Newborns use multiple cues to separate the words in speech

We know that babies learn language incredibly fast.  One question that LuCiD is trying to answer is how they do it!  We often think that to learn language infants have to figure out what words mean and how they are organised.  But, to do that, they first have to figure out where the words are!  Language is tricky in that we don’t leave silences between the words, (like we leave spaces when we write) so picking out the words is not so easy.  Think of trying to listen to a language you don’t speak, it’s quite difficult to figure out how to break it into words!

We were interested in trying to figure out if infants are born with any strategies to help them pick out words inside speech.  In our recent paper, we tested 80 newborns and found that they can pick words out using some pretty impressive strategies.  We used near-infrared spectroscopy (an easy, non-invasive brain imaging technique) to see how their brain responded after we had them listen to a mini language.   We put certain cues in our mini language to see if those cues help them pick out words.  It turns out that the babies could notice patterns in the rhythm of speech and how often sounds occur together and could use these cues to help them pick out words!

The rhythm.  Everyone uses rhythm when they speak!  There are tiny changes the length of sounds and which sounds we emphasize when we talk. We do it spontaneously and we naturally exaggerate this when we address children.  For example, think about how you say, “We are going to the park!”  versus “We are going to the park?”.   Those are the exact same words but there are tiny differences in how you say them and those tiny differences are in the rhythm! Without even thinking about it, we do things like make the last syllables of words slightly longer. We wanted to see if newborns notice these differences and use them to help pick out the words.

To test this, we made a new mini-language that had four words, mipedu, vokanu, befite, and lugola.  For each word, we used the rhythm of someone actually speaking and copied this rhythm onto our words (so the ends of our words had longer syllables, for example).  We then played newborns our language for three minutes, with the four words in the same order, with no other cues to the boundaries (no pauses between words!).  After this we had them listen to the words, now with big pauses between them (and we took out the rhythm cues) and part-words, things like peduvo and kanube.  The tricky thing about part-words is that the babies heard these part-words during learning, but they don’t follow out rhythm rules (the long syllable would be in the middle). So, if the infants know that longer and lower-pitch syllables are at the ends of words, they should have learned that mipedu is a word, but that peduvo is not.  And that is exactly what we found! Their brains responded differently to words and part-words, suggesting that the infants had learned the word boundaries using just the rhythm!

The statistics.  Our next question was whether they could use other cues as well.  In language, there are also statistical patterns to word boundaries.  What that means is that sounds that appear together often are likely to be words and sounds that don’t appear together often are more likely to be boundaries between words. For example, the sounds ba- and –by occur together quite often because they make up the word baby, and the sounds ha- and –ppy occur together quite often because they make up the word happy.  But, the sounds-ppy and ba- occur together much less frequently.  This is because many words can come before baby (happy, little, hungry) and many words can come after happy (birthday, puppy, mummy).  So sound patterns that occur together much more frequently are likely to be words and sound patterns that are less frequent are likely to come from two words!

We wanted to see if newborns can do these kinds of computations as well, to help them sort out whether the words might be.  So we did the same kind of study, with our same mini language, but this time we took out the rhythm cues and we shuffled the order of the words around randomly.  So now the only cue to the word boundaries was how often the sounds occurred together: the three sounds in each word ALWAYS occurred together and the sounds at the word boundaries only SOMETIMES occurred together.  Again, we took a look at the how the newborn brain responded to words and part-words and found a difference!  This suggests that even at birth, the baby brain is running some pretty complicated analyses on the sounds that they hear and this gives them a head start on making sense of language.

These findings show that we are born with some strategies to help make sense of the language we hear.  This might help explain why infants learn language so fast, even at birth they start to be able to process language and break it down into words!  We are working on some new research as well to see if there are other cues that newborns can use and how these cues might help them remember the words that they hear.

Further details of this research can be found in our recent paper:

Fló, A., Brusini, P., Macagno, F., Nespor, M., Mehler, J., & Ferry, A. L. (2019). Newborns are sensitive to multiple cues for word segmentation in continuous speech. Developmental science, e12802.