*
The rest of the lines include the probabilities p(word=w|topic=i) as follows:
topic_prob_given_slot.txt
-------------------------
For every predicate template there are the following three lines:
A template is denoted as:
X Y
topic_mixture_per_slot_x (and respectively for slot y) includes a tab-delimited list of the following entries:
::
where t=* refers to the smoothed probability p(topic=*|slot_x) that applies to all zero-count topics for this slot.
wt_rules_XXX.txt
----------------
For every predicate inference rule, template_lhs -> template_rhs, there are the following three lines:
(tab-delimited)
rule_type = [LIN|BINC|COS], indicating the utilized similarity measure Lin, BInc and Cosine respectively.
DIRT scores denote the context-insensitive scores for the rule.
WT_scores_per_slot_x (and respectively for slot y) includes a tab-delimited list of per-topic rule score entries:
::
The score for topics whose entry is not present is zero.
For every template_lhs, the file includes the top template_lhs->template_rhs rules (up to 500 rules), sorted high-to-low according to their DIRT score.