AI Researcher

Muhammad S. Abdo

PhD Candidate · Indiana University Bloomington

I study how language models represent, transfer, and reason about linguistic information across speech, text, and multilingual settings. My work spans mechanistic interpretability, natural language inference, and knowledge-grounded AI.

Explore Research Publications View CV

Citations

Publications

h-index

i10-index

About

AI Researcher & Computational Linguist

I am a dual-major PhD candidate in Computational Linguistics and Middle Eastern Languages & Cultures at Indiana University Bloomington, with a minor in Computer Science. My research centers on mechanistic interpretability, multilingual natural language inference, and evidence-grounded AI systems.

I investigate what language models encode internally, how reliably they transfer across languages, and how structured knowledge can ground their outputs. I also build public datasets and tools, including Rasid and the cloud-based research editor RogueTeX.

Full Biography →

Research

Areas of Focus

Connecting linguistic theory with empirical machine learning to build AI systems that are interpretable, reliable, and grounded in evidence.

01 // Interpretability

Mechanistic Interpretability

Probing internal model representations to understand how speech and language transformers encode linguistic features, nativeness, and structure.

Transformers AI Safety Speech

02 // NLI

Natural Language Inference

Studying entailment, contradiction, and pragmatic inference across languages, with a focus on evaluation reliability and label drift introduced by machine translation.

Entailment Pragmatics Multilingual

03 // Knowledge Grounding

RAG & Knowledge Graphs

Building ontology-based knowledge graphs and retrieval systems that turn complex scientific literature into structured, queryable, and evidence-grounded resources.

RAG Ontology Biomedical AI

04 // Language Technology

Multilingual & Low-Resource NLP

Developing datasets, corpora, and models for languages and domains that remain underrepresented in modern AI, including work on variation, ellipsis, entities, and discourse.

Datasets Language Variation NLP

All Research Projects →

Publications

Selected Works

All Publications →

Journal 2025 📄 Published

Ellipsis in Arabic: Using Machine Learning to Detect and Predict Elided Words

Muhammad S. Abdo, Damir Cavar, Billy Dickson, Attia Youseif

Arabic Linguistics, John Benjamins

Scholar ↗

Workshop 2025 ★ 1 cite

AMWAL: Named Entity Recognition for Arabic Financial News

Muhammad S. Abdo, Yash Hatekar, Damir Cavar

FinNLP-FNP-LLMFinLegal @ COLING 2025, Abu Dhabi

Scholar ↗

Conference 2024 ★ 12 cites

The Typology of Ellipsis: A Corpus for Linguistic Analysis and Machine Learning Applications

Damir Cavar, Louis V. Mompelat, Muhammad S. Abdo

SIGTYP 2024 @ EACL

Scholar ↗

Journal 2020 ★ 74 cites

How Do Arab Tweeters Perceive the COVID-19 Pandemic?

Basma A. Essam, Muhammad S. Abdo

Journal of Psycholinguistic Research

Scholar ↗

News

Recent Activity

May 2026

APSA MENA AI Roundtable Participated in the American Political Science Association (APSA) roundtable, discussing the regional impacts, governance, and technical challenges of artificial intelligence in the MENA region.

May 2026

Completed Technical AI Safety CourseAI Safety Fellowship — Completed BlueDot Impact's technical AI safety curriculum focusing on neural network alignment and governance frameworks.

Apr 2026

Poster at NLP @ Michigan Day — Presented mechanistic interpretability approach to detecting linguistic nativeness in Arabic speech transformers.

Apr 2026

Two Posters at MSLD 2026 — Presented at Midwest Speech and Language Days in Urbana-Champaign: (1) label drift in multilingual NLI, (2) ontology-based KG for Alzheimer's research.

Apr 2026

Best Abstract Award — Received Best Abstract Award for graduate students at the 39th Annual Symposium on Arabic Linguistics (ASAL39), which I co-organized at Indiana University.

Mar 2026

New Publication — "Ellipsis in Arabic: Using Machine Learning to Detect and Predict Elided Words" officially published in John Benjamins' Arabic Linguistics.

Feb 2026

RogueTeX Launch — Released RogueTeX, a web-based LaTeX editor with Hugging Face cloud compilation and Supabase cloud storage. Open to the public at roguetex.app.