Tokenization (word and sentence) in NLP - Model Metrics & Evaluation

Metrics & Evaluation - Tokenization (word and sentence)

Which metric matters for Tokenization and WHY

Tokenization breaks text into pieces like words or sentences. The key metric is tokenization accuracy, which measures how many tokens are correctly identified compared to a trusted reference. This matters because wrong tokens can confuse later steps like understanding or translation.

Confusion matrix for Tokenization

          | Predicted Token | Predicted No Token
    ---------------------------------------------
    Actual Token    |       TP       |        FN
    Actual No Token |       FP       |        TN

    TP = correctly found tokens
    FP = wrongly added tokens
    FN = missed tokens
    TN = correctly ignored non-tokens
    

Tradeoff: Precision vs Recall in Tokenization

Precision means how many predicted tokens are actually correct. High precision means fewer wrong splits.

Recall means how many real tokens were found. High recall means fewer missed tokens.

For example, in sentence tokenization, missing a sentence end lowers recall. Adding extra breaks lowers precision.

Good tokenization balances both. Too many splits (high recall, low precision) confuse meaning. Too few splits (high precision, low recall) lose details.

Good vs Bad Metric Values for Tokenization

• Good: Precision and recall above 95% means tokens match well with true text parts.
• Bad: Precision below 70% means many wrong tokens added.
• Bad: Recall below 70% means many tokens missed.
• F1 score near 1.0 means balanced and accurate tokenization.

Common Pitfalls in Tokenization Metrics

• Ignoring punctuation: Some tokenizers split on punctuation differently, causing metric mismatches.
• Data leakage: Testing on data used for tokenizer tuning inflates scores.
• Overfitting: Tokenizer tuned too much on one text style may fail on others.
• Accuracy paradox: Overall accuracy can be misleading if many tokens are easy to find but some critical tokens are missed.

Self Check

Your tokenizer has 98% accuracy but 12% recall on sentence breaks. Is it good?

Answer: No. The low recall means it misses most sentence boundaries. This harms tasks needing sentence understanding, even if overall accuracy seems high.

Key Result

Tokenization quality is best judged by balanced precision and recall above 95%, ensuring tokens match true text parts accurately.

Practice

1. What is the main purpose of tokenization in natural language processing?

easy

A. To remove stop words from text

B. To translate text into another language

C. To split text into smaller units like words or sentences

D. To generate new sentences from text

Solution

1. Step 1: Understand tokenization

   Tokenization means breaking text into smaller pieces such as words or sentences.

2. Step 2: Identify the main goal

   The main goal is to prepare text for further processing by splitting it into tokens.

3. Final Answer:

   To split text into smaller units like words or sentences -> Option C

4. Quick Check:

   Tokenization = splitting text [OK]

Hint: Tokenization means cutting text into pieces [OK]

Common Mistakes:

• Confusing tokenization with translation
• Thinking tokenization removes words
• Believing tokenization generates new text

2. Which of the following Python code snippets correctly tokenizes a sentence into words using NLTK?

easy

A. from nltk.tokenize import word_tokenize
sentence = 'Hello world!'
tokens = word_tokenize(sentence)

B. import nltk
sentence = 'Hello world!'
tokens = nltk.split(sentence)

C. from nltk.tokenize import sent_tokenize
sentence = 'Hello world!'
tokens = sent_tokenize(sentence)

D. sentence = 'Hello world!'
tokens = sentence.split_words()

Solution

1. Step 1: Check correct import and function

   The correct function to tokenize words in NLTK is word_tokenize from nltk.tokenize.

2. Step 2: Verify code correctness

   from nltk.tokenize import word_tokenize sentence = 'Hello world!' tokens = word_tokenize(sentence) imports word_tokenize and applies it correctly to the sentence.

3. Final Answer:

   from nltk.tokenize import word_tokenize\nsentence = 'Hello world!'\ntokens = word_tokenize(sentence) -> Option A

4. Quick Check:

   Use word_tokenize for word splitting [OK]

Hint: Use word_tokenize from nltk.tokenize for words [OK]

Common Mistakes:

• Using sent_tokenize for word tokenization
• Calling non-existent split_words() method
• Using nltk.split which does not exist

3. What will be the output of this Python code using NLTK?

from nltk.tokenize import sent_tokenize
text = 'Hello world! How are you?'
sentences = sent_tokenize(text)
print(sentences)

medium

A. ['Hello world!', 'How are you?']

B. ['Hello world! How are you?']

C. ['Hello', 'world!', 'How', 'are', 'you?']

D. ['Hello world', 'How are you']

Solution

1. Step 1: Understand sent_tokenize function

   sent_tokenize splits text into sentences based on punctuation.

2. Step 2: Apply sent_tokenize to the text

   The text has two sentences: 'Hello world!' and 'How are you?'.

3. Final Answer:

   ['Hello world!', 'How are you?'] -> Option A

4. Quick Check:

   sent_tokenize splits sentences correctly [OK]

Hint: sent_tokenize splits text at sentence ends [OK]

Common Mistakes:

• Confusing sent_tokenize with word_tokenize output
• Expecting no split for multiple sentences
• Ignoring punctuation as sentence boundary

4. Identify the error in this code snippet for word tokenization using NLTK:

import nltk
tokens = nltk.word_tokenize('Hello world!')

medium

A. The string should be a list, not a plain string

B. word_tokenize should be called as nltk.tokenize.word_tokenize

C. word_tokenize does not exist in NLTK

D. Missing import of word_tokenize from nltk.tokenize

Solution

1. Step 1: Check how word_tokenize is imported

   word_tokenize is in nltk.tokenize, not directly in nltk module.

2. Step 2: Identify correct import

   Must import word_tokenize specifically: from nltk.tokenize import word_tokenize.

3. Final Answer:

   Missing import of word_tokenize from nltk.tokenize -> Option D

4. Quick Check:

   Import word_tokenize correctly [OK]

Hint: Import word_tokenize from nltk.tokenize, not nltk [OK]

Common Mistakes:

• Assuming nltk.word_tokenize exists
• Trying to call word_tokenize without import
• Passing list instead of string to tokenizer

5. Given a paragraph with multiple sentences, how can you tokenize it into words while preserving sentence boundaries using NLTK?

hard

A. Use word_tokenize directly on the whole paragraph

B. Use sent_tokenize to split sentences, then word_tokenize each sentence separately

C. Use split() method on the paragraph string

D. Use sent_tokenize only, it also splits words

Solution

1. Step 1: Understand the need to preserve sentence boundaries

   Preserving sentence boundaries means keeping words grouped by sentences.

2. Step 2: Apply sent_tokenize then word_tokenize

   First split paragraph into sentences, then tokenize words in each sentence separately.

3. Final Answer:

   Use sent_tokenize to split sentences, then word_tokenize each sentence separately -> Option B

4. Quick Check:

   Split sentences first, then words [OK]

Hint: Split sentences first, then tokenize words inside each [OK]

Common Mistakes:

• Tokenizing words directly loses sentence grouping
• Using split() which is too simple
• Assuming sent_tokenize splits words