Goobo Labs
Research
OverviewPublicationsBenchmarks (SomBench)Open SourceRoadmap
DatasetsModels
Learn
ProgramsLearning PathsCurriculum
BlogCommunity
Sign inExplore research
Goobo Labs

An open research lab building the foundations of Somali AI.

Research

  • Overview
  • Publications
  • Benchmarks
  • Roadmap

Datasets

  • SomNLP Corpus
  • SomNLP STT
  • Wikipedia Corpus
  • Citation Guide

Programs

  • Data Science & ML Bootcamp
  • Python for Everyone
  • NLP (coming soon)
  • Curriculum

Resources

  • Docs
  • Models & Tools
  • Open Source
  • Blog
  • Soplang

Contact

  • info@goobolabs.so
  • Mogadishu, Somalia

Company

  • About
  • Careers
  • Impact
  • Press Kit
  • Verify Certificate
  • Contact

Community

  • GitHub
  • Hugging Face
  • X
  • LinkedIn

© 2026 Goobo Labs. Open research for Somali AI.

Privacy PolicyTerms of Service
HomeBlogResearch
ResearchJul 9, 2026 · 6 min read

Introducing the Somali Language Standard (SLS) – Phase 1 Complete

We've formally published the first two standards of the Somali Language Standard (SLS) — an open, machine-readable, CI/CD-validated framework for the Somali language, starting with the alphabet and the standards process itself.

Sharafdin Yusuf

Sharafdin Yusuf

Lead Engineer & Researcher

At Goobo Labs, we've spent years building the infrastructure required to bring the Somali language into the modern era of AI. Through our work on SomNLP Corpus v2, the Soplang programming language, and our morphological tokenizers, we've encountered one recurring, fundamental bottleneck: inconsistency.

In the world of NLP, inconsistency is the enemy of performance. Somali language datasets are riddled with conflicting spelling conventions, varying loanword adaptations, and erratic punctuation usage. Without a centralized, machine-readable source of truth for Somali orthography and grammar, every tech company, AI researcher, and linguist has had to guess — or invent their own rules.

What is the Somali Language Standard (SLS)?

The SLS is designed to act as the W3C or IETF for the Somali language. It is a strictly governed, open-source framework that centralizes the rules of the language — from the basic alphabet to complex grammar and machine-learning terminology.

Crucially, SLS is not a static PDF. It is a machine-readable, CI/CD-validated repository. Every standard is tracked via a JSON registry, follows a strict lifecycle (Draft → Proposed → Review → Candidate → Stable), and traces normative requirements (MUST, SHOULD) to compliance checklists. This allows developers to programmatically verify whether their datasets and tokenizers are "SLS-Compliant."

Completing Phase 1: The Bedrock

You cannot build a dictionary or an LLM instruction dataset if your systems cannot agree on what constitutes a "letter." Phase 1 was entirely focused on building the constitutional framework and locking in the alphabet. We have officially published our first two standards into the Proposed state.

2

Standards published

26

Alphabet characters

3

Digraphs defined

DAG

Dependency model

  • SLS-0000: The Standards Process — the meta-standard, analogous to IETF's RFC 2026. Defines how all future standards are numbered, structured, versioned, and governed.
  • SLS-0001: The Somali Alphabet Standard — formally defines the Somali Latin alphabet (Far Soomaali): 21 consonants, 5 vowels, collation rules for digraphs (dh, sh, kh), and vowel-length notation via doubling rather than diacritics.

The Big Technical Win: Solving Glottal Stop Tokenization

If you've ever tried to train a tokenizer or an LLM on Somali text, you are intimately familiar with the glottal stop problem. Historically, the glottal stop has been written using the ASCII apostrophe (U+0027). Standard NLP tokenizers like WordPiece or BPE classify it as punctuation (Po). When they encounter a word like su'aal (question), the tokenizer aggressively splits it into three tokens, shattering the semantic integrity of the word.

python
from transformers import AutoTokenizer

tok = AutoTokenizer.from_pretrained("xlm-roberta-base")

# ❌ ASCII apostrophe — classified as punctuation (Po)
tokens_before = tok.tokenize("su'aal")
print(tokens_before)  # ['su', "'", 'aal']  — 3 tokens, word is shattered

# ✅ SLS-0001: U+02BC MODIFIER LETTER APOSTROPHE (Lm)
tokens_after = tok.tokenize("suʼaal")
print(tokens_after)   # ['suʼaal']           — 1 token, semantic unit preserved

In SLS-0001, we made the normative ruling to standardize the glottal stop as U+02BC MODIFIER LETTER APOSTROPHE (ʼ). Because U+02BC is classified in Unicode as a letter (Lm), standard tokenizers will now treat suʼaal as a single, contiguous string. This seemingly minor typographical ruling will dramatically improve Somali tokenization efficiency and downstream LLM performance.

Looking Ahead: Phase 2

With Phase 1 complete and currently in its public comment period, we are immediately pivoting to Phase 2: The Orthography Standard. By building from the alphabet upwards, we are creating a mathematically sound, dependency-checked foundation for the future of the Somali language.

  • SLS-0002 (Spelling Rules): Standardizing gemination, assimilation, and vowel harmony.
  • SLS-0004 (Punctuation): Aligning punctuation rules with the newly locked alphabet.
  • SLS-0005 (Capitalization): Defining exact case-pairing behaviors for digraphs.

We invite linguists, developers, and researchers to review the proposed standards and participate in the public comment period on our GitHub repository. Let's build the future of Somali NLP together.

NLPSomaliStandardsTokenizationOpen Source
Sharafdin Yusuf

Sharafdin Yusuf

Lead Engineer & Researcher

Writing about Somali language technology, open data, and AI from the lab in Mogadishu.

All postsGet in touch

On this page

  • What is the SLS?
  • Completing Phase 1
  • The glottal stop fix
  • Looking ahead: Phase 2

Share

Back to Blog
More from the lab

Related posts

Research6 min read

Soplang v2.0.0: a compiler for the first Somali programming language

Our team shipped a Rust-based compiler for Soplang — Cranelift JIT and ahead-of-time builds, with syntax written entirely in Somali keywords.

Jun 30, 2026Read
Research7 min read

Why Somali needs its own tokenizer

Foreign multilingual tokenizers over-segment Somali — BERT-base spends 2.69 tokens per word. Our native BPE tokenizer, trained on 529M words of SomNLP-Corpus, gets that down to 1.53: a 1.75× improvement.

Nov 11, 2025Read
Dataset5 min read

SomNLP-Corpus v2: 887M+ tokens from six open sources

Our largest Somali text release yet: 1.77M clean documents, 591M words, and ~887M tokens from HPLT, CC100, mC4, OPUS, MADLAD, and MT560 — filtered through a six-stage pipeline.

Jun 9, 2026Read
Newsletter

Stay in the loop

Dataset releases, new bootcamp cohorts, research papers, and lab updates — straight to your inbox. No spam, unsubscribe anytime.