View Proposal

---

Proposer

Simona Frenda

Title

Towards new models aware of diverse cultures, values and beliefs in NLP

Goal

Create new methods to incorporate diverse cultural values and beliefs in existing small models, enabling fairer answers.

Description

The need to design models aware of multiple beliefs, values and cultures arises because of the relevance of AI models in society. Indeed, not all countries and societies have equal access to technologies and their development, and if, on the one hand, stereotypes about nationalities and cultures are learnt by models because of training data, on the other hand, their opinions and perspectives are not considered by AI models [1,2]. The aim of this project is to explore methods (e.g., retrieval-augmented strategies and fine-tuning) to enhance the knowledge of models towards specific cultures, values and beliefs to detect highly subjective phenomena such as hate speech detection, emotion, misinformation and irony. By leveraging multi-annotated corpora and metadata about the annotators [3], the student can design models that are able to produce multiple responses (reflecting different beliefs/values/cultures), giving the user the opportunity to decide, taking into account various perspectives. In this context, another challenge will be the evaluation of the model in terms of fairness and inclusivity of the considered cultures, beliefs and values.

Resources

[1] Dignum, V. (2022) Responsible Artificial Intelligence - From Principles to Practice: A Keynote at TheWebConf 2022. [2] Narayanan Venkit, P., Gautam, S., Panchanadikar, R., Huang, T., and Wilson, S. (2023) Unmasking Nationality Bias: A Study of Human Perception of Nationalities in AI-Generated Articles. In AIES '23. [3] Frenda, S., Abercrombie, G., Basile, V., Pedrani, A., Panizzon, R., Cignarella, A.T., Marco, C. and Bernardi, D., (2024). Perspectivist approaches to natural language processing: a survey. Language Resources and Evaluation.

Background

https://pdai.info/ https://le-wi-di.github.io/ https://github.com/anudeex/Awesome-Pluralistic-Alignment

Url

Difficulty Level

Moderate

Ethical Approval

None

Number Of Students

0

Supervisor

Simona Frenda

Keywords

pluralistic models, multilingual and multicultural dataset

Degrees

Master of Science in Artificial Intelligence
Master of Science in Artificial Intelligence with SMI
Master of Science in Data Science