New NSA patent explicity mentions machine transcription

[Julian Assange <proff@iq.org>]

In today's Indy:

   http://www.independent.co.uk/news/Digital/Features/spies151199.shtml

 By Suelette Dreyfus                                                                                                           
                                                                                                                                 
   15 November 1999                                                                                                              
                                                                                                                                 
   The US National Security Agency has designed and patented a new
   technology that could aid it in spying on international telephone
   calls. The NSA patent, granted on 10 August, is for a system of
   automatic topic spotting and labelling of data. The patent
   officially confirms for the first time that the NSA has been
   working on ways of automatically analysing human speech.
                                                                                                                                 
   The NSA's invention is intended automatically to sift through human
   speech transcripts in any language. The patent document
   specifically mentions "machine-transcribed speech" as a potential
   source.
                                                                                                                                 
   Bruce Schneier, author of Applied Cryptography, a textbook on the
   science of keeping information secret, believes the NSA currently
   has the ability to use computers to transcribe voice conversations.
                                                                                                                                 
   "One of the holy grails of the NSA is the ability automatically to
   search through voice traffic. They would have expended considerable
   effort on this capability, and this indicates it has been
   fruitful," he said.

   To date, it has been widely believed that while the NSA has the
   capability to conduct fully automated, mass electronic
   eavesdropping on e-mail, faxes and other written communications, it
   cannot do so on telephone calls.
   
   While cautioning that it was difficult to tell how well the ideas
   in the patent worked in practice, Schneier said the technology
   could have far-reaching effects on the privacy of international
   phone calls.
   
   "If it works well, the technology makes it possible for the NSA to
   harvest millions of telephone calls, looking for certain types of
   conversations," he said.
   
   "It's easy to eavesdrop on any single phone call, but sifting
   through millions of phone calls looking for a particular
   conversation is difficult," Schneier explained. "In terms of
   automatic surveillance, text is easier to search than speech. This
   patent brings the surveillance of speech closer to that of text."
   
   The NSA declined to comment on the patent. As a general policy, the
   agency never comments on its intelligence activities.
   
   Yaman Akdeniz, director of Cyber-Rights & Cyber-Liberties UK,
   warned that with the new patent and a proposed AT&T and BT joint
   venture, which will allow US law enforcement agencies to tap the
   new communications network: "We might have a picture in which all
   British communications are monitored by the NSA."
   
   The revelation of the NSA's patent is likely to cause tensions with
   the European Parliament. Over the past two years, the Parliament
   has commissioned several reports which examined whether the NSA has
   been using its electronic ears for commercial espionage,
   particularly in areas where US corporations compete with European
   and other companies.
   
   The NSA relies on an international web of eavesdropping stations
   around the world, commonly known as Echelon, to listen into private
   international communications. The network emerged from a secret
   agreement signed after the Second World War between five nations
   including Australia, New Zealand, Canada, Britain and the US. Two
   of the NSA's most important satellite listening stations are
   located in Europe, at Menwith Hill in Yorkshire and Bad Aibling in
   Germany.
   
   Julian Assange, a cryptographer who moderates the online Australian
   discussion forum AUCRYPTO, found the new patent while investigating
   NSA capabilities.
   
   "This patent should worry people. Everyone's overseas phone calls
   are or may soon be tapped, transcribed and archived in the bowels
   of an unaccountable foreign spy agency," he said.
   
   One of the major barriers to using computers automatically to sift
   through voice communications on a large scale has been the
   inability of machines to "think" like humans when analysing the
   often imperfect computer transcriptions of voice conversations.
   
   Commercial software that enables computers to transcribe spoken
   words into typed text is already on the market, but it usually
   requires the machine to spend time learning how to understand an
   individual voice in order to produce relatively error-free
   text. This makes such software impractical for a spy agency which
   might want automatically to transcribe and analyse telephone calls
   on a large scale.
   
   It is also difficult for computers to analyse voice conversations
   because human speech often covers topics that are never actually
   spoken by name. According to the NSA patent application, "much of
   the information conveyed in speech is never actually spoken
   and... utterances are frequently less coherent than written
   language".
   
   US Patent number 5,937,422 reveals that the NSA has designed
   technology to overcome these barriers in two key ways.  First, the
   patent includes an optional pre-processing step which cleans up
   text, much of which the agency appears to expect to draw from human
   conversations. The NSA's "pre-processing" will remove what it calls
   "stutter phrases" associated with speech based on text.
   
   Second, the patent uses a method by which a computer automatically
   assigns a label, or topic description, to raw data.  If the method
   works well, this system could be far more powerful than traditional
   keyword searching used on many Internet search engines because it
   could pull up documents based on their meaning, not just their
   keywords.
   
   Dr Brian Gladman, former MoD director of Strategic Electronic
   Communications, said that while he doubted the NSA had deployed the
   patented system yet, the new technology could become a "potent
   future threat" to privacy.
   
   "If the technology does what it says automatically finding and
   extracting the meaning in messages with reasonable accuracy then it
   is way ahead of what is being done now," he said.
   
   The best way for people to protect their private communications was
   to use encryption, he said. Encryption software programs scramble
   data to prevent eavesdropping. "I'm afraid widespread interception
   is a fact of life and this is what makes encryption so important,"
   he said.
   
   "The problem in the UK is that our government is working with the
   US to prevent UK citizens defending themselves using encryption,"
   he said, referring to the continuing use of export controls to
   hamper the widespread availability of encryption products.
   
   The NSA's current spy technology may be more advanced than methods
   described in the patent because the application is more than two
   years old. The US Patent Office approved the patent on 10 August
   this year, but the NSA originally lodged the application on 15
   April 1997. The US Patent office keeps all applications secret
   until it issues a patent.