Do we understand a word by the company it keeps?


Imagine you only speak your own native language. You find yourself “locked in a room”, let’s say, with a Chinese book, a Chinese dictionary, and a set of rules in your own native language. By applying the rules, you learn to relate the Chinese words in the dictionary to the book, only by recognising the shapes of the characters they contain. You learn to manipulate the strings of Chinese symbols so well that you start to resemble a native Chinese speaker who actually understands Chinese words and sentences. Yet, what you are doing is just manipulating formal symbols without knowing what they mean. Are you behaving any differently than the artificial mind of a computer?

This famous Chinese room thought experiment by the philosopher of language John Searle first fascinated me when I was a university student, because it raises the intriguing question of how the human mind relates knowledge about symbols like words to the knowledge about the meanings of concepts (Want to find out more about the difference between a word and a concept? Read Guillermo’s blog The infinite space inside your head: The birth of conceptual spaces).

The distributional hypothesis of word knowledge

One very popular idea is that the meaning of a word, like sky, freedom, or apple, is derived from the company it keeps. For instance, the meaning of the word apple would result from the number of times it is used with the words tree, ground, fields, jam, tart, knife, dinner, restaurant, market, to eat, to cook, agriculture, and so on, across many different contexts, such as books, conversations, magazines, newspapers, online media, WhatsApp messages, etc. This idea is called the “distributional hypothesis”, as it proposes that the meaning of a word stems from how it is distributed in actual language in daily life.
But you might wonder: does the human mind really understand the meanings of words based on their distribution in usage? Or is this kind of knowledge no different to the ‘knowledge’ of Chinese words you learn in the thought experiment, which is basically what artificial minds do? (Want to find out more about artificially intelligent chatbots? Read Naomi’s blog Creating androids that dream of electric sheep.)

How our experiences shape our knowledge of what words mean

In real life, an important part of our knowledge of what a word like apple means is indeed based on our knowledge of which words apple often keeps company with. This is because a lot of the information we collect about world facts, happenings, activities, and objects comes to us through language exposure, such as by reading and listening to others. Although we’re often not aware of this, we can use this information when we come across a word we don’t know the meaning of, to try to guess its meaning. Probably you’ve never heard before the Chinese word 地球, dìqiú, or the Sanskrit word पृथ्वी, pṛthvī, but now imagine you find it in contexts surrounded by words you know mean plants, climate, prairies, climate change, skies, pollution, to cultivate, soil, vegetables, fruits, and also mother, and even day, planet, green, and Greta. What is your guess? Could you grasp its meaning?

Because you know the meanings of the other words that dìqiú and pṛthvī, keep company with, you can use this information to make your guess. Then you may intuitively think of “earth”!

But your guess in this case was only made possible by the fact that you were already aware of the existence of a concept like “earth”, and you were also familiar with the meanings of the other words that dìqiú and pṛthvī are associated with. At this point, you might think, “So how did I learn the meanings of these words in the first place?”. Of course, the meanings of words like apple, earth, to smile, and happiness become familiar to us during language learning because of the experiences we have interacting with the objects, facts, and feelings they refer to. Ultimately, knowing what apple means, and that an apple is red, round and sweet, is knowledge that is gained over time by integrating all our experiences with seeing, touching, biting into, and eating apples. These experiences taken together allow us to form a coherent meaning of the word apple. Therefore, for some scholars the meanings of words cannot be fully detached from the perceptions through the senses and the emotions by which we experience the world: if the words we use were disconnected from these experiences, we would be no different than the person in the thought experiment who ‘learns’ to speak Chinese just by learning how to combine Chinese words into sentences.

Putting it all together: How do humans understand words?

Brain imaging data suggest that humans use both the knowledge they have of how words are distributed in real-life language and their perceptual and emotional experiences of the world to understand the meanings of words, as they are both reflected in brain activity. When the brain processes the meaning of a word like blue, it can retrieve both information about the related meanings of the other words that it keeps company with, such as colour, sky, winter blues, sea, and ink, and information about our experiences with the things it refers to in the real world. This is how the human mind is different to the artificial mind of a computer. The latter has no way to establish a link between the information that it could acquire by reading, talking and listening to others, and any experience of the world that allows us humans to truly understand what it means to see, touch or eat an apple. Therefore, making sense of words is itself a uniquely human experience!

Further Reading

  • Andrews, M., Frank, S., Vigliocco, G. (2014). Reconciling embodied and distributional accounts of meaning in language. Topics in Cognitive Science, 6(3), 359-370. doi: 10.1111/tops.12096. Epub 2014 Jun 17.
  • Bi, Y. (2021). Dual coding of knowledge in the human brain. Trends in Cognitive Sciences, 25(10), 883–895.
  • Binder, J. R., Conant, L. L., Humphries, C. J., Fernandino, L., Simons, S. B., Aguilar, M., & Desai, R. H. (2016). Toward a brain-based componential semantic representation. Cognitive Neuropsychology, 33, 130-174.
  • Carota, F., Nili, H., Pulvermüller, F., Kriegeskorte, N. (2020) Distinct fronto-temporal substrates of distributional and taxonomic similarity among words: evidence from RSA of BOLD signals. Neuroimage. 2021 Jan 1;224:117408. doi: 10.1016/j.neuroimage.2020.117408. Epub 2020 Oct 10. PMID: 33049407.
  • Searle, John. R. (1980) Minds, brains, and programs. Behavioral and Brain Sciences 3 (3): 417-457

Writer: Francesca Carota
Editors: Dilay Karadoller
Dutch translation: Lynn Eekhof
German translation: Natascha Roos
Final editing: Eva Poort