Pasteur’s Quadrant: Computational Linguistics, LSA, and Education

This paper argues that computational cognitive psychology and computational linguistics have much to offer the science of language by adopting the research strategy that Donald Stokes called Pasteur’s quadrant--starting and testing success with important real world problems--and that education offers an ideal venue. Some putative examples from applications of Latent Semantic Analysis (LSA) are presented, as well as some detail on how LSA works, what it is and is not, and what it does and doesn’t do. For example, LSA is used successfully in automatic essay grading with content coverage feedback, computing optimal sequences of study materials, and partially automating metadata tagging, but is insufficient for scoring mathematical and short textual answers, for revealing reasons. It is explained that LSA is not construable as measuring co-occurrence, but rather measure the similarity of words in their effect on passage meaning,