mathiascreutz commited on
Commit
4228fac
1 Parent(s): bc72843

Minor modifications

Browse files
Files changed (1) hide show
  1. README.md +6 -1
README.md CHANGED
@@ -129,7 +129,12 @@ The entries in the training sets have been ranked automatically by how
129
  likely they are paraphrases, best first, worst last. The quality
130
  parameter indicates the estimated proportion (in percent) of true
131
  paraphrases in the set. Allowed quality values range between 60 and
132
- 100, in increments of 5 (60, 65, 70, ... 100).
 
 
 
 
 
133
 
134
 
135
  ### Data Instances
 
129
  likely they are paraphrases, best first, worst last. The quality
130
  parameter indicates the estimated proportion (in percent) of true
131
  paraphrases in the set. Allowed quality values range between 60 and
132
+ 100, in increments of 5 (60, 65, 70, ..., 100). A value of 60 means
133
+ that 60% of the sentence pairs in the downloaded set are estimated to
134
+ be true paraphrases (and the remaining 40% are not). A higher value
135
+ produces a smaller but cleaner set. The smaller sets are subsets of
136
+ the larger sets, such that the `quality=95`set is a subset of
137
+ `quality=90`, which is a subset of `quality=85`, and so on.
138
 
139
 
140
  ### Data Instances