ings, and civil- and criminal-court trials. Unless a third party has already transcribed, closed-captioned, or applied speech-recognition techniques on the record, most reporters have no way to search even a rough transcript. In addition, many reporters record many of their interviews digitally but rarely have useful speech-recognition software to index them. Basic consumer software products (such as Dragon-speech from Nuance) work on simple, short recordings or trained voices. Other promising projects (such as Google’s Audio Indexing, or GAUDi) are not publicly available. GPS and voice recognition on mobile phones and voice mail could make reporters think solving their problem is simple.

Reporters could make near-daily use of technology and a user interface that would provide approximate indexing of a variety of voices and conditions, leading them to the portions they most want to review. They do not require the accuracy of, say, e-discov-ery by lawyers or official government records. Instead, they want a quick way to move to the portion of a recording that contains what may be of interest, then carefully review and transcribe it. Existing technology is probably adequate for reporters’ immediate needs, but we are unable to find reasonably simple user interfaces to the technology that would allow unsophisticated users to test the technology on their own recordings.

Extracting data from forms and re-
ports. Much of the information col-
lected by reporters arrives in two
genres: original forms submitted to
or created by government agencies,
often handwritten, and reports gener-
ated from larger systems, sometimes
electronically and sometimes on pa-
per. Examples include financial dis-
closure statements of elected officials,
death certificates, safety inspections,
sign-in sheets at government check-
points and police incident reports.
Journalists have few choices today:
retype key documents into a database;
attempt to search recognized images;
or simply read them and take notes.
An in-house programmer can occa-
sionally find the pattern of digital re-
ports intended for printing that can
be leveraged to reverse them back into
a structured database, but this time-
consuming job requires skill well be-
yond nearly all reporters.

new tools, new organizations

A handful of new services have emerged to help address journalism’s data challenges. Usually free for small-scale or non-commercial use, they facilitate analysis, visualization, and presentation of structured data: Google Refine promises to let reporters scrap their spreadsheets for filtering, viewing, and cleaning basic data sets; ManyEyes from IBM lets news organizations visualize and share data on their Web sites; Tableau Public from Tableau Software, Google Earth, and other such products are routinely used by news organizations to generate and publish visualizations. New tools (such as TimeFlow developed at Duke University as an investigative tool for temporal analysis) are being created to address some longstanding needs of reporters. 4

Another set of tools created for other purposes, often experimental or academic, shows promise for the fast-paced, ad hoc nature of reporting challenges. Several political-science scholars have created tools for clustering legislation and other public documents; homeland-security developers have created tools (such as Georgia Tech’s Jigsaw15) for visualizing the connections among documents; and the CMU Sphinx project3 has created reasonably accurate open source speech-recognition technology. Applications developed for intelligence, law enforcement, and fraud investigations by such companies as Palantir Technologies and I2 are expensive and finely tuned to specific industries, though they address similar challenges on a different scale and with different requirements for speed and accuracy.

DocumentCloud ( http://www.docu-
mentcloud.org), a nonprofit founded
in 2009 by journalists at the New York
Times and ProPublica, hopes to ad-
dress one of the most vexing issues in
documents reporting: scanned images
files. With it, reporters can annotate
their documents as they find inter-
esting or questionable sections and
see which entities appear in multiple
documents. At this writing, most news
organizations have used it most ef-
fectively to publish government docu-
ments. But the project, which includes
information extraction as a standard
feature, shows great promise helping
address some of the problems of di-
gesting large document collections.