend themselves easily into the word by word, left to right, presentation mode_ This is clearly 2o a situation where phrasal units of a sub-sentence granularity can be utilized much more effectively. In addition, psychological experiments on active reading {Y.Y.
along, Temporal typography, charaderizzaiion of time-varying typographic forms, Master's thesis, MTT Media l.,ab, 1995) show that when text is presented dynamically in the manner of a ticker, subjects' reading speeds are significantly slower than for text is presented scatisiically. Ott the other hand, dynamic presentations of text which show words or short phrases in the same location, but serially, one after the other, have reading speeds comparable to those for normal static texts.
To dare, no applications have been developed utilizing temporal typography for dynamic delivery of content abstractions. along has looked at how dynamic type in 30 general can be used for four different communicative goals: expressive messages, dialogue, active reading and real time conversation. Most relevant to this discussion are her experiments on active reading. In one of these she used a basic RSVP
(Rapid Serial 07-99 15:38 FROM: SAWYER 8 ASSOCIATES ID:650 493 4549 PAGE 32/35 W() 99126172 PCTNS9&24384 V isual Presentation) method (words or phrases presented sequentially one after another, on the same line and at the same position) to deliver a sequence of news headlines. In a second set of experiments called HIGHWAY NEWS, three dimensions are utilized, combined with a zooming motion, to present a sequence of text highlights.
°News headlines are placed ane after another in the z-dimension. Headlines are presented serially according to active input from the reader. The reader presses a mouse button to fly through the rows of headlines-as if flying over a highway of text.' These experiments show the strong feasibility of high impact, low engagement, delivery of senaanficaIly prominent text 5ragments being utilized as a powerful technique for t0 visualizing certain types of inherently linear information.
None of the work cited above relics on automatically generated meaning absu~actioz~s as its input; yet, it is clear that the topically-xxch capsule overviews generated by the document analysis t~hnology discussed in sections 2 and 3, and exemplified in section 5, are just the kind of serna~ntic highlights which Wong's t s experiments in active reading assunne. Conversely, up till now there has been no thought as to how the nature of topic-based capsule overviews in particular would fit the notion of dynamic type. This is a key feature of the present invention, and is discussed below.
8. jZynamic Doetrment Viewers 2o Below will be described three embodiments of systems or viewers for providing dynamic presentation of documents. These three embodiments will be hereinafter rcfcn~ed to as the TopicsTicker Viewer, Rapid Serial Visual Presentation (RSVP) Viewer and the'ViewTool Viewer. It should be understood that although these three viewers are descn.'bed, one of ordinary skill in the are recognizes a variety of views could be utilized 2s and they would be witb~in the spirit and scope of the present invention. To more particularly describe these err~bodiments refer now to the following discussion inn eo~njunction with the accompanying Figures.
The above three embodiments pmvide different dynamic views of document content. The difference is largely due to the variety of operational environments in 30 which the viewers have bees applied. A variation on a news ticker is designed to be deployed in situations where screen real estate may be at prerniu~an, or where several different channels of i~ciformation may be delivered simultaneously to the same 'in-box';

rJi-.» 15=JO t'iCUM_~HWYtK LSe H55U1..1H1G5 lU-d5lO y5.3 y5y~ t'Hl.at .SJ/J5 WO 99n6I'~2 PCTIUS9~81Z4384 typically such situations assume that users would only want to get a very general idea of document content. For situations where more screen teal estate might be avaiL~ble, and/or it may be known ahead of time that more detail concerning document conterrt might be reguired, a different viewer develops ideas from rapid serial visual presentation (RSVP).
~Cet another interface caters to the need to be able to get immediate access to the foil text of a document, without losing the benefits of rapid sidrntnzug through content highlights while fully maintaining information about the larger context.
All of the viewers assume an environment where incoming docurncnts get analyzed to capsule overview level (See Figure 3); the results of the analysis are to embedded into the original text by means of; for example, special propose tags.
8A. 'C~yics ici er Viewer TopicsTicker Viewer as showfi izt Figure 5 is a minimalist, hands-free, peripheral vision-directed ticker tape, with certain aspects of its display tied far serial delivery of a document's topic stamps: the string in the left panel is the document tide, and the right gavel is where the display cycles, contimlously, through the document's topic stamps_ When running over a set of documents, switching from one docttmerrt to the next is cued by a color change and a vertical scroll.
Z0 8B. $~RisL,~~rla1 Vi~~al P_resentaltion IRSVPI Viewer Figure 6 is a flow chart of ttie operation of the RSVP viewer. First, a list or plurality of consecutively ordered documents arc provided, via step 602. Next, the first document in order is selected, via step 604. Thereafter, the first topic stamp from the current document is selected from the capsule overview, via step 606. The relational z5 context for that topic staulp is selected from the capsule overview, via step 608. 'xhe canonical form for that topic stamp is retrieved, via step 610. Then the topic stamp is displayed in the background, in translucent type, and its associated relational context is displayed in the foreground, in heavier type, vita step 612. l~Iext, a time period for displaying the relational context is calculated, via step 614_ This time period is based 3o open the amount of information displayed. The topic stamp and relational context are then displayed for that time period, via step 616. Xt is then determined if there are other topic stamps, via step 6I 8. If there are other topic stam~ps,'thcn the next topic stamp is ~O /-bb 1 5 : ,3ty rtcUh! : SHw Y trC t~ H55U1. 1 H 1tS 1 U : b510 ~-1'-J.i ~-! 54b t'HCat .f4/ .ib WO 9912dI7Z PCT/US9&14384 selected, via step 620 and its associated relational context is selected, via step 608, and the cycle zs repeated. rf on the ocher hand, there arc no other topic stamps, a visual worker is displayed, indicating a doctuneat change, via step 622. It is then determined if there arc other documents, via step 624_ If tyre are no other documents then recycle s back from the fast document. If these are other documents, then the next document ins consecutive order is selected, via step 626. ThereaRer, its first topic stamp is selected, via step 606, and the cycle is repeated.
This continuous cycle defines the basic mode of operation for RSVP viewer. It can, as will be discussed below, be interrupted by the user at any time.
to In its basic mode with no user interaction, the RSVP Viewer cycles through all salient relational contexts izl a docxnment, maintaining the order in which they appear in the text As shown in Figures 7A-78, each context phrase is displayed as the prox~ain~ent object on the screen 702; at the same time the context is overlaid onto topic expansions (displayed as translucent text} 7Q4. This facilitates feather interpretations of the context ~ s strings by the user: expansions relate phrasal contractions in the string displayed tv their full canonical forces in the text, make clear ents of dangling anaphors, and so forth. Note, for instance, the backgrolmd display (Figure 7C) 704 of the full form of the antecedent for the anaphoric 'he"in the foreground 702. in this particular context, "he"
has been resolved to "Ctlbert Armeho".
2o Cycling though the complete set of salient contexts, in their original sequence, offers a good indication of aboutness at a given level of depth and detail.
Granularity of display is adjustable via a parameter. thus RSVP Viewer could be reduced to a TopicsTicker Viewer by only cycling through the document's topic stamps, or it could be used to display sequences of sentences. Relational contexts offer just the right balance ~5 between terseness (phrases are more easily perceived and assimilated than sentences) and informativeness (phrases larger than 'bare' topic stamps convey richer data}.
The amount of time a phrase is displayed is dynamically calculated, based on studies of aciivc reading and perception; the intent is to optinnize the full document display regime so that individually, each phrase can be processed by the user, while globally, the entire set of 30 salient contexts can be cycled through rapidly.
'lbere are provisions for maintainitxg overall context, by continuously displaying the title of the current document T06, as well as for allow ng context svi~itching, by user 15=J~ tlCUM:SHWfGtC Cn H55U1..1H~GJ ~U.O~~O -1JJ -tJwtJ ~ r-vw: vJi vv WO 99JZ61T1 PCTJU.S9811~384 3t selection of a document from a pop-up menu 708 (co-located with the title), via for example by a mouse click.
RSVP Viewer is designed as an entirely autonomous viewer. a,Rer all the content highlights in a document have been displayed, the next document will be loaded and the cycle repeated (just like in TopicsTicker Viewer, a suitable visual cue signals document change)_ This makes it very appropriate for situations where readers do not have much time, bandwidth, or opportunity, to interact with the display, but they would want to be peripherally aware of new documents that come into the system. On the other hand, if a particular context catches the user attention, a 'zoom' mechanism makes use of the t0 multiple levels of analysis of the document (as defined via the containment hierarchy discussed in Section 3, "Capsule Overviews as Document Abstractions" ).
Tbas will reveal, on demaad, progressively larger and more well detailed document fragments: sentences, paragraphs and segments. Figures 7B, 7D and 7E
viewed in this sequence give an indication of hove progressively informative contexts are t5 revealed on demand: '~ displays a togic and its content, in Figure 7D the latter is contextualized to the sentence containing it and in Figure TE, this sentence is displayed in the content of the relevant (containing) paragraph. Foreground aad background information are always differentiated by hierarchies of type. The transition between displays - e.g_, from 7B to 7x?, or from 7D to 7E, is accentuated by a visual "zoom"
zo mechanism. In such a hierarchy, about the larger context is always easily available: for inst~ce, further specifics concerning what 'fee says "(see Figure 7c) is imz~nediately available by a single click in the display area, the result of which is shown in Figure 7F..
At any given point of time, and depth of detail, the display uses a combination of visual cues to highlight the information-bearing unit which is in focus, and associate this 25 with the larger context in which it appears in the original sow~ce. In particular, properties of type, both static and dynamic, come to convey various aspects of the document analysis: primary focus of attention is denoted by using heavy black typeface;
background context, by using translucent text; semantic relatedness, by overlaying the focus and context onto the same display area; different level of attention to detail by 30 visually and percepu'bly, zooming in when more detail is requested, and by zooming nut when the user retreats back into the default "fly-through, fivm bird's eye view," mode.
Note that while such visual devices are very effective for delivering document highlights, 07-99 15-4A FROM: SAWYER 8 ASSOCIATES ID:650 493 4549 PAGE 1/20 WO 99~16t'TZ PCTIUS9~824384 they rely crucially on being able to carry out the layered analysis described with respoct to capsule overviews.
The RSVP Viewer is particularly well-suitod far deployment in a screen saver mode, in a background window on a desktop machine, or on a large screen projection in communal areas. In any of these situations, a topic or context might catch a reader's peripheral attention, and then they eau decide to take a further look. RSVP
Viewer thus naturally extends, sad fits into, the emergitzg 'push' model of information delivery.
8C_ Vievrl'ool V',~erver to Referring now to Figures 8A - 8C, what is shown is an embodiment of the ViewTool viewer. The ViexrTool viewer freely borrows some o~the ideas of RSVP
viewer. However, the emphasis here is to present a fuller overview of the salient topic stamps in a document, one glare, a single "overview" panel 804. This overview is coutextua,lized to a document 'thumbnail' 816, indicative of the distribution of these t5 highly salient objects in the text. At the same time, a separate 'detaais' area 806 constantly displays additional information pertitzeat to the cement information-seelCirag context deployed by the user. The details area 806 is used both for dynamic display of richer contexts, as in the RSVP viewer for providing access to the fuh text, or topically coherent segments from it, on de,~nand_ Thus, the aim of this viewer is to develop a more 2o elaborate notion of context, while maintaining permanent focus on the salient highlights (topic scamps) in the document The View'fooI viewer further seeks to offer more i>ateractivity to the user, in ways which z~nake the targeted expiorazion of portions of the document nanual and transparent.
The ViewTool views places the capsule overview of a document within the 25 comext of the document itself This is maintained by synchronized display of discourse se~nents, topic stamps, and relational contexts in three panels. The whole document 816 is displayed in the left panel; this is defberately unreadable, and is intended to function as a document thumbnail serving as a contextual referent for the topics presented in the central panel. With the use of an appropriate color coding scheme, it also serves as. an 3o indicator of the distribution of topically prominent phrases in the document. The central panel 804 lists the highly salient topic stau~ps. Contextuaiizarion for these is achieved by aligning the topic stamps for a given discourse segment with the textual span of that adz-99 15:48 FROM: SAWYER 8. ASSOCIATES ID:650 493 4549 PAGE 2/20 w0 99a6tn PC'rr~IJ5981~3s4 segment in the thumbnail as indicated by 808 and 810 is Figure 8A and discussed below.
'1"l~is offers an immediate overview of, for instance, what is being discussed in the beginning of the doewnnent, or in the end, or which topics keep recurring throughout, and so forth.
'T'he central panel is sensitive to the user's Focus of attention: as the mouse rolls over a topic stamp 804, the discourse segment from which this topic has been extracted is highlighted in the left panel shown at 810. The highlighting also indicates the segmentation of the source documents into topically different, and distinct, text sections.
This design makes it easy tv do rapid selection of areas of interest in the document, as it to is mediated by the topic stamps per segment display. Again, the granularity of analysis and the layered contextual information in the capsule overview make it cosy to offer immediate and more detailod infvzmation about any given set of topic stair~ps:
simultaneously with highlighting the appropriate diiscourse segment 810 in the left panel 802, relational contexts for the same set of topic stamps 808 and 812 are displayed is cyclically, in RSVP-like fashion 814, in the right panel 806. "fhis ensures that topic stamps are always related with contextual cue phrases. Thus an additional Level of detail is made available to the user, with very Little 'prompting' on their part 4n the other hand, as Figure 8B illustrates, if ii is still the case that the full text of the segment would be required, clicking on its 'proxy' topic stamps 808 (in the middle panel) would display 2o this in the right panel 818_ 'fhe larger area available there, as well as an automatic readjustment of the size of types ensures that the text is readable.
Rcfe~cring now to pigure 8C, as a natural exte!asion of the same metaphor, clicking on the document proxy 816 in the Left panel brings up the full document text in the right panel 820. The full text always uses color markup to indicate, in yet another way, 25 topically salient phrases and their relational contexts.
By always ma~autaining the larger document context for any grain of information of relevance and interest in the documents, ViewTool viewer is an ideal skimming tool, because it provides additional i~aformatiarl that may be important in deciding whether looking more closely at the document would be required_ For example, users can get a 3o sense of the size of the document, whether it contains any pictures, and other visual cue features. They can see the density of tvpies and the relevant ordering of the topics in relation to the different sections of the doctmient. The tool offers the ability to see 07-99 15:48 FROM: SAWYER 8 ASSOCIATE5 ID:650 493 4549 PAGE 3/20 WO 99/t6192 PCT/US9~IZ4384 arbitrarily detailed contextual information relevant to a topic, while leveraging that same containment hierarchy of layered information units to prevent overload.
8D. View,~r'~, y mmarv The viewers in a preferred embodnmwt can be fully implemented in Java for cross-platform use, and can be deployed in a variety of ways withirx a suite of iutranet tools for collaboration and communication within communities- In one particular example of use, an on-line newspaper' has been configured as the primary news source within a learning community. By means of a variety of web spiders and document filters, 1o external news stories are collected and published' in the newspaper. RSVP
is used far primary delivery of the external news, on a dedicated page, projected on large display in a shared common area. TopicsTicker Viewer offers a brief overview of the latest news on the finnt page of the newspaper. ViewTool Viewer is available as am alternative browser, for more pro-active and focused access to the document particularly in situations where s5 the newspaper is being viewed in personal workstations. ViewTool Viewer is also used for browsing of personal information feeds, sent tn a document analysis server engine via a simple e-mail protocol. Tn general, any configuration of viewers can be deployed for pe~sonatized windows' into continuous news feeds, combining a variety of scz~een delivery modes.
2o The notior~ of temporal typography and dynamic delivery of content zrlediated via content highlights offer an appealing synergy of form and content, which not only alleviates inevitable (given the current state-of the-art of text processing technology) shortcomings of summarization technologies today, but also suggests that additional utility, and user satisfa~ion, can be demred from imperfect analysis technologiesr-if 25 usability and interface issues are addressed from this pezspcctive-A system and method in accordance with the present invention is disclosed which can present thematic capsule overviews of documents to users. For each document a capsule overview is derived which will depict the core content of an average length article in a more accurate and representative manner than utilizing conventional 30 techxziques. The capsule overviews, delivered in a variety of dynamic presentation modes, allow the user to duickty get a sense of what a document is abort, and decide whether they want to read it in more detail. Thus, the system and method greatly ~ / -~1r7 l r7 -'fJ C fCVf'l - JHW 1 GlC Js HJJL/Lr 1 H ~L.J a U - t-JJ1U
~tJ.Wi J-1J w~-fVV. ai <au WO 99I26IT2 PCI'IUS9&L4384 facilitate the process of focused navigation into the parts of the document which may be of particular interest to the user. Tn a preferred embodiment the display of capsule overviews leverage novel presentation metaphors for the dynamic delivery of content.
This delivery can be mediated by mmnilayered abstractions of document content, making 5 heavy use of ideas of temporal typography, in particular for exploiting the interactions between form and content.
Although the present invention has been described in accordance with the embodiments shown, one of ordinary skill in the art will readily recognize that there could be variations to the ennbodilnents and those variations would be within the spirit to aad scope of the present invention. Accordingly, many modifications may be made by one of ordinary skill ili the art without departing from the spirit and scope of the appended claims.

Claims (21)

What is claimed is:

1. A method for dynamic presentation of contents of a plurality of documents on a display; comprising the steps of:
a) receiving a plurality of documents;
b) providing a plurality of capsule overviews corresponding to the plurality of documents; and c) dynamically delivering document content as encapsulated within the plurality of capsule overviews.

2. The method of claim 1 in which a plurality of capsule overviews include a containment hierarchy.

3. The method of claim 2 in which the containment hierarchy of information-bearing units provides different levels of abstractions of the document.

4. The method of claim 3 in which the abstractions comprise at least one topic stamp which is embedded within a plurality of layers of progressively more informative text fragments related to the document.

5. The method of claim 2 wherein the at least one topic stamp is highly salient.

6. The method of claim 2 in which the containment hierarchy is accessed automatically.

7. The method of claim 2 in which the containment hierarchy is accessed manually.

8. The method of claim 5 in which the document content is delivered via a TopicsTicker viewer.

9. The method of claim 6 in which the document content is delivered via a Rapid Serial Visual Presentation viewer.

10. The method of claim 6 in which the document content is delivered via a ViewTool viewer.

11. The method of claim 7 in which the document content is delivered via a Rapid Serial Visual Presentation viewer.

12. The method of claim 7 in which the document content is delivered via a VeiwTool viewer.

13. The method of claim 9 in which the Rapid Serial Presentation viewer comprises the cycling through a plurality of topic stamps and their associated relational contexts for the plurality of documents.

14. The method of claim 13 in which the cycling step comprises:
a) selecting a document from the plurality of documents;
b) selecting a topic stamp from the document from the capsule overview;
c) selecting a relational context relating to the topic stamp from the capsule overview;
d) retrieving a canonical form of the topic stamp from the capsule overview;
e) displaying the topic stamp in background and the relational context in the foreground for a time period that is calculated automatically; and f) repeating steps a-e for the plurality of documents.

15. The method of claim 14 in which the time period for display is calculated based on amount of the relational context.

16. The method of claim 15 wherein the topic stamp and relational context are displayed at different intensities.

17. The method of claim 16 wherein the topic stamp is dark and relational context is translucent.

18. The method of claim 14 wherein the topic stamp and relational context zoom in and out

19. The method of claim 1 wherein the data is displayed via a screen saver.

20. The method of claim 1 wherein the data is displayed on a large screen projection.

21. The method of claim 1 wherein the data is displayed in a background window on a computer.