Predicting sounds helps the brain to recognise words

New research on listening shows how the human brain uses its own form of predictive text.

In a recent study published in the journal Current Biology, three Cambridge researchers at the MRC Cognition and Brain Sciences Unit, Pierre Gagnepain, Rik Henson and Matt Davis, show that the brain is constantly using knowledge of familiar words to predict what speech sounds will be heard next. Matt Davis explains: "Many of us are familiar with using predictive texting on mobile phones – these systems try to guess what you want to say to save you the trouble of typing it. What we've shown is that the human brain uses a similar, but more sophisticated form of prediction in making sense of speech. The Superior Temporal Gyrus, a part of the brain involved in hearing, is constantly predicting what sounds will come next when listening to speech. So, having heard the syllable 'form…' rather than trying to guess whether the word is 'formal', 'formidable' or 'formula', the brain predicts which sounds would come next if each of these words were said. By predicting which sounds will be heard, the brain can respond to incoming speech extraordinarily quickly."

In the future, the group plans to look at how prediction might help make sense of speech sounds that are difficult to hear. Dr Davis continues: "We know that when listening to speech in a noisy environment - a crowded pub for instance - being able to predict what someone is trying to say can really help with understanding speech. Now that we know how the brain uses these predictions in processing speech, we might be better placed to help people with a hearing impairment who struggle to make sense of speech when there is background noise."

The picture, generated by Simon Strangeways, illustrates predictive coding as a neural mechanism that allows listeners to use knowledge of familiar words and their constituent sounds when understanding speech. Each speech sound is a star and familiar words are constellations. A human listener predicts future sounds when listening, so as to recognise the constellation ("formula") once the sequence of sounds ("formu...") uniquely identifies a single word.

Details of the paper can be found here. There is more information about this, and other related research, on Matt Davis's webpage.