How we build Bayan

Old scans of the Official Gazette

Reading Arabic a machine struggles to see.

[date] · [author] · 5 min read

In Egypt, the law is what was published in the Official Gazette, or in the Egyptian Gazette supplement, al Waqai al Misriya. Many of the texts lawyers work with today were published decades ago: the Real Estate Registration Law appeared in al Waqai al Misriya on 24 August 1946, and the Civil Code in 1948. What we have of those issues are scans of old paper, and reading them by machine is harder than it looks.

What makes a page hard

  • Dots. Many Arabic letters differ only by their dots, such as ب ت ث ن ي, and dots are the first thing to fade in old print.
  • Joined letters. Arabic is written joined up, so a computer cannot see where one letter ends as easily as it can with separate Latin letters.
  • Old typefaces and columns: pages in two or three columns, tables, margins and stamps.
  • Numbers: Arabic Indic digits in faded print, fractions and dates.
  • The issue numbers themselves. «العدد 31 مكرر», issue 31 bis, and «العدد 9 مكرر (أ)», issue 9 bis (a), are separate issues published alongside the regular ones. Law 164 of 2025 was published in issue 31 bis, and Law 9 of 2022 in issue 9 bis (a).

The errors that matter

A reading error in a legal text is not a typo. A dropped line can drop a «لا», a "not", and reverse the rule. A missing dot can turn one word into another. A misread digit can change a period, an amount or an article number.

Why a general model gets this wrong

Modern text recognition systems do not read letters alone. They use a language model to guess an unclear letter from its context. Most of the time that improves the reading. But it carries the same risk we describe in every article: the model picks the most likely option. If a rare legal word has faded, it puts a common word that looks similar and fits the context in its place, and the sentence comes out grammatical and different in meaning. A text that reads smoothly gives no sign that it was misread. This is fluency before accuracy, at the reading stage rather than the answering stage.

How Bayan reads old texts

[Sherif confirms before publication.]

  • The original digital text is preferred wherever it exists; scans are used only where nothing else exists.
  • Legislation read by machine is checked against the image of the original page, and corrected by hand where needed.
  • Corrected places are marked, and every text stays linked to the image of the page it came from, so you can see the original yourself.
  • Publication details are kept for every text: Official Gazette or al Waqai al Misriya, the issue number as issued, bis included, and its date.
Share

The month's key legal updates and what's new in Bayan, once a month.

Ask Bayan your first question.

Or book a team demo and bring a real contract: thirty minutes is enough to know whether Bayan fits your practice.

Book a team demo