The research, led by William Villegas-Ch of Universidad de Las Américas in Ecuador together with colleagues from Universidad Internacional del Ecuador and Universidad Estatal de Milagro, focused on transformer-based models: multilingual BERT, known as mBERT, and XLM-R, a larger model built on the RoBERTa architecture. The team also tested a fine-tuned variant, XLM-R-FT, adapted to culturally diverse training data. These models sit behind countless applications, from customer-service chatbots to social media monitoring tools, and their perceived fluency in dozens of languages has made them the default choice for multilingual sentiment and emotion analysis.

The problem, the researchers argue, is that standard evaluations of these systems rely on structured, homogeneous datasets that bear little resemblance to real-world communication. Everyday emotional expression is messy. Speakers switch languages mid-sentence, lean on idioms whose meanings cannot be assembled from their parts, and convey feelings indirectly through irony and understatement. “Me hierve la sangre,” a Spanish phrase that literally translates to “my blood boils,” expresses anger through a figurative construction. “Estoy sad AF hoy” blends English and Spanish in a single clause. “Oh, great… just what I needed” communicates frustration through positive words. None of these patterns are well represented in the canonical corpora on which multilingual models are trained.

To measure how well the models cope, the team assembled an evaluation set of 4,892 instances drawn from public sources including the BBC Forum Dataset, the Brazilian offensive-comment corpus OFFCOMBR, Italian SENTIPOLC data and Twitter-based corpora. The set was balanced across four languages: English, Spanish, Portuguese and Italian. English served as a reference point because of its dominance in pre-training data, while the three Romance languages were chosen for their structural similarity paired with sharply different pragmatic and colloquial conventions around irony, attenuation and intensification. Each instance was categorized as idiomatic, code-switched, or indirect in structure.

Because the source datasets carried inconsistent labels, the researchers manually annotated a representative subset according to Ekman’s six basic emotions: joy, sadness, anger, fear, surprise and disgust. Native speakers with linguistic training performed the annotation, with two annotators independently labeling each sample. Agreement was required to reach a Cohen’s kappa of at least 0.8, and disagreements were resolved through adjudication or the samples were discarded, ensuring that the emotional ground truth was itself robust.

The experimental protocol was rigorous. All models ran on the HuggingFace Transformers library atop PyTorch, executed on NVIDIA A100 GPUs in a Linux computing cluster. Fine-tuning used the AdamW optimizer with a learning rate of 2 × 10⁻⁵, a batch size of 32, up to ten epochs, dropout regularization at 0.1 and early stopping when validation Macro-F1 stalled for three consecutive epochs. Evaluation used stratified five-fold cross-validation, with folds balanced jointly by language and emotion class so that no category was underrepresented in either training or testing.

The headline result is a pattern of consistent degradation on culturally marked input. When models moved from clean, monolingual sentences to code-switched or idiomatic ones, F1 scores fell by as much as 10 points and, in some configurations, between 13 and 18 points. In Spanish, mBERT dropped from an F1 of 0.82 on monolingual inputs to 0.67 on code-switched sentences and 0.66 on idiomatic expressions. Portuguese and Italian showed even larger declines. English, by contrast, retained average accuracy above 0.84 under clean conditions and degraded only modestly under noise.

The language gap tracks closely with how much of each language the models saw during pre-training. English, massively represented in training corpora, proved most stable. Spanish occupied an intermediate position. Portuguese and Italian, less richly represented, exhibited the largest drops and the widest variability, with Italian showing the greatest instability of all. The researchers computed performance deviations relative to English of −0.06 for Spanish, −0.09 for Portuguese and −0.12 for Italian, a gradient that reinforces what they describe as representation-driven bias: models are most reliable precisely where their training data was densest.

Specific emotional categories proved especially fragile. Joy and disgust showed the highest variability across languages and conditions. In Portuguese, accuracy on disgust fell from 0.82 on clean text to 0.69 on noisy text, while in Italian it reached the study’s lowest observed value at 0.67. The team also documented systematic confusion among anger, fear and sadness, with confusion rates of 0.25 between anger and fear and 0.20 between fear and sadness, suggesting that when emotional states are conveyed informally, the models’ decision boundaries between negative emotions blur badly.

Noise alone was enough to trip up the baseline models. When the researchers introduced perturbations mimicking real digital communication, emoji substitutions, social media abbreviations, minor spelling errors and emphatic reduplicated constructions, mBERT suffered accuracy losses of 0.13 in Portuguese and 0.15 in Italian. In one striking example, the Portuguese sentence “Fiquei muito fps com esso [disgusted face],” a deliberately corrupted expression of disgust, was classified as joy by mBERT, apparently because surface cues such as the emoji overwhelmed contextual meaning. The fine-tuned XLM-R-FT fared considerably better, keeping accuracy losses to within 0.05 across languages, but even it could not eliminate sensitivity to distortion.

Cultural interference from English emerged as a distinct failure mode. The researchers constructed adversarial examples using false cognates, expressions in Spanish, Portuguese or Italian that superficially resemble English words but carry different meanings. The Spanish phrase “estoy constipado,” which means “I have a cold” rather than what an English speaker might assume, could push the model toward incorrect negative emotional categories purely through lexical similarity. Across culturally ambiguous sentences, mBERT and XLM-R reached error rates of up to 25 percent in Italian and Spanish due to such interference, while the fine-tuned variant reduced this to below 15 percent.

Perhaps the most illuminating part of the study is its use of interpretability tools, LIME and SHAP, which attribute a model’s prediction to individual input tokens. These analyses showed that models frequently anchor their judgments on lexically salient words while ignoring context. In the Spanish expression “Estoy re quemado con esta vaina,” the model correctly weighted “quemado” as emotionally negative but also amplified the contribution of “vaina,” a context-dependent filler word with neutral semantic load. In Italian, multi-word idioms were decomposed into independent tokens, producing erroneous emotional assignments whenever the expression resisted compositional interpretation. By contrast, the Portuguese colloquialism “massa” was correctly associated with positive emotion, suggesting that models succeed mainly where colloquial usage happens to align with patterns in training data.

The performance gap between direct and indirect emotional expression proved remarkably uniform. All models showed F1 reductions of between 0.08 and 0.10 when moving from explicit statements of feeling to indirect or sarcastic formulations, with mBERT showing the largest swings. This, the authors conclude, reflects a bias toward explicit emotional patterns aligned with Anglophone linguistic structures, an Anglocentric tendency that persists even when the model is operating in another language.

Fine-tuning helps, but it is not a cure. XLM-R-FT outperformed both baselines across every emotional category, improving Macro-F1 by up to 0.14 points over mBERT in categories such as fear and disgust, and it limited degradation on idiomatic and code-switched input to less than 8 F1 points relative to monolingual sentences. Yet the gap never closed entirely. Adaptation, the study finds, mitigates sensitivity to linguistic variation without removing it, and the fine-tuned model retains whatever structural biases were baked into its original pre-training.

The authors are careful to note the study’s limits. Only three target languages were examined, without full coverage of dialectal and regional variation within each. Human annotation of the culturally sensitive subset introduces subjectivity despite consensus mechanisms, and interpretability methods such as LIME and SHAP rest on assumptions that may not fully capture a transformer’s internal computation. The evaluation also focused on classification, not generative models or interactive settings.

Even so, the implications reach well beyond the laboratory. Conversational agents, content moderation systems and social media analytics tools deployed in multilingual environments routinely encounter exactly the kind of language this study shows the models mishandling. A benchmark score earned on clean, structured data, the researchers warn, says little about behavior in the wild. Their proposed framework, combining controlled perturbations, cross-linguistic comparison and token-level interpretability, offers a way to test for those hidden fragilities before deployment. Future work, they argue, should focus on datasets that explicitly capture cultural variation across regions and registers, and on integrating external semantic resources that encode idiomatic and context-dependent meaning, so that the next generation of multilingual systems can finally read between the lines the way humans do.

The research team embarked on a comprehensive examination of 1.2 billion social media entries posted during 2019, harnessing data from two of the world’s largest platforms: Twitter and Weibo. Employing cutting-edge natural language processing techniques, specifically the Bidirectional Encoder Representations from Transformers (BERT) model, in 65 different languages, the scientists were able to quantify sentiment scores across a diverse range of cultural and linguistic contexts. Each post was rated on a scale from 0.0, indicating extremely negative emotions, to 1.0, representing highly positive expressions. This large-scale sentiment analysis was then spatially aggregated across nearly 3,000 distinct geographical locations and cross-referenced with local temperature data, enabling a robust correlation between heat extremes and shifts in expressed mood.

A striking discovery emerged when the dataset was dissected in the context of economic disparity. The analysis found that in lower-income countries, where average annual per-capita income falls below $13,845 according to World Bank standards, the negative impact of temperatures exceeding 95 degrees Fahrenheit (35 degrees Celsius) on sentiment increased by approximately 25 percent. In wealthier nations, the effect was discernibly muted, with sentiment negativity rising by around 8 percent under similar heat conditions. This disparity underscores the amplified vulnerability of economically disadvantaged populations to climate-induced psychological stressors, highlighting a critical social equity issue in climate adaptation strategies.

The study’s co-author, Professor Siqi Zheng of MIT’s Department of Urban Studies and Planning and the Center for Real Estate, emphasizes the broader implications of these findings. Zheng notes that this research “opens up a new frontier in understanding how climate stress is shaping human well-being at a planetary scale.” Unlike previous studies that narrowly focused on physical health and economic metrics, this investigation introduces emotional well-being as an integral dimension of climate impact assessment. By integrating sentiment analysis with environmental data, the work pioneers a novel interdisciplinary approach that bridges urban studies, climatology, and psychology.

Technically, the application of BERT—one of the most sophisticated transformer-based language models—enabled nuanced understanding of emotional content across a vast linguistic spectrum. BERT’s ability to grasp complex syntax and semantic subtleties allowed the researchers to accurately decode sentiments embedded in short social media posts, which traditionally pose challenges for natural language processing due to their brevity and informal structure. This methodological innovation permitted a scalable assessment that surpasses conventional survey techniques, enabling real-time, global emotional monitoring over extended periods.

Beyond immediate correlations, the research also ventured into long-term projections, leveraging global climate models to estimate how emotional well-being might evolve under continued warming trends. Assuming some degree of human adaptation to heat stress, the models forecast a 2.3 percent decline in global emotional positivity attributable solely to high-temperature days by the year 2100. Although this figure may appear modest, it represents a significant psychological burden when extrapolated across billions of individuals, with potential cascading effects on social cohesion, productivity, and mental health infrastructures worldwide.

One of the notable challenges and caveats acknowledged by the researchers pertains to the demographic representativeness of social media users. The platforms analyzed tend to underrepresent certain population segments, such as very young children and the elderly, who are simultaneously known to be disproportionately sensitive to heat-related health risks. Consequently, the study’s estimates of emotional impacts may be conservative, as these vulnerable groups could experience even more pronounced declines in well-being during extreme temperature events but are less likely to express these sentiments online.

Moreover, the study elucidates that while heat stress broadly influences sentiment, the underlying mechanisms may be multifaceted. Physiological discomfort, disrupted sleep patterns, social tensions, and economic pressures exacerbated by climatic extremes all converge to shape emotional responses. Such complexity beckons further multidisciplinary inquiry to disentangle causal pathways and identify mitigating interventions. The authors advocate for integrating emotional resilience into societal adaptation frameworks, recognizing that fostering mental robustness will be crucial alongside traditional physical and infrastructural measures.

The global scope and scale of the dataset stand out as pioneering features. This study represents one of the first attempts to harness social media at a planetary level for environmental psychology research. By analyzing sentiments across nearly 3,000 locations distributed worldwide, the researchers achieved unprecedented geographic granularity, capturing heterogenous cultural and economic contexts that color climate-induced emotional experiences. This breadth affords policymakers and scientists a more detailed understanding of where and how to target support and adaptation efforts.

The research was conducted under the aegis of MIT’s Sustainable Urbanization Lab, a hub for interdisciplinary investigations into the interactions between urban environments, climate change, and human behavior. The team’s collaborative approach brought together experts from MIT, Chinese Academy of Sciences, Harvard University, Duke University, Maastricht University, and the Laureate Institute for Brain Research. This diverse expertise ensured rigorous analysis bridging computational linguistics, urban planning, climatology, and behavioral science.

Importantly, the dataset produced by this investigation has been made publicly accessible through the Global Sentiment project, enabling scientists, urban planners, and policymakers worldwide to explore intricate connections between climate variables and human sentiment. This open resource represents a valuable step toward democratizing climate data and fostering informed, evidence-based responses to the growing emotional and social challenges posed by global warming.

As the planet continues to warm, this research underscores the necessity of broadening our understanding of climate impacts beyond the tangible to include the intangible yet profound alterations in human emotional landscapes. By illuminating the silent suffering induced by extreme heat through the lens of social media, the study calls for urgent action to incorporate psychological well-being into global climate strategies, ensuring not only survival but also the emotional health of societies in the decades ahead.

Subject of Research: The impact of rising global temperatures on human emotional well-being, analyzed through large-scale social media sentiment data.

Article Title: Rising Temperatures Are Altering Human Sentiment Globally

News Publication Date: 21-Aug-2025

Web References: https://www.globalsentiment.mit.edu/

References: DOI 10.1016/j.oneear.2025.101422 (One Earth)

Keywords: Social media, Human behavior, Climatology, Earth climate, Climate change, Climate change adaptation