Sequence Models - NLPs

Question 1

What are word embeddings

Answer 1

Features of a word, say ‘gender’,’royalty’ etc etc

Question 2

Learning word embeddings need a huge text corpus. True/False?

Answer 2

True…sometimes, billions of words.

Question 3

How do we use are analogies in NLP? Man:Woman=King:??

Answer 3

Minimise with each word in the vocab (e_king - e_man + e_woman)

Question 4

What is cosine similarity?

Answer 4

Explain

Question 5

What is negative sampling?

Answer 5

It is trying to find the next word “juice” in a “….orange juice. “Orange” in the context and “Juice” is the positive match. We will also have 5 negative matches.

Question 6

Rule based systems have _____ precision and _____ recall (high/log)

Answer 6

high precision and low recall.

Question 7

What is Semantic Slot Filling in NLP?

Answer 7

We have slots (like “Find flights between A to B”), which we try to fill using rule based approaches.

Question 8

What are the steps in Semantic Slot Filling?

Answer 8

1. Feature Engineering 2. Probabilistic Graphical Model

Question 9

What are the 4 levels of NLP Pyramid

Answer 9

Morphology (words) -> Syntax -> Semantics (meanings)-> Pragmatics

Question 10

What is co-reference in a text?

Answer 10

Which reference the same entity in a text

Question 11

What are constituency Trees?

Answer 11

The various parts of speech organized as a tree in a sentence.

Question 12

1. Why should we ignore high frequency n-grams
2. Why should we ignore low frequency n-grams

Answer 12

High frequency n-grams lead to overfitting

Low frequency n-grams are like stop-words, which do not add any value