Question 2– Dependency Parsing (17 pts)
A transition-based dependency parser analyzes the sentence ‘I booked a flight from L.A.’. Here
is the gold standard tree from this sentence.
a) Suppose that the parser starts in the initial configuration for the sentence and takes the
transitions Shift, Shift, Left-Arc. State the new configuration. To represent the partial
dependency tree, list the arcs contained in it.
b) State a complete sequence of transitions that takes the parser all the way from the
initial configuration to a terminal configuration, and that recreates all arcs of the gold
c) Provide a modified transition sequence where the parser mistakenly predicts the arc
‘booked -> from’, but gets the other dependencies right except for the change of the
parent of ‘from’.
Question 5 – Neural Models (22 pts)
Consider the following architecture to classify sentiment at the sentence level. Each sentence
1- How does hi (the hidden state of the Bi-LSTM) compare to xi (the embedding)? (what is
the advantage of this over using xi directly)
2- If we didn’t want to use attention scores what would be an alternative approach to get
the sentence representation. Describe the role of the attention scores.
3- Provide a generic equation to compute the attention scores and indicate the part that
represents the trainable parameters in the equation
4- Given that this is a classification problem, please:
a. indicate the name of the loss function and its equation
b. Indicate the name and formula for the activation function at the last layer
considering that the vector at its inputs is called ‘z’
本网站支持淘宝 支付宝 微信支付 paypal等等交易。如果不放心可以用淘宝交易！
E-mail: firstname.lastname@example.org 微信:itcsdx