An Arabic linguistics research body bringing open tools to an international academic audience
- Sector
- Arabic Language
- Work
- Research Software
- Duration
- 8 weeks
28,536 visitors
reached worldwide: we built the platform, launched it, and brought its audience
Five instruments
morphology, articulation, analysis, a bilingual assistant and a publishing surface, on one core
46 of 54
analysis features discovered in the source prose, 8 encoded by hand
Byte-identical
the same analysis re-run six months later, output for output
Fig. 1
Every letter pairing in one grid
All letter pairings in a single grid, ordered by where in the mouth each sound is made, so the patterns can be explored directly.
- 1. Every combination in a single view: nothing to page through, and no filter to set up before you can look.
- 2. Rows and columns both run from the back of the throat to the lips, so neighbours in the picture are neighbours in the mouth.
- 3. The gaps between blocks are the boundaries where one place in the mouth gives way to the next.
- 4. Roughly two cells in five are empty. Those pairings have no attested pattern, and they stay in the picture.
01The starting point
Valuable scholarship, with no practical way to explore it.
The source is a spreadsheet of Arabic prose. Each letter's phonetic description is written as a sentence rather than held as a set of fields, so nothing in it can be counted, grouped or plotted until those sentences are made readable by a machine, which required Arabic prose to be normalised before it could be analysed reliably.
And there is a second requirement that matters more in scholarship than in software: a finding intended for citation must remain the same when another scholar checks it.
Fig. 2
A spreadsheet with no overview
The research lived in a hand-maintained spreadsheet. Its central question, whether where a sound is made predicts a verb’s pattern, needs the whole picture at once.
- 1. A long list can be read one entry at a time, but never taken in at once.
- 2. Each entry belongs at one point in an overall picture that the list could not show.
02What we built
An open platform for Arabic linguistic research.
The whole letter-pair space is one picture. All 676 letter-pair combinations in a single grid, rows and columns ordered by where in the mouth each letter is made, so neighbours in the picture are neighbours in the mouth. Each cell is coloured by the pattern its verbs take, and the boundaries between articulation zones are drawn on. Roughly two cells in five carry no attested pattern, and they are left in the picture rather than dropped: an absence is a result too.
The analysis features behind the colours were found in the source descriptions rather than listed out in advance: 46 of 54 columns are discovered after normalisation, with eight explicitly encoded as derived features. Hundreds of candidate associations are tested, which means some will look real by chance, so the analysis corrects for that and drops them before anything is reported. Where the separation between clusters is modest, the platform says so rather than presenting the groups as obvious.
The part that matters most is the least visible. Run the pipeline again on a clean copy and the outputs are byte-identical: the feature table, the clusters, the summaries and the plots. Run it against the outputs stored six months earlier and they match too, byte for byte. For a research tool that is the whole value: a result a scholar can cite is a result that does not move.
Beside the grid sit a lookup that conjugates a root and breaks it down by radical, and an assistant wired to the same dataset that answers in Arabic or English. The research body updates every word of the public site itself, without a developer.
Fig. 3
Conjugation, broken down letter by letter
A lookup returns a root’s full conjugation, broken down letter by letter, beside an assistant that answers from the same data in Arabic or English.
- 1. Type a root and get its full conjugation.
- 2. Each of its letters is broken down by where and how the sound is made.
- 3. An assistant answers questions in Arabic or English.
- 4. Both read from the same dataset.
03How the platform works
From source prose to reproducible tools.
Fig. 4
Features extracted from written descriptions
Each letter’s phonetic description was written as Arabic prose. Most of the analysis features were extracted from that text, after the Arabic spelling was cleaned up.
- 1. Each letter’s description, as written in the source: a sentence of prose.
- 2. Most of the features, 46 of 54, were found in those sentences rather than listed in advance.
- 3. The other 8 were added by hand.
Fig. 5
Results that reproduce exactly
Run twice on a clean copy, the analysis gives exactly the same output, and it matches results stored six months earlier.
- 1. The outputs from the client’s own run, stored in March 2026.
- 2. Two fresh runs on a clean copy, six months later, match each other exactly, plots included.
- 3. Their features, clusters and summaries match the March outputs exactly, down to the last character.
04What it reached
Built, launched, and carried to its audience.
The same analysis, run again six months later
Byte-identical
Byte-identical, six months on
Letter-pair combinations in a single view
All 676, nothing paged
Analysis features found in the source prose rather than listed by hand
46 of 54
46 of 54 discovered, 8 encoded
Full analysis, cold start to every output
~80 seconds
About 80 seconds, cold start
Built with
ViteReactTypeScriptPythonpandasscikit-learnUMAPTailwind CSS
Have something worth making public?
Bring us the research, material or public tool you are starting with. We will help work out what it needs to reach the people it is meant for.
Keep reading.
Staffing & Recruitment
From Incoming CVs to Client-Safe Candidate Profiles
Manlytic Group’s recruiters work from their phones. CVs reach them on WhatsApp as scans, photos and two-column tables, passport details have to be exact, and anything sent to an employer has to leave the agency as the only route to the candidate.
- Every file made for an employer is built without the candidate’s contact details, then searched again before it can be shared.
- CVs are read without guessing: gaps stay visible, and anything unsure waits for a person to confirm.
- A likely duplicate is reported and held for a recruiter, never merged on its own.
Regulated Services
Building an AI Assistant You Can Control
How a private AI product company turned difficult regulatory material into a controlled assistant for professional firms.
- Your own material, not a public model's: administrators decide what is in the knowledge base and what is not
- Add a new regulation yourself: upload the PDF and the system cleans, splits and indexes it without a developer
- Provision a company, invite named people into it, and cut their access server-side when it ends
Commerce
Making Interior Products Easier to Find
How a MENA AI studio turned an interior photograph into searchable product language, UAE results and a shortlist customers could export.
- Photograph something you saw and get back things you can buy nearby, without typing a word
- Keep what you like, set quantities against it, and export the shortlist as a PDF you can hand to someone else
- Hand it a photograph of something that is not an interior item and it stops, rather than confidently returning the wrong furniture