The Turing Test and Translation

Many people have heard of the Turing Test, but not everyone knows about Alan Turing and his background. Alan Turing was a British mathematician, computer scientist, cryptanalyst and philosopher. He developed the principles of artificial intelligence and his work was instrumental for the creation of computer science. Turing also played an important part in breaking the Enigma code in the Second World War, a major contribution to Allied victory.

  • In 1950, he suggested that artificial intelligence could be answered with an imitation game. The referee talks to a person on one line and a computer imitating a person in the other. If the human judge cannot distinguish between the two, then there is no reason to say that the machine does not think.
  • Forty years later, the game was played. In 1990, The Cambridge Center for Behavioral Studies designed a contest to implement what has come to be known as the Turing Test. Dr. Loebner pledged a prize of $100,000 for the first computer indistinguishable from a human, and each year, a cash prize is awarded to the most human-like computer.
  • For now, it is only a game, left in the hands of amateurs.  Not a single professional involved in the development of artificial intelligence can assign a date in any predictable future to the replication in machines of the human ability to think.

What machines are missing for the Turing Test is relevant for machine translation too:

  • Machines are bad at making decisions. They have the most impressive memory, but when it comes to making choices, they are helpless. Since decision-making is of the essence to parse a sentence, they could never know if “Chinese apeman dated” involves a microscope or a table for two.    
  • Computers lack common sense. The human language skill of analyzing a sentence into its component categories and functions relies heavily on our experience of what are the odds of something happening in the real world.
  • People favor certain structures and word pairs. Unlike machines, we have the instinctive ability to follow certain paths that give us a head start in parsing sentences.
  • Humans have expectations while machines do not. Humans frequently rely on their own expectations and the multiple layers of meaning that distinguish them as crafty, duplicitous, cunning, intriguing, calculating, ingenious social animals.

Computers are still a long way off from passing the Turing Test, even in just one language, let alone two.