Ever faced difficulties to write good rebuttals for
#PeerReview
? Jiu-Jitsu argumentation comes to the rescue 🥋 !
In our
#EMNLP2023
paper we draw inspiration from Jiu-Jitsu argumentation to generate rebuttals that align with a reviewer’s attitude roots and themes 📚 (1/🧵)
In the world of arguments, it's all about attitude roots and themes 🌱 What are these?
They’re the hidden drivers behind our arguments. Instead of countering surface-level reasoning, understanding these can revolutionize how we engage in conversations – like peer review. (2/🧵)
Peer-reviewing for a scientific paper is composed of two phases – review and rebuttal. Reviewers provide feedback in the reviewing phase; authors need to mitigate concerns raised in the review using carefully designed counter-arguments in the rebuttal phase. (3/🧵)
#EMNLP2023
It has been proposed that writing effective rebuttals can induce a positive score change. However, writing rebuttals is difficult, especially for junior researchers and non-native speakers. (4/🧵)
#EMNLP2023
Previous efforts in counter-argument generation focussed on surface-level reasoning e.g., falsifying premises. However, underlying beliefs that drive arguments were never considered.
This is where the idea of attitude roots and attitude themes come into the picture. (5/🧵)
Imagine countering arguments like a
#JiuJitsu
master! 🥋 Instead of attacking the surface, focus on understanding an arguer's attitude roots and themes. By doing so, we can craft rebuttals that resonate with their beliefs. (6/🧵)
#EMNLP2023
Does this concept even apply to peer reviews? Do the reviewers have underlying attitude roots and themes while judging the standards of scientific papers?
Spoiler alert!😲 Yes, they do. Let's learn more! (7/🧵)
🌟 In our paper on Jiu-Jitsu Argumentation for Writing Peer Review Rebuttals, we explore this new paradigm in the domain of peer reviews with a novel task, dataset, and generation methods! 📚 (8/🧵)
📃
📖 Enriching Discourse Structure 📝
To achieve attitude and theme-guided rebuttal generation, we've enriched an existing dataset for peer reviews and proposed a new dataset,
#JITSUPEER
. (9/🧵)
#EMNLP2023
#JITSUPEER
comes with different attitude roots and themes, recast from existing concepts along with customized rebuttals. These customized rebuttals align with the reviewer's attitude roots and themes and are known as canonical rebuttals. A sneak peek into JITSUPEER 🧐 (10/🧵)
We've develop strong rebuttal generation strategies on
#JITSUPEER
that align with attitude roots and themes. 🚀
We however observe that the models still underperform on this task due to the sheer variety of the canonical rebuttals in our dataset. (11/🧵)
This underscores the complexity of this task and points to the potential of data-efficient approaches in this area.
Consider following our authors
@i_sukannya
,
@anne_lauscher
(
@unihh
) and
@IGurevych
for more info. See you at
#EMNLP2023
in 🇸🇬! (12/12)
📃