25 May 1999


Date: Tue, 25 May 1999 05:59:45 +0200 (CEST)
From: Anonymous <nobody@replay.com>
To: cypherpunks@toad.com

High-tech snooping tools developed for spy agency

The Vancouver Sun (May 24, 1999)

Jim Bronskill
Southam Newspapers

OTTAWA -- Canada's electronic spy agency is quietly bankrolling the development of cutting-edge systems that can identify voices, analyze printed documents and zero in on conversations about specific topics.

Documents show the Communications Security Establishment has enlisted the help of several leading Canadian research institutes to devise state-of-the-art snooping tools.

CSE, an agency of the defence department, collects and processes telephone, fax and computer communications of foreign states, corporations and individuals. The federal government uses the intelligence gleaned from the data to support troops abroad, catch terrorists and further Canada's economic goals.

CSE and counterpart agencies in the United States, Britain, Australia and New Zealand share intercepted communications of interest with one another, effectively creating a global surveillance web, according to intelligence experts.

CSE's interest in high-tech devices that help locate specific conversations and documents is a clear indication the five-member alliance collects and sifts large volumes of civilian traffic, said Bill Robinson, a researcher in Waterloo, Ont., who has long studied the spy agencies.

"This technology is needed to process vast communications streams when you're hunting for nuggets within it."

Robinson said the devices have legitimate uses, but hold "potentially frightening" implications for people's privacy as the technology advances.

The Centre for Pattern Recognition and Machine Intelligence, located at Concordia University in Montreal, received $355,000 to develop two systems for CSE that automatically analyze printed documents, such as faxes, once they are digitally captured in a computer data bank.

The first system, completed early last year, quickly determines the language of a document, said the centre's C. Y. Suen.

"Some humans may have problems in distinguishing Spanish from Portuguese, for example, or Spanish from Italian," he said. "So what we have developed is a system that can do it automatically."

The second device electronically searches captured documents for distinct features, including logos, photos, text or signatures.

Combining the two systems enables a user, for example, to search a data bank for Japanese documents containing photos, or Russian faxes with signatures.

Records obtained by Southam News under the Access to Information Act show CSE commissioned several other projects during the last two years. They include:

- An $84,981 contract with the University of Waterloo in Ontario for the "development of multilingual computer speech recognition systems."

- A $115,000 agreement with the University of Quebec at Chicoutimi to research "speaker identification" procedures.

- Work by the Centre de Recherche Informatique de Montreal on "topic spotting" -- a means of identifying the subject of a conversation. The $150,393 contract was the most recent of several awarded to CRIM.

CSE spokesman Kevin Mills did not provide information on specific goals of the projects, but allowed: "In general, any research that we're funding has some kind of interest for CSE."

The agency has been working on voice and phrase-detection systems for at least a decade. The documents, however, show the research continues, with some devices yet to be perfected.

CSE and its four international partner agencies use computers capable of recognizing intercepted messages containing specified names, addresses, telephone numbers and other key words or numbers, says a new report on surveillance technology, by Scottish researcher Duncan Campbell.

However, Campbell found the agencies lack systems for homing in on conversations featuring particular words.

CSE would have trouble picking out a phone call with the words "assassination" or "revolution" because the speech recognition systems developed to date cannot instantly recognize an unknown person's voice traits.

"The key problem, which is familiar to human listeners, is that a single word heard on its own can easily be misinterpreted, whereas in continuous speech the meaning may be deduced from surrounding words," says Campbell's report.