Who's Who of AI

Debora Nozza

79 trust researcher @deboranozza.bsky.social · 451 followers
AI research NLP & language Responsible AI
Why they matter

Researcher with public evidence across AI research, NLP & language, Responsible AI.

AI signals
13
past 30d
Sources
7
distinct domains
Discussions
0
past 30d
Latest signal
1d ago
View every signal from Debora Nozza →
Assistant Professor at Bocconi University in MilaNLP group • Working in #NLP, #CSS and #Ethics • She/her • #ERCStG PERSONAE
What they're sharing

Articles & links

↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

For today's reading group, @marlutz.bsky.social presented "State media control influences large language models" by Waight et al. (2026) Paper: www.nature.com/articles/s41... #NLProc

State media control influences large language models | Nature nature.com
AI Weekly's analysis →
  • Chinese state-media content appears in typical LLM training sets at roughly 41 times the rate of Chinese-language Wikipedia.
  • Across 37 countries, models prompted in the local language produce more regime-favorable responses in countries with lower press freedom.
  • A pretraining experiment with just 6,400 state-scripted documents pushed an open-weight model to pro-government responses nearly 80 percent of the time.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

For today's reading group, @veraneplenbroek.bsky.social presented "Old Habits Die Hard: How Conversational History Geometrically Traps LLMs" by Simhi et al. (2026) Paper: arxiv.org/abs/2603.03308 #NLProc

[2603.03308] Old Habits Die Hard: How Conversational History Geometrically Traps LLMs arxiv.org
AI Weekly's analysis →
  • A new arXiv paper introduces History-Echoes, a framework showing conversation history biases what large language models generate next.
  • Across three model families and six datasets, the authors find gaps in latent space form a 'geometric trap' confining a model's response trajectory.
  • The work suggests hallucinations from earlier turns can influence later responses, implying an early mistake tends to persist inside a session.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

#MemoryModay #NLProc Fornaciari et al.'s 2022 'Hard and Soft Evaluation of NLP models with BOOtSTrap SAmpling - BooStSa' is a useful Python tool for benchmarking NLP predictions with bootstrapped sampling. aclanthology.org/2022.acl-dem...

Hard and Soft Evaluation of NLP models with BOOtSTrap SAmpling - BooStSa aclanthology.org
AI Weekly's analysis →
  • BooStSa is an ACL 2022 demo tool that automates bootstrap-sampling significance tests for comparing NLP model results across many conditions.
  • Unlike prior helpers, it handles both standard categorical predictions and soft-label outputs expressed as probability distributions across classes.
  • Tommaso Fornaciari, Alexandra Uma, Massimo Poesio and Dirk Hovy presented the tool at ACL 2022 in Dublin, Ireland.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

For today's reading group, @esradonmez.bsky.social presented "RLHF May Not Reflect Genuine Preferences" by Ghafouri et al. (2026). Interesting thoughts on whether annotations are actually real preferences! Paper: arxiv.org/abs/2604.03238 #NLProc #RLHF

RLHF May Not Reflect Genuine Preferences arxiv.org
AI Weekly's analysis →
  • A new arxiv preprint argues RLHF annotator responses may not represent genuine preferences at all, but responses constructed on the spot.
  • Filtering high-inconsistency annotators in two RLHF datasets flipped majority harm classifications for 18.6% of prompts.
  • The same filtering shifted mean ratings by more than 13 points on a 100-point scale, suggesting systematic rather than random noise.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

#TBT #NLProc Hessenthaler et al.'s 2022 work delves into AI's link with fairness & energy reduction in English NLP models, challenging bias reduction theories. #AI #NLP #sustainability aclanthology.org/2022.emnlp-m...

Bridging Fairness and Environmental Sustainability in Natural Language Processing aclanthology.org
AI Weekly's analysis →
  • An EMNLP 2022 paper reports that knowledge distillation, a common efficiency technique, can actually decrease model fairness rather than preserve it.
  • The case study evaluates distilled models on natural language inference and semantic similarity, with gender bias measured via the Word Embedding Association Test.
  • The authors argue fairness and environmental sustainability are studied in isolation, and that an exclusive focus on one can hinder the other.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

For today's reading group, @pia-p.bsky.social presented "When Personalization Meets Reality: A Multi-Faceted Analysis of Personalized Preference Learning" by Yijiang River Dong et al. (2025) Paper: aclanthology.org/2025.finding... #NLProc

aclanthology.org View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

#MemoryModay #NLProc 'My Answer is C' by Wang et al. (2024) underscores the scrutiny needed for full text responses in LLMs multi-choice evaluations. aclanthology.org/2024.finding...

“My Answer is C”: First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models aclanthology.org
AI Weekly's analysis →
  • A Findings of ACL 2024 paper reports mismatch rates over 60% between first-token log-probability rankings and the model's actual text answer.
  • The gap holds across final option choice, refusal rate, choice distribution and robustness under prompt perturbation, according to the authors.
  • Models heavily fine-tuned on conversational or safety data are especially impacted, and constraining prompts to force an option letter does not close the gap.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

#TBT #NLProc 'Language is Scary when Over-Analyzed...' by @arimuti.bsky.social et al. explores argumentative reasoning in misogyny detection (2024). Detecting implicit misogyny proves challenging for language models. aclanthology.org/2024.emnlp-m...

Language is Scary when Over-Analyzed: Unpacking Implied Misogynistic Reasoning with Argumentation Theory-Driven Prompts aclanthology.org
AI Weekly's analysis →
  • The paper reframes misogyny detection as an argumentative reasoning task and tests whether LLMs can supply the missing link to implied meaning.
  • Authors conclude LLMs 'fall short on reasoning capabilities' about misogynistic comments and default to internalized stereotypes about women.
  • The study covers Italian and English, using zero-shot and few-shot prompts with chain-of-thought reasoning and augmented knowledge techniques.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

#TBT #NLProc Bergman et al.'s 'Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design' explores AI launch with a value-sensitive lens. aclanthology.org/2022.sigdial...

Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design aclanthology.org
AI Weekly's analysis →
  • A 2022 SIGDIAL paper proposes a framework for deciding whether and how to release end-to-end conversational AI, grounded in value-sensitive design.
  • The authors argue such models trained on internet data may learn toxic or otherwise harmful language, forcing tradeoffs between positive impact and harm.
  • The contribution is not a mitigation technique but decision guidance for practitioners, surveying tensions between values, potential positive impact, and potential harms.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

#MemoryMonday #NLProc 'Entropy-based Attention Regularization Frees Unintended Bias Mitigation from Lists' by Attanasio et al. redefines bias reduction in #AI, sans prior term knowledge. #2022Publication aclanthology.org/2022.finding...

Entropy-based Attention Regularization Frees Unintended Bias Mitigation from Lists aclanthology.org
AI Weekly's analysis →
  • EAR adds an objective that penalizes tokens with low self-attention entropy, discouraging BERT from overfitting to specific training terms.
  • Across three benchmark corpora in English and Italian, EAR matches or exceeds state-of-the-art for hate-speech classification and bias metrics.
  • Because the method needs no list, it also surfaces the terms most likely to induce bias as a diagnostic byproduct.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

#MemoryMonday #NLProc 'BERTective: Language Models and Contextual Information for Deception Detection' by Fornaciari, T. et al. (2021) explores AI's ability to detect deceit through context. www.aclweb.org/anthology/20...

BERTective: Language Models and Contextual Information for Deception Detection - ACL Anthology aclweb.org
AI Weekly's analysis →
  • Fornaciari, Bianchi, Poesio and Hovy combine BERT with attention over surrounding text to identify deceptive statements in Italian dialogues.
  • Only context near the target utterance helps, and only when it comes from the same speaker rather than from an interlocutor's questions.
  • The authors report a new state of the art on the task and release the dataset and code for reproducibility on GitHub.
Read full analysis →
 View on Bluesky →
↻ Debora Nozza reposted
MilaNLP Lab @milanlp.bsky.social

#TBT #NLProc 'Exploring challenges in Zero-shot Cross-lingual Hate Speech Detection, @deboranozza.bsky.social (2021) reveals how current models may inaccurately label non-hateful, language-specific interjections as hate speech signals.' aclanthology.org/2021.acl-sho...

Exposing the limits of Zero-shot Cross-lingual Hate Speech Detection aclanthology.org
AI Weekly's analysis →
  • Debora Nozza's 2021 ACL-IJCNLP short paper tests transferring an English hate speech model to Italian and Spanish without target-language labels.
  • Post-hoc explanations show the model misreads non-hateful, language-specific taboo interjections as signals of hate speech.
  • The paper concludes zero-shot cross-lingual models cannot be used as they are and need to be carefully designed.
Read full analysis →
 View on Bluesky →
Their network

In Debora Nozza's orbit

Center = Debora Nozza. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.