It C
Resume:
Shamos / J Zhejiang Univ SCI **** *,*(**).'117*-****-****
i : ii
J ournal of Zhejiang University SCIENCE
I SSN 1009-3095
h ttp://www.zju.edu.cn/j zus
E -mail: ****@***.***.** : II
M achines as readers: A solution to the copyright problem
S HAMOS Michael I.
(School of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213, USA)
(Department of Computer Science, Universityof Hong Kong, Hong Kong, China)
E -mail: ******@**.***.***
R eceived Aug. 5, 2005; revision accepted Sept. 10, 2005
Copyright and its international complications have presented a significant barrier to the Universal Digital Library
A bstract:
(UDL)'s mission to digitize all the published works of mankind and make them available throughout the world. We discuss the
effect of existing copyright treaties and various proposals, such as compulsory licensing and the public lending right that would
allow access to copyrighted works without requiring permission of their owners. We argue that these schemes are ineffective for
purposes of the UDL. Instead, making use of the international consensus that copyright does not protect facts, information or
processes, we propose to scan works digitally to extract their intellectual content, and then generate by machine synthetic works
that capture this content, and then translate the generated works automatically into multiple languages and distribute them free of
copyright restriction.
K ey words: Universal Digital Library (UDL), Copyright, Digital rights management, Compulsory license, Berne convention,
Public lending right, Synthetic documents, Machine translation
d oi: 10.1631/jzus.2005.A1179 D ocument code: A C LC n umber: TP391
p ublished since Man began writing them. Something
I NTRODUCTION
o ver half o f them can be found in the combined li-
b raries o f the world. As o f September 2005, OCLC's
E veryone associated with digital libraries, and
W orldCat listed over 57 million records on items
e specially participants in the joint China-India-USA
s panning the last 3 000 years. While 100 million
M illion Book Project, assumes that it will be benefi-
b ooks is a vast corpus, if digitized it would be easily
c ial to mankind to digitize all the works ever pub-
m anageable with present technology. Even assuming
l ished and make them available over the Internet,
t hat each book requires 100 MB o f memory, the total
w hether for fee or otherwise. However, it is difficult
s torage requirement would be 109 GB. With current
t o find a clear printed agenda explaining how society
r etail prices for disk storage hovering at $0.50 U.S.
w ill be enriched aside from the obvious advantages of
p er GB in small quantity, the total cost would be
p reservation, indexing and efficient digital distribu-
u nder $500 million to store every book ever written,
t ion. I propose that the ultimate impact o f a Universal
w hich is much less than the original cost of purchas-
D igital Library (UDL) cannot be realized unless
i ng, scanning or even storing one copy o f each o f
c omputers are able to read, process, paraphrase and
t hem 1.
t ranslate its contents. The reasons involve copyright
T he question I want to address is how to use such
l aw, human behavior and the limitations o f our cog-
a c orpus once it is created. One o f our central prob-
n itive power to assimilate information.
l ems as a society is how to allow developing nations
I t is estimated that 100 million books have been
t o reap the benefit o f the world's technological de-
v elopments. The challenge o f feeding the world,
T he typical cost to store a book in a circulating library, all charges
p roviding it with fresh water and keeping it healthy
i ncluded, is about U.S. $6.00 per book per y,ear. At this rate, the digital
s torage would be paid for in a month. d epends on distribution o f knowledge in usable form
S hamos / J Zhejiang Univ SCl 2005 6.4(11):1179-1187
1180
t o people who need it. For example, the spread of p leasure from reading for other purposes. The joy of
m odem farming techniques to areas of chronic famine r eading poetry, or novels, or even well-expressed
o r overpopulation would be of incalculable value. t echnical material, will continue unabated as it has for
This requires much more than distributing copies of t housands of years. Satisfying an appetite for pleasure
is not one of the principal objectives of the Universal
E nglish books on agriculture to people who cannot
r ead English. Yet the historical function of libraries D igital Library, however. Its greater concern must
u ltimately be for those who need information and
has been primarily to allow selected populations to
k nowledge for practical purposes to better themselves,
v iew or borrow a small set of chosen publications.
a nd this implies a far different manner of use of its
T he media visionary Marshall McLuhan ob-
c ontent.
s erved that the first use of new technology is imitation
T he reality is that most references humans make
o f the old. History is so replete with examples of this
t o textual materials are not the result of large-scale
p henomenon that no one seems to have asked whether
r eading or assimilation, as one might observe in col-
it is simply an observed human reaction or a funda-
l ege students reading assigned textbooks, but are in
m ental necessity. It appears to be fundamental. When
t he form of directed lookup, that is, the result of
a n ew technology becomes available, it cannot dis-
s earch. When one has a question, one needs the an-
p lace the old unless it is able to replicate every one of
s wer, and often does not have sufficient time to read
the old functions. Otherwise, it will not be adopted.
entire works to obtain it. The farmer who needs to rid
This creates a low threshold for acceptance. All the
h is field of a specific pest is not interested in the his-
n ew technology must do is to duplicate the old, pos-
t ory of agriculture, nor even of prescriptions for kill-
s ibly faster, cheaper or with enhanced performance,
i ng the multitude of insect species that have plagued
and it is likely to be embraced. There is no economic
f armers for centuries. He wants to know what to do
i mperative to have it do more than that.
t oday to solve his problem, and that is a search ques-
T ranslating McLuhan's maxim into the digital
tion.
l ibrary context, we can expect that the first large
I f the answer is contained in one paragraph of
digital libraries will simply replicate traditional ones.
o ne book written in a foreign language, the farmer is
T hat is, they will accumulate carefully selected works,
n ot interested in paying for a full translation of the
i ndex them, and deliver them, or allow them to be
f oreign book into his own language so he can read the
v iewed, under controlled conditions to authorized
r elevant paragraph. He just wants the answer. Un-
patrons. The indexing will be excellent and the de-
f ortunately, the copyright system does not provide an
l ivery remarkably rapid, but nothing reallynew will
e fficient mechanism to allow him to pay the propor-
h appen. While this is undoubtedly a desirable de-
t ional value of the text he wants to read in comparison
v elopment, I seriously question the degree to which it
w ith the price of the whole work. Under the present
w ill benefit mankind.
m echanism, he must locate the book, which is cur-
T he reason is that mere access to books does not
r ently not easy unless it has been digitized, then either
s olve societal problems. The lifetime reading capacity
b uy or borrow the book and commission of translation
o f a human being is about 3 000 books. This assumes
o f the portion he desires. It is illegal for his library to
r eading one book per week for 60 years. A very dili-
m ake a digital copy of the paragraph available to him
g ent reader might push the number to 10 000 books in
w ithout permission of the copyright owner, and the
a l ifetime, which is still only 1/100 of one percent of
c ost of even asking for such permission will often
w hat has been published. And how much can the
e xceed the price of the work.
r eader actually assimilate from all this reading that
T hat method might succeed if copies of books
w ill train him for a new job, improve his crop yields
w ere always readily and cheaply available. The UDL
o r teach him mathematics?
It is important here to distinguish reading for c an eventually scan all public domain 1works to solve
p art of the problem. Without the need to restrict dis-
t ribution or account for royalties, every literary work
" Public domain" is an often misunderstood term that means "free of
o n Earth or any portion of any such work will be free
c opyright and available for any otherwise legal use without charge." It
t o us all. This is a major advance.
d oes not simply mean "accessible to the public".
Shamos / J Zhejiang Univ SCI 2005 6A(11):117*-****-****
H owever, because of the gradual lengthening of t o the use of their works irl such a manner. A secon-
c opyright terms around the world, particularly in the d ary problem is that the OPIC proposal does not ad-
U nited States, which has extended the term consid- d ress uses of the work other than manufacture and
d istribution of whole copies.
e rably beyond what is required by international con-
v entions, by far the majority of works so far published
o n Earth are still in copyright. To estimate the per-
c entage at 90% would not be amiss. Of even greater C OPYRIGHT ISSUES
c oncern is that technological materials having the
m ost relevance are very recent and hence most likely It is clear that copyright is a major impediment to
t he development of digital libraries. There is also
t o be in copyright.
E ven copyright itself would not be such a huge c onsiderable uncertainty in most countries as to the
p roblem were it not for the fact that a large percentage s cope of fair use of copyrighted works and it is un-
o f works that are still in copyright are out of print. certain whether any form of digitization, even just for
T his puts the potential user in an impossible situation. indexing, is permissible. Even in the United States,
E ven if he is willing and able to buy the book, he w hich has produced the most extensive set of legal
literally cannot do so. The publisher is unable to d etermination on this issue, there is still reluctance by
s upply a copy, and to duplicate a library copy would small organizations to test the limits of the law.
c onstitute copyright infringement. Even if the poten- T o demonstrate that the question does not admit
tial user were willing to risk an infringement suit, the o f a simple answer, consider scanning a copyrighted
U DL cannot put itself in the position of contributing w ork only for the purpose of developing an online
to the infringement. k eyword index of its text. The U.S. Digital Millen-
n ium Copyright Act (DMCA) provides that, under
Prof. Raj Reddy of Carnegie Mellon University,
w ho conceived of the UDL, has suggested that own- s pecified conditions, a "service provider shall not be
liable ... for infringement of copyright by reason of
ers of out-of-print works ought to consent to an ar-
r angement in which a user of out-of-print the provider referring or linking users to an online
l ocation containing infringing material or infringing
i n-copyright (OPIC) material would pay a fee to
d ownload or print it, which would be split between a ctivity, by using information location tools, includ-
ing a directory, index, reference, pointer, or hypertext
t he UDL and the owner. In case the user wanted a
link... This is the statutory justification for
h ard copy of the work, the need could be satisfied by
an on-demand printing house that would sell him a l arge-scale indexes. Even if the index points to an
infringing copy of a work, the indexer is not liable
single copy (produced from the digital version at the
u nless it knows of the infringement.
U DL). The proceeds would be divided among the
O ne might assume that, if it is legal to create an
o wner, the UDL and the printing house.
index of a copyrighted work, it ought to be legal to
This solution has encountered obstacles, the
f oremost of which I refer to as the "Titanic" problem. d istribute the index to others. The index itself is not
c opyrightable, and, after all, others would be privi-
N umerous books on the sinking of the Titanic have
a ppeared since 1912. Most go out of print in short l eged to generate an index themselves, so what could
order. Others, such as Walter Lord's "A Night to b e improper about sharing indexes? The difficulty is
R emember," became best-sellers before sales dimin- t hat an index to a work, which typically contains the
i shed to a negligible level. When the movie "Titanic" e xact position of every word in the work, can be used
t o reconstruct the work in its entirety, and distributing
w as released in 1997, several of these books returned
t o the best-seller list. The publishers argue that if they t he entire index, as opposed to simply allowing online
queries, can be regarded as equivalent to distributing
h ad a revenue-sharing contract with the UDL, they
t he work itself.
w ould miss out on the revenue spike associated with
s uch events. A digital library site has great difficulty com-
p lying with copyright law, even if it wants to. A first
T he result is that most publishers will not agree
p roblem is that it is extremely hard to determine
w hether a work is still subject to copyright, and, if so,
I 1 7 U.S.C. w
Shamos / J Zhejiang Univ SCI 2005 6A(11):1179-1187
1182
m eantime, development is proceeding rapidly on
w ho the copyright owner might be. In the U.S., dili-
d igital rights management (DRM) technology, which
g ent examination of copyright registration and re-
c opyright owners expect will make piracy impossible.
n ewal records is insufficient to produce an answer,
I b elieve this to be a futile hope. Any work in-
e ven for works for which registration was sought.
t ended for humans must be presented in a form that
S ince the elimination of copyright formalities by
t he human senses can perceive. For books and videos,
i nternational agreement, there is no office or database
t his means that the content must be visible. For music,
t o which one might refer that gives the copyright
i t must be audible. Therefore, at the point that the
s tatus of a work. One reason is that the ownership of
d igital material is transformed into analog form for
c opyright may have been transferred to another party.
p resentation to the human, it can be captured. It
W hile such transfers must be in writing, there is no
c annot be protected from capture because the brain
r equirement that the Copyright Office, the repository
d oes not have a direct digital interface. The unen-
o f copyright records, need be informed of the transfer.
c rypted signals must go through the air so the content
T here is thus no place one may look for a definitive
c an be acquired by the human. This is the reason that
d etermination of ownership, so a digitizer who wants
D RM is doomed to failure. Nevertheless, it will surely
t o seek permission is frustrated form doing so.
p resent barriers to the use of digital materials.
S uppose one learns that a work is still in copy-
r ight, but copyright owner no longer exists or cannot
b e located. Is it legal to copy the work, or is it simply
C OMPULSORY LICENSING
t aking a risk that is unlikely to produce adverse con-
s equences?
D ifferent solutions to the copyright problem
h ave been proposed. Some are promising only within
M cLuhan's inhibitory prescription, that is, so long as
P IRACY
t he lending library model dominates. The United
K ingdom has implemented a "public lending right",
T he book world has remained remarkably un-
( PLR) which recognizes the fundamental unfairness
s cathed by the sort of digital piracy that has plagued
o f purchasing one copy of a book, placing it in a
t he movie and record industries. There is no Napster
p ublic library, and having large numbers of people
f or books. I believe that the reason is not a low level
b enefit from it through borrowing without any addi-
o f interest in reading, but the fact that it is still painful
t ional payment to the copyright owner. The owner is
t o read from computer terminals while it is not at all
c ompensated only once, at the initial purchase of the
d ifficult to watch DVDs or listen to music on digital
b ook. The UK Parliament each year allocates money
d evices.
t o a common fund, the proceeds of which are dis-
T he situation will change radically when an
t ributed to copyright owners on a pro rata basis de-
e lectronic book (eBook) artifact is available that
p ending on how often each book is checked out of
m imics the properties of a paperback book, namely
p ublic libraries. The success of the scheme depends,
t he ability to flip and bend pages, mark one's place,
o f course, on the amount allocated. It might be de
w rite notes in the margins and put it in a convenient
minimis, o r it might constitute a windfall.
p ocket. O f course, this device will be able to store
R egardless, the measurement is based on
10 000 books (by our count, requiring less than a
T erabyte)--a lifetime supply for our diligent reader. c hecking out entire books and is unable to measure
A las, McLuhan tells us that the first of these will do p artial uses, such as repeated reading of a single
l ittle more than pretend they are traditional books--at c ritical paragraph or section. However, the public
l east until they win acceptance. l ending right is increasing in popularity around the
O nce the true eBook exists, the publishing in- w orld, as it deserves to be, since it corrects a basic
d ustry will fight precisely the same battle now being i nequity.
w aged by music and movie companies: it will be T he digital analog of a public lending right is
d ifficult to formulate, since it, too, suffers from an
i mpossible to put the e-genie back in the bottle and
p iracy will run rampant in this domain as well. In the i nability to measure, or provide compensation for,
S hamos / J Zhejiang Univ SC! 2005 6A(I I): I179-1187 1183
partial uses. A critical assumption underlying the PLR O f course the difficulty with any compulsory
is that the publisher benefits from revenue for each l icensing scheme is collecting revenue and managing
c opy of the work that is placed in a library, and that p ayments. If governments were to contribute sub-
e ach copy can only be lent out to one person at a time. stantial amounts of money to a common fund to pay
T herefore, if the demand for a book increases, more c opyright owners for the use of their works, the
libraries will have to buy it, and the publisher will p roblem would be far simpler. Few, if any, govern-
b enefit directly. m ents understand that paving the information high-
T his assumption does not hold for digital works. w ay is as important as maintaining the traditional one.
O ne copy in digital form can be distributed at essen-
t ially no cost to an arbitrary number of users in es-
s entially no time. Furthermore, many users will only S CIENTIFIC JOURNALS
w ant to view, use or download only a single page
f rom a lengthy work, and even if proportional com- P ossibly the most persuasive argument against
p ensation were provided to the publisher, the result c opyright arises when the use of the law acts in a
m ight be unfairly low. m anner directly contrary to its stated intent. The un-
A d ifferent sort of solution is the compulsory d erpinning of copyright law in the U.S. is the clause
license. This is a permission to reproduce copyrighted in the Constitution giving Congress the power to
m aterial which the owner cannot refuse, but the user e nact copyright legislation: "Congress shall have
m ust pay for. How much the user pays can be deter- p ower ... to promote the progress of science and
m ined by a statutory formula, as in the United States, u seful arts by securing for limited times to authors
o r by an independent tribunal, as in Japan. and inventors the exclusive right to their respective
T he U.S. has an effective, but narrowly limited, w ritings and discoveries" (Pollack, 2002). The term
c ompulsory licensing scheme for two kinds of works: " science," as used at the time, was not restricted to
p honorecords of copyrighted songs and redistribution s cientific activity as we now understand it, but re-
o f television signals over cable systems. Anyone who f erred to learned studies in general. The definition is
w ishes to sell copies of a sound recording of a song thus broader than it might appear.
m ay do so without prior permission by simply noti- T he Constitutional purpose behind copyright is
f ying the copyright owner and remitting a statutory i nhibited if publishers &scientific material are able to
r oyalty based on the length of the song and the k eep the results of research out of the hands of aca-
n umber of copies produced 1. d emics by charging exorbitant (actually, prohibitive)
T hese compulsory licenses work because the p rices for scientific journals. An extreme example is
q uantum of use is easily measurable--it is determined E lsevier's oft-cited journal Brain Research, w hose
b y the number of copies made and the duration of the o ne-year subscription price for 2005 was U.S.
w ork. However, no ancillary rights are included in the $23 483, as reported on the company's website.
c ompulsory license, such as the right to make deriva- F ew would begrudge a publisher the opportunity
t ive works like translations. Again, it is difficult to see to make a profit, and although the number of sub-
w hat the digital equivalent of such a license would s cribers to Brain Research is undoubtedly low, the
e ncompass since there is no useful digital analog of e conomics of academic journal publishing set it apart
" copy". in that the authors, reviewers and editors receive no
T he Japanese compulsory license is much more c ompensation for their work---only the publisher gets
e xtensive, but also more cumbersome. A far larger p aid--and the original purpose for which the work
c ategory of works is subject to compulsory licensing, w as created, which is to achieve widespread dis-
b ut the royalty is determined by an adjudicatory body s emination rather than earn money, is thwarted, and
r ather than statute, which introduces delay and un- e ven choked off, by the fee structure.
c ertainty in the process. T here was a time when scholars were completely
d ependent on print publishers to distribute their work,
and many researchers remember a day when they
T he royalty in the U.S. for each copy as of January 1, 2006 will be
w ould await anxiously the arrival in the post of the
1.75 cents per minute with a minimum payment of 9 cents per song.
S hamos / J Zhejiang Univ SCI 2005 6A(11):1179-1187
1184
n ext issue of a journal in their field. Print publication w ere copyrightable, then the owner might extract a
to achieve dissemination is no longer required, or r oyalty from anyone who published such an instruc-
e ven efficient. What costs $23 483 to buy on paper t ion until the expiration of copyright, in effect grant-
c osts $0 to produce on the Internet, since everyone i ng the protection afforded by a patent, but for a much
c onnected with the creation of a paper works for free. l onger period of time. Since this is undesirable, the
P ublishers now typically require camera-ready copy i dea and expression are said to merge into the idea
o f papers, or at least an electronic format, so even the a lone, and no copyright protection is possible.
t ypesetting is without charge. This means that the
u seful business future for expensive print journals can
b e measured by an hourglass, and I hope their death
I NTERNATIONAL COPYRIGHT AGREEMENTS
c an be hastened by the UDL.
C opyright law is largely territorial--acts that
t ake place in a given country are subject to the copy-
T HE IDEA/EXPRESSION DICHOTOMY
right law of that country and no others 2 (Pa, 2000).
H owever, large numbers of nations have entered into
A s a general rule, subject to international com-
a s eries of treaties that provide for certain minimum
p lexities discussed below, copyright law does not
l evels of copyright protection and accord to foreign-
p rotect ideas, facts, or processes, but only the manner
ers the same level of protection as that provided to
in which they are expressed. For example, the U.S.
n ationals of the country, a concept referred to as "na-
C opyright Act states, "In no case does copyright
t ional treatment." The principal agreements are the
p rotection for an original work of authorship extend
B eme Convention, the Universal Copyright Conven-
to any idea, procedure, process, system, method of
t ion (UCC), the Paris Convention for the Protection of
o peration, concept, principle, or discovery, regardless
I ndustrial Property, the WIPO Copyright Treaty and
o f the form in which it is described, explained, illus-
t he Agreement on Trade-Related Aspects of Intel-
trated, or embodied in such work ''1. This is a won-
l ectual Property Rights (TRIPS). Any sort of analysis
d erful exception, since what we want from copy-
o f the effect of these agreements on copyright law
r ighted works for the purpose of practical benefit to
a round the world is far beyond the scope of this article,
m ankind is precisely what is uncopyrightable, namely
s ave for two points: the treatment of ideas vs. ex-
the knowledge, ideas, concepts and operational in-
p ression and compulsory licensing.
s tructions they contain. The question, explored below,
T he Paris Convention of 1883 established an
is the extent to which the uncopyrightability of ideas
international union now known as the World Intel-
is recognized outside the U.S.
l ectual Property Organization (WIPO), which cur-
In some cases, facts or ideas admit essentially of
r ently has 169 members. The Paris Convention itself
o nly one possible expression. In such cases, the idea
d oes not address copyright, but the Berne Convention,
w ins out over expression and the expression becomes
a dministered by WIPO and having 156 signatories,
u ncopyrightable according to the doctrine of
d oes so. The WIPO Copyright Treaty (WCT), which
" idea/expression merger". To hold otherwise would
m odifies the Berne Convention, has 56 members. The
a llow the discoverer of a fact to obtain a monopoly
U CC has been adopted by 64 nations. TRIPS is an
o ver it for the full term of copyright. For example,
a greement between WIPO and the World Trade Or-
s uppose someone discovered that "eating peaches
g anization (WTO), which has 148 members. China,
c ures cancer" and published a pamphlet containing
I ndia and the U.S. are signatories to all of these
that phrase. All ways of expressing that discovery are
a greements except that China and India have not
e ssentially equivalent. In the parlance of copyright
law, they are "substantially similar". If the phrase r atified the WCT.
The Berne Convention specifically allows na-
t ions to impose compulsory licenses on certain types
17 U.S.C. w (b). o f works (e.g., musical works), forbids such licenses
2 E xceptions are recognized when acts committed in one country result
o n others (e.g. cinema) and is silent on other works
i n infringement in another country.
S hamos / J Zhejiang Univ SCI 2005 6A(11):117*-****-****
s uch". Even though China and India are not signato-
s uch as literary works, including books and computer
p rograms (where they are copyrightable). r ies to the WCT, TRIPS contains an equivalent pro-
The UCC does not address copyright in ideas, v ision. As we shall see, these treaty provisions fur-
n ish a legal pathway to universal benefit from the
b ut contains several clauses intended to promote
U DL.
c ompulsory licensing. For example, Article V allows
s ignatory countries to grant a compulsory license to
m ake translations. This is a very important license
s ince without it no one has an independent right to
translate a work. The practical effect of keeping the R ESPONSE TO COPYRIGHT
right of translation with the copyright owner is to
v astly restrict the spread and utility of the vast ma- T he foregoing sections have been a lengthy
j ority of works. Even if the publisher has no interest in p relude to my central proposal, which is that since
p ublishing a Polish translation of a French work, a c opyright poses so many obstacles to the goals of the
P olish citizen cannot do so without permission 1, and U DL, and we cannot ignore copyright, the only al-
t hus for the entire term of copyright the work will t ernative is to circumvent it.
V arious proposals have been made to modify the
r emain inaccessible to Poles.
c opyright statutes in various ways to promote assis-
The ability to have works translated is essential to
the goals of the UDL. Unfortunately, the UCC only t ance to developing nations, such as expanding the
s cope of compulsory licenses, relating license fees to
permits compulsory licenses--it does not mandate
a n ation's per capita income, providing tax benefits to
them, and relatively few nations have enacted a com-
d onors who dedicate their copyrights to the public
p ulsory license for translation. As a general principle,
d omain, implementing micropayment schemes to
w here intellectual property treaties specifically allow
p rovide compensation for partial use of a work, and
w ealthy nations to grant benefits to poor nations, they
t he like. Such proposals, even if they found favour,
d ecline to do so, and the effect is a continual widening
w hich they do not, would take a very long time to
o f the gap between the industrial and the developing
e nact because of the fierce debate that accompanies
world. It is this deliberately maintained inequality I
a ny modification to the economic effect of copyright.
p ropose to correct later in this paper.
W hile the public lending right is gaining favour
T RIPS Article 9.2 reads, "Copyright protection
a round the world, it only benefits citizens of the en-
shall extend to expressions and not to ideas, proce-
a cting country, and only applies to publicly-available
dures, methods of operation or mathematical concepts
p hysical copies of works.
as such". Article 10 provides that "Compilations of
I p ropose instead that we at the UDL operate
d ata or other material, whether in machine readable or
c ompletely within the existing statutes and interna-
o ther form, which by reason of the selection or ar-
t ional agreements and make full use of the exemption
r angement of their contents constitute intellectual
o f facts, concepts and principles from the scope of
creations shall be protected as such." It then goes on to
c opyright protection, as provided in international
c larify that protection does not extend to the data or
treaties.
material itself, unless the material is separately copy-
It is established in various jurisdictions (e.g., the
rightable.
U .S.) that one is privileged to use a copyrighted work
The WCT Article 2 states, "Copyright protection
t o extract its unprotected content 2. This is the basis on
e xtends to expressions and not to ideas, procedures,
w hich reverse engineering and digital indexing are
m ethods of operation or mathematical concepts as
p ermitted. This means that it is legal to have com-
p uters process works to obtain their essential infor-
B oth France and Poland are signatories to the Berne Convention and
m ation, provided that this can be done without
t hus Poland must accord as much protection to the French author as it
c opying their expression. However, this principle by
d oes to its own nationals.
2 I t is uncertain whether one may actually copy the work in order to
i tself accomplishes nothing, since the effort of dis-
e ngage in this activity, but it is certainly permissible to use an au-
thorized copy. tilling expression from a work is very substantial and,
3 The resulting expression would also be separately copyrightable, if
u p to now, has required human labor3.
p roduced by a human, which only compounds the problem.
S hamos /JZhejiang Univ SCI 2005 6.4(11):1179-1187
1186
i nfringing copyright b2t extracting the essentials from
S YNTHETIC DOCUMENTS
a v ariety of sources and synthesizing an entirely new
a rticle from the given ones.
A " synthetic document" is one that is produced
E ventually we want to reach the stage of auto-
b y a machine based on other inputs, typically textual
m atic creation of encyclopedia articles, so that a user
articles. This arena has been the subject of much re-
c an request an analytical article of a given length on a
search. Perhaps the simplest example of a synthetic
s pecific topic, such as "Tsunami Preparedness in the
d ocument is an index, which is very laborious to
I ndian Ocean". The program would absorb a large
g enerate by hand but can be created very quickly by a
c orpus of documents, analyze them, evaluate their
machine. Such an index is synthetic because it did not
s ources, check for inconsistencies and consensus, and
e xist beforehand.
p repare a critical article of the required length.
T he reason indexes are easy to make automati-
U sing digital collections as fodder for synthetic
c ally is that they are almost purely syntactic. No se-
d ocument generators is not a new idea (Wactlar,
mantic understanding of the text being indexed is
1996). However, it does not appear to have been
n eeded to create the index. There are some interesting
r ecognized previously that synthetic documents, if
s yntactic challenges, such as recognizing whole
t hey are created without copying original expression,
p hrases and idioms, and determining whether a word
a re free of copyright restrictions. Not only can they be
that occurs at the beginning of a sentence should
f reely distributed, but they can be translated into an
r emain capitalized in the index, but highly useful
a rbitrary number of languages.
i ndexes can be produced even ignoring such issues.
G reat strides have been made recently in auto-
G oogle is perhaps the extreme example of how useful
m ated translation (Kanellos, 2005). While translation
an index can be.
o f unrestricted, pre-existing text remains difficult,
A bstracts and summaries are more advanced
n early perfect machine translations can be made in
f orms of synthetic documents since they require
r estricted domains if controls can be placed on the
p rocessing at the sentence, rather than the word, level.
t ext to be translated at the time it is originally gener-
D etection of topic sentences and elimination of re-
a ted (Carbonell, 2000). The reason is that ambiguities
d undant content are required.
in syntax and vocabulary can be resolved at genera-
A t the next level up is synthesis from multiple
t ion time to yield fully translatable text. Therefore, an
d ocuments, in which a program ingests different texts
a utomated translation system, working in cooperation
and processes them to generate summaries. Google
w ith a document synthesizer, would be able to pro-
N ews is an excellent example. It scours online
d uce accurate output in multiple languages without
n ewspaper feeds and produces essentially a front page
t he need for human editing.
c ontaining brief headlines, along with links to the
A ll of the above systems, from indexing soft-
s ource publications. It was described at a very general
w are to full treatise generators and automated trans-
l evel in the Web publication "Digital Inspiration" on
l ation systems, if properly structured, avoid the com-
M ay 31, 2005: "Google News basically crawls news
p lications of copyright law and permit the informa-
sites, finds 'story clusters,' ranks the sources, figures
t ional content of copyrighted works to be digested,
o ut how prominently each source is running the story,
t ranslated and distributed worldwide without en-
f igures out whether its a big story or a little story,
c umbrance.
f igures out geographic references, and builds the
p ages for the various geographic and language
e ditions". Most people are surprised to learn that the
C ONCLUSION
p rocess is entirely algorithmic. No human input is
utilized, save for the effort of the newspaper reporters
I p ropose that the UDL undertake the scanning
w ho wrote the original stories and the editors all over
o f all works, even those that are in copyright. Such
t he world who independently decided which stories
s canning is legal for the purpose of creating finding
w ere important enough to publish.
aids, such as indexes, and for extracting informational
A t a greater level of sophistication a system
c ontent. Works that are in copyright cannot be pro-
c ould produce summaries of news articles without
1187
S hamos /JZhejiang Univ SCI 2005 6.4(11):1179-1187
sen, P.M., Bauer, M.lVI.,Busdiecker, III. R.F., Hayes, P.J.,
v ided to the public without permission, but, as we
Huettner, A.K., McLaren, B.M., Nirenburg, I., Riebling,
h ave seen, doing so is unnecessary for the works to be
E.H., Schmandt, L.M., Sweet, J.F., Baker, K.L.,
o f significant use. It is essential in all UDL activity
Brownlow,N.D., Franz, A.M., Holm, S.E., Leavitt, J.R.R.,
t hat copyrighted works be protected against theft, Lonsdale, D.W., Mitamura, T., Nyberg, E.H., 2000. Inte-
p iracy and inadvertent distribution and used only for grated Authoring and Translation System. U.S. Patent
t he purposes discussed above. 6,163,785.
Kanellos, M., 2005. Google Dominates in Machine Transla-
A ll digitized works can then be used as data to
tion Tests. CNET News.corn, August 22. http://news.com.
s oftware that will produce synthetic documents. Be-
com/2100-1038 3-5841819.html.
c ause these documents will not be copyrighted, they
Pa, W.D., 2000. Twentieth Century Fox Film Corp. v.
c an be translated into any language, especially by iCraveTV, 53 U.S.
Machines as readers: A solution to the copyright problem
Contact this candidate