Michele Papucci

PhD Student in Natural Language Processing at the University of Pisa #NLPROC

[ItaliaNLP Lab @ItaliaNLP_Lab]

🧶 1/3

We’re excited to share our second paper presented at #LREC2026 in Palma: “Controllable Sentence Simplification in Italian: Fine-Tuning Large Language Models on Automatically Generated Resources” by @mpapucci_, Giulia Venturi and Felice Dell'Orletta

[ItaliaNLP Lab @ItaliaNLP_Lab]

In this work we present UGLD, a decoding method that nudges LLMs toward or against a predefined vocabulary at inference time. No fine-tuning needed and without loss of fluency.

📄 Paper: https://t.co/aJisq2rKTK
🔧 GitHub: https://t.co/xjTVJZQB9i
🐍 Install UGLD: pip install ugld

[ItaliaNLP Lab @ItaliaNLP_Lab]

🚀 Excited to share our latest work at READIxTSAR workshop at #LREC2026

"Lexical Conditioning of Model's Distribution through Uncertainty-gated Soft-Mixing of Probabilities" by @mpapucci_ Giulia Venturi and Felice Dell'Orletta.

[Michele Papucci @mpapucci_]

Last week I presented my latest work at the Findings Poster Session of ACL 2025 in Vienna!

If you missed it, check it out 💥

https://t.co/HDhwEqpQfa

[Michele Papucci @mpapucci_]

@ItaliaNLP_Lab @CLiC_it_conf @AILC_NLP 🔥🔥

[ItaliaNLP Lab @ItaliaNLP_Lab]

🎉 Great news! We got 9 papers accepted at CLiC-it 2025! Looking forward to presenting them this year in Cagliari! 🇮🇹
#CLiCit2025 #NLProc @CLiC_it_conf @AILC_NLP

[ItaliaNLP Lab @ItaliaNLP_Lab]

We are at the Lectures on Computational Linguistics in Milan organized by @AILC_NLP ! 🔥 @mpapucci_ and @workerplacemint are presenting their PhD work during the poster session! 🥳 #NLProc

[Michele Papucci @mpapucci_]

@koksing456 @pdrndr @AlessioMiaschi @gpuccetti92 Gemma is going to be uploaded here: https://t.co/JUqkmHQX3p

[Michele Papucci @mpapucci_]

🧵1/ Machine-Generated Text (MGT) detection is failing.

Our paper, accepted at Findings of ACL 2025, shows that LLMs can easily fool generated-text detectors.
https://t.co/tGVAwQKSZR

@pdrndr, Cristiano Ciaccio, @AlessioMiaschi, @gpuccetti92, Felice dell'Orletta, Andrea Esuli

[Michele Papucci @mpapucci_]

@koksing456 @pdrndr @AlessioMiaschi @gpuccetti92 We've yet to upload the models, but the Llama one will be uploaded on HuggingFace in the next days, here: https://t.co/ROo00bX9qn

[Michele Papucci @mpapucci_]

8/ TL:DR;
🚨 State-of-the-art Detectors today are too shallow
📉 A bit of style alignment makes them crumble
🧠 We need stronger benchmarks
🛠 We develop a way to create hard, in-domain texts for making and evaluating the next generation of more robust and reliable MGT Detectors

[Michele Papucci @mpapucci_]

7/ What about Humans?
Human performance was unaffected: they performed poorly in detecting machine-generated text (around 50% accuracy in a binary task) both before and after our alignment.

[Michael Hanna @michaelwhanna]

@mntssys and I are excited to announce circuit-tracer, a library that makes circuit-finding simple!

Just type in a sentence, and get out a circuit showing (some of) the features your model uses to predict the next token. Try it on @neuronpedia: https://t.co/JYmcZz1f1J

[ItaliaNLP Lab @ItaliaNLP_Lab]

4) Findings: Stress-testing Machine Generated Text Detection: Shifting Language Models Writing Style to Fool Detectors (@pdrndr, @mpapucci_, Ciaccio C., @AlessioMiaschi, @gpuccetti92, Dell'Orletta F. and Esuli A.)