So, in the end, it’s finished—and now it’s my turn to share that I have successfully defended my PhD thesis entitled "Unmasking Implicit and Subtle Hate Speech: NLP Approaches for Detecting and Countering Online Harm." It’s been an amazing three years in France 🇫🇷 at Université Côte d'Azur, 3IA Côte d'Azur , Centre Inria d'Université Côte d'Azur, where I learned absolutely A LOT—both professionally and personally.

Thanks to Elena Cabrio and Serena Villata for trusting me, for their patience, and for their excellent guidance. Also, thanks to all the members of the jury for reviewing my work and being part of my defense: Farah Benamara, Paolo Rosso, Catherine Faron, and Tommaso Caselli. I also want to thank all the members of MARIANNE and WIMMICS for the huge support, tons of brainstorming moments, and the everyday welcoming vibes that made the office amazing.

Also to the members of LanD at Fondazione Bruno Kessler - FBK that made my period abroad in Italy 🇮🇹 very special. And finally, the huge support of my family, friends, and everyone who was part of this journey—I will be eternally grateful to all of you.

What’s next? A PostDoc as an ERCIM fellow at HCDA@Centrum Wiskunde & Informatica in Amsterdam, Netherlands 🇳🇱, where I've been working during the past two months. I think the best way to show my gratitude to everyone that helped me is by doing the best PostDoc that is possible. Until the next conference, event, or coffee break 👋🏼.

#phd #computerscience #defense #nlp #marianne

Photos of my PhD Defense

🚆 Last week, I started a 3-months visit to the Language and Dialogue Technologies (LanD) research team at Fondazione Bruno Kessler - FBK in Trento, Italy 🗻 🇮🇹 . It's been an amazing first week, meeting plenty of great colleagues and working on my PhD in collaboration with the group.

Thanks a lot to the EUR Digital Systems for Humans, Université Côte d'Azur, and the DocWalker Mobility Program for considering me as one of the selected candidates for the internship! 💻

You can take a look at my story published by the Doctoral School of my University: https://ds4h.univ-cotedazur.eu/studying-at-ds4h-graduate-school/students-story-nicolas-benjamin-ocampo-i3s-docwalker-2024-laureate

💻 Really happy to have been part of the Fête des Startups. It was a great chance to learn more about the Inria Startup Studio program and chat with other researchers and PhD students who have been part of this journey and are building their own companies 📊. Thanks also to Stephanie Morales for all the help and guidance during this mission. Definitely a great experience!

💻 Happy to have participated in the "Natural Language Argumentation Day" and contributed to its organization with friends and colleagues from the WIMMICS team. It was a great chance to meet great researchers and learn more about argumentation technologies. Also, thanks to Elena Cabrio and Serena Villata for their efforts in leading this event and all the institutions that made it possible.

📺🎙️ I was interviewed on the show 'AHI LO VES' on Canal 2 of Salta, Argentina, where I had the opportunity to share my research in hate speech detection using artificial intelligence. A special thanks to Raúl Vargas for the interview and to the AHI LO VES team for providing me with the space.

📰 Hoy Día Córdoba recently interviewed me about my research on Implicit and Subtle Hate Speech Detection at Université Côte d'Azur—many thanks to Marcelo Lucero for the post and to Hoy Día Córdoba for the space.

The full post can be found here: https://hoydia.com.ar/sociedad/un-egresado-de-la-unc-rastrea-discursos-de-odio-en-las-redes-con-inteligencia-artificial/

I was invited to give a seminar at Universidad Nacional de Cordoba about abusive language detection, where I had the chance to present the work of two recent publications: "An In-depth Analysis of Implicit and Subtle Hate Speech Messages" and "Playing the Part of the Sharp Bully: Generating Adversarial Examples for Implicit Hate Speech Detection."

During the talk, I introduced how implicit and subtle hateful messages are more challenging to detect than explicit ones and possible data strategies to improve detection.

At the beginning of May, I participated in the EACL 2023 conference, where I presented my first publication during my Ph.D. entitled "An In-depth Analysis of Implicit and Subtle Hate Speech Messages". Here, I share my experience at one of the top conferences in NLP.

Volunteering

Volunteering at the conference was an excellent opportunity to connect with fellow Ph.D. students working in similar research areas. We had some obligations, such as check-in registration and helping with the plenary or poster sessions. Normally, the high workload is during the first registration day (where volunteers shine the most). Then, for shifts after the first day, sporadically, have people willing to register or need some guidance. I think it is a good chance to get to know other students working in similar domains. So, for me, applying for volunteer work is a must at any conference.

Venue and Organization

The choice of the conference venue, Valamar Lacroma Dubrovnik Hotel, for EACL 2023 struck a good balance between offering various amenities and ensuring attendees stayed engaged on-site during breaks. The overall organization of the event was excellent, with only a few minor changes made on the go. Additionally, the city of Dubrovnik itself was beautiful and small enough to explore its main attractions within one or two evenings.

Keynotes

I want to highlight one of the three invited talks given by Edward Grefenstette entitled "Going beyond the benefits of scale by reasoning about data", which focused on the effectiveness of Large Language Models (LLMs). He discussed how machine learning models needed only a few data samples reflecting a new behavior/application to output behavior like it. Therefore, in this direction, the approach of increasing the data size and model architecture remains promising as long as we provide new data/tasks to train our LLMs on. Additionally, he showed that LLMs are less affected by "recency bias", easy to optimize, hard to overfit, and commonly a single epoch (with maybe only a fraction of the data) is sufficient to train them.

Papers

I really liked papers oriented on ethical issues and responsible AI. This conference had much to offer in this direction, with many outstanding contributions that raised interesting discussions.

Below are some papers I could look through in detail:

Generation-Based Data Augmentation for Offensive Language Detection: Is It Worth It?
Camilla Casula and Sara Tonelli

They presented an evaluation of existing and novel data augmentation setups based on generative large language models for offensive language detection. They tested their approach in a cross-evaluation setting, showing that data augmentation's positive effect may not be consistent across multiple setups.

Why Don't You Do It Right? Analysing Annotators' Disagreement in Subjective Tasks
Marta Sandri, Elisa Leonardelli, Sara Tonelli, and Elisabetta Jezek

They study multiple types of disagreement (taxonomies) when annotating subjective tasks in order to observe the relation between disagreements and the performance of classifiers trained on that annotated data. (Conclusion: high disagreement => harder to classify)

Don't Blame the Annotator: Bias Already Starts in the Annotation Instructions

Mihir Parmar, Swaroop Mishra, Mor Geva, and Chitta Baral

The same conclusion was reached above, but here, the authors focused on identifying biases in annotation guidelines (they called it instruction bias) that can lead to overestimation of model performance.

In-Depth Look at Word Filling Societal Bias Measures
Matúš Pikuliak, Ivana Beňová, and Viktor Bachratý

They explore word-filling techniques to measure bias in Language Models, showing their current limitations.

Nationality Bias in Text Generation

Pranav Narayanan Venkit, Sanjana Gautam, Ruchi Panchanadikar, Ting-Hao Huang, and Shomir Wilson

A study on how GPT2 presents positive and negative biases to certain nationalities based on internet usage and economic status. They also present a debiasing approach using Adversarial Triggers.

Looking for a Needle in a Haystack: A Comprehensive Study of Hallucinations in Neural Machine Translation

Nuno M. Guerreiro, Elena Voita, and André Martins

They present how machine translation models can hallucinate content, which I didn't expect to happen in such a task as it seems pretty straightforward.

There are way more articles that seemed exciting and are on my TODO list to read. I encourage you to take a look!

I presented my work "Subtle Abusive Language Detection" at "Ma thèse en 180 secondes", a francophone competition where PhD students have to explain their thesis topic in 180 seconds to a broad audience. It was definitely an amazing experience where I tested my presentation skills in French!

🏆 Very happy to receive the price as one of the "10 Jovenes Sobresalientes de Córdoba" in 2022. Thanks, Bolsa de Comercio de Córdoba, for keeping this initiative every year.

VIVA CÓRDOBA!

Presentation video during the award ceremony.