Department of Computing

COMP348 Document Processing and the Semantic Web

Tutorial Week 6

There are fewer exercises this week, so you can finish off uncompleted ones from last week.

Text Classification and MT

In lectures we briefly looked at the Bleu metric for evaluating MT output. Some more recent approaches use classifiers to help decide on MT output quality. As training data, some of these use human translations as positive data, and machine translations (on the assumption that these are all worse than human translations) as negative data.

Consider the English translations of the Korean data from lectures. What kind of features might you use to build such a classifier?

HUMAN: A total of eight interviewees include Joseph S. Nye, 
a professor of Harvard University’s Kennedy School of Government
who set up national development strategies for the United States 
via his theory of “smart power,” and experts on the 
Korean Peninsula coming from Washington’s conservative and 
progressive think tanks. 

MACHINE 1: To interview su E thu power objection it led and 
Joseph age Harvard large Kennedy who establishes the 
national growth strategy of the United States su khwul the professor 
back world-wide great scholar and progress and conservativeness 
of Washington the younger lu representative Sing khu 8 person 
strong earthquake class Korean Peninsula specialist backs of the 
tanks responded.

MACHINE 2: Smart power of the United States through the 
interview theories established national development strategies, 
the Kennedy School of Harvard University Professor Joseph age 
and scholars worldwide, including liberals and conservatives, 
the Washington think tank's senior representative across the 
Korean peninsula, including eight professional matter.

Grammars

1. You are given the following constituent-based NP grammar:

   NP -> Det N1 PP
   NP -> N1 PP
   NP -> Det N1 
   NP -> N1
   NP -> ProperName
   ProperName -> ProperNoun ProperNoun
   ProperName -> ProperNoun
   N1 -> AdjP N1
   N1 -> N N
   N1 -> N
   N1 -> N1 N
   AdjP -> Adv Adj
   AdjP -> Adj
   PP -> Prep NP
   

   For each of the following NPs, determine the parts of speech of the individual words, then work out whether the NP is covered by the above grammar. If so, determine the grammatical structure of the NP. If not, determine what rules would extend the coverage sufficiently.

   • the lawyer for General Electric
   • incredibly greedy capitalist lackeys
   • cat
   • the cow that jumped over the moon
   • the noses on the faces of the ladies of the harem of the court of King Caractacus

2. (optional) Build up a VP grammar to handle the following sentences.

   • The cat slept quietly.
   • Fairy penguins migrate to Philip Island.
   • I will demonstrate the device.
   • I know that the butler killed Miss Havisham.