Gains:
- Know the difference between simplification and transcription and be able to use AI only for limited, verifiable tasks (comprehension support, lexicon hypothesis, simplification draft).
- Ability to verify each output with the original text and reliable sources, avoiding traps such as completing the missing text and straightening the content
- Being able to understand that the human philologist is indispensable in works such as transcription, prosody and scientific edition.
A large part of Turkish literature was written in a language that today's readers cannot read directly: Ottoman Turkish (Ottoman Turkish - Turkish written in the Arabic alphabet, used during the Ottoman Empire, enriched with Arabic and Persian words and phrases). Most of the Divan poetry, history books, newspapers and archive documents are in this language. Making them available to today's readers requires two processes: transcription (translating a text from one alphabet to another, preserving the sounds with special signs; for example, changing the text in Arabic letters to Latin letters) and simplification (replacing the old and heavy words of the text with words that today's readers can understand).
In this unit we will learn to use AI as a helper in these tasks — but here the warning is stronger than ever: In legacy text work, the risk of AI errors and hallucinations is very high; Each output cannot be used without verification by a competent person in the field by comparing it with the original text. AI can misread a word, "complete" a couplet, make up a meaning that doesn't exist. Here AI is an outline and discussion tool; It will never replace the expert philologist.
Why is the risk higher in this area?
For three reasons:
- Ambiguous reading. In Arabic script, short vowels are often not written; The same string of letters can be multiple words. Correct reading requires context and expertise. AI can present the most likely prediction as “sure.”
- Heavy vocabulary. The subtleties of Arabic-Persian phrases, metaphors (stereotyped images in classical poetry) and aruz (measure based on syllable length) in Divan texts are lost or distorted in a superficial simplification.
- Tendency to complete. AI can automatically fill in missing or worn text according to what it "should" be. This is a grave error in philology; is to make up a text that does not exist.
Attention: Telling the AI to "translate this Ottoman couplet" and publishing the output without verifying it is the most dangerous mistake in the field. Transcription and simplification outputs should always be compared with the original text, a reliable dictionary and, if possible, an existing edition (scholarly publication).
How to use AI safely
In older texts, use AI for these limited but useful tasks:
- Comprehension (gist) support: Learning today's equivalent of heavy words of a text in Latin letters as a draft.
- Dictionary hypothesis: Having the possible meanings of a word listed — then verified against a reliable dictionary.
- Simplification draft: Taking a rough draft of a paragraph in current Turkish — then having the expert revise it.
- Comparison: Comparing your simplification with that of the AI and seeing the overlooked areas.
On the other hand, do not let the AI do the following tasks alone: transliteration from Arabic script image; scientific edition; completion of missing text; claim to definitive meaning.
Tip: Ask for two versions of simplification: (1) close simplification that preserves meaning, (2) a list that explains word equivalents. The second list makes it easier for the expert to quickly confirm and makes the AI's concocted responses visible.
three mini cases
Case 1 — The dictionary hypothesis saved time. A graduate student received the draft equivalent of 30 heavy words in a history text in Latin letters from AI. 30 equivalent confirmed from reliable dictionary; 4 were wrong and have been corrected. Dictionary browsing time was significantly shortened, but the verification step was not skipped.
Case 2 — Completion error caught. A researcher asked AI about the missing part of a worn-out couplet. The AI gave the prosody a fitting, fluent, but completely fabricated completion. When the researcher looked at the edition notes, he saw that this chapter was actually different. The "AI fills in the missing text" trap has become concrete.
Case 3 — Simplification distorted meaning. An editor took an AI simplification of a divan couplet; Since AI translated the metaphor (stereotyped image) literally, the original meaning of the couplet was lost. The editor consulted with a field expert and made a simplification that preserved the content.
Four copyable templates
1) Heavy word equivalent (conditional on confirmation):
Give a LIST of today's possible equivalents of the words aggravated in the old Turkish text with Latin letters below. For each word, make a note of "not certain, should be verified from the dictionary". Adding words that are not in the text, completing the text.Text: [paste text here]
2) Two-version simplification:
Simplify the following text in two ways: (1) a close version that preserves the meaning and imagery, (2) a table showing the old/new equivalent for each word you changed. Filling in the missing parts, adding meaning that doesn't exist. Mark where you are not sure.Text: [paste text]
3) Simplification control:
Below is the actual text and my simplification. Mark places where meaning is distorted, omitted, or misrepresented in my version. Don't impose your own version; just point out the problematic areas and link your reasoning to the text. Original: [...] My version: [...]
4) Plot/concept explanation:
Explain the classical images and metaphors in the following couplet. For each: possible meaning, usage in tradition, and note "must be verified from expert source". Do not make a definitive judgment, also state alternative readings.Couplet: [write the couplet]
Weak prompt / Strong prompt
Weak: "Translate this Ottoman text into today's Turkish."
Strong: "Simplify the old Turkish text with Latin letters below; preserve the images and metaphors, do not reduce it to literal meaning. Show each word you have changed in an old/new table. Complete the missing or ambiguous part; mark it as 'uncertain, must be verified by an expert'. Do not add any words or verses that are not in the text."
The powerful prompt closes the completion trap, prevents content loss, and prompts a table that makes verification easier. The weak prompt leaves the door open for the AI to fit and smooth.
Task and security table
business
AI alone
Correct use
Meaning of the word heavy in Latin letter text
risky
Draft + dictionary confirmation
transliteration from Arabic script
not done
expert philologist
Simplification
draft
Expert overhauls
Complete missing text
Absolutely not done
Edition/expert
Explanation
hypothesis
Confirmation from expert source
scientific edition
not done
philologist
Aruz and meter: Why does AI have particular difficulty?
Most of the classical Turkish poetry was written in aruz (a system of measurement based on the length and shortness of syllables, based on patterns). Determining which aruz pattern a couplet is in requires deciphering whether the syllables are long or short one by one, and knowing some subtleties (such as counting one long syllable as two short ones or dropping a vowel). This is a regular but very subtle job, and this is where the AI often goes wrong: it can misname a pattern, mismeasure a syllable, or suggest a pattern that "sounds good" but is wrong.
Therefore, determining the aruz meter is not a job that can be blindly done by AI; At best it gives a "candidate", and this candidate must be verified on the couplet itself by someone with knowledge of the meter. The same applies to the analysis of stops and patterns in syllabic meter (folk poetry meter based on the number of syllables in each line); AI can even make mistakes in counting syllables. Meter is the place where the accuracy of literary analysis is tested: a false claim to meter makes an essay unreliable from the start.
Caution: Do not put a prosodic pattern name given by the AI into a study without verifying it (such as "failatün failatün..."). Measurement is both technical and cultural knowledge, and human expertise is indispensable in this field. Use AI at most as a guide to the question "where should I look for the measure of this couplet, what patterns are possible?"
Common mistakes
- Publishing the transcription/simplification without verifying it. Each output is compared to the original text.
- Having the AI fill in the missing text. This is making up non-existent text; It is definitely not done.
- To reduce the meaning to literal meaning. Classical images must be preserved.
- Thinking that only reading is certain. Multiple readings are common in Arabic script; Ask about alternatives.
- Concluding critical text without consulting an expert. The human philologist is indispensable in this field.
In summary
AI in old texts; It is a useful but high-risk tool for comprehension support, lexical hypothesis, and simplification sketching. Tasks such as transcription, edition and completion of missing texts belong to the expert philologist. Verify each printout with the original text, reliable dictionary, and current editions; Don't fall into the traps of completion and flattening. In this field, human expertise comes before speed.
Application task
Find a short, unsimplified ancient text in Latin letters (a divan couplet or a history paragraph). Get the blueprint from AI with the “two-version simplification” template. Then verify each word in the word table with a reliable dictionary; Tick the wrong ones. Check whether you reduce an image to its literal meaning and correct it if necessary.
checklist
- [ ] I did not have AI do the Arabic script transcription on its own.
- [ ] I did not fill in the missing/ambiguous part.
- [ ] I confirmed the word equivalents from a reliable dictionary.
- [ ] I checked that the images and images are preserved.
- [ ] I consulted the philologist who specializes in the critical text.