NeLePe

A Contect Free Language can be parsed with

• Finite State Automata with a stack

• Finite State Transition Networks

• Recursive Transition Networks

• Augmented Transition Networks

• Finite State Automata

What is an illocutionary act

• None of the other options

• The act that how hearer is affected

• A speech art

• An act performed by means of uttering the words

• The act of uttering something with a certain "meaning"

Syntactical analysis is difficult, in general, due to the need to solve:

• Topicalization

• Sense amiguities

• Long distance dependences

• All other options

• Agreements

The difference between constituency and dependency parsing is that the former

• is intractable

• none of the other options

• is not detecting noun phrases

• is generating a parsing tree with more nodes

• is using tree adjoining grammars

A difference between RNN and Transformers for NLP is

• None of the above options

• RNNs don't allow parallelization during training

• Transformers better identify long-term dependencies

• Transformers completely eliminate recurrence and replace it with self-attention

Part of speech tagging may be done with

• Hidden Markov Models

• Recurrent Neural Networks

• Co-reference resolution

• Latent Dirichlet Allocation

• Porter Stemmer

Which are the differences between tf*idf and word2vec

• The former captures synonymy better

• The former counts, rather than predicts

• The former has a shorter, denser representation

• The former predicts, rather than counts

• The former has a longer, sparser representation

Which of the following sentences is generating a Winograd schema?

• I wanted to insert the ball into the bag but it was too small.

• I wanted to insert the idea into the bag but the phone ringed.

• I wanted to insert the ball into the bag but the phone ringed.

• I wanted to insert the ball into the bag the bag was too small

A difference between LSTMs and GRUs for NLP is:

• LSTMs cannot be used to construct a language model

• None of the other options

• LTSMs do not try to solve the vanishing gradient problem

• GRUs have less trainable to learn, thus are faster to compute

What other processings are useful for co-reference resolution

• Clustering

• Part-of-speech tagger

• Parser

• None of the other options

• Named Entity Recognition