Language learning by numbers

In order to understand a language, we need to learn to spot the individual words it contains, and we need to figure out whether these words are following any particular sets of rules. This is difficult to do (especially as there are no pauses between spoken words to help us), but these tasks are really important - without them, we would never be able to decode the sentence thegirliscrying, to know who was doing what (thegirliscrying) or when (thegirliscrying). Although these two tasks seem incredibly complex, both infants and adults are great at learning how to do them.

One of the things that helps us with this is our expert ability to spot patterns. Even without trying to, we can learn to identify new words from language by figuring out the likelihood that they are part of, or separate to, the words that they appear next to in speech¹. Likewise, we can also figure out whether these words are being presented in particular order, and are following certain rules². For instance, after enough experience with English we come to know that the, girl, is, and crying are all separate words, and that is and –ing appear together to describe something that is currently happening. These two tasks are critical parts of language learning, as they help us to understand what exactly is being said.  

Previously, research has suggested that these two tasks may occur in isolation, with learners first identifying individual words in speech, before figuring out the rules that those words adhere to³. However, we have reason to think that these tasks are in fact related, and can occur together at the same time – with the same type of learning process applying to both tasks.

To test this, in a recent study⁴ we trained adults on a new language that contained words and rules. The words were played for 10 minutes in a random order, with no pauses between them. The rules involved a relationship between the first and last sound of each word, in that certain sounds were always paired together, regardless of what sound came in between them. An example of this type of rule in natural language is the pairing of the word is and –ing; these often go together, regardless of what sound intervenes them (e.g. is crying, is running). Importantly, the only way that learners could identify these words and rules was by calculating the frequency with which sounds occurred, and in which order.

We tested whether adults had been able figure out what the words by playing them words and non-words, and asking them to say which ones seemed to fit the language they had heard. We tested rule learning in the same way, using new words that were either grammatically correct or grammatically incorrect. Learners performed both of these tasks successfully, and were just as good at identifying words as they were at identifying rules. Contrary to prior suggestion³ our findings indicate that learners can use the same information to identify the complex structure of language at the very same time that we’re trying to figure out what the words in the language actually are. These results indicate that the processes involved in the very first stages of children’s language learning may be richer and more complex than we previously thought.

References

1. Saffran, J. R, Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old  infants. Science, 274, 1926-1928.

2. Gómez, R. L. (2002). Variability and detection of invariant structure, Psychological Science, 13(5) 431-436.

3. Peña, M., Bonatti, L., Nespor, M., & Mehler, J. (2002). Signal-driven computations in speech processing. Science, 298, 604-607.

4. Frost, R. L. A & Monaghan, P. (2016).  Simultaneous segmentation and generalisation of non-adjacent dependencies from continuous speech. Cognition, 147, 70-74.