| LSA News |
Benoît Lemaire & Philippe Dessus (http://www.upmf-grenoble.fr/sciedu/blemaire/lsa.html) have a new LSA page with lots of references: Readings in Latent Semantic Analysis for Cognitive Science and Education
From now on, the one2many web page will produce warnings for each word that is not in the corpus used.
I have changed the names of the spaces built from samples of general readings. They were called tasaXX where the XX was 03, 06, 09, 12, or ALL. The tasa stands for Touchstone Applied Science Associates, the source of the text samples. The numbers stand for the (American) grade level to which the samples were intended to cover -- i.e. tasa09 includes cumulative general readings for students up to and including 9th grade. I changed the names of these spaces to "General_Reading_up_to_XX_Grade" and reduced the number of available dimensions to 300. It seemed like a lot of new users weren't reading the semantic space information page, so they didn't know what the tasa spaces were. Instead they would choose the encyclopedia space, which isn't very good for modeling student behavior.
I reduced the dimensionality to 300 for all the general spaces as that is our best estimate right now for where the optimal dimensionality should be set. You can now leave the "Number of Factors to Use" box blank to get optimal, rather than maximum saved dimensionality. You should think about each of these general spaces as representing what 1 average child would have read by the time she has reached that grade level.
In addition, the Psychology spaces based on Myers 4th and 5th edition textbooks have been renamed "Psychology_Myers_Xth_ed" with dimensionsreduced to 400.
We have received a number of requests from people who want to do more substantial experiments using LSA than those that can easily be done through the web interface. We had hoped to provide such functionality, so when I created the new guestbook interface and mailing lists, I added a checkbox for people to request a login account. This was a bit premature on my part. After re-assessing our resources, we decided that neither our machine nor our personnel could support a large outside user base at this time. We can't provide the resources needed for unix/perl scripting support, which is how you do things with LSA outside of the web interface. We are very sorry we can't provide this functionality and we will let you know if this changes in the future. You will still continue to be able to use the applications on the website without a login account.
We have had several requests for an application which will give you sentence to sentence coherence measures for texts. Peter Foltz and I have already developed perl scripts that do this. Sometime over Holiday break we will adapt this code to the web environment. Expect this application to be available in January.
Darrell Laham, LSA Webmaster
Here are links to a few of the articles that are available online: