The following is an excerpt from the Estimating LAT Confidence section of this paper:
... Since some LAT detection rules are more reliable than others, we would like to have a confidence value in each LAT that can be used to weight each LAT appropriately during answer scoring. To accomplish this, we trained a logistic regression classifier using a manually annotated gold standard. The classifier uses the focus and LAT rules that have fired as features, along with other features from the parse, NER, and the prior probability of a particular word being a LAT...
It's not clear to me what is the target variable of the logistic regression classifier. It sounds to me that it is binary variable indicating whether the fired rule was correct (1) or incorrect (0). Therefore, the prediction of the logistic regression classifier is the probability that the word is an LAT given that a certain rule has been hired (and some other features). Does this application of the logistic regression classifier make sense?
The details of the paper are as follows:
A. Lally et al., "Question analysis: How Watson reads a clue," in IBM Journal of Research and Development, vol. 56, no. 3.4, pp. 2:1-2:14, May-June 2012. doi: 10.1147/JRD.2012.2184637