Elamur Research⌕
☰
home / authors / Seraphina Goldfarb-Tarrant

Seraphina Goldfarb-Tarrant

Seraphina Goldfarb-Tarrant has 4 papers in our library, in 2021–2026, most of them on Language models & pretraining, Bias, fairness & safety, Multilingual & cross-lingual NLP; often with Adam Lopez, Arash Ahmadian, Ayomide Odumakinde.

4 papers in our library · 2021–2026

No OpenAlex profile matched yet.

📄 4 works in our library · 🔗 Cited 169 times in our library · ⭐ Paper Stars ranking: #5832

The profile is the library's match of the name; two people with one name are not always told apart.

Elsewhere: Google Scholar · Semantic Scholar · dblp · ACL Anthology

 Fields, topics and concepts

Of its works in our library, with the number of works on each.

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 4 worksArtificial Intelligence · 4 worksSafety Research · 3 worksSocial Sciences · 3 worksArts and Humanities · 3 worksLanguage and Linguistics · 3 worksInformation Systems · 1 work

🧭 Topics 4Research subjects and languages the work is about.Evaluation & benchmarks · 1 workBias, fairness & safety · 2 worksLanguage models & pretraining · 2 worksMultilingual & cross-lingual NLP · 1 work

 Works by year

2020s

20261 paper20251 paper20241 paper20211 paper

 Papers

The most cited first (OpenAlex and our library).

  1. Intrinsic Bias Metrics Do Not Correlate with Application Bias (2021 · 117 citations)
    Research on debiasing word embeddings usually assumes that lowering bias inside the vector space will also lower unfair behaviour in downstream systems.
  2. The Multilingual Alignment Prism: Aligning Global and Local Preferences to Reduce Harm (2024 · 49 citations)
    Most LLM safety alignment targets English and relies on Western-centric data. This leaves non-English users more exposed and lets other languages serve as a route around guardrails. The paper asks three things: whether harm mitigation can work across many languages at once, whether it transfers across languages, and…
  3. Safer or Luckier? LLMs as Safety Evaluators Are Not Robust to Artifacts (2025 · 13 citations)
    Many teams let an LLM judge decide which of two models gives safer output, but this practice has rarely been stress-tested for safety.
  4. Lost in Simulation: LLM-Simulated Users are Unreliable Proxies for Human Users in Agentic Evaluations (2026 · 5 citations)
    Agentic benchmarks commonly replace real people with an LLM that plays the user, but this approach has rarely been checked against actual human behaviour.

 Co-authors here

10 co-authors with a page, the most papers together first.

Venues

Where their works here were published, the most works first, with the editions of those that have a page.

 Cited by

Papers of our library cite their works 169 times (of the 149,439 papers whose bibliographies we read); 36 citing papers with a page here →.

Updated

This page in Markdown → https://md.research.elamur.ai/free/en/authors/seraphina-goldfarb-tarrant/