Natural Language Processing Assessment: Sample Questions | Bryq
Beginner
Intermediate
Advanced
These are three sample questions from Bryq’s Natural Language Processing (NLP) assessment, a technical test for hiring NLP engineers, machine learning engineers and data scientists who work with text. The full assessment covers three levels: beginner, intermediate and advanced, from text preprocessing and linguistic foundations to NLP algorithms, transformer models and language model applications. Each question is multiple choice, with one correct answer out of four options.
Beginner
Intermediate
Question
Why can removing all stop words harm a sentiment analysis model?
Select your answer
Some stop words, such as “not,” can reverse a sentence’s meaning.
It makes texts too long to process efficiently.
Stop words carry most of a text’s sentiment.
Models can’t process text once stop words are removed.
WHY IS THIS THE CORRECT ANSWER
Many default stop word lists include negations such as “not” and “no,” so removing them can turn “not good” into “good” and flip the sentiment. Whether to remove stop words should depend on the task, not be applied automatically.
Intermediate
Intermediate
Question
In TF-IDF, why does a word like “the” receive a low score even when it appears many times in a document?
Select your answer
Because it appears in almost every document, so its inverse document frequency is low.
Because it is shorter than most other words.
Because TF-IDF only scores nouns.
Because its term frequency is always set to zero.
WHY IS THIS THE CORRECT ANSWER
TF-IDF multiplies how often a word appears in a document by how rare that word is across the whole collection. A word found in almost every document has an inverse document frequency close to zero, so it scores low even when frequent, while more distinctive words score higher.
Advanced
Intermediate
Question
Why do transformer models add positional encodings to their input embeddings?
Select your answer
Because self-attention on its own doesn’t capture word order.
To reduce the size of the model’s vocabulary.
To prevent the model from overfitting.
To replace the attention mechanism in the early layers.
WHY IS THIS THE CORRECT ANSWER
Self-attention compares every token with every other token at the same time, so without extra information it treats a sentence as an unordered set of words. Positional encodings add each token’s position to its embedding, letting the model tell “dog bites man” apart from “man bites dog.”
