Abeer Aldayel

p.s. 🌿The title “Hatred Stems from Ignorance” reflects the study’s theoretical grounding in Aristotle’s Rhetoric. which aligns with insights of Ibn Rushd known for reintroducing Aristotle’s texts. The phrase is inspired by one of Ibn Rushd’s reflections on the roots of hatred 📖
shorturl.at/u0PXI

3-This work highlights why ''how you respond'' matters as much as ''what you say" by showing that counterspeech operates through distinct modes of persuasion: reason, emotion and credibility
This work has been accomplished by MSc Graduate Ghadi Alyahya 🌟
📄 Paper: arxiv.org/pdf/2403.154...
#ICWSM2025

2- We examined 9K+ turns across closed (chat-style) and open (X-style) conversations
*Key findings*
- Humans rely more on reason 📈 counterspeech
- LLMs rely more on emotion 🎭 counterspeech
- Reason-based counterspeech gets more supportive stance replies from others 💬 >>

1- While most research asks “Was it persuasive?” ❌✅
we dive deeper into how persuasion works in counterspeech
Instead of binary outcomes of “convincing or not” we explore the rhetorical appeal used
Reason, Credibility, Emotion
a coarse-grained lens on persuasion modes, not fine-grained techniques

📢**Persuasion takes different modes!**
Instead of just asking "Is this counterspeech persuasive?" we asked "How is it persuasive?"
In our new #ICWSM25 paper, with Ghadi Alyahya 🌟 we distill 3 persuasion modes: reason, emotion, and credibility to better examine how humans and LLMs counter hate >>

Covert Bias: The Severity of Social Views’ Unalignment in Language Models Towards Implicit and Explicit Opinion Abeer Aldayel, Areej Alokaili, Rehab Alahmadi. Proceedings of the Third Workshop on NLP for Positive Impact. 2024.

with key findings highlighting the discrepancies between implicit/explicit opinions on social topics and downstream task level
🔈Hear more from @areej_okaili tomorrow! Meanwhile, take a look at the exciting insights in the paper > aclanthology.org/2024.nlp4pi-...
#EMNLP2024 🌴 #NLP4PositiveImpact

In our study "Covert Bias The Severity of Social Views Unalignment in Language Models Towards Implicit and Explicit Opinion” with
@areej_okaili @RehabAlahmadi2 we show how stress testing 🏋️ serves as an evaluation medium to uncover the extent of bias amplification toward implicit opinions in LLMs >>

💡👥How does implicit language that subtly conveys a viewpoint amplify social bias in large language models?
Want to learn more? If you’re attending #emnlpm2024
visit our work tomorrow Friday at #NLP4PosImpact
(in-person posters session)
🗓️15th Nov, 11:00 a.m ET, Location: Foster Hall
🧵>>>