Figurative and Cultural Knowledge in LLMs: Investigating Cross-Domain Transfer through Fine-Tuning

Mena Attia; Mona Diab; Thamar Solorio

cs.CLAug 18, 2026

Figurative and Cultural Knowledge in LLMs: Investigating Cross-Domain Transfer through Fine-Tuning

Authors: Mena Attia, Mona Diab, Thamar Solorio

Organizations: MBZUAI · Carnegie Mellon University

Abstract

Figurative language is deeply culturally embedded; fluent use requires not just linguistic competence but cultural immersion. We ask whether LLMs can learn this link: does fine-tuning on cultural data improve figurative language understanding, and vice versa? We conduct a systematic study across four models (ALLaM-7B, Fanar-1-9B, Qwen3-8B, Llama-3.1-8B) and six Arabic datasets spanning cultural commonsense, proverbs, and poetry across diverse dialects and regions. Fine-tuning on poetry improves idiom comprehension (+2.33%, p<0.05), a gain our ArabicMMLU control does not reproduce, indicating that it stems from figurative content rather than Arabic language adaptation and pointing to a sensitivity to non-literal meaning that transfers across figurative types. Cultural fine-tuning, by contrast, lowers proverb-interpretation accuracy in both Arabic-centric models. Transfer between the two domains is otherwise indistinguishable from noise, with Arabic models frequently regressing after fine-tuning, suggesting prior saturation of relevant knowledge, while multilingual models show greater adaptation headroom. Error analysis further reveals that fine-tuning reinforces experiential cultural knowledge while destabilizing historically grounded factual knowledge. Our findings suggest that the relationship between culture and figurative language, though conceptually natural, is not straightforwardly captured through fine-tuning alone.

Figures & tables

Figure 1: Scene from the Egyptian TV show Nelly and Sherihan.

Dataset	Description	Size	Coverage	Usage
AraDiCE-Culture mousi2024aradicebenchmarksdialectalcultural	Multiple-choice cultural commonsense reasoning benchmark	180	6 Arab countries	Test
ArabCulture sadallah2025commonsensereasoningarabculture	Cultural knowledge and practices across the Arab world	3,482	13 Arab countries	Train
Palm alwajih-etal-2025-palm	Cultural commonsense and social reasoning in Arabic	15.5k (train), 1.93k (test)	22 Arab countries	Train
FannOrFlop alghallabi2025fannflopmultigenremultiera	Poem–explanation pairs capturing poetic preference and aesthetic judgment across 14 genres, 12 eras	6,984	Arabic poetry	Train
Jawaher magdy-etal-2025-jawaher	Arabic proverb understanding and interpretation	800 (train), 198 (test)	20 Arabic varieties	Train, test
Kinayat attia-etal-2026-beyond	Egyptian Arabic idiom–explanation pairs	150	Egyptian Arabic	Test

Table 1: Datasets used for training and evaluation, covering figurative language and cultural knowledge across Arabic varieties and regions. Cultural datasets (AraDiCE-Culture, ArabCulture, Palm) and figurative language datasets (FannOrFlop, Jawaher, Kinayat) are visually distinguished; ArabicMMLU serves as a control.

Figure 2: Average performance difference.

	Model	Jawaher	Kinayat	AraDiCE
Base	ALLaM-7B-Instruct	0.8990	0.8400	0.7537
	Qwen3-8B	0.8131	0.6800	0.5037
	Fanar-1-9B-Instruct	0.8906	0.7667	0.6870
	Llama-3.1-8B-Instruct	0.6886	0.5956	0.5204
Fine-tuned on Palm Subset	ALLaM-7B-Instruct	0.8620	0.8333	0.7167
	Qwen3-8B	0.8300	0.7178	0.5204

Table 2: Average evaluation results across three runs on Jawaher, Kinayat, and AraDiCE datasets. Results show accuracy scores (

\uparrow

) for base models and models fine-tuned on different subsets.

Model	Train Set	Avg $\Delta$ %	Impr	Regr	Net
ALLaM	Jawaher	-1.30	21	28	-7
ALLaM	FannOrFlop	-1.87	15	25	-10
Fanar	Jawaher	+0.37	22	20	+2
Fanar	FannOrFlop	+0.73	23	19	+4
Llama	Jawaher	-0.77	38	42	-4
Llama	FannOrFlop	+0.37	43	41	+2

Table 3: Aggregate performance changes on culture by model and fine-tuning dataset.

	Impr	Regr	Net
By Country
Jordan	50	18	$+32$
Lebanon	30	29	$+1$
Qatar	26	28	$-2$
Palestine	30	35	$-5$
Egypt	22	34	$-12$

Table 4: Performance changes on cultural evaluation by country and topic category, aggregated across all models and fine-tuning configurations.

Task	Impr	Regr	Net	Avg $\Delta\%$
Cultural Fine-tuning
Idioms	195	159	$+36$	$+1.00$
Proverbs	149	162	$-13$	$-0.27$
Poetry Fine-tuning
Idioms	100	58	$+42$	$+2.33$
Proverbs	97	73	$+24$	$+1.01$

Table 5: Overall fine-tuning effect on idiom and proverb interpretation, aggregated across all models. Cultural fine-tuning uses ArabCulture and Palm; Poetry fine-tuning uses FannOrFlop.

		Idioms		Proverbs
Model	Train Set	Avg $\Delta$ %	Net	Avg $\Delta$ %	Net
ALLaM	ArabCulture	$-1.33$	$-6$	$-0.84$	$-5$
	Palm	$-0.67$	$-3$	$-3.70$	$-22$
	Poetry	$-1.11$	$-5$	$-1.01$	$-6$
Fanar	ArabCulture	$-1.33$	$-6$	$-1.01$	$-6$
	Palm	$-2.00$	$-9$	$-3.03$	$-18$

Table 6: Aggregate fine-tuning results by model and fine-tuning dataset for both idiom and proverb interpretation. Cultural fine-tuning datasets (ArabCulture, Palm) and Poetry fine-tuning (FannOrFlop) are visually distinguished. Net refers to the total net improved predictions across all three seeds.

Appendix figures & tables33 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 3 — Figure 3: Prompt used for the MCQ understanding task.

Panel of Figure 4 — Figure 4: Prompt template used for AraDiCE-Culture evaluation.

Panel of Figure 5 — Figure 5: Prompt used for fine-tuning on FannOrFlop poetry explanation.

Panel of Figure 6 — Figure 6: Prompt used for fine-tuning on the Jawaher proverb explanation task.

Panel of Figure 7 — Figure 7: Prompt used for fine-tuning on ArabCulture open-ended completion.

Panel of Figure 8 — Figure 8: Prompt used for fine-tuning on the Palm instruction-following task. No task-specific preamble is used; the instruction field carries full context.

		Run 1 (seed=0)			Run 2 (seed=42)			Run 3 (seed=21)
	Model	Jawaher	Kinayat	AraDiCE	Jawaher	Kinayat	AraDiCE	Jawaher	Kinayat	AraDiCE
Base	ALLaM-7B-Instruct	0.8889	0.8600	0.7667	0.9091	0.8267	0.7667	0.8990	0.8333	0.7278
	Qwen3-8B	0.8030	0.6733	0.5278	0.8333	0.6867	0.5056	0.8030	0.6800	0.4778
	Fanar-1-9B-Instruct	0.8788	0.7733	0.7222	0.9040	0.7267	0.7000	0.8889	0.8000	0.6389
	Llama-3.1-8B-Instruct	0.7020	0.5933	0.5056	0.6919	0.5933	0.5167	0.6717	0.6000	0.5389
Fine-tuned on Palm Subset	ALLaM-7B-Instruct	0.8687	0.8533	0.7333	0.8737	0.8133	0.7111	0.8434	0.8333	0.7056

Appendix

Table 7: Zero-shot evaluation results across three runs with different random seeds on Jawaher, Kinayat, and AraDiCE datasets. Results show accuracy scores (

\uparrow

) for base models and models fine-tuned on different subsets.

Figure 9: Performance difference on different datasets for models fine-tuned on the ArabCulture dataset (diff = fine-tuned accuracy - base accuracy).

Figure 10: Performance difference on different datasets for models fine-tuned on the FannOrFlop dataset (diff = fine-tuned accuracy - base accuracy).

Figure 11: Performance difference on different datasets for models fine-tuned on the Jawaher dataset (diff = fine-tuned accuracy - base accuracy).

Figure 12: Performance difference on different datasets for models fine-tuned on the Palm dataset (diff = fine-tuned accuracy - base accuracy).

Figure 13: Performance difference on different datasets for models fine-tuned on the ArabicMMLU baseline dataset (diff = fine-tuned accuracy - base accuracy).

Figure 14: Average performance difference between models fine-tuned on the full FannOrFlop dataset vs. FannOrFlop subset.

Figure 15: Average performance difference between models fine-tuned on the full ArabCulture dataset vs. ArabCulture subset.

		Run 1 (seed=0)			Run 2 (seed=42)			Run 3 (seed=21)
	Model	Jawaher	Kinayat	AraDiCE	Jawaher	Kinayat	AraDiCE	Jawaher	Kinayat	AraDiCE
Fine-tuned on FannOrFlop	ALLaM-7B-Instruct	0.8990	0.8600	0.7889	0.9343	0.8333	0.7167	0.8990	0.8467	0.7222
	Qwen3-8B	0.8333	0.6867	0.5556	0.8586	0.7333	0.4944	0.8586	0.7200	0.5278
	Fanar-1-9B-Instruct	0.8687	0.7533	0.7167	0.8636	0.7267	0.6889	0.8586	0.7867	0.6278
	Llama-3.1-8B-Instruct	0.7424	0.5733	0.5167	0.6919	0.5467	0.4889	0.6970	0.6200	0.5556
Fine-tuned on ArabCulture	ALLaM-7B-Instruct	0.8990	0.8467	0.7722	0.9040	0.8333	0.7444	0.8838	0.8333	0.7389

Appendix

Table 8: Zero-shot evaluation results across three runs with different random seeds on Jawaher, Kinayat, and AraDiCE datasets. Results show accuracy scores (

\uparrow

) for models finetuned on FannOrFlop and ArabCulture full datasets.

Configuration	Subset	Seed	Jawaher	Kinayat	AraDiCE
Default	Palm	0	0.8687	0.8533	0.7333
		42	0.8737	0.8133	0.7111
		21	0.8434	0.8333	0.7056
	FannOrFlop	0	0.8838	0.8667	0.7778
		42	0.9040	0.8133	0.7167
		21	0.8788	0.8067	0.7111

Appendix

Table 9: Zero-shot evaluation results for ALLaM-7B-Instruct on Jawaher, Kinayat, and AraDiCE datasets. Results show accuracy scores (

\uparrow

) comparing the default LoRA configuration against two ablation settings.

Configuration	Dataset	Jawaher	$\Delta$ Jawaher	Kinayat	$\Delta$ Kinayat	AraDiCE	$\Delta$ AraDiCE
r=4, $\alpha$ =8, lr=5e-5	Palm Subset	0.8620	—	0.8333	—	0.7167	—
r=4, $\alpha$ =8, lr=5e-5	FannOrFlop	0.8889	—	0.8289	—	0.7352	—
r=16, $\alpha$ =32, lr=1e-5	Palm Subset	0.8737	+0.0118	0.8222	$-$ 0.0111	0.7370	+0.0204
r=16, $\alpha$ =32, lr=5e-4	Palm Subset	0.8182	$-$ 0.0438	0.8222	$-$ 0.0111	0.7426	+0.0259
r=16, $\alpha$ =32, lr=5e-4	FannOrFlop	0.8788	$-$ 0.0101	0.7756	$-$ 0.0533	0.6870	$-$ 0.0481

Appendix

Table 10: Average zero-shot evaluation results for ALLaM-7B-Instruct across three runs on Jawaher, Kinayat, and AraDiCE datasets.

\Delta

values indicate the difference relative to the default configuration for the corresponding finetuning dataset. Results show accuracy scores (

\uparrow

Task	Finetuned on	$\Delta$	95% CI	$p$	ICC	Sig.
AraDiCE	Jawaher	$-0.19$	$[-1.76,+1.39]$	0.8101	$\phantom{-}0.018$
AraDiCE	Poetry	$+0.00$	$[-1.48,+1.48]$	0.9940	$-0.006$
Kinayat	ArabCulture	$+0.78$	$[-1.22,+2.72]$	0.4353	$-0.011$
Kinayat	Palm	$+1.22$	$[-1.00,+3.44]$	0.2881	$\phantom{-}0.027$
Kinayat	Poetry	$+2.33$	$[+0.39,+4.28]$	0.0207	$\phantom{-}0.009$	$*$
Jawaher	ArabCulture	$-0.08$	$[-1.56,+1.39]$	0.9195	$\phantom{-}0.078$

Appendix

Table 11: Clustered aggregate effects, pooled across all four models (items resampled jointly). ICC is the intraclass correlation coefficient.

\Delta

and CI bounds in percentage points.

*

denotes

p<0.05

Model	Finetuned on	Task	Seed	$\Delta$	95% CI	$p$
ALLaM	Poetry	AraDiCE	42	$-5.00$	$[-8.89,-1.11]$	0.0225
Qwen	Palm	Idioms	42	$+4.67$	$[+1.33,+8.00]$	0.0156
Llama	Poetry	Idioms	21	$+10.00$	$[+3.33,+16.67]$	0.0081
ALLaM	Palm	Proverbs	21	$-5.56$	$[-9.60,-1.52]$	0.0127
Fanar	Palm	Proverbs	42	$-3.54$	$[-6.57,-1.01]$	0.0391

Appendix

Table 12: Per-run results significant by exact McNemar’s test (

p<0.05

Count	Country	Question
Top 10 Most Frequently Improved
9	Jordan	؟ندرألاب ناضمر راهنب راطفإلاب رهجلا ةبوقع وش
7	Jordan	؟ندرألاب جاوزلا ديلاقتو تاداع يه وش
6	Syria	؟ايروس لهأ اهيف لفتحيب يللا دايعألا يه وش
6	Jordan	؟ندرألاب لافطألل ةيبعشلا باعلألا يه وش
6	Jordan	؟ندرألاب ةروهشم راجشأو تاتابن رتكأ وش

Appendix

Table 13: Top 10 most frequently improved and regressed questions from the AraDiCE dataset, with counts across all models fine-tuned on figurative language (proverbs and poetry).

Idiom (Arabic)	Impr	Regr
ْهشِو ْلَكَأ	9	3
ْهَدْلِجْلا ىَلَع	8	4
نِرِي ْهاَّلَخ	2	7
ْهَضاَخَمْلا ِّسَج	2	5
ْبلْقِتِو ْبَرْضِتِب اَيْنُّدلا	4	5
بياس هفك	6	2

Appendix

Table 14: Representative subset of the 21 unstable idioms (improved in some seeds, regressed in others), with counts across all fine-tuned models.

Proverb (Arabic)	Dialect	Impr	Regr
لمج عسوي بابلا	Kuwaiti	1	8
اهنب نم هسفن لماع	Egyptian	2	7
اودعا بيجت ام نطبلا	Libyan	1	6
صوردلا وارعني ام ژاژغتلا تعاس	Mauritanian	5	3
مادك نم ديأو هرو نم ديأ عجر	Iraqi	4	2
اشاب نايمعلا ىلع روعألا	Omani	1	2

Appendix

Table 15: Representative subset of the 15 unstable proverbs (improved in some seeds, regressed in others).

Idiom (Arabic)	Impr	Regr
ْهشِو ْلَكَأ	3	3
نِرِي ْهاَّلَخ	1	4
ْهَدْلِجْلا ىَلَع	4	2
ْبلْقِتِو ْبَرْضِتِب اَيْنُّدلا	4	1
ْهَضاَخَمْلا ِّسَج	1	2
يِدْنه ْرْمَت ْنَبَل ْكَمَس	1	2

Appendix

Table 16: Representative subset of the 15 unstable idioms (improved in some seeds, regressed in others) under poetry fine-tuning, with counts across all models.

Proverb (Arabic)	Dialect	Impr	Regr
حازم ةودعلا	Algerian	3	2
اهنب نم هسفن لماع	Egyptian	1	3
لبج لوقت تناو لمج لوقا انا	Omani	1	1
نيَعلا يحَتسِت مَفلا مَعطِا	Palestinian	2	1
يراوشلا زنخت ةدحو ةتوح	Moroccan	2	1
مادك نم ديأو هرو نم ديأ عجر	Iraqi	2	1

Appendix

Table 17: The 7 unstable proverbs (improved in some seeds, regressed in others) under poetry fine-tuning.

Dialect	Impr	Regr	Net
Mauritanian	30	10	$+20$
Yemeni	15	1	$+14$
Iraqi	7	2	$+5$
Tunisian	7	2	$+5$
Jordanian	11	7	$+4$
Kuwaiti	12	8	$+4$

Appendix

Table 18: Net fine-tuning effect by Arabic dialect variety on the proverb task.

Model / Dataset	Base%	FT%	$\Delta\%$	95% CI	$p$	Impr	Regr
ALLaM
ArabCulture (seed 0)	86.0	84.7	$-1.33$	$[-5.33,+2.67]$	0.754	4	6
ArabCulture (seed 21)	83.3	82.0	$-1.33$	$[-6.00,+3.33]$	0.774	5	7
ArabCulture (seed 42)	82.7	81.3	$-1.33$	$[-5.33,+2.67]$	0.754	4	6
Palm (seed 0)	86.0	85.3	$-0.67$	$[-5.33,+4.00]$	1.000	6	7
Palm (seed 21)	83.3	83.3	$\pm 0.00$	$[-4.67,+4.67]$	1.000	6	6

Appendix

Table 19: Improvement and regression breakdown on Kinayat idioms across cultural fine-tuning (ArabCulture and Palm) and poetry fine-tuning (FannOrFlop). Base% and FT% are accuracy before and after fine-tuning;

\Delta\%

is the percentage-point change, reported with 95% paired-bootstrap confidence intervals over test items and exact McNemar

p

-values; Impr and Regr are the number of individual predictions improved or worsened. ∗ marks

p<0.05

Model / Dataset	Base%	FT%	$\Delta\%$	95% CI	$p$	Impr	Regr
ALLaM
ArabCulture (seed 0)	88.9	88.9	$\pm 0.00$	$[-2.53,+2.53]$	1.000	3	3
ArabCulture (seed 21)	89.9	87.4	$-2.53$	$[-5.05,+0.00]$	0.125	1	6
ArabCulture (seed 42)	90.9	90.9	$\pm 0.00$	$[-3.03,+3.03]$	1.000	5	5
Palm (seed 0)	88.9	86.9	$-2.02$	$[-5.56,+1.52]$	0.388	4	8
Palm (seed 21)	89.9	84.3	$-5.56^{*}$	$[-9.60,-1.52]$	0.013	3	14

Appendix

Table 20: Improvement and regression breakdown on Jawaher proverbs across cultural fine-tuning (ArabCulture and Palm) and poetry fine-tuning (FannOrFlop). Base% and FT% are accuracy before and after fine-tuning;

\Delta\%

is the percentage-point change, reported with 95% paired-bootstrap confidence intervals over test items and exact McNemar

p

-values; Impr and Regr are the number of individual predictions improved or worsened. ∗ marks

p<0.05

Model / Dataset	$\Delta\%$	95% CI	Impr	Regr
ALLaM
Jawaher (seed 0)	$+1.11$	$[-3.33,+5.56]$	10	8
Jawaher (seed 21)	$-1.11$	$[-5.56,+3.33]$	8	10
Jawaher (seed 42)	$-3.89$	$[-7.78,+0.00]$	3	10
FannOrFlop (seed 0)	$+1.11$	$[-3.33,+5.56]$	9	7
FannOrFlop (seed 21)	$-1.67$	$[-5.56,+1.67]$	4	7

Appendix

Table 21: Improvement and regression breakdown on AraDiCE-Culture across figurative fine-tuning (Jawaher) and poetry fine-tuning (FannOrFlop).

\Delta\%

is the percentage-point change from base to fine-tuned accuracy, reported with 95% paired-bootstrap confidence intervals over test items; Impr and Regr are the number of individual predictions improved or worsened. ∗ marks

p<0.05

under an exact McNemar test.

Count	Idiom
Top 15 Most Frequently Improved
16	ْدوُع ىَلَع ْدوُد
9	ْهَديِدَحْلا ىَلَع
9	ْهشِو ْلَكَأ
8	ْهَدْلِجْلا ىَلَع
6	ْعاَرِّدلاو ْعاَبْلاِب

Appendix

Table 22: Top 15 most frequently improved and regressed idioms across all models fine-tuned on cultural data (ArabCulture and Palm).

Count	Dialect	Proverb
Top 15 Most Frequently Improved
10	Mauritanian	.ْكلْكَّلَعْي ال ْكاعْما ْدِراو ُهاَمْلا
9	Mauritanian	لازغا وما نيعفا سونخب لك
6	Yemeni	قزري كبرو قزخ
6	Lebanese	سبايلاو رَضْخألا لكأ
6	Kuwaiti	.رحب هطقو ريخ وس

Appendix

Table 23: Top 15 most frequently improved and regressed proverbs across all fine-tuned models, with counts and dialect variety.

Dialect	Impr	Regr	Net
Yemeni	10	2	$+8$
Mauritanian	11	3	$+8$
Jordanian	8	1	$+7$
Kuwaiti	6	0	$+6$
Tunisian	5	1	$+4$
MSA	4	0	$+4$

Appendix

Table 24: Net poetry fine-tuning effect by Arabic dialect variety on the proverb task.

Count	Idiom
Top 15 Most Frequently Improved
8	ْدوُع ىَلَع ْدوُد
6	ّْبَد ْنِمْو ِّبَه ْنِم
4	ْهَدْلِجْلا ىَلَع
4	ْبلْقِتِو ْبَرْضِتِب اَيْنُّدلا
3	ْهَديِدَحْلا ىَلَع

Appendix

Table 25: Top 15 most frequently improved and regressed idioms across all models fine-tuned on poetry data (FannOrFlop).

Count	Dialect	Proverb
Top 15 Most Frequently Improved
4	MSA	ادرمت ميئللا تمركأ تنأ نإو هتكلم ميركلا تمركأ تنأ اذإ
4	Egyptian	اهيلَع سِّسَحِي ةَحطَب هُسار ىلَع يللإ
3	Mauritanian	.ْكلْكَّلَعْي ال ْكاعْما ْدِراو ُهاَمْلا
3	Algerian	.حازم ةودعلا
3	Yemeni	قزري كبرو قزخ

Appendix

Table 26: Top 15 most frequently improved and regressed proverbs across all models fine-tuned on poetry data (FannOrFlop), with counts and dialect variety.

Figurative Language Understanding

Figurative language understanding studies how people and AI interpret nonliteral meaning in metaphors, idioms, proverbs, puns, and related forms.

See all 45 papers

Wisdom in Unity: The Role of Multilingual Training in Figurative Language Identification in Proverbs
Aug 8, 2026Rama Alomair, Remas Alsubaie, Walaa Saifalislam +5Figurative Language Understanding Multilingual Language Models
College of Computer and Information Sciences, King Saud University
VIVID: A Culturally Grounded Benchmark Exposing the Figurative Language Gap in Vietnamese NLP
Aug 4, 2026Tu Tran Do, Nhat Ngoc Nguyen, Khanh-Tung Tran +3Figurative Language Understanding Multilingual Language Model Evaluation
Posts and Telecommunications Institute of Technology (PTIT), Vietnam · School of Computer Science and Information Technology, University College Cork, Ireland
Metaphors We Compute By: A Computational Audit of Cultural Translation vs. Thinking in LLMs
Apr 6, 2026Yuan Chang, Jiaming Qu, Zhu LiCultural Bias in Language Models Cultural Knowledge in Language Models
Meta, Burlingame, CA · Amazon, Seattle, WA

Aug 8, 2026J/K move · Enter open · S save

Wisdom in Unity: The Role of Multilingual Training in Figurative Language Identification in Proverbs

Rama Alomair, Remas Alsubaie, Walaa Saifalislam, Rima Alsonbul+4

College of Computer and Information Sciences, King Saud University

Figurative Language Understanding Multilingual Language Models Multilingual Language Model Evaluation Cross-Lingual Transfer

Open paper AskPDF arXiv

Although multilingual approaches to figurative language identification are not new, the shift beyond language-homogeneous training data requires a clearer understanding of the contribution of translated multilingual supervision. We examine this question using 742 proverb concepts across 6,787 translated instances for seven languages. We evaluate five models including multilingual encoders and instruction-tuned LLMs through progressively increasing levels of multilingual supervision. Moreover, we introduce multidimensional annotation framework for proverbs that characterizes proverbs through four complementary figurative forms: Metaphorical, Moral/Advisory, Cause-Effect, and Culture-Specific. Our findings show that overall, adding multilingual training data beyond 50% provides only limited additional improvement, although the best supervision level varies across models and languages. Also, we show that combining diverse figurative forms yields the strongest overall performance. A notable finding is that the least frequent figurative form culture-specific exhibits the largest performance gains under multilingual supervision. Furthermore, the moral/advisory and culture-specific forms of proverb contribute more to instruct tuning LLM overall figurative identification performance. These findings motivate multilingual figurative identification to move beyond metaphor-centric taxonomies toward concept-level multidimensional frameworks that explicitly model complementary forms of figurative meanings that are context representative.

Abstract

Figurative Language Understanding

Explore similar work

Wisdom in Unity: The Role of Multilingual Training in Figurative Language Identification in Proverbs

VIVID: A Culturally Grounded Benchmark Exposing the Figurative Language Gap in Vietnamese NLP

Metaphors We Compute By: A Computational Audit of Cultural Translation vs. Thinking in LLMs