id stringlengths 16 22 | prompt stringlengths 13 29.7k | task stringclasses 3
values | question_lang stringclasses 24
values |
|---|---|---|---|
sum-sum_13_eng_eng | The following is the full text of an *ACL paper in English. Write the paper's abstract in around 200 words, using formal third-person academic prose.
PAPER:
## 1 Introduction
Large Language Models (LLMs) (Yan et al. , 2025; Wei et al. , 2025; Yu et al. , 2026; Xin et al. , 2025; Chen et al. , 2024; Gong and Sun , 202... | sum-sum | eng |
sum-sum_66_eng_eng | The following is the full text of an *ACL paper in English. Write the paper's abstract in around 200 words, using formal third-person academic prose.
PAPER:
## 1 Introduction
The chain-of-thought prompting (CoT; Wei et al. , 2022; Nye et al. , 2022) has become the de facto standard for achieving the best performance ... | sum-sum | eng |
sum-sum_51_eng_eng | Write a polished abstract for the *ACL paper provided below. Aim for around 200 words in formal third-person academic prose.
PAPER:
## 1 Introduction
Large Language Models (LLMs) advance a wide range of language tasks (Devlin et al. , 2019; Radford et al. , 2019; Zhang et al. , 2022b; Team et al. , 2024; Touvron et a... | sum-sum | eng |
sum-sum_55_eng_eng | Read the full text of an *ACL paper below and write a ~200-word abstract for it. Use formal, third-person academic prose.
PAPER:
## 1 Introduction
Traditional spoken dialogue systems are typically half-duplex, alternating turns between user and system. While simple, this design struggles to capture the natural dynami... | sum-sum | eng |
sum-sum_15_eng_eng | Read the full text of an *ACL paper below and write a ~200-word abstract for it. Use formal, third-person academic prose.
PAPER:
## 1 Introduction
Text embedding models (Wang et al. , 2024a; Sturua et al. , 2024; Nussbaum et al. , 2024) encode textual inputs into vector spaces. These models enable efficient semantic ... | sum-sum | eng |
sum-sum_43_eng_eng | Read the full text of an *ACL paper below and write a ~200-word abstract for it. Use formal, third-person academic prose.
PAPER:
## 1 Introduction
So called "Big AI" (Muldoon et al. , 2024, p12) – the corporations, the state capture, the "various kinds of automation sold as AI" (Bender and Hanna , 2025, p162) – is es... | sum-sum | eng |
sum-sum_16_eng_eng | The following is the full text of an *ACL paper in English. Write the paper's abstract in around 200 words, using formal third-person academic prose.
PAPER:
## 1 Introduction
Event Factuality Prediction (EFP) is the cornerstone of robust Natural Language Understanding, aiming to determine the veridical status of even... | sum-sum | eng |
sum-sum_53_eng_eng | Summarize the following *ACL paper into a well-formed abstract of around 200 words. Write in formal third-person academic prose.
PAPER:
## 1 Introduction
Transformers (Vaswani et al. , 2017) frequently concentrate attention on an early position in a way that is largely insensitive to content. This attention sink has ... | sum-sum | eng |
sum-sum_25_eng_eng | The following is the full text of an *ACL paper in English. Write the paper's abstract in around 200 words, using formal third-person academic prose.
PAPER:
## 1 Introduction
Transformers (Vaswani et al. , 2017) routinely display an attention sink: a persistent tendency to allocate disproportionate attention mass to ... | sum-sum | eng |
sum-sum_22_eng_eng | "Summarize the following *ACL paper into a well-formed abstract of around 200 words. Write in formal(...TRUNCATED) | sum-sum | eng |
End of preview. Expand in Data Studio
WMT26 MIST Shared Task Test Release
Submit by 01 August 2026, AoE. For more details, please refer to our website.
Update log
15 July 2026: We fixed an issue where the English QA-OEG prompts are empty. If you downloaded the data on or before 15 July, please kindly re-download it just in case.
14 July 2026: Test set release: 24 languages and 3 sub-tasks.
Test data breakdown
Our test data is released in the JSONL format hosted on Hugging Face, where each line is a JSON object with the following fields:
{
"id": # this is the unique identifier for the test instance;
"prompt": # this is the input to your participating system - it is self-contained, including all contexts, constraints, and the question;
"task": # the task type, one of "qa-oeg", "qa-context", or "sum-sum";
"question_lang": the language of the question.
}
Each id consists of four parts, for example "qa-oeg_44_zho_zho", "qa-context_2_ita_spa", or "sum-sum_64_ind_eng". This is a concatenation of the task field, an integer identifier, the language of the question, and the language of the context (if available):
- task, this is one of "qa-oeg" (open-ended QA), "qa-context" (context-based QA), or "sum-sum" (summarization);
- an integer identifier;
- language of the question (and almost always the language of the expected output);
- language of the context, if available; otherwise, the same as the question language.
Each prompt is self-contained and includes all the context and question. If you are participating in a subset of the tasks or language, you can filter the test set by task or question_lang.
- Downloads last month
- 31