Part-of-speech tagging in NLP - ML Experiment: Train & Evaluate

Experiment - Part-of-speech tagging

Problem:Build a model to assign part-of-speech tags to words in sentences. The current model uses a simple neural network but shows signs of overfitting.

Current Metrics:Training accuracy: 98%, Validation accuracy: 75%

Issue:The model overfits the training data, resulting in poor generalization on validation data.

Your Task

Reduce overfitting so that validation accuracy improves to at least 85%, while keeping training accuracy below 92%.

You can only modify the model architecture and training hyperparameters.

Do not change the dataset or preprocessing steps.

Hint 1

Hint 2

Hint 3

Hint 4

Solution

NLP

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Embedding, Bidirectional, LSTM, Dense, Dropout
from tensorflow.keras.callbacks import EarlyStopping

# Assume X_train, y_train, X_val, y_val are preprocessed and ready

vocab_size = 10000  # example vocabulary size
embedding_dim = 64
max_len = 50  # max sentence length
num_tags = 17  # number of POS tags

model = Sequential([
    Embedding(vocab_size, embedding_dim, input_length=max_len),
    Bidirectional(LSTM(64, return_sequences=True)),
    Dropout(0.5),
    Dense(num_tags, activation='softmax')
])

model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.0005),
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

early_stop = EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True)

history = model.fit(X_train, y_train,
                    epochs=20,
                    batch_size=32,
                    validation_data=(X_val, y_val),
                    callbacks=[early_stop])

Added a Dropout layer with rate 0.5 after the LSTM layer to reduce overfitting.

Reduced LSTM units from 128 to 64 to decrease model complexity.

Lowered learning rate to 0.0005 for more stable training.

Added EarlyStopping callback to stop training when validation loss stops improving.

Results Interpretation

Before: Training accuracy was 98%, validation accuracy was 75%, showing strong overfitting.

After: Training accuracy dropped to 90%, validation accuracy improved to 87%, indicating better generalization.

Adding dropout, reducing model size, lowering learning rate, and using early stopping help reduce overfitting and improve validation accuracy in sequence tagging tasks.

Bonus Experiment

Try using a Conditional Random Field (CRF) layer on top of the LSTM to improve sequence tagging accuracy.

💡 Hint

CRF layers model tag dependencies and can improve POS tagging by considering tag sequences rather than independent predictions.

Practice

1. What is the main purpose of part-of-speech tagging in natural language processing?

easy

A. To label each word with its grammatical role in a sentence

B. To translate text from one language to another

C. To count the number of words in a sentence

D. To generate new sentences automatically

Solution

1. Step 1: Understand the role of part-of-speech tagging

   Part-of-speech tagging assigns labels like noun, verb, adjective to each word, showing its grammatical role.

2. Step 2: Compare with other options

   Translation, word counting, and sentence generation are different NLP tasks unrelated to POS tagging.

3. Final Answer:

   To label each word with its grammatical role in a sentence -> Option A

4. Quick Check:

   POS tagging = labeling word roles [OK]

Hint: POS tagging means labeling words by their grammar role [OK]

Common Mistakes:

• Confusing POS tagging with translation
• Thinking POS tagging counts words
• Assuming POS tagging generates sentences

2. Which Python code correctly uses NLTK to perform part-of-speech tagging on the sentence 'I love AI'?

easy

A. import nltk
nltk.pos_tag(['I', 'love', 'AI'])

B. import nltk
nltk.tag_pos(['I', 'love', 'AI'])

C. import nltk
nltk.pos_tag('I love AI')

D. import nltk
nltk.pos_tag(['I love AI'])

Solution

1. Step 1: Check correct function and input type

   The correct function is pos_tag and it expects a list of words, not a string.

2. Step 2: Analyze each option

   import nltk nltk.pos_tag(['I', 'love', 'AI']) uses pos_tag with a list of words, which is correct. import nltk nltk.tag_pos(['I', 'love', 'AI']) uses a wrong function name. import nltk nltk.pos_tag('I love AI') passes a string instead of a list. import nltk nltk.pos_tag(['I love AI']) passes a list with one string, not separate words.

3. Final Answer:

   import nltk nltk.pos_tag(['I', 'love', 'AI']) -> Option A

4. Quick Check:

   pos_tag + list of words = correct syntax [OK]

Hint: Use pos_tag with a list of words, not a single string [OK]

Common Mistakes:

• Passing a string instead of a list
• Using incorrect function name
• Passing a list with one combined string

3. What is the output of the following Python code using NLTK's pos_tag?

import nltk
sentence = ['She', 'runs', 'fast']
tagged = nltk.pos_tag(sentence)
print(tagged)

medium

A. [('She', 'DT'), ('runs', 'VB'), ('fast', 'RB')]

B. [('She', 'NN'), ('runs', 'NN'), ('fast', 'JJ')]

C. [('She', 'PRP'), ('runs', 'VBZ'), ('fast', 'RB')]

D. [('She', 'PRP'), ('runs', 'VBD'), ('fast', 'RB')]

Solution

1. Step 1: Understand POS tags for each word

   'She' is a pronoun (PRP), 'runs' is a verb in present tense third person singular (VBZ), 'fast' is an adverb (RB).

2. Step 2: Match tags with options

   [('She', 'PRP'), ('runs', 'VBZ'), ('fast', 'RB')] matches these tags exactly. Other options have incorrect tags like noun (NN), determiner (DT), or past tense verb (VBD).

3. Final Answer:

   [('She', 'PRP'), ('runs', 'VBZ'), ('fast', 'RB')] -> Option C

4. Quick Check:

   Pronoun + present verb + adverb = [('She', 'PRP'), ('runs', 'VBZ'), ('fast', 'RB')] [OK]

Hint: Know common POS tags: PRP=pronoun, VBZ=verb present, RB=adverb [OK]

Common Mistakes:

• Confusing verb tenses VBZ vs VBD
• Mixing pronouns with nouns
• Mislabeling adverbs as adjectives

4. The following code throws an error. What is the most likely cause?

import nltk
sentence = 'He is happy'
tagged = nltk.pos_tag(sentence)
print(tagged)

medium

A. The nltk module is not imported correctly

B. The sentence variable should be a tuple, not a string

C. pos_tag requires a second argument specifying the language

D. The input to pos_tag should be a list of words, not a string

Solution

1. Step 1: Check input type for pos_tag

   pos_tag expects a list of words, but here a single string is passed, which causes an error.

2. Step 2: Verify other options

   nltk is imported correctly, pos_tag does not require a language argument, and input as tuple is not required.

3. Final Answer:

   The input to pos_tag should be a list of words, not a string -> Option D

4. Quick Check:

   pos_tag input must be list, not string [OK]

Hint: Always pass a list of words to pos_tag, not a string [OK]

Common Mistakes:

• Passing a string instead of a list
• Assuming pos_tag needs language argument
• Confusing input types (tuple vs list)

5. You want to tag parts of speech in a sentence but also handle unknown words gracefully. Which approach best improves POS tagging accuracy for new words?

hard

A. Manually assign tags to each unknown word before tagging

B. Use a POS tagger with a built-in model trained on large diverse text

C. Ignore unknown words during tagging to avoid errors

D. Replace unknown words with a fixed placeholder before tagging

Solution

1. Step 1: Understand handling unknown words in POS tagging

   Taggers trained on large, diverse datasets can predict tags for new words based on context and patterns.

2. Step 2: Evaluate other options

   Manually tagging unknown words is impractical, ignoring them loses information, and replacing with placeholders removes context.

3. Final Answer:

   Use a POS tagger with a built-in model trained on large diverse text -> Option B

4. Quick Check:

   Robust model with training data handles unknown words best [OK]

Hint: Choose taggers trained on big data to handle new words well [OK]

Common Mistakes:

• Trying to manually tag unknown words
• Ignoring unknown words instead of tagging
• Replacing words loses sentence meaning