Worksheet: Sarcasm and Irony Detection

Background

Sarcasm and irony are forms of figurative language where the intended meaning differs from the literal meaning. Detecting these is crucial for sentiment analysis, content moderation, and conversational AI.

Sarcasm: Using irony to mock or convey contempt. Often involves saying the opposite of what one means.

Irony: A broader concept where reality differs from expectation, or words convey opposite meaning.

Why This is Hard

Sarcasm often looks identical to sincere text without context
Cultural and individual differences in sarcasm usage
Same words can be sincere or sarcastic depending on tone
Written text lacks vocal cues present in speech

Part 1: Basic Sarcasm Recognition

Example 1

"Oh great, another meeting that could have been an email."

Sarcastic Sincere Unclear

Example 2

"I love working weekends!"

Sarcastic Sincere Unclear

Example 3

"What a beautiful day for a hike!"

Sarcastic Sincere Unclear

Question 1

For Example 2 and 3, what additional information would you need to be certain?

Part 2: Context Dependence

Example 4a

"Nice job on the presentation."

Context: Said after colleague delivered an award-winning pitch

Sarcastic Sincere

Example 4b

"Nice job on the presentation."

Context: Said after colleague forgot their slides and rambled for 30 minutes

Sarcastic Sincere

Question 2

The same sentence has opposite meanings in different contexts. How should annotation handle this?

Always provide context for annotation Annotate text in isolation (without context) Mark context-dependent cases as "unclear" Annotate both literal and contextual interpretation

Explain your choice:

Part 3: Sarcasm Markers and Cues

Example 5

"Wow, what a BRILLIANT idea. I can't POSSIBLY imagine what could go wrong."

Sarcastic Sincere

Example 6

"Sure, that plan will definitely work 🙄"

Sarcastic Sincere

Question 3

What textual cues suggest sarcasm? (Check all that apply)

ALL CAPS emphasis Eye-roll emoji 🙄 Exaggeration (BRILLIANT) Rhetorical questions Quotation marks Ellipsis (...)

Are these markers reliable? Can they be used sincerely?

Part 4: Cultural and Platform Differences

Example 7 (Twitter/X)

"Day 47 of my diet. Still fat. This is fine."

Sarcastic Sincere/Self-deprecating humor Both (humorous but sincere frustration)

Example 8 (Product Review)

"This vacuum cleaner is so good at collecting dust that it broke after one use. Highly recommend!"

Sarcastic (negative review) Unclear

Question 4

Does the platform/genre affect how you interpret sarcasm?

Part 5: Confidence Annotation

Example 9

"Thanks for the feedback. Really helpful."

Question 5

Label this example and rate your confidence:

Sarcastic Sincere Unclear

Confidence level:

1
Very uncertain

2

3
Moderate

4

5
Very certain

Should confidence be part of sarcasm annotation?

Part 6: Sentiment and Sarcasm Interaction

Question 6

Consider the review: "This product exceeded all my expectations. It fell apart after two days. Just what I was hoping for."

What is the sentiment?

Positive (literal text) Negative (intended meaning) Label both literal and intended

Should sarcasm detection be done before or after sentiment analysis?

Part 7: Group Discussion

Question 7

Compare your annotations with your group. Where did you disagree?

Sarcasm detection itself Context interpretation Confidence levels Cultural differences Subtle vs. obvious sarcasm

Describe one specific disagreement:

Part 8: Reflection

Question 8

Why is sarcasm detection one of the hardest NLP tasks?

Requires extensive context Cultural variation Individual style differences Surface form identical to sincere Requires inferring speaker intent

Key Takeaway

Sarcasm annotation requires modeling speaker intent, not just text—and intent depends on context, culture, and relationship.

The same text can be sincere or sarcastic depending on context
Annotation guidelines must decide: literal meaning or intended meaning?
Confidence annotation may be as important as the label itself
Some texts are genuinely ambiguous—and that's useful information