IABAC Certified Natural Language Processing Expert (CNLPE) — Questions and Answers

Question 1: Which parsing strategy processes input left-to-right and builds the parse bottom-up using a stack and input buffer?

CYK parsing
Earley parsing
Shift-reduce parsing (Correct answer)
Recursive descent parsing

Correct answer: Shift-reduce parsing

Shift-reduce parsing maintains a stack and an input buffer, repeatedly shifting tokens onto the stack or reducing a stack sequence using grammar rules.

Question 2: What is the purpose of the OntoNotes corpus in NLP research?

Providing large-scale annotations for multiple tasks including NER, coreference, semantic roles, and parse trees across multiple genres (Correct answer)
Hosting pre-trained word embeddings for download
Storing a list of English ontological concepts and their definitions
Benchmarking machine translation between major world languages

Correct answer: Providing large-scale annotations for multiple tasks including NER, coreference, semantic roles, and parse trees across multiple genres

OntoNotes is a multi-layer annotated corpus used to train and evaluate models on NER, coreference resolution, SRL, and parsing simultaneously.

Question 3: Which pre-training objective does BERT use to learn bidirectional representations?

Causal language modeling left-to-right
Masked language modeling and next sentence prediction (Correct answer)
Next sentence prediction only
Contrastive sentence embedding

Correct answer: Masked language modeling and next sentence prediction

BERT pre-trains using masked language modeling (predicting randomly masked tokens) and next sentence prediction (determining if two sentences are consecutive).

Question 4: In the context of word embeddings, what is the primary advantage of FastText over Word2Vec?

Uses bidirectional context windows
Faster training speed on GPUs
Produces higher-dimensional vectors
Handles out-of-vocabulary words using subword information (Correct answer)

Correct answer: Handles out-of-vocabulary words using subword information

FastText decomposes words into character n-grams, allowing it to generate embeddings for words not seen during training.

Question 5: What is the Naive Bayes classifier commonly used for in NLP?

Text classification tasks such as spam detection and sentiment analysis based on word frequency features (Correct answer)
Aligning source and target words in machine translation
Generating text sequences token by token
Parsing the syntactic structure of a sentence

Correct answer: Text classification tasks such as spam detection and sentiment analysis based on word frequency features

Naive Bayes applies Bayes' theorem with a conditional independence assumption between features, making it a fast and effective baseline for text classification.

Question 6: What is back-translation in machine translation?

A technique to reverse the order of tokens in training data
Augmenting training data by replacing words with their antonyms
Translating a document and then translating it back to verify accuracy
Translating monolingual target-language text into the source language to create synthetic parallel training data (Correct answer)

Correct answer: Translating monolingual target-language text into the source language to create synthetic parallel training data

Back-translation uses an existing translation model to generate source-side sentences from target-language monolingual data, creating cheap additional training pairs.

Question 7: What is the copy mechanism in sequence-to-sequence models?

A technique for initializing decoder weights from the encoder
A caching layer that stores previously computed hidden states
A mechanism that allows the decoder to directly copy tokens from the source input sequence rather than always generating from the vocabulary (Correct answer)
A method for duplicating training examples to balance classes

Correct answer: A mechanism that allows the decoder to directly copy tokens from the source input sequence rather than always generating from the vocabulary

The copy mechanism (pointer networks) lets the model point to and copy source tokens, which is critical for tasks like summarization where proper nouns or rare words should be reproduced exactly.

Question 8: In NLP, what is a 'corpus'?

A type of neural network layer
The output vocabulary of a language model
A single annotated sentence
A large structured collection of texts used for training or analysis (Correct answer)

Correct answer: A large structured collection of texts used for training or analysis

A corpus (plural: corpora) is a large, structured collection of texts used for linguistic analysis, training, or evaluating NLP models.

Question 9: What is the BIO tagging scheme's limitation compared to BIOES?

BIO requires a separate model for each entity type
BIO uses more tags than BIOES
BIO cannot distinguish single-token entities from multi-token entity starts, while BIOES adds S (Single) and E (End) tags for clarity (Correct answer)
BIO cannot label the outside of entity spans

Correct answer: BIO cannot distinguish single-token entities from multi-token entity starts, while BIOES adds S (Single) and E (End) tags for clarity

BIOES extends BIO with E (End of entity) and S (Single-token entity) tags, providing richer boundary signals that can improve NER model performance.

Question 10: The CYK (Cocke-Younger-Kasami) algorithm requires the grammar to be in which normal form?

Chomsky Normal Form (Correct answer)
Backus-Naur Form
Extended Backus-Naur Form
Greibach Normal Form

Correct answer: Chomsky Normal Form

CYK requires Chomsky Normal Form (CNF), where every rule is either A → BC or A → a, enabling dynamic programming over spans.

Question 11: What is template-based information extraction?

Parsing XML or JSON data formats
Using neural networks to learn extraction patterns from data
Generating summaries from structured databases
Using predefined patterns or templates to identify and fill structured information slots in text (Correct answer)

Correct answer: Using predefined patterns or templates to identify and fill structured information slots in text

Template-based IE uses hand-crafted or learned patterns (e.g., regular expressions, trigger words) to populate structured records from unstructured text.

Question 12: What is perplexity as a metric for language models?

The average sentence length in the training corpus
The exponentiated average negative log-likelihood of a test corpus; lower perplexity means better prediction (Correct answer)
The percentage of tokens the model classifies incorrectly
The number of parameters in the model

Correct answer: The exponentiated average negative log-likelihood of a test corpus; lower perplexity means better prediction

Perplexity measures how surprised a language model is by unseen text — a perplexity of N means the model is as confused as if it had to choose uniformly among N options at each step.

Question 13: What is the difference between closed-domain and open-domain question answering?

Closed-domain QA is restricted to a specific topic or dataset, while open-domain QA answers questions about any topic using a large corpus or the web (Correct answer)
Closed-domain QA generates answers while open-domain extracts spans
Open-domain QA requires a knowledge graph while closed-domain does not
Closed-domain QA uses neural networks while open-domain QA uses rules

Correct answer: Closed-domain QA is restricted to a specific topic or dataset, while open-domain QA answers questions about any topic using a large corpus or the web

Closed-domain systems (e.g., medical QA) operate within a bounded knowledge base, while open-domain systems retrieve from broad corpora like Wikipedia to answer any question.

Question 14: What does the BLEU score measure in NLP?

Named entity recognition accuracy
Machine translation quality (Correct answer)
Sentiment classification precision
Parsing tree depth

Correct answer: Machine translation quality

BLEU (Bilingual Evaluation Understudy) measures machine translation quality by comparing n-gram overlaps with reference translations.

Question 15: An application of Google Translate is ___.

Information extraction
Information retrieval (Correct answer)
Sentiment analysis

Correct answer: Information retrieval

While Google Translate's primary function is machine translation, it serves as a powerful enabler for information retrieval across language barriers. By translating web pages, documents, or search queries, it allows users to access and comprehend information originally published in foreign languages. This capability effectively expands the scope of retrievable information, making previously inaccessible content understandable and searchable.

Question 16: What is the time complexity of the CYK parsing algorithm for a sentence of length n with a grammar of size |G|?

O(2ⁿ)
O(n)
O(n³ · |G|) (Correct answer)
O(n²)

Correct answer: O(n³ · |G|)

CYK runs in O(n³ · |G|) time because it fills an n×n triangular table by considering all possible split points for every span.

Question 17: What is open information extraction (OpenIE)?

Extracting entities only from Wikipedia
Parsing text using only rule-based patterns
Extracting relational triples (subject, relation, object) from text without a predefined ontology (Correct answer)
Performing NER restricted to a closed set of entity types

Correct answer: Extracting relational triples (subject, relation, object) from text without a predefined ontology

OpenIE systems like AllenNLP's extract arbitrary (subject, relation, object) triples from sentences, enabling scalable knowledge acquisition without labeled training data.

Question 18: What is 'attachment ambiguity' in NLP parsing?

Uncertainty about which constituent a modifier or prepositional phrase should attach to (Correct answer)
Uncertainty in word segmentation
Ambiguity caused by unknown words
Ambiguity in POS tags

Correct answer: Uncertainty about which constituent a modifier or prepositional phrase should attach to

Attachment ambiguity occurs when a modifier (e.g., a prepositional phrase) can grammatically attach to more than one constituent, yielding different meanings.

Question 19: In dependency parsing, what is the 'head' of a dependency relation?

The root node in a constituency parse tree
The word that another word (the dependent) grammatically modifies or depends on (Correct answer)
The subject of the main clause in the sentence
The first token of the sentence that governs all other tokens

Correct answer: The word that another word (the dependent) grammatically modifies or depends on

In a dependency parse, each word (except the root) has a head — the word it syntactically depends on, connected by a directed arc.

Question 20: What does RLHF (Reinforcement Learning from Human Feedback) enable in language model training?

Enabling zero-shot learning without any pre-training
Faster convergence by replacing gradient descent with policy gradients
Compressing model size through preference-based pruning
Aligning model outputs with human preferences using a reward model trained on human rankings (Correct answer)

Correct answer: Aligning model outputs with human preferences using a reward model trained on human rankings

RLHF trains a reward model on human preference comparisons and then uses RL (typically PPO) to fine-tune the LLM to maximize that reward, improving helpfulness and safety.

Question 21: What is the main purpose of the CRF (Conditional Random Field) layer commonly added on top of BERT for sequence labeling tasks?

To apply dropout regularization during inference
To model dependencies between consecutive output labels (Correct answer)
To convert subword tokens back to word-level predictions
To reduce the dimensionality of token representations

Correct answer: To model dependencies between consecutive output labels

CRF models global label sequence dependencies, ensuring predictions like B-I-O tagging constraints are respected across tokens.

Question 22: What is the primary purpose of stop word removal in NLP preprocessing?

To correct spelling errors in text
To convert text to lowercase
To eliminate common words that carry little meaningful information (Correct answer)
To segment text into sentences

Correct answer: To eliminate common words that carry little meaningful information

Stop word removal filters out very common words like 'the,' 'is,' and 'in' that typically don't contribute meaningful information for many NLP tasks.

Question 23: In transformer models, what does 'positional encoding' solve?

It encodes part-of-speech tags into input vectors
It normalizes token embeddings to unit length
It adds word frequency information to embeddings
It gives the model information about the order of tokens, since attention itself is order-agnostic (Correct answer)

Correct answer: It gives the model information about the order of tokens, since attention itself is order-agnostic

Since the self-attention mechanism treats all positions equally, positional encodings are added to token embeddings so the model can distinguish word order.

Question 24: Which tagging scheme is commonly used for NER to mark entity boundaries?

BIO (Beginning, Inside, Outside) tagging (Correct answer)
POS (Part-of-Speech) tagging
Dependency labeling
Chunking with IOB2 only

Correct answer: BIO (Beginning, Inside, Outside) tagging

BIO tagging marks the first token of an entity with B-, continuation tokens with I-, and non-entity tokens with O, clearly delineating entity boundaries.

Question 25: What is the role of part-of-speech (POS) tagging in the NLP pipeline?

It segments sentences into clauses
It removes stop words from text
It assigns grammatical categories (noun, verb, adjective, etc.) to each token (Correct answer)
It generates syntactic parse trees directly

Correct answer: It assigns grammatical categories (noun, verb, adjective, etc.) to each token

POS tagging labels each token with its grammatical category, which is a prerequisite for most syntactic parsing approaches.

Question 26: What is a probabilistic context-free grammar (PCFG)?

A grammar that uses neural networks to score parses
A CFG where rules are applied randomly
A CFG where each production rule has an associated probability, and probabilities for a non-terminal sum to 1 (Correct answer)
A grammar restricted to a fixed vocabulary

Correct answer: A CFG where each production rule has an associated probability, and probabilities for a non-terminal sum to 1

A PCFG augments each CFG rule with a probability so that the parser can rank alternative parse trees by their overall probability.

Question 27: In ____, natural language processing is used.

Topic modeling
All of the above (Correct answer)
Text classification
Chatbots

Correct answer: All of the above

Natural Language Processing (NLP) is applied in a wide range of real-world scenarios. Chatbots utilize NLP to understand user input and generate appropriate responses, text classification categorizes documents based on their content, and topic modeling identifies abstract themes within a collection of texts. All these applications demonstrate the versatility and utility of NLP in processing and understanding human language.

Question 28: What is Byte-Pair Encoding (BPE) used for in NLP?

Compressing model weights to reduce memory
Subword tokenization to handle rare and out-of-vocabulary words (Correct answer)
Encoding text as binary for storage
Splitting paragraphs into sentences

Correct answer: Subword tokenization to handle rare and out-of-vocabulary words

BPE is a subword tokenization algorithm that iteratively merges frequent character pairs, enabling models to handle unseen words by breaking them into known subunits.

Question 29: What is relation extraction in NLP?

Finding synonyms for words in a sentence
Extracting noun phrases from a parse tree
Splitting compound sentences into simple ones
Identifying semantic relationships between entities mentioned in text, such as 'works at' or 'located in' (Correct answer)

Correct answer: Identifying semantic relationships between entities mentioned in text, such as 'works at' or 'located in'

Relation extraction goes beyond NER to determine how entities are related, populating knowledge bases with structured (subject, relation, object) triples.

Question 30: What is 'word sense disambiguation' in NLP?

Determining which meaning of a polysemous word is used in context (Correct answer)
Translating words between languages
Splitting compound words into components
Converting words to their base form

Correct answer: Determining which meaning of a polysemous word is used in context

Word sense disambiguation (WSD) identifies which meaning of a word with multiple meanings is intended based on surrounding context.

Question 31: What is a knowledge graph in the context of NLP?

A directed acyclic graph of dependency arcs
A structured representation of entities and their relationships stored as a graph of nodes and edges (Correct answer)
A lookup table of word embeddings
A parse tree produced by a constituency parser

Correct answer: A structured representation of entities and their relationships stored as a graph of nodes and edges

Knowledge graphs (e.g., Wikidata, Freebase) store real-world facts as (entity, relation, entity) triples and power question answering, recommendation, and search systems.

IABAC Certified Natural Language Processing Expert (CNLPE)

The CNLPE certification validates expertise in NLP techniques including text processing, syntactic analysis, named entity recognition, sentiment analysis, and modern language model architectures. It covers both foundational NLP concepts and applied machine learning approaches for real-world text-based AI systems.

Questions

35 min

Time Limit

60%

To Pass

Checking for saved progress...

Exam Rules

You can skip questions and return to them later
Flag questions for review before submitting
No feedback shown until you submit the entire exam
Unanswered questions count as wrong — answer everything
10 pretest questions are mixed in and don't affect your score
Timer auto-submits when time runs out
Your progress is auto-saved every 30 seconds