Home / Services / Practice 01
Arabic language processing
Systems that read Modern Standard Arabic accurately and Gulf dialect natively — deployable inside your own network, with no text leaving the country.
The problem
Your text is not the Arabic these tools were trained on
Commercial NLP is trained overwhelmingly on newswire Modern Standard Arabic. Real institutional text is not that. Citizen feedback, social posts, support tickets and internal correspondence are written in dialect, with inconsistent spelling, mixed script, and grammatical particles that do not exist in MSA at all.
The failures are not subtle. Below is a single Kuwaiti sentence run through a standard MSA pipeline and through ours.
| Token | Standard MSA pipeline | Our analyser |
|---|---|---|
| الخدمات | noun, plural ✓ | noun, plural ✓ |
| صارت | verb, past ✓ | verb, past ✓ |
| وايد | unknown token — dropped | intensifier adverb, Kuwaiti — carries the sentiment |
| هالسنة | unknown token — dropped | demonstrative + noun, Kuwaiti — resolves the time reference |
Two dropped tokens out of six. Both of them the ones that told you how the citizen actually felt and when they meant. Multiply that across a hundred thousand comments and the report you brief your minister on is measuring something other than public opinion.
Capabilities
What we build
Morphological analysisالتحليل الصرفي
Root, pattern, part of speech and diacritisation for every token, including dialectal forms. This is the layer everything else sits on, and the reason our downstream accuracy holds up on real text.
Sentiment analysisتحليل المزاج العام
Polarity and intensity scored against Kuwaiti and wider Gulf usage rather than newswire Arabic, so the posts carrying the strongest opinion are not returned as neutral. Automatic dialect labelling lets you segment an audience by how they write, not only by what they say. This is the layer behind our social media analytics, where it scores public conversation at national scale.
Topic & entity extractionاستخراج الموضوعات والكيانات
The dominant subjects across a body of text over a given period, clustered by theme rather than by hashtag or keyword — and the names of people, entities, laws and places, normalised against your own reference lists so the same ministry is not counted under four spellings.
Retrieval for Arabic archivesالبحث الدلالي
Semantic search and question answering over your own documents, wired to an LLM of your choosing — including models that run entirely inside your data centre.
How we work
Three stages, and you can stop after the first
Sample assessment
You send a representative extract of your text. We run it through our pipeline and return a written assessment: what is extractable, what accuracy to expect, what would need building, and whether an off-the-shelf tool would in fact serve you better.
Pilot on live data
A working system on a bounded slice of your operation — one department, one campaign, one archive — with measured accuracy against a human-annotated benchmark you can audit.
Deployment & handover
Production install inside your environment, integration with your existing dashboards, documentation in Arabic and English, and training for the team that will own it after we leave.
Proof
Where this has already run
Extending a global text-mining platform
We worked with SAS R&D on the Arabic processing inside SAS Text Miner. When a vendor of that size brings in an outside team for a language, it is a reasonable signal.
Read more →National AssemblyDialect-aware public opinion analysis
Our Social Intelligence Analyzer ran as an add-on to Pulsar for parliamentary and broadcast use, and was written up as a published Pulsar case study.
Read more →Kuwait TVAnalytics behind the news bulletin
Outputs from our systems were used to generate trend segments broadcast on national television news.
Read more →Start here
Tell us what your text is doing wrong.
Send us a sample — a set of citizen comments, a document archive, a support inbox — and we will come back with a written read on what is achievable, what it would take, and whether you need us at all.
Send a sample of your text with the form. We reply in writing.