intro’d to Sharon here:
New Scientist (@newscientist) tweeted at 5:36 AM – 4 Apr 2017 :
Google uses neural networks to translate without transcribing https://t.co/glRhZ1pbIJ https://t.co/2Z76KWNsGV(http://twitter.com/newscientist/status/849224178020626432?s=17)
Machine translation of speech normally works by first converting it into text, then translating that into text in another language. But any error in speech recognition will lead to an error in transcription and a mistake in the translation.
Researchers at Google Brain, the tech giant’s deep learning research arm, have turned to *neural networks to cut out the middle step. By skipping transcription, the approach could potentially allow for more accurate and quicker translations
The system could be particularly useful for translating speech in *languages that are spoken by very few people, says Sharon Goldwater at the University of Edinburgh in the UK.
*ie: idiosyncratic jargon
International disaster relief teams, for instance, could use it to quickly put together a translation system to communicate with people they are trying to assist. When an earthquake hit Haiti in 2010, says Goldwater, there was no translation software available for Haitian Creole.
Goldwater’s team is using a similar method to translate speech from Arapaho, a language spoken by only 1000 or so people in the Native American tribe of the same name, and Ainu, a language spoken by a handful of people in Japan.
The system could also be used to translate languages that are rarely written down, since it doesn’t require a written version of the source language to produce successful translations.
And text translation service Google Translate already uses neural networks on its most popular language pairs, which lets it analyse entire sentences at once to figure out the best written translation. Intriguingly, this system appears to use an “interlingua” – a common representation of sentences that have the same meaning in different languages – to translate from one language to another, meaning
it could translate between a language pair it hasn’t explicitly been trained on.
The Google Brain researchers suggest the new speech-to-text approach may also be able to produce a system that can translate multiple languages.
like 7 bn.. toward a nother way to live..
But while machine translation keeps improving, it’s difficult to tell how neural networks are coming to their solutions, says Bahdanau. “It’s very hard to understand what’s happening inside.”
Sharon Goldwater is a Reader in our Institute for Language, Cognition and Computation. She is an internationally leading researcher who has made major interdisciplinary research contributions across natural language processing, machine learning, and computational cognitive science. In particular, her work on Bayesian models for probabilistic machine learning have broken new ground in unsupervised learning of linguistic structure.
She explains: “Current speech and language technology uses supervised methods: the computer system learns about a specific language based on examples that are annotated by humans. This method works well for the relatively few languages where substantial linguistic annotations are available, but my focus on unsupervised language learning has the potential to make speech and language technology available to the more than 5,000 languages used across the world.”
Her work has made an impact in areas as diverse as speech technology (Bayesian language models for speech recognition) and child language acquisition (computational models of how children learn words and phonemes). Her research addresses the fundamental question of how a computational system can learn to analyse the structure of whatever human language it is exposed to, in an unsupervised or minimally supervised setting. Her examination of this question takes inspiration from both child language development, which suggests that successful language learning integrates multiple probabilistic sources of information in a structured way, and probabilistic machine learning, which provides the tools to develop systems that can do so.
I’m also grateful that the work on this domain has received so much attention. Studying how people use and learn language, and how machines can mimic this behaviour, can help us understand our own cognitive abilities and help create better systems that will be simple and efficient for people to use.
Regina Dugan on facebook’s non-invasive brain-computer interface that will let you type at 100 wpm — by decoding neural activity devoted to speech.
because – what is legible ness.. who decides..