Errata for "Web-Scale Features for Full-Scale Parsing"

This addendum/errata provides some extra clarification regarding the affinity and parahrase features used:

1) Section 2.1: The final version only contains the highest specificity/complexity affinity features (and not all of the increasingly complex features).

2) Section 2.2: For the paraphrase features, we collect the most frequent context words of a given head and argument word pair (and not "of a given POS pair" as mentioned in the paper); and then the actual features contain the POS tags of the head and argument and the top context words found for that head and argument.