At team of scientists lead by the Caroline Lyon at the University of Hertfordshire, UK have developed a “baby robot” that learns human speech in much the same way a human child learns. The robot converses with a human teacher and soon a string of incoherent babble from the robot beings forming actual words, in the same way a baby’s noises come together to form speech.
The project may lead to robots and AI that form words and sentences on their own, actually learning a language in the same way humans do. This would lead to more natural responses in AI and less errors when using programs like Siri.
It can also be very helpful in understanding how humans learn language in a more profound way. It is known that between the ages of 6 and 14 months, children begin to go from babbling strings of nonsensical syllables to actually forming words. Once they have a few base words, these then provide contextual clues for learning other words, eventually giving children a full vocabulary of words to choose from and form complete sentences.
The team at Hertfordshire was motivated by their understanding of this process to program a humanoid robot called iCub with all of the syllables in the English language. This allows the robot (who they call DeeChee) to babble (rather creepily) like a baby. Researchers then treated DeeChee like a child, speaking and attempting to teach the robot about rudimentary ideas like color and shape.
DeeChee was programmed in listen to the teacher and then respond with its own syllables. Each time it heard a syllable repeated by the teacher it would give that syllable a higher score in its 40,000 syllable lexicon. It would then use these syllables more often, and even begin forming words to match what the teacher said.
When the teacher showed approval for recognizable words, these syllables where given even more value in the computers programming. As the process continued, more and more recognizable words would begin to show up amid the babble. The scientists believe this closely replicates the early stages of speech development. Babies often become sensitive to repeat words and attempt to replicate them in their own speech.
“That words can emerge from babble using a statistical learning process not specific to language demonstrates that this stage of language acquisition does not require hard-wired grammar faculties”, said Lyon in the research publication PLoS One. In other words, this method is the absolute first step humans take in the process of learning speech. Grammar and the cognitive association of words follow in humans, but not with DeeChee, who will never use those syllables to form complete sentences or understand their meaning.
But it is the first step in what could eventually lead to advanced AI that develops its own language from the ground up. As Lyons puts it: “If you want the robot to work with natural speech, then you might need to teach it from the very beginning.”