|
What is common practice for dealing with the ambiguous parts-of-speech tags (things like "NNS|NN", "VBD|VBN", etc) tags found in corpora such as the Penn Treebank. I'm specifically interested in how people typically deal with this when reporting accuracy, precision, recall, and other performance metrics in publications, but would also be interested in how in hearing about how it might come into play in other part of the process, like during learning. |
