How do babies learn what words mean?

By 12 months, babies understand about 50 words (Fenson et al., 1994) and by age 6 a typically developing child knows between 9,000 and 14,000 words!! That's learning about 9 words a day on average! Linguist Stephen Pinker called them 'lexical vacuum cleaners' (Pinker, 1994) and we know that children can 'fast map' a word in only one exposure.

That's why they sometimes get compared to parrots I think, since they can repeat a word straight after hearing it! It takes more like 10 exposures to learn a word more meaningfully and remember it, but still it's pretty incredible how quickly children can learn new words.

But the reference problem is a puzzling one. If a child sees a fish for the first time, and we label it 'fish', do they assume it means the whole object? Or could it be just the fin?

And if they know that 'dog' refers to their four-legged pet, how do they stop themselves from referring to all animals with four legs as 'dog' (that's the "extension" problem - where do words stop? Is that other creature - a cat - called 'dog' too?)

They are learning that their dog might have other names like Fido, puppy, doggy, woof-woof, etc. And they might learn that in French it is 'chien', which is what Papa calls it...oh la la, c'est compliqué!

These puzzles have been keeping linguists and psychologists busy for at least 50 years and there is no single, perfect answer. There is only evidence to support several different theories of word acquisition.

Sometimes you might hear people talk about the way babies learn something as a fact, but really the scientists are still looking for the solution and the only thing everyone agrees on is that there isn't a simple answer! 

1) Constraints theory (strong version): children have in-built boundaries for words. There are various constraints proposed, such as assuming that a new word refers to the whole object (hence rabbit = whole animal, not its ears), or that if rabbit refers to a fluffy animal with long ears it can't also refer to a scaly wet animal. That needs another label (fish). Other suggested constraints include using labels only for objects in the same category (so dog can refer to other dogs, but not to bones), or labels for objects are shared by function (e.g. all round objects that you play with are called ball). Sometimes the nouns seem more obvious in English but in Japanese, you can leave nouns out entirely and just use verbs. There is thus evidence for and against the strong constraints theory, as it needs to work across *all* languages, not just English where a lot of the research has focused and where nouns are kind of a big deal.

2) Constraints principles (less strong version): this theory proposes that children learn to refer to objects, actions and attributes with a label; that the label extends to a category of similar objects; and that object labels take precedence over actions or events. This softer version allows for lots of error margin (e.g. when 'ball' is used to refer to the sun or moon). But again, it is hard to find evidence that allows for universal constraints in languages other than English. 

For example, in Korean, there are different categories for the prepositions IN and ON in English. The Korean prepositions refer to something being tight or loose, or against, or on a head or a horizontal surface like a table. So a ring on a finger is KKITA, but a hat on a head is SSUTA. Thus we can't conclude that these constraints, even though not as strong as in (1), are universal across languages. Something must be happening in context as a baby starts hearing a specific language, rather than being in-built and applicable across any language. 

3) Enter the 'social-pragmatic' account: this is a fairly broad group of research that converges on the idea that children learn from social cues. It's such a huge topic we will just give you some headlines: babies might follow the intention, gaze or pointing finger of the person they are speaking to. We share attention by saying 'Oh look, a rabbit!' This theory suggests that babies might know what we are intending to say based on the fact that a new thing has appeared, or is surprising to the speaker rather than something of interest to the baby. It's a socially constructed (i.e. we are doing it together) account of how babies learn to speak. The problem with it is that a lot of the research has been carried out in America or Germany, and if we are to prove that gestures and sharing attention are essential catalysts for language learning, they need to work in other cultures too. So (as usual with science) more evidence from a wider range of cultures and families is needed before we can draw a strong conclusion that language emerges out of social behaviours.

So however they do it, and the jury is still out on the exact mechanisms – it could be a combination of in-built universals and learned-on-the-job, language-specific skills – we can all agree that babies are INCREDIBLE at learning words!

What does your instinct tell you? Personally, I have a feeling it's more of a socially constructed thing than an in-built rules thing, but I am open to the evidence and can't wait to see where this fascinating research conversation takes us next!