Dependency parsing in NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding Dependency Parsing Output

Given the sentence: "The cat sat on the mat." and its dependency parse, which of the following correctly represents the head of the word "sat"?

A "sat" is the root of the sentence, so it has no head.

B "sat" has the head "cat" because the cat is doing the action.

C "sat" has the head "on" because it shows location.

D "sat" has the head "mat" because it is the object.

💡 Hint

Think about which word is the main verb that connects the sentence.

In dependency parsing, the main verb of a sentence is usually the root and has no head. Here, "sat" is the root.

❓ Predict Output

intermediate

Output of Dependency Parsing with spaCy

What is the output of the following code snippet that uses spaCy to parse the sentence "She enjoys reading books." and prints each token's text and its dependency label?

NLP

import spacy
nlp = spacy.load('en_core_web_sm')
doc = nlp('She enjoys reading books.')
output = [(token.text, token.dep_) for token in doc]
print(output)

A [('She', 'nsubj'), ('enjoys', 'ROOT'), ('reading', 'xcomp'), ('books', 'dobj'), ('.', 'punct')]

B [('She', 'dobj'), ('enjoys', 'nsubj'), ('reading', 'ROOT'), ('books', 'xcomp'), ('.', 'punct')]

C [('She', 'ROOT'), ('enjoys', 'nsubj'), ('reading', 'dobj'), ('books', 'xcomp'), ('.', 'punct')]

D [('She', 'punct'), ('enjoys', 'dobj'), ('reading', 'nsubj'), ('books', 'ROOT'), ('.', 'xcomp')]

💡 Hint

Remember the subject usually has the label nsubj and the main verb is ROOT.

The subject "She" is labeled nsubj, "enjoys" is the root verb, "reading" is an open clausal complement xcomp, and "books" is the direct object dobj.

❓ Model Choice

advanced

Choosing a Dependency Parsing Model for Low-Resource Language

You want to build a dependency parser for a low-resource language with very limited annotated data. Which model approach is most suitable?

A Train a large transformer-based parser from scratch on the small dataset.

B Use a rule-based parser handcrafted for the language without machine learning.

C Use a pre-trained multilingual model and fine-tune it on the small dataset.

D Train a simple linear classifier on raw text without any linguistic features.

💡 Hint

Think about leveraging existing knowledge from other languages.

Pre-trained multilingual models have learned general language patterns and can be fine-tuned effectively on small datasets, making them suitable for low-resource languages.

❓ Metrics

advanced

Evaluating Dependency Parsing Accuracy

Which metric best measures how well a dependency parser predicts the correct head for each word in a sentence?

A BLEU score

B Perplexity

C F1 score for named entity recognition

D Unlabeled Attachment Score (UAS)

💡 Hint

Focus on metrics that evaluate syntactic structure correctness.

Unlabeled Attachment Score (UAS) measures the percentage of words whose predicted head matches the gold standard, ignoring dependency labels.

🔧 Debug

expert

Debugging Incorrect Dependency Parse Output

You trained a dependency parser, but it often predicts the root word incorrectly, assigning root to punctuation marks. Which is the most likely cause?

A The evaluation metric is not suitable for dependency parsing.

B The training data has incorrect root annotations for punctuation.

C The tokenizer splits words incorrectly causing misalignment.

D The model uses a too high learning rate causing overfitting.

💡 Hint

Check the quality of your training annotations carefully.

If punctuation tokens are annotated as roots in training data, the model will learn to predict them as roots, causing errors.