Table of Contents
Universal Tagset
1. Why do you think that the best results are obtained by training on original tagsets and testing on the universal one (why are the numbers in the O/U column higher than in O/O and U/U)? (Table 1)
2. What is the advantage of USR-I over USR-G? (Why would someone use the former, when the latter gives better results?)
Universal Treebank
3. How does the annotation scheme of the Universal Treebank differ from the original Stanford Dependencies?
4. Which of the labels (from Table 1 on page 3) do you not understand? (Name 3-5 labels you understand the least or are the most interested in.)
Both
5. (Credit MP)
Let us have the following two sentences:
“John gave Mary nice books.”
“Benefits include gas central heating.”
a) Assign the correct universal POS tags and dependency trees for both the sentences.
b) How do you think the sentences would be parsed by a delexicalized cross-lingual parser with French as the source language?