Why your brain may work like a dictionary:
DOES your brain work like a dictionary? A mathematical analysis of the connections between definitions of English words has uncovered hidden structures that may resemble the way words and their meanings are represented in our heads.
“We want to know how the mental lexicon is represented in the brain,” says Stevan Harnad of the University of Quebec in Montreal, Canada.
As every word in a dictionary is defined in terms of others, the knowledge needed to understand the entire lexicon is there, as long as you first know the meanings of an initial set of starter, or “grounding”, words. Harnad’s team reasoned that finding this minimal set of words and pinning down its structure might shed light on how human brains put language together.
The team converted each of four different English dictionaries into a mathematical structure of linked nodes known as a graph. Each node in this graph represents a word, which is linked to the other words used to define it – so “banana” might be connected to “long”, “bendy”, “yellow” and “fruit”. These words then link to others that define them.
This enabled the team to remove all the words that don’t define any others, leaving what they call a kernel. The kernel formed roughly 10 per cent of the full dictionary – though the exact percentages depended on the particular dictionary. In other words, 90 per cent of the dictionary can be defined using just the other 10 per cent.
But even this tiny set is not the smallest number of words you need to produce the whole dictionary, as many of these words can in turn be fully defined by others in the kernel. This is known as the minimal grounding set (MGS), which Harnad explores in his most recent work. Unlike the kernel, which forms a unique set of words for each dictionary, there are many possible word combinations that can be used to create an MGS – though it is always about half the size of the kernel.
What’s more, the kernel has a deeper structure. The team found that half of its words made up a core group in which every word connects to every other via a chain of definitions. The other half was divided into satellite groups that didn’t link to each other, but did connect with the core (see diagram).
And this structure seems to relate to meaning: words in the satellites tend to be more abstract than those in the core, and an MGS is always made up of words from both the core and satellites, suggesting both abstract and concrete words are needed to capture the full range of meaning.
So what, if anything, can this tell us about how our brains represent words and concepts? To find out, Harnad’s team looked at data on how children acquire words and found a pattern: as you move in from the full dictionary towards the kernel and finally the MGS, words tend to have been acquired at a younger age, be used more frequently, and refer to more concrete concepts (arxiv.org/abs/1308.2428). “The effect gets stronger as you go deeper into the kernel,” Harnad says.
That doesn’t mean children learn language in this way, at least not exactly. “I don’t really believe you just have to ground a certain number of things and from then on close the book on the world and do the rest by words alone,” says Harnad. But the correlation does suggest that our brains may structure language somewhat similarly to a dictionary. To learn more, the team has created an online game that asks players to define an initial word, then define the words in those definitions. The team then compares whether their mental dictionaries are similar in structure to actual ones.
Phil Blunsom at the University of Oxford isn’t convinced word meanings can be reduced to a chain of definitions. “It’s treating words in quite a symbolic fashion that is going to lose a lot of the meaning.” But Mark Pagel of the University of Reading, UK, expects the approach to lead to new insights – at least for adult brains. “This will be most useful in giving us a sense of how our minds structure meaning,” he says. For example, one question raised by the relatively small size of the MGS is why we burden ourselves with so much extraneous vocabulary.