Teaching machines to notice a metaphor in Arabic.
AraMIP is a two-stage annotation procedure that extends the Metaphor Identification Procedure VU (MIPVU) with the classical Arabic rhetorical tradition — distinguishing isti‘āra, kināya, and tashbīh at both the word and the construction level.
كان الرجل مفاجأة لا تنتهي
“The man was a surprise that never ends.”
Metaphors are everywhere in language — “time is money,” a leader who is a “lion” — and computers are increasingly good at spotting them in English. Arabic has been left behind, partly because it has its own, much older toolkit for talking about figurative language, developed over a thousand years by Arabic rhetoric scholars. AraMIP is a step toward closing that gap: it's a rulebook that tells human annotators — and, eventually, AI systems — how to consistently mark which words in an Arabic sentence are being used figuratively, and which of three classical categories (metaphor, an indirect “implied meaning,” or a simile) that figurative language belongs to. The team tested the rulebook on 300 real sentences and used what went wrong to write a clear list of fixes for the next version.
Two stages: first find it, then classify it.
AraMIP keeps full compatibility with MIPVU while adding a second layer grounded in ʿilm al-bayān, the Arabic science of eloquence.
Lexical‑unit MIPVU annotation
Every word is checked against its “basic” dictionary meaning and coded for how it's being used in context.
Rhetorical span classification
Metaphor‑bearing words are grouped into full rhetorical constructions and rated for how conventional or novel they are.
The literal, “basic” meaning is impossible in context — only the figurative reading survives.
Both the literal and figurative meanings stay plausible; the figurative one is simply implied.
An explicit comparison, usually signalled by a lexical marker such as “like” or “as.”
AraMIP: Extending MIPVU Towards Metaphor Identification in Arabic
Marathe, Mandar, Ali, Manar, Nabhani, Sara, Abu Ahmad, Raia, Baroud, Ibrahim, and Hassan, Omar Abdelmomen Amin. Accepted. “AraMIP: Extending MIPVU Towards Metaphor Identification in Arabic.” In Proceedings of the Fourth Arabic Natural Language Processing Conference. Budapest, Hungary: Association for Computational Linguistics.
@inproceedings{aramip2026,
title = {AraMIP: Extending {MIPVU} Towards Metaphor Identification in {Arabic}},
author = {Marathe, Mandar and Ali, Manar and Nabhani, Sara and Abu Ahmad, Raia and Baroud, Ibrahim and Hassan, Omar Abdelmomen Amin},
booktitle = {Proceedings of the Fourth Arabic Natural Language Processing Conference (ArabicNLP 2026)},
year = {2026},
address = {Budapest, Hungary},
publisher = {Association for Computational Linguistics},
note = {Accepted}
}
ArabicNLP 2026
28–29 October 2026 · Budapest, Hungary · Fourth Arabic Natural Language Processing Conference, an official ACL workshop.
Presentation slides
Available after the conference.
Session handout
Available after the conference.
Talk recording
Available after the conference.