Multilingual security · Security research ethics · Lower-resourced NLP
Heather Lent
Assistant Professor
Heather studies security and privacy in multilingual language models, and how to conduct NLP security research responsibly.
Her work on embedding inversion tests whether an attacker can recover text from multilingual model representations. In NLP Security and Ethics, in the Wild, she and her coauthors examine how NLP security papers handle potential harms and propose guidance on responsible disclosure.
Heather also develops resources and evaluations for Creole languages. CreoleVal provides multilingual benchmarks, while What a Creole Wants, What a Creole Needs examines how community priorities should shape the tasks researchers choose. Her work on multilingual data quality connects these concerns to the resources used to train and evaluate models.
Multilingual model security
Testing privacy leakage through text embeddings and how vulnerabilities differ across languages.
Security research ethics
Examining harm minimisation and responsible disclosure in NLP security research.
Lower-resourced NLP
Developing and evaluating language technology for Creole and other lower-resourced languages, with attention to community priorities.
Selected work
5 representative papers
NLP Security and Ethics, in the Wild
Audits ethical practice in NLP security and develops concrete guidance for harm minimisation and responsible disclosure.
- Security, Privacy & Safety
- Communities, Domains & Consequences
Text Embedding Inversion Security for Multilingual Language Models
- Security, Privacy & Safety
- Multilingual & Lower-Resource NLP
How Good is Your Wikipedia? Auditing Data Quality for Low-resource and Multilingual NLP
Audits all non-English Wikipedia editions and shows why data quality—not language coverage alone—must be a first-class concern in multilingual NLP.
- Multilingual & Lower-Resource NLP
- Factuality, Reliability & Evaluation
CreoleVal: Multilingual Multitask Benchmarks for Creoles
- Multilingual & Lower-Resource NLP
- Communities, Domains & Consequences
What a Creole Wants, What a Creole Needs
- Multilingual & Lower-Resource NLP
- Communities, Domains & Consequences
Supervision and collaboration
Research with Heather Lent
Heather welcomes projects on multilingual model security, privacy risks, and security research ethics, as well as lower-resourced NLP and evaluation.