Technologies begin catching up with market expectations

Many people have experienced the fun or frustration of playing with technologies that are not yet ready for prime time. Just consider Newton’s pilloried handwriting recognition, mobile phones in the days of spotty networks, or early attempts at plug-and-play (AKA “plug and pray”) software. Machine translation (MT) and automated speech recognition (ASR) fall into that category of innovations that have taken a while to establish their credentials and gain market acceptance. Who hasn’t laughed at MT output or been frustrated by the interactive voice response (IVR) systems that don’t seem to understand that you really do want to talk to a human representative?

Text by Donald A. DePalma

Inhaltsübersicht

Technologies begin catching up with market expectations

My first experience with MT was writing my own simple English<>Russian translation program in the late 1970s for a course in computational linguistics. Boasting a small lexicon and a relatively simple grammar that included support for dependent clauses and variable sentence structures, it earned me a passing grade as it demonstrated the complexity of handling unpredictable natural language and gave me reasons to empathize with MT developers. My introduction to ASR was at Digital Equipment Corporation in the early 1980s where we experimented with the software that became DECtalk. The contemporaneous speech recognition software was clunky at best and very dependent on the speaker’s clear, slow enunciation and patience.

Over the last few years, both MT and ASR have entered mainstream use. Several years ago Google’s translation tool began offering users tolerable output that has gotten ...