Speak a little too haltingly and with long pauses, and a speech-to-text transcriber might put harmful, violent words in your mouth, Cornell researchers have discovered.