Emotion detection is a central problem in NLP, with recent progress driven by transformer-based models trained on established datasets. However, little is known about the linguistic regularities that characterize how emotions are expressed across different corpora and labels. This study examines whether linguistic features can serve as reliable interpretable signals for emotion recognition in text. We extract emotion-specific linguistic signatures from 13 English datasets and evaluate how incorporating these features into transformer models impacts performance. Our RoBERTa-based models enriched with high level linguistic features achieve consistent performance gains of up to +2.4 macro F1 on the GoEmotions benchmark, showing that explicit lexical cues can complement neural representations and improve robustness in predicting emotion categories.
Figures & tables
Title
Year
Source
Domain
Format
Size
# of labels
Affective Text ( Strapparava and Mihalcea, 2007 )
2007
News Headlines
News
Snippets
1,250 headlines
6
BU-NEmo + ( Gao, 2022 )
2022
News Headlines
News
Snippets
1,297 headlines
8
Cancer Emo ( Sosea and Caragea, 2020 )
2020
Online Cancer Survivors Network
Health
Snippets
25,000 online comments
8
CARER ( Saravia et al., 2018 )
2018
Twitter
Social Media
Snippets
664,462 tweets
6
EmoContext ( Chatterjee et al., 2019 )
2019
Conversation with Conversational Agent
Conversational Agent
Dialogues
38,424 dialogues
4
EmoProgress ( Wemmer et al., 2024 )
2024
Self-report
Psychological Self-report
Dialogues
149 dreams + 339 dialogues
10
Table 1. Overview of the datasets
Figure 1. Comparisons of the signature for the Anger and Joy labels in the GoEmotions Dataset
Model
F1 (macro)
Precision
Recall
SOTA baseline (BERT ( Kocoń et al., 2023 ) )
52.75
N/A
N/A
Reproduction baseline
49.67
51.87
48.57
RoBERTa-LexEnhance (ours)
55.18 ± 0.25
55.85 ± 0.50
57.60 ± 0.45
RoBERTa-EarlyFusion (ours)
54.98 ± 0.59
54.95 ± 0.77
57.79 ± 1.02
LLaMA 3.2 (0-shot)
0.23
9.52
0.12
LLaMA 3.2 (1-shot)
5.01
21.51
3.31
Table 2. Overall results on the GoEmotions dataset.
Model
F1 (macro)
Precision
Recall
SOTA baseline (BERT ( Prasad M., [n. d.] ) )
90.00
90.00
90.00
RoBERTa-LexEnhance (ours)
88.68
88.15
90.08
RoBERTa-EarlyFusion (ours)
88.52
87.35
90.18
Table 3. Overall results on the CARER dataset.
Figure 2. Comparison of F1 scores per emotion label across GoEmotions and Carer datasets.
Emotion recognition uses language, voice, face, body movement, and physiological signals to infer emotional categories or dimensions such as valence and arousal.
Recognition of signers' emotions suffers from one theoretical challenge and one practical challenge, namely, the overlap between grammatical and affective facial expressions and the scarcity of data for model training. This paper addresses these two challenges in a cross-lingual setting using our eJSL dataset, a new benchmark dataset for emotion recognition in Japanese Sign Language signers, and BOBSL, a large British Sign Language dataset with subtitles. In eJSL, two signers expressed 78 distinct utterances with each of seven different emotional states, resulting in 1,092 video clips. We empirically demonstrate that 1) textual emotion recognition in spoken language mitigates data scarcity in sign language, 2) temporal segment selection has a significant impact, and 3) incorporating hand motion enhances emotion recognition in signers. Finally we establish a stronger baseline than spoken language LLMs.
Kotaro Funakoshi, Yaoxiong Zhu
FIRST, Institute of Integrated Research · Institute of Science Tokyo · ICT, School of Engineering
Prosodic emphasis varies across languages, emotions, and speaking styles, yet existing emphasis detection models are largely trained and evaluated on monolingual neutral read speech. We introduce MMEE (Multilingual Multi-Emotion Emphasis), a corpus of 10,000 professionally recorded expressive utterances (14.13 hours) across 7 languages and 34 emotion/style categories, with three-level perceptual labels (10 annotations per sample). We benchmark two state-of-the-art architectures under monolingual, cross-lingual, multilingual, cross-emotion, cross-dataset, and data-scale settings. Monolingual models show limited zero-shot transfer, degrading across typologically distant languages, while multilingual training substantially improves robustness. Models transfer robustly between high- and low-arousal emotions; bidirectional transfer between synthetic and perceptual benchmarks suggests shared prosodic structure; and performance stays robust even at smaller training scales.
Large language models are routinely deployed on text that varies widely in emotional tone, yet their reasoning behavior is typically evaluated without accounting for emotion as a source of representational variation. Prior work has largely treated emotion as a prediction target, for example in sentiment analysis or emotion classification. In contrast, we study emotion as a latent factor that shapes how models attend to and reason over text. We analyze how emotional tone systematically alters attention geometry in transformer models, showing that metrics such as locality, center-of-mass distance, and entropy vary across emotions and correlate with downstream question-answering performance. To facilitate controlled study of these effects, we introduce Affect-Uniform ReAding QA (AURA-QA), a question-answering dataset with emotionally balanced, human-authored context passages. Finally, an emotional regularization framework is proposed that constrains emotion-conditioned representational drift during training. Experiments across multiple QA benchmarks demonstrate that this approach improves reading comprehension in both emotionally-varying and non-emotionally varying datasets, yielding consistent gains under distribution shift and in-domain improvements on several benchmarks.