ARABIC AI · EVALUATION · RESEARCH

Abdessamad Nafissi

AI Language Engineer
& Arabic NLP Researcher

I help teams improve Arabic AI through linguistic evaluation, high-quality training data, and speech and text quality assurance. My independent, published NLP research explores converting Moroccan Arabizi into Arabic script.

Abdessamad Nafissi
Professional Learning

Certifications & Coursework

  • Microsoft Azure AI Fundamentals Microsoft · 2022
  • Data Science: Transformers for NLPUdemy · 2023
  • Natural Language Processing with PythonUdemy · 2022
  • Linear Algebra · Statistics & ProbabilityUdemy · 2022
About Me

Arabic expertise. Applied AI & research.

Language engineering experience and independent research grounded in Arabic linguistics.

I’m an AI Language Engineer and independent Arabic NLP researcher specializing in language evaluation, speech and text quality, and training data for AI systems. I combine native Arabic expertise with Python-based analysis to identify linguistic errors, improve data consistency, and evaluate model outputs.

My experience includes supporting Meta through TEKsystems and Apple through Appen, working on Arabic training and evaluation data, annotation review and adjudication, linguistic guidelines, and speech and text quality. My expertise spans Modern Standard Arabic and Moroccan Arabic, including morphology, pronunciation, diacritics, code-switching, and contextual meaning.

My published research on Moroccan Arabizi de-romanization combines computational methods with native-speaker audits and detailed error analysis. I use Python, regex, pandas, SQL, PyTorch, and Hugging Face alongside a background in linguistics and translation. My working languages are Arabic, English, and French.

Arabic
Native language expertise
MSA · Moroccan Arabic · dialect variation
3
Working languages
Arabic · English · French
Speech + Text
Language-data evaluation
Annotation · adjudication · quality review
Arabic NLP
Published independent research
Arabizi · benchmarking · human audits
Skills & Practice

Language expertise, applied.

The capabilities I bring to AI language engineering, Arabic AI evaluation, and independent NLP research.

Arabic Linguistics

  • Modern Standard Arabic & Moroccan Arabic
  • North African dialect variation
  • Phonetics & phonology
  • Morphology, diacritics & code-switching

Applied in Arabic language-data evaluation and linguistic error analysis.

Speech & Text Evaluation

  • Speech/audio data & transcription validation
  • LLM output & training-data evaluation
  • Annotation, adjudication & golden datasets
  • ASR/TTS data quality & guideline refinement

Experience with Siri speech/text data and Arabic AI evaluation workflows.

Data & NLP Tools

  • Python, pandas & SQL
  • Regular expressions & data validation
  • PyTorch & Hugging Face
  • Model benchmarking & error analysis

Used in quality workflows and the four-system de-romanization study.

Multilingual Work

  • Arabic: native
  • English: fluent / native-level
  • French: fluent / professional
  • Translation, proofreading & localization

Professional language work across documents, documentaries, software, and games.

Experience & Education

My Professional Journey

From multilingual language work to Arabic AI data and evaluation.

Experience

Oct 2024 – Apr 2026

AI Language Engineer

Meta · through TEKsystems · Remote, FL

Built and evaluated Arabic training and evaluation datasets across MSA and dialects. Created, reviewed, and adjudicated annotations and golden datasets. Used Python, SQL, regex, tools, and APIs for extraction and quality checks, and provided feedback on guidelines, prompts, and evaluation metrics.

Jan 2022 – Aug 2023

Senior Data Linguist

Apple · through Appen · Richardson, TX

Evaluated speech and text language data for Siri command applications. Reviewed model-generated responses for meaning, wording, context, and consistency, applying phonetics, syntax, semantics, and morphology to annotation and quality review.

May 2021 – Dec 2021

Data Linguist

Apple · through Appen · Richardson, TX

Reviewed language and search data for accuracy, usefulness, relevance, and contextual interpretation. Collaborated with team members to resolve ambiguous content and quality issues.

Apr 2012 – May 2021

Senior Translator / Proofreader

Freelance · Remote

Translated and proofread French, English, and Arabic content for clients including VICE/HBO and Sensata Technologies. Work spanned documents, documentaries, localization, and video games, with research into slang, idioms, and specialized terminology.

Past translation work & client reviews on Smartcat

Education

Oct 2021 – Oct 2024

Self-taught NLP Engineering

Online study & applied projects

Studied linear algebra, statistical modeling, NLP, NLU, and Transformers. Developed and applied a custom NLP pipeline for sentiment analysis.

Completed Jun 2013

Master of Arts · Fundamentals of Translation & Linguistics

Translation & Linguistic Studies · Casablanca, Morocco

Advanced study in linguistics, audiovisual translation, digital-content localization, and translation project management, with analysis of morphological, syntactic, semantic, and contextual challenges.

WES · Verified International Academic Qualifications U.S. equivalencies: bachelor’s degree (English Studies) and master’s degree (Translation — Arabic/English/French).
Completed Jun 2011

Bachelor of Arts · English Literature & Applied Linguistics

University Hassan II · Mohammedia, Morocco

Study of phonetics, syntax, semantics, sociolinguistics, and practical applications of linguistic theory.

My Portfolio

Featured Research & Projects

Published Arabic NLP research and practical language-data tooling.

Moroccan Arabizi Research Breaking the
Script Barrier
Rules · MLE · LSTM · mT5 Published research · PDF available
Published researchDe-romanizationArabizi

Breaking the Script Barrier

Automatic De-Romanization of Maghribi Arabizi to Arabic Script in Social Media

An empirical benchmark comparing an ordered rule mapper, MLE word mapping with regex fallback, a word-level character LSTM, and a full-passage mT5-small Transformer on the same 805-passage test set. Native-speaker audits and linguistic error analysis examine reference quality, alignment noise, and dialectal variation.

No single system leads on every metric. mT5 has the highest observed BLEU; MLE has the strongest character- and word-level form fidelity.

Published independent research on Moroccan Arabizi de-romanization. Read the paper above, or explore the public research code and aggregate evidence on GitHub. Dataset access remains subject to upstream conditions.

Four-system benchmark

Frozen 805-passage test set · higher BLEU, chrF, and PosEM are better; lower CER and WER are better.
SystemBLEU ↑chrF ↑CER % ↓WER % ↓PosEM % ↑
Rule mapper7.2446.7324.5072.508.69
MLE + regex35.4469.6015.1742.4816.76
Word LSTM9.6251.9844.9469.009.42
mT5-small36.7568.8642.8664.0321.54

PosEM measures position-wise token agreement. Bold values mark the best observed score in each column. The mT5–MLE BLEU difference is 1.31 points (95% paired-bootstrap interval: −0.55 to 3.24; p = 0.074), which does not establish superiority at the 0.05 level or equivalence.

Dataset, human audits & interpretation

Dataset: 8,045 passage pairs from UBC-NLP / NileChat Arabizi-Morocco, split into 6,436 training, 804 validation, and 805 test pairs. Its Arabizi was generated from Arabic-script Moroccan material with Command R+; performance on naturally authored Arabizi remains untested. Source data is subject to upstream access conditions and CC BY-NC 4.0.

Human audits: two native Moroccan Arabic speakers independently accepted all 300 sampled test references. All-positive labels make Cohen’s κ undefined. A separate 100-pair audit found 75 aligned and 25 misaligned positional training pairs. Review of 50 apparent LSTM errors yielded 42 true errors, 5 acceptable variants/non-errors, and 3 uncertain cases; this error-conditioned sample is not an accuracy estimate.

Interpretation: mT5 uses full passages, external pretraining, and more eligible training data than the word-level systems. This is not an architecture-controlled comparison. All 78 mT5 outputs that failed to emit EOS within the generation limit remain in the primary scores. Split provenance and source-document independence retain limitations described in the manuscript. The exploratory human mT5–MLE review does not establish a human-preference winner.

Python Jupyter Notebook Data Cleaning Social Media NLP

Social Media Text Preprocessing & Sentiment Analysis

An educational Python notebook exploring URL removal, custom emoji-to-text mappings, English-oriented normalization, and TextBlob sentiment scoring. Word clouds and label counts help inspect the processed sample; the project does not claim benchmarked classification accuracy.

Get in Touch

Start a Conversation

Interested in AI language engineering, Arabic NLP research, LLM evaluation, or language-data quality? Let’s connect.

Location

Tampa Bay Area, FL

Connect Internationally

For open-source developments, resume deep-dives, or professional network inquiries, find me on these networks.