EXTENDING KNOWLEDGE ACCESS TO
UNDERSERVED CITIZENS

Wallace Feurzeig
BBN Systems and Technologies

ISSUES AND GOALS

If we are serious about making available the rich human and information resources of the national information infrastructure (NII) to all citizens, we have to address the great, and growing, knowledge gulf between the "haves" and "have-nots" in American society. This disparity poses a real threat to our political and social lives. It is at variance with our history and our democratic ideals. Many citizens who are economically underserved are also "informationally disadvantaged" (National Telecommunications and Information Administration, 1995). The knowledge empowerment made possible by the new information technologies must become available to all. We must work toward democratizing access to effective use of networking technology by all Americans who are capable of benefiting from its use. We need to provide the underserved not only with access to these potentially empowering information resources but also support in their use through education, training, and acculturation. Researchers at BBN and elsewhere have done preliminary work along these lines to better understand the issues. Though neither systematic nor comprehensive, this work identifies some key research questions and suggests specific directions for more substantial action-oriented research efforts.




INTERNET USE BY LOW-INCOME FAMILIES

A recent study conducted from December 1994 to January 1996 probed the barriers, benefits, and perceived worth of the Internet to six low-income urban families in Florida, a group representative of the traditionally underserved and informationally disadvantaged population (Bier, 1996). The researchers asked what these families would actually do on-line given unrestricted Internet access in their homes. Each family was lent a home computer, high-speed modem, and printer; was provided with dial-up point-to-point Internet access; and was given training on the use of the mouse and keyboard. The computers were equipped with an interface security program, an integrated productivity package, several educational games, a typing tutorial, and a set of Internet utilities. Families were taught how to communicate with each other electronically and how to locate and acquire resources from the Internet. Additional training and technical support were available on demand for the duration of the project. Through interviews, visits, and telephone and e-mail interactions, researchers obtained data on the amount of time participants spent on-line, the sites they visited, the information they sought, and the obstacles they encountered. The participants made use of virtual hospitals, medical dictionaries, and physicians' desk references. They joined support groups, investigated scholarships, and made local transportation arrangements. They investigated appliances, employment listings, and local calendars of events. They e-mailed, chatted, and surfed the World Wide Web; made friends; felt personally empowered as learners; and gained a new sense of community. The results showed that Internet access enabled "powerful emotional and psychological transformations" on the part of the participants.




EDUCATIONAL SOFTWARE USE BY EDUCATIONALLY DISADVANTAGED STUDENTS

During the past two years, my colleagues at BBN have used the computer program RelLab (for Relativity Laboratory) to teach the concepts of relative motion to educationally disadvantaged inner-city high school students in Boston (Horwitz, 1995). The program enables users to construct and run relativity "thought experiments." The inner-city students worked on the same sequence of activities as the educationally advantaged suburban students with whom we worked earlier—and with almost identical results. They advanced steadily from simple Galilean (low-speed) relativity problems to complex ones involving the frame independence of the speed of light and its implications for simultaneity, time dilation, and length contraction. Along the way they exhibited the same frustrations and overcame the same obstacles as the suburban students, and they progressed at about the same rate. Moreover, they showed as much interest and had as many breakthroughs as their advantaged peers. The only real differences were that the inner-city students were considerably less verbally communicative than the suburban students and almost all were severely educationally deprived. This was particularly evident in the case of math. Their knowledge of the decimal system was spotty and unreliable, and they had great difficulty graphing data. Despite their competence on the computer, they were hopeless at pencil-and-paper tasks with the same material, particularly written tests. Based on posttest interviews of the students, it appeared that their poor performance was due not only to knowledge gaps but also to significant deficits in reading comprehension.

During the past year BBN researchers have also been using another educational computer program, GenScope, to teach the concepts of genetics to suburban and inner-city students in high school biology classes. GenScope enables users to explore and experiment with genetics models. It has proved engaging to both male and female students of widely differing ages, backgrounds, and ability levels. The most interesting results have come from the inner-city population, where students who had learned very little from a conventionally taught biology class learned a great deal about complex genetic processes working with GenScope and used their new knowledge effectively to accomplish a variety of analytical and constructive tasks. Moreover, they remembered what they had learned, and some were able to give accurate and detailed explanations weeks later in interviews conducted away from the computer. However, we found the same results as those noted above with RelLab—though the inner-city students often learned to do sophisticated genetics on the computer, their knowledge rarely transferred to paper-and-pencil tests. They were unable to express in writing the complex knowledge they exhibited on the computer. Given the software tools and support, educationally disadvantaged students acquired and applied complex reasoning, but their knowledge was locked in. They lacked the basic reading comprehension and communications skills that are fundamental for success in education and skilled occupations. The challenge is to help students like these acquire the verbal literacy that will enable them to participate more fully as citizens in the knowledge society.




RESEARCH ISSUES

The positive outcomes shown in the study of Internet use by low-income families are indeed impressive and encouraging. This work cries out for replication and suggests new questions for further research. What are the key requirements for making such Internet use productive? What skills and motivation must users possess to begin with? What kind of training and ongoing support are required? How much on-site help is required, and how practical and effective is on-line mentoring in augmenting initial training? Can the facilities, training, and support be provided not only or not primarily in the home but also in social settings like community organizations, churches, local schools (after-school programs), and particularly libraries—places with expert human information resources?

The results from the use of sophisticated learning technology by inner-city high school students—their success in acquiring and using complex knowledge contrasting strongly with their difficulty in communicating that knowledge—are more problematic and troubling and raise another set of research questions. Can high school students who fail to acquire competence in reading, writing, and communications skills overcome these deficits later? Are these difficulties remediable by tutoring? Or is there a literacy learning barrier for English (analogous to the putative foreign-language age block) that makes the acquisition of fluency in reading and writing English enormously difficult after a certain age or developmental level?

These questions provide a rich source of cognitive, educational, ethnographic, and technological research investigations, from intensive small-scale studies of individual development to long-term longitudinal programs involving large populations. Rather than providing brief summaries of possible research projects across this wide spectrum, I will focus on a single project that is motivated by both of the preliminary studies discussed above and that has important implications for advancing the goal of universal citizen participation and empowerment through educational networking.

The productive use of the Internet by low-income citizens would not have been possible if they did not bring to these activities a fairly high level of functional literacy. The stunted development of literacy skills among the inner-city high school students severely limited their social participation and learning opportunities, despite their inherent intellectual abilities. Effective utilization of the NII, no matter how rich its user interface and information resources, will require that its users bring a modicum of literacy. But our nation has an enormous number of citizens who do not meet that test. I describe a research project that seeks to address, through the use of new interface technology and associated instructional activities, a most severe literacy problemthat of American adults who are functionally illiterate.




THE ADULT ILLITERACY PROBLEM

The development of a technologically literate work force is a national imperative for the United States in the face of increasing international competition. Major federal initiatives such as the School-to-Work Opportunities program and the Advanced Technology Education program have been created to address this critical need. However, these efforts will be severely hampered if the country's massive adult illiteracy problem is not solved. In confronting our formidable economic challenges we need to recognize that the most central and essential prerequisite for virtually all types of new jobs is functional literacy—the ability to read and comprehend text and to use various forms of word processing and other communications software.

Workplace literacy surveys indicate that over 90 percent of current occupations require reading. A recent study by the U.S. Department of Labor and the American Society of Training and Development found that, on average, workers spend more than 1.5 hours each day reading such materials as forms, charts, manuals, electronic display screens, and general literature. Further, these requirements are increasing: only 4 percent of new jobs can be filled by people with the lowest levels of literacy, as compared with 9 percent of existing jobs. According to the study, even the one-third of American workers who perform production and service delivery will need to read at an eighth-grade level. The remaining two-thirds will need to read at postsecondary and higher levels. Levels of literacy that were once acceptable will be marginal by the year 2000.

However, the stark reality is that the number of adults in the U.S. population with unacceptable levels of literacy is enormous. Already, more than one out of five adult Americans are functionally illiterate, and their ranks are swelling by about 2.3 million persons each year. Nearly 40 percent of minority youth and 30 percent of semiskilled and unskilled workers are illiterate. Illiteracy costs the United States over $225 billion annually in corporate retraining, lost competitiveness, and industrial accidents. The implication is clear: our goal of providing a modern competitive work force hinges very directly on our ability to achieve a massive improvement in adult functional literacy during the next decade. This cannot be accomplished through the use of human teaching alone. There simply are not enough reading instructors. Their teaching must be augmented by the creation and widespread application of an effective technology for automating literacy tutoring.

Although for a small fraction of illiterates the ability to read is impeded because of neurological problems, and for others there are learning difficulties that are not associated with sensory or motor problems, the primary cause of illiteracy among Americans is a failure to learn to read. Learning to read requires time and practice. Research indicates that once the basics of learning to read are in place, a grade-level gain in reading ability takes approximately 100 hours of engaged literacy training time. Further, at beginning levels of reading, individual feedback, motivation, and guidance are critical. For most adult illiterates a major obstacle to effective reading development lies in two simple facts—the human resources do not exist to provide the teaching support that is needed, and there is no way to adequately increase their number to provide such support during the next several years. A sufficient force of trained professionals and paraprofessionals at the level of expertise required cannot be developed, even with a massive injection of funding. The only option is the effective introduction of appropriate technology.




RESEARCH PROJECT ON ADULT ILLITERACY

My thesis is that one technology in particular—computer-based speech recognition—provides the central and essential capability required for launching a significant attack on a major segment of the adult illiteracy problem. The reason for this view is straightforward. Many adults who have serious reading difficulties can speak English intelligibly, even when their speech is in a dialect other than standard U.S. dialects. We at BBN plan to begin from their strength—their ability to speak English. They know how easy it is to talk and how hard it is for them to read. They do not know that their own speech can be transformed into text, that there is a direct, albeit complex, correspondence between the two forms of language. Even though they cannot initially read the text that is produced, they know that it is their own, that they created it from their own speech. So in a very real sense they know what the text "says." Further, they can repeat the utterance, either in its entirety or partially, a word or phrase at a time. In doing so, they begin to get a handle on the translation problem through a procedure they substantially control, in ways and at rates they find comfortable.

The starting point in this instructional strategy is to have the computer generate a speech utterance. The trainee is then asked to repeat the utterance, to "say the same thing." The speech recognition system prints the correct text corresponding to the utterance, but it remembers the trainee's articulations of the constituent words and phrases, including those not spoken in standard English. So the system develops a knowledge of the trainee's idiosyncratic pronunciations and the text of the words and phrases to which they correspond. This information will be used to generate the correct English text in activities where the trainee takes the initiative in generating the utterances.

This training strategy is distinctly different from traditional approaches based on giving trainees a reading task from the start. Instead, reading skills are developed in a more natural and gradual way from speaking and listening skills that, to begin with, are a great deal stronger. The transition from the scaffolding of speech-driven pattern matching to text-driven decoding and interpretation is not simple, but the confidence inspired in the trainee by being able to begin to make sense of text should substantially help in confronting and bridging this gap. In later phases of the training, students will be engaged more in practice activities to extend their vocabulary and to apply reading in contexts of use through task-oriented activities. The training sessions will involve the use of workplace literacy materials, including interactive simulations, games, and extended stories, gradually increasing in the level of challenge and complexity. The functional literacy tasks will be based on those defined by the National Adult Literacy Survey, particularly those in the first three levels (e.g., interpreting bus or airline schedules; following written and illustrated instructions from a manual or display for such tasks as repairing a paper jam in a copying machine). Interactive computer simulations of such tasks will involve the development of procedural reasoning and problem-solving skills in addition to reading comprehension skills.

This approach requires continuous, real-time, high-accuracy, dialect-sensitive speech recognition. With the introduction of such new and powerful facilities, the development of a literacy trainer incorporating two-way interactive speech technology has become feasible. Earlier speech recognition systems were not capable of recognizing with high accuracy, naturally spoken utterances involving large vocabularies and complex grammars. Moreover, they were not "speaker independent" in that their accuracy for any speaker was highly dependent on whether the system was tuned to the speech characteristics of that speaker, through a tedious, time-consuming "training" process. All this has changed with the introduction of systems such as the BBN HARK recognizer. Given the vocabulary and grammar associated with a student session, HARK can recognize utterances with a word accuracy of around 95 percent. Furthermore, the system can readily be configured to recognize speakers with Haitian, Cambodian, Hispanic, and other common dialects.

This experimental research and development project would be conducted jointly by industry and university researchers. I envisage a 10-year program to implement a comprehensive yet inexpensive training system, demonstrate its instructional capabilities, and evaluate its learning benefits with a representative group of trainees chosen to exemplify a wide range of literacy problems. A two-year small-scale pilot to develop a prototype system and demonstrate the instructional approach might constitute the first stage of the project. It would endeavor to show how speech-mediated literacy training technology can effectively be used to make major inroads on our national illiteracy problem. Speech scientists, instructional researchers, and software developers are keenly interested in participating in the proposed work. The project would have the long-term goal of substantially reducing the adult illiteracy problem in the United States and opening the way for fuller participation of all citizens in the knowledge society.




REFERENCES

Bier, Melinda. 1996. Personal Empowerment in the Study of Home Internet Use by Low-Income Families. World Wide Web article, http://www.educ.ksu.edu/projects/jrce/v1/Bier/article.html.

Horwitz, Paul. 1995. Electronic communication, Department of Education discussion group on a National Technology Plan, March 20, Kirk Winters, moderator. Archived in inet.ed.gov under NTPlan.

National Telecommunications and Information Administration. 1995. Falling Through the Net: A Survey of the "Have-Nots" in Rural and Urban America. Washington, D.C.: U.S. Department of Commerce.




ELECTRONIC ACCESS TO SERVICES FOR
LOW-INCOME POPULATIONS

Adam Porter
University of Maryland





INTRODUCTION

Much of the Internet's popular appeal stems from its potential to provide a global virtual marketplace for goods and services. Already high-profile projects are under way to develop video on demand, virtual malls, and massive digital libraries. In addition to these glamorous products, more mundane government services will some day be delivered (in whole or in part) via the Internet. These services might include Medicare, welfare and unemployment, immigration, and job placement and training assistance. Since many of these services exist to serve low-income populations, it is important to determine whether low-income users differ from other Internet users and, if so, what implications this has on user interface design.

For the past several years I have been working with a nonprofit group called Raising Hispanic Academic Achievement (RHAA). RHAA is located in the Washington, D.C., metropolitan area and provides academic tutoring and mentoring to several hundred Latino children and their parents. Over 70 percent of these children come from families whose annual income is below $20,000 and many receive government services such as those described above.




SURVEYING THE USER COMMUNITY

I conducted an informal survey of this user community to help understand its characteristics and needs. This is not a scientific sample, and it is clearly incomplete. Nevertheless, it provides some insight into issues to consider when designing interfaces for these types of government services. The survey identified several characteristics that might not be found in other user communities:

• Non-English speaking. Many respondents are first- or second-generation Americans. Consequently, many do not speak English at all or have limited proficiency.

• Limited educational background. Many of the respondents have not completed high school. Moreover, they have had little exposure to computers and computer programming.

• Limited literacy. Since many of the respondents do not speak English and have limited education, functional illiteracy in English is high. Furthermore, 20 percent (estimated) of the parents I surveyed are also illiterate in Spanish.

• Limited access to computers. Few of the respondents have computers in their homes. Those who do tend to have lower-end machines with limited storage and printing capabilities. Almost all have phone or pager service. Some of those without home computers have access to the Internet through schools and/or public libraries.





CONSIDERATIONS FOR USER INTERFACE DESIGN

Based on this survey I have identified several issues and concerns that should be considered when developing user interfaces. Many of these, of course, will apply to other communities as well. I have grouped these issues into three categories: (1) computer literacy, (2) language literacy, and (3) limited computer resources.




COMPUTER LITERACY

• Simple predictable interfaces. Users often have to formulate queries to search large information spaces. Frequently, they must write these queries using SQL or logic programming languages. Since many of these users have no programming experience, this type of interface leads to errors and should be avoided when possible. Also, interfaces should be predictable, not changing from use to use, since this can lead to confusion.

• Rapid, incremental, and reversible control. Users should be encouraged to navigate large information spaces quickly and easily. Rather than programming queries, users should be able to visually define queries and then refine them rather than recomputing from scratch. Also, each operation should be reversible.

• Training. One research area that should be explored is the development of novice versus expert interfaces. To support novice users better, error recovery and prevention schemes will be important, as will further study into architectures for "help" systems.





LANGUAGE LITERACY

• Direct manipulation. Since even experts make frustrating typing mistakes when using textual interfaces, the illiterate will find these interfaces unusable. Although direct manipulation interfaces alleviate some of these problems, there is still a strong need for more effective visual presentation approaches.

• Audio support. For people who cannot read or whose language is unwritten (e.g., certain Creole languages), audio support may be necessary. In these cases, text-to-speech systems may provide low-cost conversion of written content.

• Culturally appropriate presentation. Different languages are presented and organized differently. They have different accent marks, character sets, and orientations in which words are read and written. (This problem also appears in commercial software development where products must be internationalized.) Rather than expending scarce resources to translate content, reformat output, and redevelop character-processing code, some research should be devoted to low-cost translation systems, flexible software architectures, reconfigurable browsers, and so forth.





Limited Computer Resources

• Presentation on low-bandwidth devices. Many people do not have computers in their homes and have limited access elsewhere. As interface developers we cannot assume that everyone is using Netscape. We need to explore methods for low-cost conversion of content for different devices (e.g., phones, beepers). For example, Lucent Technologies has developed a language that allows HTML documents to be presented in one way for a phone and in another way for a standard browser.

• Users without fixed Internet addresses. Low-income people may have computer access only through public channels, such as libraries and schools, rather than at home. Notification services that assume a fixed or a forwarding address will be inadequate. Some research issues to consider are secure identification mechanisms and flexible locator services.

• Long-running services. Providing services may require many steps, with human intervention at several points. Therefore, users may need to suspend long-running services and resume them later (possibly from a different physical location, with different computer resources, user names, etc.). Consequently, architectures need to be developed for incremental interaction and interfaces that summarize interaction histories.





SUMMARY

This paper explores the design of user interfaces for systems that provide electronic access to government services. Since these systems exist to serve low-income populations, I conducted an informal survey of one low-income population to understand their needs. The survey indicated that low-income populations will have many novice computer users, require support for languages other than English, and have limited access to computer resources. Since redeveloping content for every user community would be prohibitively expensive, this situation presents a wide variety of research challenges. Three of the most interesting topics are (1) visual presentation and searching of large information spaces, (2) low-cost translation and conversion of content, and (3) flexible software architectures and interaction patterns.




ACCESS FOR PEOPLE WITH DISABILITIES

Larry Goldberg
WGBH Educational Foundation





INTRODUCTION

The barriers that prevent persons with limited sight, hearing, or mobility from gaining access to the fruits of the emerging national information infrastructure (NII) are not all that different from the barriers that face all the other millions of Americans who have, voluntarily or not, opted out of participation. Limited bandwidth, limited technological resources, limited technological facility, limited income, limited time, and limited interest are the restraints bifurcating our society. This division is self-perpetuating in that without interventions like the promotion of every-citizen interfaces the gap will widen and feed on itself.

For people with disabilities, input/output issues are the core concerns. Disabled people share with their able-bodied colleagues concerns about the complexity of systems, training needs, flexible and intelligent interfaces, and so forth. But without the ability to input commands and data and receive appropriate output, the rest of the issues are irrelevant.




TODAY'S MEDIA

For the more basic, one-way components of the NII (i.e., television in its broadcast, cable, satellite, video-on-demand formats), provision of access to the content flowing over these pipelines continues to be an after-the-fact retrofit. That is, upon finalization of programming, closed captions or video descriptions are added. In the case of the former, caption data are encoded into the vertical blanking interval (VBI) of the television signal and delivered and displayed to the end user via a set-top box or built-in decoder chip. In the case of the latter, an additional audio track is added to the master video and delivered to the end-user via an auxiliary audio channel (the Second Audio Program) on stereo televisions and video cassette recorders.

Though widely utilized for decades (closed captioning) and years (descriptions), these technologies are by no means 100 percent reliable, owing to both system errors and human errors. Research into fail-safe mechanisms for delivery of these data would provide the benefit of consumer confidence and programmer satisfaction.

In addition, the distribution plants of the major broadcasters and cable system operators often either cannot traffic the extra audio channel for video description or will strip those descriptions prior to delivery to the home. Preliminary research into digitization of the auxiliary audio channel and encoding into the VBI (like in the captioning process) has been conducted with promising but inconclusive results. The added question requiring an extra user appliance for decoding VBI-encoded audio needs to be examined. Research may determine means of adding such capability into today's or next-generation television systems.

Production of access services also is time and labor intensive. Though trivial in terms of the overall cost of mass media production, budgets for access services are carefully scrutinized. This has become an even greater concern as the sources of information and entertainment continue to grow. As we approach the ability for every citizen to become a programmer/distributor, the need to facilitate and lower the cost of producing and adding access adjuncts or alternate means of output (text and audio) becomes more dire.

Potential research in this area needs to focus on the utilization of speech-to-text and text-to-speech technologies. Neither has yet been developed to the point of aiding in the production or distribution process of either captions or descriptions, but both show tremendous promise. Today, most of the burden for providing captioning on live programming falls on a extremely small group of highly skilled "stenocaptioners" who can outgun any speech recognition engine available today.

Captioning has clearly proven itself to be a service that serves populations far beyond the originally intended audience. Early optimistic explorations into the use of captioned television for early and remedial readers has been followed by exploitation of captioning for students of English as a second language.

Computer manufacturers are now building caption displays into their "personal computer/televisions" with the added capability of downloading caption data for archiving and indexing purposes. And researchers in the field of digital video storage have long been excited about the use of captions as indices for large-scale video databases for research, archiving, and production uses.

Perhaps more trivial but ever more evident, captioned television has become widely used in environments that require an alternative to audio output, such as health clubs, bars, libraries, and for late-night television watching.

Video description uses for additional populations have just begun to be explored and would certainly benefit from significant research projects. Preliminary work has begun to determine the potential of description to assist students and adults with learning disabilities, especially attention deficit disorder. Additional similar benefits may be discovered through research into video description's impact on information retrieval where masses of data and information clutter could be reduced through "descriptive guides."

Embedded description has also been suggested as a potentially useful way of indexing large visual databases if coupled with intelligent speech recognition. Little or no research has yet been conducted in this area, but that may point the way to another repurposing of an access technology.




NEW MEDIA

In the worlds of nonlinear and digital multimedia, interactive media, and advanced television, many problems remain to be solved. Standards are either evolving or have not yet been considered for incorporation of captioning and description in all of these venues. Early efforts at creating such standards ignored the ancillary uses of these technologies and threatened to lock in bad designs.

For example, the ability to display text in the emerging DVD (Digital Video Disk or Digital Versatile Disk) format has been designed with language translation as the primary purpose for text display. Since the DVD format has mostly been designed in Japan, the text is being created through bit-mapped graphics files, not ASCII text as in broadcast closed captioning. This has resulted in many unforeseen design flaws: the inability of a DVD player to address the closed-captioning circuitry in television sets, the inability to flexibly change fonts or type sizes, and the inability to use the text as a search engine.

These decisions are made as new media products rush to market, with little attention paid to the best-possible design for the built-in access services and how they can be configured for universal applicability. Similar concerns are being directed toward the designs of the alternate audio features of advanced television and DVD.

As great a concern has been expressed about the "appliances" people will be using to access the Internet or other digital information and entertainment systems. With personal computers the ability to add software or hardware to assist those with special needs is relatively straightforward (though not without significant programming and development efforts). But if every citizen will be accessing the new services via "thin clients," smart phones, set-top boxes, or low-end, low-cost browsing boxes attached to television sets, the ability to add such services as speech synthesis or output to refreshable Braille or large-print displays is problematic.

When digital sound files are sent over the Internet and received via these boxes, how will the boxes turn the sound into text for deaf people?

Research into the ability to incorporate these alternate output modes into inexpensive information appliances can result in more accessible low-end browsers for every citizen, including blind and deaf people. Incorporation of new access standards into digital media formats (such as QuickTime, Active Video, and RealAudio) will facilitate the ability of low-end browsers to display incorporated access technologies. All of these research challenges can be addressed in the short term (one to five years), except perhaps for the need for a fully capable speech recognition technology, which, after 20 or more years of effort, still requires many years to approach the speed, accuracy, and other capabilities needed for use by deaf people.

It has been the experience of those in the world of media access that these design and research challenges are not complex (certainly not as complex as the creation of the new media themselves). When focused attention and resources are applied to the problems, solutions are readily discovered, especially when approached by consortia of public and private practitioners and researchers. Early awareness and design-from-the-blueprint-stage thinking obviate the need for expensive and inefficient retrofits that are resisted by producers and consumers alike.




SUGGESTED UNIFORM RESOURCE LOCATORS

http://www.wgbh.org/caption

http://www.wgbh.org/dvs

http://www.wgbh.org/ncam

http://trace.wisc.edu