March 01, 2005

The dogs of speech technology

The association of dogs and speech research in Geoff Pullum's recent post reminds me of a story. The time was 1977, about 3:00 a.m. one cold winter night. The place was Murray Hill, N.J., the home of AT&T Bell Labs research. Building 2, wing D, 4th floor, in the console room of the DDP-224 interactive computer. Joe Olive and I had been programming since dinner, and our new speech synthesis system was pronouncing its first phrases: "The birch canoe slid on the smooth planks." "Mesh wire keeps chicks inside." "The spot on the blotter was made by green ink." It sounded really good to us, maybe not totally natural, but clear as crystal.

One of the night-shift cleaning crew came in to empty the trash can. He looked around, puzzled, for the source of the sound. "Y'all got dogs in here?"

This was a good lesson on the role of expectations in speech perception, and the importance of formal intelligibility testing.

A similar point is one of the themes of Solzhenitsyn's novel "The First Circle", which deals with speech research in a Stalinist labor camp. But that's a story for another time.

Posted by Mark Liberman at March 1, 2005 12:03 AM