SBE-RCUK Lead Agency - The Development of Lexical Flexibility

One of the most striking features of human communication is our ability to use words in flexible and creative ways: We often use the same sound to mean multiple different things. For instance, we hammer using a hammer, eat chicken that comes from a chicken, and drink from glasses that are made of glass. These flexible uses of words provide us with a vivid expressive power: we can use a word with one meaning while alluding to all the rest. But while flexibility may make language more expressive, researchers have often assumed that it also makes language harder to learn. In particular, if a word's meaning is a flexible, moving target, then how are children ever supposed to learn it in a reasonable amount of time?

This has led a number of researchers to conclude that the words of an "ideal" language would be unambiguous. But in recent work we have developed a theory that suggests this assumption is backwards: Flexible uses of words do not impede children from learning words, in fact, they appear to facilitate it. The theory is best explained by an example. Imagine that you are a parent of a small child, and you want your child to avoid hitting their head on the leg of a nearby table. Your child does not yet know what the leg is called, but has learned that long, straight, supportive body parts are called "legs". When they hear you say "Watch out for the leg" and do not see any relevant body parts, they can guess that "leg" refers to something related to the body part, and thus might have a reasonable chance of guessing what you mean. By contrast, if you had to use a different, unambiguous word for each concept that you wanted to express, you might have to say something like "Watch out for the dax". Since your child has no idea what "dax" means, they are unlikely to be able to heed your warning.

Consistent with this, we have uncovered initial evidence that children are able to master flexible language with minimal difficulty. This project aims to understand how children are able to do that, based on a combination of tightly-controlled behavioural research on developing children, and the annotation and analysis of transcripts of children's and parents' speech.

First, we will build on and analyze existing databases of children's conversations with parents and others. We will code these databases for some of the particular features of flexible words, and then use advanced statistical methods to examine the linguistic environment that adults provide for children, and how children respond to that environment.

Second, we will conduct a series of studies that assess how children between 2- and 4-years manage to learn and understand the different meanings of flexible words, while at the same time keeping these meanings distinct in their heads. For example, how do children learn that "chicken" can label both an animal and a kind of meat, while also understanding that things that are true of the animal need not be true of the meat (and vice versa)?

Third, we will conduct experiments to explore how children learn to use flexible words in creative ways that they could not have heard before. Linguists have provided tantalizing examples of children's coinages, from "Broom my mess" to "Cracker my soup". Our experiments will explore how children learn to use words in these ways, providing insight into the origins of the type of linguistic creativity that infuses poetry and oratory.

Our final series of studies will examine what the consequences of flexibility are for children's cognitive development more generally. In particular, our theory raises the surprising possibility that the particular ways that flexible words are used in a language may encourage children to learn about the world in specific ways. For example the fact that we "shovel" snow using a "shovel" suggests that shovels are made for shoveling. Our last studies assess whether such linguistic clues affect how children explore and learn about objects.

1) Who will benefit from the research, and how?
Our research investigates the ways that children become flexible and creative users of language. It also examines the consequences of this flexible and creative language for their broader conceptual development. The resulting major outputs are twofold. First, an increased understanding of children's remarkable learning abilities. Second, a set of newly-annotated corpora of children's conversations with adults and other children. These corpora will be annotated for the semantic content of the words that are used, a project that is unique, as far as we aware. These two outputs will provide a number of important benefits to different groups.

A) Impact on Public and Parents
Linguistic and cognitive development are some of the most fascinating topics in psychology, especially for parents. We will work to inform the natural curiosity of families about how their children learn. For example, we will summarise our research findings on our respective websites, we will submit general readership articles to magazines focused on parenting (e.g., Junior) and general science and psychology (Psychology Today, Scientific American Mind), and we will make direct contact with local families through public lectures and engagement events (e.g., through public lectures at museums and pre-schools, and potentially at the Edinburgh Fringe and Science Festivals depending on interest).
In addition, we will create matched exhibitions to be shown in museums in both Edinburgh and San Francisco, including posters describing our research projects, and surveys that parents can take to tell us about their children's language development. As this latter data is collected, it will be entered into the exhibition, highlighting differences and similarities in how children learn about language across two sides of the world.

B) Impact on Software Development
There is a large and growing market for software aimed at children and toddlers. Meanwhile, software and hardware are both placing an increasing emphasis on natural language processing (the ability to understand and respond to a user's verbal instructions), as can be seen in features like Apple's Siri or Amazon.com's Echo. These markets are likely to interact, requiring us to develop software that is able to accurately process the speech of young children and then interpret what it means. This requires an understanding of how children use words in meaningful ways. As of now, there are no publicly available corpora of child language in which the meanings of words are tagged for their senses, but this is a vital component of any natural language processing system (for instance, in the mid-to-late 2000s, there were severe limitations to automatic translation systems, because they could not accurately identify the meaning of the word that was being used). Our proposed corpora will therefore provide a resource that is not only beneficial for researchers, but also for companies and non-profit organisations. It will allow them to train statistical natural language processing systems on data that accurately mimics children's environments.

2) How will we evaluate impact?
For lectures, website posts, magazines, etc, we will judge impact by the size of the audience/readership. For our exhibitions, we judge impact by the number of people interacting with the surveys and recording their data, and by the museum's estimate of interest in our posters. For our corpora, we will evaluate impact by requesting information about the number of times that they are downloaded from the CHILDES website.