Present: Tanya Clements, Vered Gored, Bill Parod, Steve Ramsay, Sara Steger

We talked about implementing Dunning's log likelihood ratio in Monk, and the particular discussion led to a confirmation of two general points:

1. All analytics should flow through D2K or its successor.

The particular consequence of that decision is that Vered will be in touch with Phil Burns about getting the Java code for Dunning's from Phil Burns for implementing it in a D2K environment.

#2 We want to continue the practice, begun with Sara's Naive Bayes use case for sentimentality, of constructing real work flows for real use cases from the XML files through the data store and analytics to the user interface. This may involve pinchitting of various kinds. The interface may be a command line, the datastor may be the WordHoard model, pending completion of the Monk datastor. It may even be the case that some temporary data flow is rigged up directly from the XML files.

The chief objective is to execute some real operations, with whatever temporary bridges are necessary, to get substantive results that can be evaluated in terms of what they do for the scholarly end user and what they tell us about likely performance problems.

The particular consequence of that decision is to implement the chain of operations that lets Sara run Dunning's tests on the corpus of nineteenth-century fiction