Discussion Questions: 22 April

1. (DE) On pg 22, Landauer and Dumais write: "The question, then, is whether (LSA), when combined with the statistics of experience, will produce a faithful reflection of human knowledge." What exactly do they mean by "human knowledge"?
2. (DE) Landauer and Dumais claim (on pg. 13) that simply finding robust constraints on language acquisition is not enough. One must also somehow show that every constriant "sufficiently helps to solve the overall inductive problem." In other words, "Note to science: don't bother to show me a robust constraint on language unless you can in parallel solve the age-old probelm of induction." But later they claim that constraints work together in unexpected, nonlinear ways ("...the effects of constraints may emerge...[pg. 12]") which leads me to believe that the existence of a robust constraint should be very interesting to them. Sophistry?
3. (CP) When they talk about how the average person must be learning 7-15 words per day, even during junior high. What about the naming explosion? Would this change their numbers?
4. (MG) How might the LSA framework accommodate real-world reference as well as usage?
5. (MG) How do the word and the context vectors in the basic LSA model differ? Is it necessary that things work this way? What does this have to do with the unitization question?
6. (MG) How can words be learned from texts in which those words do not occur?