Dual Perspectives in Emotion Attribution: A Generator-Interpreter Framework for Cross-Cultural Analysis of Emotion in LLMs

Aizirek Turdubaeva; Uichin Lee

cs.CLMar 30, 2026

Dual Perspectives in Emotion Attribution: A Generator-Interpreter Framework for Cross-Cultural Analysis of Emotion in LLMs

Authors: Aizirek Turdubaeva, Uichin Lee

Organizations: KAIST / Daejeon, Republic of Korea

Abstract

Large language models (LLMs) are increasingly used in cross-cultural systems to understand and adapt to human emotions, which are shaped by cultural norms of expression and interpretation. However, prior work on emotion attribution has focused mainly on interpretation, overlooking the cultural background of emotion generators. This assumption of universality neglects variation in how emotions are expressed and perceived across nations. To address this gap, we propose a Generator-Interpreter framework that captures dual perspectives of emotion attribution by considering both expression and interpretation. We systematically evaluate six LLMs on an emotion attribution task using data from 15 countries. Our analysis reveals that performance variations depend on the emotion type and cultural context. Generator-interpreter alignment effects are present; the generator's country of origin has a stronger impact on performance. We call for culturally sensitive emotion modeling in LLM-based systems to improve robustness and fairness in emotion understanding across diverse cultural contexts.

Figures & tables

Figure 1: Cross-cultural emotion attribution. An event labeled as anger is interpreted as shame or sadness.

Figure 2: Emotion attribution accuracy for each model, averaged across all personas.

Figure 3: Confusion matrix aggregated across personas in the GPT-4o model.

	Interpreter
Generator	AU	AT	BR	BG	CN	FI	HN	IN	MW	NL	NZ	ES	SE	US	ZM	Avg
AU	0.74	0.74	0.71	0.75	0.73	0.71	0.71	0.72	0.71	0.74	0.73	0.74	0.74	0.74	0.73	0.73
AT	0.81	0.81	0.79	0.81	0.80	0.77	0.80	0.79	0.78	0.81	0.79	0.81	0.79	0.81	0.79	0.80
BR	0.78	0.78	0.75	0.78	0.78	0.76	0.77	0.77	0.76	0.77	0.77	0.78	0.77	0.78	0.77	0.77
BG	0.77	0.76	0.76	0.76	0.77	0.75	0.75	0.76	0.76	0.76	0.75	0.75	0.76	0.75	0.75	0.76
CN	0.63	0.63	0.62	0.63	0.67	0.60	0.61	0.62	0.59	0.62	0.62	0.62	0.62	0.65	0.63	0.62

Table 1: GPT-4o Model: Generator-Interpreter Accuracy Results. Red indicates the lowest performance value per interpreter column. Note that AU = Australia, AT = Austria, BR = Brazil, BG = Bulgaria, CN = China, FI = Finland, HN = Honduras, IN = India, MW = Malawi, NL = Netherlands, NZ = New Zealand, ES = Spain, SE = Sweden, US = USA, ZM = Zambia.

Appendix figures & tables20 assets

Supplementary material from the paper’s appendix.

Appendix

Model	Release	Size	Open	License
gpt-4o-2024-05-13	May 2024	–	X	Proprietary
gemini-pro-1.5	Apr 2024	–	X	Proprietary
claude-3.5-sonnet	Oct 2024	–	X	Proprietary
mistral-7b-instruct-v0.3	May 2024	7.3B	O	Apache 2.0
qwen-2.5-7b-instruct	Oct 2024	7B	O	Apache 2.0
deepseek-v3	Dec 2024	671B	O	MIT

Appendix

Table 2: Model specifications including release dates, parameter sizes, open-source availability, and license types.

Category

Examples of excluded responses

Null / Invalid

“Nothing.”, “None.”, “Not applicable.”, “NO RESPONSE.”, “Not applicable to myself.”, “Doesn’t apply.”, “Can’t remember having had this feeling.”, “Can’t remember that feeling.”, “Can’t remember any episode of disgust.”, “Can’t remember any such feelings now.”, “I cannot recall feeling guilty in the near past. REMARK - SUBJECT MAY NOT HAVE TAKEN THE EXERCISE VERY SERIOUSLY.”, “Blank.”, “I can’t remember ever feeling shame.”, “Not included on questionnaire.”, “Cannot recall the emotion with any force.”, “Haven’t felt shame for ages.”, “Cannot remember such a situation.”, “DO NOT REMEMBER.”, “I do not recall being scared in the near past.”, “Can’t remember exact example.”, “I think I have hardly had a situation where I felt disgusted. Perhaps once, when I saw a big spider.”, “Can think of no time that I have felt fear.”, “When I considered writing a load of crap for this emotion.”, “Sadness – … I can’t remember a concrete example.”, “Never really had the experience just yet – N/A.”, “I can’t remember being truly angry. Usually it’s just disappointment that I feel.”, “Disgust is a feeling that I have never experienced with people but only with repulsive animals (e.g., snakes on T.V.).”, “I cannot remember a situation in which I felt ashamed… knocked down my own glass.”, “[ Do not know.]”, “[ No response.]”, “[ Can not think of anything just now.]”, “[ Can not think of any situation.]”, “[ Can not remember.]”, “[ Cannot recall any incident when I felt shame.]”, “[ No answer as I see very little difference between shame and guilt. Therefore see answer for "shame".]”, “[ Do not remember any situation of that kind.]”, “[ Can not think of anything.]”, “[ I can not remember anything in particular. What I can remember is feelings of having done or said something which then had negative consequences.]”, “[ No description.]”, “[ I have not felt this emotion.]”, “[ Never felt the emotion.]”, “[ I do not remember any event when I felt guilty.]”, “[ I have never felt this emotion.]”, “[ I can not recall one.]”, “[ The previous incident holds good here also.]”, “[ ….has not really happened.]”, “[ Never experienced.]”, “[ Never.]”, “[ Sorry, I was never ashamed about anything in my life.]”, “[ Do not remember any incident.]”, “[ Normally I do not feel disgusted.]”, “[ I have not felt this emotion yet.]”, “[ I have not felt this emotion in my life.]”, “[ I have felt shame but am unable to remember any particular incident.]”, “[ I do not recall one here either.]”, “[ I was not, I am not, I will never ever feel ashamed of myself for something I have done.]”, “[ There have been very few instances of disgust, as far as I can remember they are not important or worth mentioning.]”, “[ Never felt guilty, as I have never done anything which could make me feel guilty.]”, “[ Honestly, I have never felt disgust at any situation in my life.]”, “[ I can positively say that I have never done anything that made me feel guilty.]”, “[ There are many instances which are all equally irritating.]”

Cross-Referenced

“The same as in ‘shame’.”, “The same event described under ‘shame’.”, “The same situation as with sadness. I was afraid about what my parents would further do to destroy my relationship.”, “The same as for SHAME and DISGUST. I was asked to resign my sorority which I did because I was ashamed.”, “The same as in SHAME.”, “The same for guilt would apply.”, “Same situation as before – having my car stolen.”, “As for disgust.”, “I do not seem to feel emotion such as shame, guilt or fear – far out.”, “Robbery mentioned under sadness.”, “[ Same as in anger.]”, “[ Same as above – friends who torture animals.]”, “[ The same as in shame.]”, “[ The same as in anger.]”, “[ The same as in guilt.]”, “The same event described under FEAR”

Appendix

Table 3: Examples of responses excluded during ISEAR dataset cleaning.

	Mistral		Qwen2		Gemini		GPT-4o		DeepSeek		Claude
Persona	Acc.	F1	Acc.	F1	Acc.	F1	Acc.	F1	Acc.	F1	Acc.	F1
Australia	0.63	0.06	0.70	0.14	0.73	0.14	0.74	0.25	0.74	0.27	0.70	0.04
Austria	0.63	0.05	0.70	0.15	0.75	0.16	0.74	0.23	0.75	0.25	0.70	0.06
Brazil	0.64	0.06	0.70	0.14	0.72	0.11	0.73	0.22	0.74	0.23	0.69	0.05
Bulgaria	0.63	0.06	0.70	0.15	0.75	0.18	0.75	0.31	0.75	0.25	0.70	0.07
Sweden	0.63	0.05	0.70	0.15	0.72	0.13	0.74	0.25	0.74	0.30	0.67	0.04

Appendix

Table 4: Emotion attribution performance (accuracy and F1-score) of six LLMs across 15 interpreter personas. The highest accuracy and F1-score for each persona are shown in bold.

	Mistral			Qwen2			Gemini			GPT-4o			DeepSeek			Claude
Emotion	P	R	F1	P	R	F1	P	R	F1	P	R	F1	P	R	F1	P	R	F1
fear	0.91	0.70	0.79	0.86	0.78	0.82	0.75	0.87	0.81	0.83	0.87	0.85	0.83	0.86	0.84	0.81	0.82	0.81
sadness	0.39	0.95	0.55	0.53	0.90	0.67	0.68	0.78	0.73	0.56	0.89	0.69	0.58	0.89	0.70	0.71	0.73	0.72
joy	0.89	0.95	0.92	0.88	0.97	0.92	0.93	0.95	0.94	0.89	0.99	0.93	0.88	0.99	0.93	0.93	0.93	0.93
shame	0.73	0.32	0.44	0.70	0.53	0.60	0.64	0.65	0.65	0.76	0.55	0.64	0.75	0.49	0.60	0.49	0.79	0.61
guilt	0.62	0.69	0.65	0.77	0.52	0.62	0.72	0.70	0.71	0.70	0.81	0.75	0.67	0.83	0.74	0.78	0.47	0.59

Appendix

Table 5: Emotion attribution performance (P: precision, R: recall, F1) of six LLMs across seven target emotions.

	Interpreter
Generator	AU	AT	BR	BG	CN	FI	HN	IN	MW	NL	NZ	ES	SE	US	ZM	Avg
AU	0.73	0.77	0.73	0.77	0.72	0.75	0.73	0.72	0.71	0.76	0.74	0.71	0.75	0.73	0.72	0.74
AT	0.83	0.84	0.81	0.83	0.79	0.78	0.79	0.79	0.76	0.82	0.81	0.78	0.81	0.82	0.79	0.80
BR	0.76	0.77	0.75	0.77	0.74	0.74	0.75	0.76	0.74	0.77	0.76	0.72	0.76	0.74	0.75	0.75
BG	0.76	0.76	0.74	0.79	0.74	0.74	0.74	0.74	0.72	0.75	0.75	0.67	0.74	0.74	0.74	0.74
CN	0.61	0.59	0.58	0.61	0.63	0.56	0.60	0.61	0.58	0.58	0.59	0.50	0.55	0.58	0.62	0.59

Appendix

Table 6: Gemini Model: Generator-Interpreter Accuracy Results. Note that AU = Australia, AT = Austria, BR = Brazil, BG = Bulgaria, CN = China, FI = Finland, HN = Honduras, IN = India, MW = Malawi, NL = Netherlands, NZ = New Zealand, ES = Spain, SE = Sweden, US = USA, ZM = Zambia.

	Interpreter
Generator	AU	AT	BR	BG	CN	FI	HN	IN	MW	NL	NZ	ES	SE	US	ZM	Avg
AU	0.61	0.62	0.62	0.62	0.63	0.60	0.61	0.62	0.60	0.60	0.61	0.63	0.60	0.63	0.61	0.61
AT	0.65	0.66	0.64	0.66	0.65	0.65	0.65	0.64	0.63	0.65	0.65	0.66	0.66	0.67	0.65	0.65
BR	0.66	0.66	0.65	0.66	0.67	0.66	0.66	0.65	0.64	0.65	0.65	0.67	0.66	0.68	0.65	0.66
BG	0.67	0.65	0.66	0.67	0.66	0.65	0.65	0.66	0.63	0.65	0.65	0.65	0.66	0.67	0.64	0.65
CN	0.47	0.46	0.48	0.46	0.45	0.46	0.44	0.46	0.43	0.46	0.47	0.47	0.45	0.47	0.46	0.46

Appendix

Table 7: Mistral Model: Generator-Interpreter Accuracy Results. Note that AU = Australia, AT = Austria, BR = Brazil, BG = Bulgaria, CN = China, FI = Finland, HN = Honduras, IN = India, MW = Malawi, NL = Netherlands, NZ = New Zealand, ES = Spain, SE = Sweden, US = USA, ZM = Zambia.

	Interpreter
Generator	AU	AT	BR	BG	CN	FI	HN	IN	MW	NL	NZ	ES	SE	US	ZM	Avg
AU	0.74	0.74	0.74	0.73	0.74	0.73	0.73	0.74	0.72	0.74	0.73	0.75	0.74	0.75	0.73	0.74
AT	0.80	0.81	0.81	0.81	0.82	0.79	0.80	0.79	0.77	0.82	0.80	0.82	0.82	0.80	0.79	0.80
BR	0.78	0.77	0.75	0.77	0.78	0.77	0.77	0.75	0.76	0.77	0.76	0.77	0.78	0.77	0.76	0.77
BG	0.76	0.77	0.75	0.76	0.78	0.75	0.76	0.76	0.76	0.76	0.76	0.75	0.76	0.77	0.75	0.76
CN	0.64	0.62	0.61	0.65	0.65	0.62	0.64	0.64	0.62	0.63	0.62	0.64	0.62	0.63	0.62	0.63

Appendix

Table 8: DeepSeek Model: Generator-Interpreter Accuracy Results. Note that AU = Australia, AT = Austria, BR = Brazil, BG = Bulgaria, CN = China, FI = Finland, HN = Honduras, IN = India, MW = Malawi, NL = Netherlands, NZ = New Zealand, ES = Spain, SE = Sweden, US = USA, ZM = Zambia.

	Interpreter
Generator	AU	AT	BR	BG	CN	FI	HN	IN	MW	NL	NZ	ES	SE	US	ZM	Avg
AU	0.68	0.69	0.69	0.69	0.68	0.70	0.69	0.71	0.68	0.70	0.69	0.68	0.69	0.69	0.68	0.69
AT	0.73	0.74	0.74	0.73	0.73	0.73	0.73	0.75	0.73	0.75	0.73	0.72	0.73	0.73	0.74	0.73
BR	0.69	0.70	0.70	0.70	0.71	0.71	0.70	0.70	0.69	0.70	0.71	0.71	0.70	0.70	0.68	0.70
BG	0.73	0.74	0.72	0.73	0.73	0.73	0.73	0.73	0.74	0.73	0.72	0.74	0.74	0.71	0.72	0.73
CN	0.56	0.56	0.56	0.55	0.52	0.55	0.55	0.57	0.53	0.56	0.55	0.55	0.55	0.55	0.54	0.55

Appendix

Table 9: Qwen2 Model: Generator-Interpreter Accuracy Results. Note that AU = Australia, AT = Austria, BR = Brazil, BG = Bulgaria, CN = China, FI = Finland, HN = Honduras, IN = India, MW = Malawi, NL = Netherlands, NZ = New Zealand, ES = Spain, SE = Sweden, US = USA, ZM = Zambia.

	Interpreter
Generator	AU	AT	BR	BG	CN	FI	HN	IN	MW	NL	NZ	ES	SE	US	ZM	Avg
AU	0.76	0.77	0.73	0.75	0.73	0.71	0.72	0.71	0.69	0.75	0.74	0.74	0.71	0.75	0.70	0.73
AT	0.77	0.77	0.77	0.77	0.73	0.72	0.74	0.75	0.69	0.77	0.77	0.77	0.75	0.79	0.71	0.75
BR	0.70	0.67	0.69	0.69	0.66	0.66	0.68	0.67	0.64	0.68	0.69	0.68	0.67	0.71	0.66	0.68
BG	0.73	0.71	0.70	0.73	0.71	0.69	0.69	0.71	0.68	0.71	0.72	0.71	0.69	0.72	0.68	0.71
CN	0.59	0.59	0.59	0.59	0.61	0.57	0.58	0.59	0.57	0.58	0.59	0.59	0.55	0.61	0.57	0.58

Appendix

Table 10: Claude Model: Generator-Interpreter Accuracy Results. Note that AU = Australia, AT = Austria, BR = Brazil, BG = Bulgaria, CN = China, FI = Finland, HN = Honduras, IN = India, MW = Malawi, NL = Netherlands, NZ = New Zealand, ES = Spain, SE = Sweden, US = USA, ZM = Zambia.

Figure 4: Worst-performing personas in the GPT-4o model (Generator: Austria).

Figure 5: Best-performing personas in the GPT-4o model (Generator: Austria)

Figure 6: Worst-performing personas in the GPT-4o model (Generator: Malawi)

Figure 7: Best-performing personas in the GPT-4o model (Generator: Malawi)

Figure 8: Worst-performing personas in the GPT-4o model (Generator: China)

Figure 9: Best-performing personas in the GPT-4o model (Generator: China)

Figure 10: Worst-performing personas in the GPT-4o model (Generator: Sweden)

Figure 11: Best-performing personas in the GPT-4o model (Generator: Sweden)

ISEAR Event Examples	Gold-label	US	AU	BG	NL	IN
Failing in an exam because of laziness	shame	guilt	guilt	shame	guilt	guilt
South Korea winning more gold medals in the Asian Olympics (more than China)	anger	joy	joy	joy	joy	joy
Being sexually assaulted on a bus and no one helped (people only laughed)	anger	fear	fear	sadness	fear	shame
Being insulted in front of my parents, and they believed other people (those insulting me)	anger	sadness	sadness	shame	sadness	shame
Not being able to do well to be chosen for the English lecture	sadness	sadness	sadness	shame	sadness	shame
Saw some girls dressed up like Americans	disgust	joy	joy	joy	joy	joy

Appendix

Table 11: Examples of emotion predictions for Chinese-origin events in the GPT-4o model. The gold-label column shows the expected emotion, while the remaining columns display predictions from different personas. Note that US = USA, AU = Australia, BG = Bulgaria, NL = Netherlands, IN = India.

Model	Emotions
gpt-4o-2024-05-13	anguish, annoyance, anxiety, awkwardness, betrayal, concern, confusion, curiosity, disappointment, discomfort, disgust, embarrassment, frustration, grief, hate, indifference, irritation, regret, relief, shock, stress, surprise, worry
gemini-pro-1.5	annoyance, anticipation, anxiety, awkwardness, boredom, concern, confusion, curiosity, disappointment, discomfort, doubt, embarrassment, envy, exasperation, frustration, gratitude, grief, hope, horror, humiliation, impatience, indifference, indignation, injustice, interest, intrigue, irritation, jealousy, nervousness, neutral, pain, panic, pressure, pride, regret, relief, resignation, respect, shock, stress, surprise, terror, worry
claude-3.5-sonnet	acceptance, amusement, annoyance, anticipation, anxiety, boredom, betrayal, concern, confusion, curiosity, disappointment, discomfort, embarrassment, envy, gratitude, grief, hope, humiliation, frustration, indifference, irritation, jealousy, loneliness, mix, neutral, nostalgia, pain, peace, pride, regret, relief, respect, satisfaction, skepticism, stress, surprise, worry, unease
mistral-7b-instruct-v0.3	affection, amusement, annoyance, anticipation, anxiety, apprehension, astonishment, attentiveness, awe, boredom, calmness, caution, compassion, concern, confusion, contentment, curiosity, determination, disappointment, discomfort, disinterest, disagreeableness, embarrassment, empathy, excitement, exhaustion, frustration, focus, gratitude, grief, hate, humility, indifference, interest, jealousy, melancholy, nervousness, neutral, patience, peace, pride, responsibility, relaxation, relief, respectful awe, respectful anxiety, shock, shyness, surprise, sympathy, uncertainty, unease
qwen-2.5-7b-instruct	anxiety, annoyance, attention, boredom, compassion, concern, confusion, contentment, curiosity, disappointment, discomfort, distraction, embarrassment, envy, excitement, focus, frustration, gratitude, grief, hate, horror, jealousy, neutral, nervousness, nostalgia, pain, patience, peace, pride, regret, relaxation, relief, shock, stress, surprise, uncertainty
deepseek-v3	annoyance, betrayal, boredom, calmness, compassion, concern, confusion, contentment, curiosity, disappointment, discomfort, envy, frustration, gratitude, grief, indifference, irritation, jealousy, neutral, patience, relief, stress, surprise

Appendix

Table 12: Unique emotion labels generated by large language models

	Interpreter
Generator	AU	AT	BR	BG	CN	FI	HN	IN	MW	NL	NZ	ES	SE	US	ZM	Avg
AU (Before)	0.74	0.74	0.74	0.73	0.74	0.73	0.73	0.74	0.72	0.74	0.73	0.75	0.74	0.75	0.73	0.74
AU (After)	0.72	0.73	0.73	0.74	0.73	0.73	0.74	0.73	0.73	0.74	0.73	0.73	0.73	0.74	0.72	0.73
Effect	-0.02	-0.01	-0.01	+0.01	-0.01	0.00	+0.01	-0.01	+0.01	0.00	0.00	-0.02	-0.01	-0.01	-0.01	-0.01
AT (Before)	0.80	0.81	0.81	0.81	0.82	0.79	0.80	0.79	0.77	0.82	0.80	0.82	0.82	0.80	0.79	0.80
AT (After)	0.62	0.82	0.64	0.81	0.83	0.81	0.82	0.80	0.81	0.79	0.82	0.80	0.83	0.81	0.82	0.79

Appendix

Table 13: Effect of explicit generator country specification on emotion attribution accuracy for the DeepSeek model. Red values indicate performance decreases, while blue ones indicate performance increases. Note that AU = Australia, AT = Austria, BR = Brazil, BG = Bulgaria, CN = China, FI = Finland, HN = Honduras, IN = India, MW = Malawi, NL = Netherlands, NZ = New Zealand, ES = Spain, SE = Sweden, US = USA, ZM = Zambia.

Emotion Representation in Language Models

Emotion representation in language models studies how models encode, interpret, and use affective concepts such as valence, arousal, and empathy.

See all 64 papers

Expressing Social Emotions: Misalignment Between LLMs and Human Cultural Emotion Norms
Apr 18, 2026Sree Bhattacharyya, Manas Mehta, Leona Chen +4LLM Evaluation Cultural Bias in Language Models
The Pennsylvania State University · Duke University · Northeastern University +1
LLMs Infer Cultural Context but Fail to Apply It When Responding
Jun 16, 2026Yisong Miao, Jian Zhu, Vered ShwartzPragmatic Reasoning in Language Models Cultural Bias in Language Models
National University of Singapore · University of British Columbia · Canada CIFAR AI Chair, Vector Institute
Cultural Authenticity: Comparing LLM Cultural Representations to Native Human Expectations
Apr 3, 2026Erin MacMurray van Liemt, Aida Davani, Sinchana Kumbale +2Cultural Alignment Cross-Cultural Language Model Evaluation
Google Research, USA · Google, USA

Apr 18, 2026J/K move · Enter open · S save

Expressing Social Emotions: Misalignment Between LLMs and Human Cultural Emotion Norms

Sree Bhattacharyya, Manas Mehta, Leona Chen, Cristina Salvador+3

The Pennsylvania State University · Duke University · Northeastern University+1

LLM Evaluation Cultural Bias in Language Models Language Model Bias Evaluation

Open paper AskPDF arXiv

The expression of emotions that serve social purposes, such as asserting independence or fostering interdependence, is central to human interactions and varies systematically across cultures. As LLMs are increasingly used to simulate human behavior in culturally nuanced interactions, it is important to understand whether they faithfully capture human patterns of social emotion expression. When LLM responses are not culturally aligned, their utility is compromised -- particularly when users assume they are interacting with a culturally attuned interlocutor, and may act on advice that proves inappropriate in their cultural context. We present a psychologically informed evaluation framework of cross-cultural social emotion expression in LLMs. Using a human study comparing European American and Latin American participants' expression of engaging and disengaging emotions, we evaluate six frontier LLMs on their ability to reflect culturally differentiated patterns for expressing social emotions. We find systematic misalignment between model and human behavior: all models express engaging emotions more than disengaging ones, with particularly stark differences observed for the generally well-represented European American persona. We further highlight that LLM responses are highly concentrated and deterministic, failing to capture the diversity of human responses in expressing social emotions. Our ablation analyses reveal that these patterns are robust to sampling temperatures, partially sensitive to prompt language, and dependent on the response elicitation format. Together, our findings highlight limitations in how current LLMs represent the interaction of cultural and emotional axes, particularly when expressing social emotions, with direct implications for their deployment in cross-cultural affective contexts.

Abstract

Emotion Representation in Language Models

Explore similar work

Expressing Social Emotions: Misalignment Between LLMs and Human Cultural Emotion Norms

LLMs Infer Cultural Context but Fail to Apply It When Responding

Cultural Authenticity: Comparing LLM Cultural Representations to Native Human Expectations