Landauer discusses in this article a computer model developed by his team based on a psychological theory called Latent Semantic Analysis (LSA). The primary emphasis in this model is on simulating several aspects of human cognition including understanding word and passage meaning and comprehending text. It simulates the learning process of school children, makes human-like assessment of semantic relation between words, and demonstrates its ability to answer multiple choice questions based on certain texts.
The LSA, based on mathematical and computational models, expresses ideas about word meaning based on the relations between word meanings in a semantic space. This approach is based on the psychological theory that humans associate perceptual objects and experience (including words) that are located near to each other in time. LSA creates a semantic space and establishes links between each word type in that space. By combining all the links in a common semantic space, it assures a unique location to each word and passage within it. To do this, the model uses an analogy of the method used in mapping the earth, while it assumes that the brain adopts a similar method in human cognition. It makes an initial estimate of closeness between words (by measuring how often they occur in the same meaningful context as compared to how often they occur in different contexts) and computes a correlational index. In the next step, it fits all the separate pieces of relations within a common space with as little distortion as possible.
The model has been tested in simulating vocabulary learning tasks of school children. While doing so, the model has demonstrated its ability in indirect learning also. It shows good performance in its knowledge in word meaning, and could demonstrate the influence of context in case of ambiguous words. Its performance in multiple choice questions based on psychology texts has been shown to be comparable to the performance of actual students in two major universities. Among other applications, LSA has been shown to represent passage meanings as well as word meanings as a single point in its semantic space. Similarly, the model has been used in essay answers about factual subjects to measure the quantity and quality of conceptual contents.