Technology Quarterly: Computers have got better at translation, voice recognition and speech synthesis but still don't understand the meaning of language
05.01.2017

Technology Quarterly: Computers have got better at translation, voice recognition and speech synthesis but still don't understand the meaning of language

5 January 2017- As consumers begin to use new technology products received as gifts over the holidays, many will test out the speech recognition or translation capabilities available on their devices. Talking and listening to computers like Apple’s “intelligent assistant” Siri, first introduced in 2011, has been fraught with problems until recently.  Communications have been improving rapidly, helped by “deep learning” enabled by digital neural networks. 

A new report from The Economist examines the state of speech recognition, speech synthesis, machine translation, sentiment analysis and voice recognition. It looks at products nearly ready for the consumer market as well as more basic research that offers the hope of breakthroughs later on.

The next trend could be the large-scale adoption of the use of speech for interactions between people and computers, according to Lane Greene, The Economist’s language columnist and author of this week’s Technology Quarterly titled, Finding a Voice.

Commenting on the report, Lane Greene said, “Consumers are not yet used to talking to their computers. Although most have tried the voice-recognition features and automated assistants on their smartphones, fewer than a third use them regularly. With steady improvements in quality, that will change. The best systems can understand even unstructured spontaneous conversations with over 90% accuracy, and do even better with careful speech and after brief training for a specific speaker. It may not be long before people will be routinely talking to their smartphones rather than typing on a tiny keyboard.”

This special report also looks at potential benefits and costs of increased communication between man and machine. For the blind or deaf, high-quality text-to-speech or automatic speech transcription promise to be life-changing. Speakers of rare tongues may no longer have to wait for years for software to run in their languages. Language technology will also interact with other artificial-intelligence-powered technologies like self-driving cars. Inevitably, it will eliminate some jobs currently done by people. For routine tasks such as summarising news or answering frequently asked questions over the phone, machines may replace humans in as little as a decade. But tasks that require a subtle understanding of meaning are unlikely to be automated away in the foreseeable future.

To interview Lane Greene on this topic, please contact:  HollyDonahue@economist.com