• Use HAL
  • Tutorial
  • Web HAL
• About HAL
  • HAL the Theory
  • HAL the Model
  • HAL the History
• HAL Research
  • HAL Researchers
  • HAL Papers

Welcome to the HAL Website

The Hyperspace Analogue to Language model is a computational instantiation of a psychological theory of word meaning. The simplist description of the theory is that a word's meaning is a function of the contexts in which that word has occurred in a person's learning environment. Put another way, two words that have occured in similar contexts will have similar meanings.

The HAL model instantiates this theory by recording word co-occurrences in a large corpus of text. For a given word, a vector is created. Each vector element is the probability of that word co-occurring with each other word in the corpus. These vectors can then be compared to each other, giving an overall measure of the similarity of two words' co-occurrence patterns. For example, the words dog, cat, and child will all tend to have high probabilities of co-occurring with the words small, home, and chase, and low probabilities of co-occurring with purple, aiplane, and drive. In the HAL model, the vectors for dog, cat, and child are relavtively similar.

At the HAL website, you can use the HAL model. You can also learn more about the HAL model and theory, download published research on the HAL model, and learn about researchers who are currently using HAL. If you have questions about HAL itself, you are encouraged to contact one of the researchers on the HAL Researchers page. If you have technical questions about the website itself, please email Jon Willits. Enjoy!