BESSTIE: A Benchmark for Sentiment and Sarcasm Classification for Varieties of English

Authors: Dipankar Srirag, Aditya Joshi, Jordan Painter, and Diptesh Kanojia.

In Findings of the Association for Computational Linguistics: ACL 2025, Vienna, Austria.

---

Abstract

Despite large language models (LLMs) being known to exhibit bias against non-mainstream varieties, there are no known labeled datasets for sentiment analysis of English. To address this gap, we introduce BESSTIE, a benchmark for sentiment and sarcasm classification for three varieties of English: Australian (en-AU), Indian (en-IN), and British (en-UK). Using web-based content from two domains, namely, Google Place reviews and Reddit comments, we collect datasets for these language varieties using two methods: location-based and topic-based filtering. Native speakers of the language varieties manually annotate the datasets with sentiment and sarcasm labels. To assess whether the dataset accurately represents these varieties, we conduct two validation steps: (a) manual annotation of language varieties and (b) automatic language variety prediction. We perform an additional annotation exercise to validate the reliance of the annotated labels. Subsequently, we fine-tune nine large language models (LLMs) (representing a range of encoder/decoder and mono/multilingual models) on these datasets, and evaluate their performance on the two tasks. Our results reveal that the models consistently perform better on inner-circle varieties (i.e., en-AU and en-UK), with significant performance drops for en-IN, particularly in sarcasm detection. We also report challenges in cross-variety generalisation, highlighting the need for language variety-specific datasets such as ours. BESSTIE promises to be a useful evaluative benchmark for future research in equitable LLMs, specifically in terms of language varieties.

---

Dataset Statistics

The table below reports counts and label distributions for publicly available splits. The test split is withheld.

Variety	Source	Train	Validation	% Pos. Sentiment	% Pos. Sarcasm	Avg. Words
en-AU	Google	946	130	73%	7%	63.98
en-AU	Reddit	1,763	241	32%	42%	52.74
en-IN	Google	1,647	225	75%	1%	44.14
en-IN	Reddit	1,685	229	25%	13%	26.65
en-UK	Google	1,817	248	75%	0%	71.51
en-UK	Reddit	1,005	138	11%	22%	37.55
Total		8,863	1,211

---

Data Fields

Field	Type	Description
text	string	The raw review or post text
source	string	Data source — google (Google Place reviews) or reddit
variety	string	English variety — en-AU, en-IN, or en-UK
sentiment	int	Sentiment label — 0 (negative) or 1 (positive)
sarcasm	int	Sarcasm label — 0 (not sarcastic) or 1 (sarcastic)

---

Dataset Configs

Use the subset/config dropdown in the dataset viewer to filter by English variety. Each config contains train and validation splits.

Config	Variety	Description
en_AU	Australian English	Google reviews and Reddit posts from Australia
en_IN	Indian English	Google reviews and Reddit posts from India
en_UK	British English	Google reviews and Reddit posts from the United Kingdom

from datasets import load_dataset

# Load a specific variety
ds = load_dataset("unswnlporg/BESSTIE", "en_AU")

# Load all varieties
for config in ["en_AU", "en_IN", "en_UK"]:
    ds = load_dataset("unswnlporg/BESSTIE", config)

---

Intended Use

BESSTIE is intended for:

• Benchmarking sentiment analysis and sarcasm detection models across World Englishes
• Studying variety-specific linguistic phenomena in NLP
• Evaluating cross-variety generalisation of pre-trained language models

The test split is withheld to maintain benchmark integrity. Researchers wishing to evaluate on the test set should contact the authors.

---

Citation

If you use BESSTIE in your research, please cite:

@inproceedings{srirag-etal-2025-besstie,
    title = "{BESSTIE}: A Benchmark for Sentiment and Sarcasm Classification for Varieties of {E}nglish",
    author = "Srirag, Dipankar  and
      Joshi, Aditya  and
      Painter, Jordan  and
      Kanojia, Diptesh",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Findings of the Association for Computational Linguistics: ACL 2025",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.findings-acl.441/",
    doi = "10.18653/v1/2025.findings-acl.441",
    pages = "8413--8429",
    ISBN = "979-8-89176-256-5",
    abstract = "Despite large language models (LLMs) being known to exhibit bias against non-mainstream varieties, there are no known labeled datasets for sentiment analysis of English. To address this gap, we introduce BESSTIE, a benchmark for sentiment and sarcasm classification for three varieties of English: Australian (en-AU), Indian (en-IN), and British (en-UK). Using web-based content from two domains, namely, Google Place reviews and Reddit comments, we collect datasets for these language varieties using two methods: location-based and topic-based filtering. Native speakers of the language varieties manually annotate the datasets with sentiment and sarcasm labels. To assess whether the dataset accurately represents these varieties, we conduct two validation steps: (a) manual annotation of language varieties and (b) automatic language variety prediction. We perform an additional annotation exercise to validate the reliance of the annotated labels. Subsequently, we fine-tune nine large language models (LLMs) (representing a range of encoder/decoder and mono/multilingual models) on these datasets, and evaluate their performance on the two tasks. Our results reveal that the models consistently perform better on inner-circle varieties (i.e., en-AU and en-UK), with significant performance drops for en-IN, particularly in sarcasm detection. We also report challenges in cross-variety generalisation, highlighting the need for language variety-specific datasets such as ours. BESSTIE promises to be a useful evaluative benchmark for future research in equitable LLMs, specifically in terms of language varieties. The BESSTIE dataset is publicly available at: \url{https://huggingface.co/datasets/unswnlporg/BESSTIE}."
}

---

License

This dataset is released under CC BY 4.0.