Inferring word meaning from surrounding context, examined through the distributional hypothesis that underlies both psycholinguistics and modern language models like Word2Vec.
INTRODUCTION
Language is such a fascinating concept to me that not even four years of devoted study in university could answer the questions I have, nor could a single research project in my final year allow me explore all the directions I find intriguing. And no, I'm not just saying this. I literally switched the topic and research area for my undergraduate research project just a few months away from my graduation. This frustrated my supervisor deeply, but he permitted it after seeing my research proposal.
The research project examined Receptive Bilingualism: the gap between what people can understand in a language and what they can actually produce in it. Following the project, I have continued to ponder on the mechanics of comprehension, and how the mind arrives at meaning with incomplete information.
Between May and July 2026, I volunteered with Mafaku's Summer Literacy Program as a chance to interact with acquisitors of a new language, to observe how they interact with language, and ultimately help these pupils in a Nigerian public primary school build a firmer relationship with the English language.
In doing so, an unexpected observation became apparent over the weeks. The pupils defined words based on the context the words appeared in, and did so confidently, even when their definitions were not always correct. I find this to be an important aspect of language use and acquisition, and have explored the research surrounding this.
THE PROGRAM
Mafaku ran a six-week literacy club across three schools, built around mini libraries stocked with books centred on African history and culture. Each school received one session a week. Every 80-minute session followed a gradual-release structure: a warm-up, a read-aloud modeled by a volunteer ("I do"), guided group discussion where the class worked through the text together ("we do"), and independent or paired reading ("you do"). Vocabulary work happened within the "we do" stage: when an unfamiliar word came up in the stories, the volunteer lead storyteller of the day explained it, contextualised it, and had the children try using it themselves.
Week 3's theme was "Overcoming Obstacles," built around stories of Nigerian historical figures, including Queen Amina, chosen specifically to get children reasoning through a character's conflict rather than just following the plot.
THE OBSERVATION
During the Queen Amina session, the story included a funeral and burial scene, and the word "mourn" came up as an unfamiliar word to unpack. I was leading this session.
When we asked the children what they thought the word meant, some of the more accurate answers described it as "when you're going to bury someone." As you may know, this definition is incorrect, however it caused me to pause and say "Hmmmnnn, well...not really." They were in the right neighbourhood, but knocking at the wrong door. This kind of word meaning from context happened often during the sessions we had.
The children were able to locate the right scene and the right domain (i.e death, funeral, loss), just not the specific psychological content the word actually names. Their definition was built out of what was happening around the word, rather than out of the word's own precise meaning. And this isn't a bad thing. Neither is it unique to these children.
THE RESEARCH, THE LITERATURE
Yarlett 2008, explores a concept called Similarity-Based Generalisation which describes the idea that children can figure out how to use words they've rarely encountered before by generalising about their use from distributions of similar words.
This example from Brenndoerfer (2025) builds understanding of this concept.
"Suppose you encounter the word "wug" in these sentences:
- "The wug ran across the field."
- "She fed the wug a carrot."
- "The wug's fur was soft and brown."
- "A wug can run up to 20 miles per hour."
You've never seen the word before, but you already know quite a lot about it. A wug is an animal. It is probably medium-sized, furry, and fast. It eats carrots, which hints at an herbivore. From pure context, with zero dictionary access, you've inferred a rich meaning."
This similarity-based generalisation is built upon a hypothesis posited a while ago; The Distributional Hypothesis.
Zellig Harris first introduced the distributional method as a technique for linguistic analysis, observing that morphological and syntactic regularities could be discovered by examining distributions of phonemes and words without reference to meaning at all (Brenndoerfer, 2025; Harris, 1954). Three years later, John Rupert Firth redirected this method towards semantics (the study of meaning), arguing that the distributional patterns in language constitutes meaning (Firth, 1957).
The distributional hypothesis posits that meaning emerges from patterns of use. Words that appear in the same kinds of sentences, surrounded by the same kinds of neighbours, tend to mean similar things (Brenndoerfer, 2025). This is the concept that was at play when the children at the primary school I volunteered at relied on context to decipher the meaning of a word.
Firth is famously quoted as saying "you shall know a word by the company it keeps," thus suggesting that by examining the context, the meaning of a word can be inferred. Words with similar meanings tend to appear in similar contexts, and those contexts are what a listener draws on to work out what a word means.
This however, is not unique to humans. It formed the basis for Natural Language Processing, giving rise to models such as Word2Vec, a model that generates static numerical representation of words. This model is implemented using two architectures: Continuous Bag of Words (CBOW), where the model predicts the target word based on its context, and Skip-gram, where the model predicts the context words using the target word (Taha, 2025).
Language models as we know today, are built on the distributional hypothesis; the same one the children at the primary school were employing.
THE END
Observing the relatedness between a random language learning event and the foundations for AI which I use and study daily seems like a first step toward a direction that is yet unclear. I've explored research around Language, Psychology and AI over the last few months so my findings are not a surprise, they are, however, unexpected. Pleasantly.
REFERENCES
- Brenndoerfer, M. (2025). The distributional hypothesis: Word meaning from context. In Language AI Handbook. https://mbrenndoerfer.com/writing/distributional-hypothesis-word-meaning-context
- Firth, J. R. (1957). A synopsis of linguistic theory, 1930–1955. In Studies in Linguistic Analysis (pp. 1–32). Philological Society. (Reprinted in F. R. Palmer, Ed., 1968, Selected papers of J. R. Firth, 1952–1959, Longman)
- Harris, Z. (1954). Distributional structure. Word, 10(23), 146–162. https://doi.org/10.1080/00437956.1954.11659520
- Taha, F. (2025). Understanding Word2Vec: CBOW & Skip-Gram [Kaggle notebook]. Kaggle. https://www.kaggle.com/code/ftaham/understanding-word2vec-cbow-skip-gram
- Yarlett, D. (2008). Language learning through similarity-based generalization [Doctoral dissertation, Stanford University]. Retrieved July 23, 2026, from https://web.archive.org/web/20140419012951/http://psych.stanford.edu/~michael/papers/Draft_Yarlett_Similarity.pdf