[forzagreen]

Adds khatib_scientific_technical: معجم المصطلحات العلمية والفنية والهندسية الجديد (لبنان ناشرون، 2018), the English-Arabic dictionary of Ahmed Shafiq Al-Khatib (Q12244028, the newer "الجديد" edition, not the 6th edition).

Before merging

Stacked on #60 (the Military Dictionary), because the dictionary and term ids follow from it; that is why it is a draft and targets feat/military-dictionary. After #60 is squash-merged, rebase with git rebase --onto main feat/military-dictionary feat/khatib-scientific-technical; the four files then apply as they are. If #60 changes its number of rows, this one has to be rebuilt.

Dumps

Both dumps are regenerated and included.

• SQLite: sqlite3 3.53.4 (the official build); it reproduces the dump committed on main byte for byte from main's database.
• MariaDB: dictionary 833 and its terms were added, in committed batches of 2000, to the MariaDB copy built for feat: add the Military Dictionary of Librairie du Liban (Q138767450) #60 (same rows and columns as migrate_to_mariadb.py). The copy was checked against arabterm.db (row counts per dictionary and checksums of every column), and full-text search returns the new terms.
• The MariaDB dump was written by mariadb-dump 11.8.9 on x86_64 instead of 11.8.5 on aarch64. Compared with feat: add the Military Dictionary of Librairie du Liban (Q138767450) #60's dump, every existing row is unchanged; besides the new rows, only the header comments, the AUTO_INCREMENT values and the autocommit statements around the inserts differ.

What is in it

• 71561 terms (ids 606402-677962), dictionary id 833, extracted from the voice dictionary on mou3jam.com (dictionary 7: 71525 entries and 6827 sub-entries).
• Three entries have no Arabic and are left out: areostyle and DVD (pointers to another headword) and shower pan = shower tray (empty on the site).
• One row per headword. An entry carries its further senses as sub-entries, each with its own part of speech or field; arabic holds the terms of all senses joined with ؛, and ، between synonyms is kept as printed.
• description is the full text of the senses, numbered (1) (2) when there are several, when it says more than arabic (15847 terms).
• The 39 sub-entries that have a keyword of their own are rows of their own: abandon a claim, or a form of the parent such as medium (pl. media), ward (with off).
• Headwords are kept as printed, including X = Y synonyms and inverted forms (pool, oil); medical_terms_* already has both.
• Superscripts and subscripts are written with Unicode characters (10⁻¹⁹); پ, ڤ and ڠ are kept as in the source.

Repairs of the source

The digitised text is damaged in places. Every repair is keyed by entry id in the assemble script; these are worth a look:

• 44 definitions whose first letter is missing or that start with a stray vowel mark (ُعاصِر for مُعاصِر under contemporary, َطَل for بَطَل under champion). The letter is supplied from the English headword; these are readings, not source text.
• 91 entries, 88 of them under B, where the last printed line of the first sense was stored as the headword's own definition (bastard: شاذّ، غيرُ in one place and قِياسيّ. غيرُ مُتناسِق in the other). The two parts are rejoined only where the break is evident (an open parenthesis, a trailing colon or preposition, an explanation in progress); otherwise they stay two senses of the same row.
• 6 mangled headwords: 0 for center, arءte for arête, Debye-Hدckel, vis-ب-vis, wدstite, Celsiusor.
• About 20 single typos: two definitions stored with their lines in reverse order (deep, demand), bent and 90° bent swapped, broken <sup> markup, a ? where the site lost λ.

Known limits

• The longest entries (set, strike, run) keep in arabic only the terms that fit in 255 characters; the description has all of them.
• Part of speech and the field label (Chem., Elec. Eng., two thirds of the entries) are not mapped to any column.
• Sense boundaries inside a multi-sense entry are sometimes wrong in the source (a line of one sense attached to its neighbour); since all senses share one row, the terms are still there.

To check

• tier is set to 2 and dict_type to terminology; change the tier if you rank it differently.
• name_english and name_french follow the Wikidata labels, without their "Arabic-English" (the dictionary is English-Arabic) and with the publisher written as in feat: add the Military Dictionary of Librairie du Liban (Q138767450) #60.

make validate passes locally on the new database.