Interested? Come by our Poster @ Poster Session (NLP) 1 on January 22nd, 12–2 PM, Hall 2!
See you in Singapore! 🇸🇬
🧵9/9
@dh-fbk.bsky.social @land-fbk.bsky.social @mobs-fbk.bsky.social
Interactional analysis and qualitative evaluation further demonstrate that LLMs are suitable for synthesizing large-scale, diverse Written MPC datasets, which can be potentially used to fine-tune smaller models for tasks like next speaker or addressee prediction.
🧵8/9
We found that some LLMs (i.e., Llama3.1, Qwen2.5) can generate well-structured multi-party conversations that conform to constraints.
Comparing generation strategies, Turn-by-Turn adheres more closely to constraints and also produces greater linguistic variability.
🧵7/9
We also propose a novel evaluation framework combining quantitative and qualitative metrics.
Such metrics measure how well generated MPCs adhere to content and structural constraints and whether they capture the complexity of real-world multi-party interactions.
🧵6/9
In this paper, we propose synthetic MPC generation testing four different LLMs guided by explicit constraints.
We explore two strategies:
1 - One-Long (OL): generate the entire conversation in one step
2 - Turn-by-Turn (TT): generate sequentially, one turn at a time
🧵5/9
By constraining LLMs, we can generate MPCs that: 1) exhibit diverse interaction structures and 2) capture rich speaker–addressee relationships, including multi-addressee turns
🧵4/9
One way to address the lack of structural diversity in social media–derived MPCs is to synthesize multi-party conversations.
🧵3/9
For current LLMs, multi-party conversations (MPCs) potentially represent a clear distributional shift, since they are mostly tested and optimized for two-party (human–assistant) interactions.
🧵2/9
Social media platforms typically enforce a one-to-one reply structure, ignoring implicit addressees and oversimplifying interaction dynamics.
In natural conversations, turns are often directed to multiple participants, generating much richer structures.
🧵1/9
🎶One more “song”?
Come see our poster next week at #AAAI2026 on our paper:
📣 “Don’t Stop the Multi-Party! On Generating Synthetic Written Multi-Party Conversations with Constraints”
With @marcoguerini.bsky.social Bruno Lepri @gglavas.bsky.social @satonelli.bsky.social
🧵
arxiv.org/abs/2502.13592
Many thanks to @milanlp.bsky.social folks for the opportunity to present some of my works 🥳
⚠️ Online Seminar - April 9th ⚠️
Today, @maxluca.bsky.social and I will give two talks about our latest works at @ellisalicante.org
and you can follow it online: ellisalicante.org/fbk-mobs-wor...
Thanks, @adigulati.bsky.social and @nuriaoliver.com for setting this up!
Congrats to @daniel-russo.bsky.social for the Best Student Paper Award at #CLiCit2024! Congrats also to the co-authors Oscar Araque and @marcoguerini.bsky.social! 🥳
Give a look at @dh-fbk.bsky.social posters at
#CLiC-it2024: Multi-label classification using EuroVoc thesaurus, machine translation for Ladin and Language Varieties of Italy! @ziorufus.bsky.social @jo-valer.bsky.social @alanramponi.bsky.social Proceedings at clic2024.ilc.cnr.it/wp-content/u...
To click it, or not to click it: that is the question.
Our #CLiC-it2024 contribution is finally out: "To Click It or Not to Click It: An Italian Dataset for Neutralising Clickbait Headlines" authored by @daniel-russo.bsky.social, Oscar Araque, and Marco Guerini.
clic2024.ilc.cnr.it/wp-content/u...
Today we will present three of our recent works at #CLiCit2024 with @helenabon.bsky.social, @docchipinti8.bsky.social, @daniel-russo.bsky.social.
Don't miss the poster session today at 10 👀
Hey, if you're in Pisa and interested to connect with other people at #CLiCit2024 check out this starter pack. Raise a hand 👋 and I will add you to this list!
Hello Bluesky! We would like to kick off this account with the DH @ FBK starter pack, which you can use to follow the members of our research unit 🌟
go.bsky.app/5feLo44
👋
