README.md

---
license: mit
task_categories:
- text-generation
language:
- en
tags:
- dialect
- english-varieties
- dpo
- preference-dataset
size_categories:
- 10K<n<100K
---

# DiaLLM — Pooled Preference Dataset (Implicit Thread)

Part of **DiaLLM: An Investigation into the Robustness-Generation Gap in
English Dialect Adaptation** (EMNLP 2026 Main).

![DiaLLM pipeline](https://raw.githubusercontent.com/surrey-nlp/diallm/main/assets/diallm_pipeline.png)

45,690 preference pairs, pooling all three variety-specific sets
([Australian](https://huggingface.co/datasets/surrey-nlp/alignment-australian-final),
[Northern British](https://huggingface.co/datasets/surrey-nlp/alignment-british-final),
[Indian](https://huggingface.co/datasets/surrey-nlp/alignment-indian-final)) without
variety targeting. Used for **implicit-thread** DPO training, where the three
varieties are pooled rather than targeted individually, preserving the
variety-agnostic objective of that thread.

## Construction

Built from the UltraFeedback preference dataset ([Cui et al., 2023](https://arxiv.org/abs/2310.01377)),
via Argilla's cleaned/binarized release ([`argilla/ultrafeedback-binarized-preferences-cleaned`](https://huggingface.co/datasets/argilla/ultrafeedback-binarized-preferences-cleaned), MIT licensed):
the originally-preferred completion is transformed into a dialectal variant
using [Multi-VALUE](https://github.com/SALT-NLP/multi-value) ([Ziems et al., 2023](https://arxiv.org/abs/2212.08011)),
based on eWAVE morphosyntactic features. Code blocks are preserved verbatim
during conversion.

## Columns

| Column | Description |
|---|---|
| `prompt` | Original UltraFeedback prompt (standard English, unmodified) |
| `chosen` | Multi-VALUE dialectal variant of the preferred completion |
| `rejected` | Original standard-English form of that completion |
| `prompt_dialect_density` | eWAVE-based dialect feature density of the prompt |
| `chosen_dialect_density` | eWAVE-based dialect feature density of the chosen completion |

**Code, checkpoints, linguistic-analysis toolkit:** https://github.com/surrey-nlp/diallm

**Paper:** https://arxiv.org/abs/2607.07669

## Citation

```bibtex
@article{painter2026diallm,
  title     = {DiaLLM: An Investigation into the Robustness-Generation Gap in English Dialect Adaptation},
  author    = {Painter, Jordan and Srirag, Dipankar and Kappiyath, Adarsh and Kanojia, Diptesh and Joshi, Aditya and Yin, Lu},
  year      = {2026},
  eprint    = {2607.07669},
  archivePrefix = {arXiv}
}
```