Exploring prompting for dialectical machine translation: a focus on north Jordanian Arabic
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Dialectal variations are common across many languages, and dialectical machine translation to the standard form of the language or other languages is crucial for effective communication with speakers of these dialects. Prompting Large Language Models (LLMs) for Machine Translation (MT) has gained popularity. However, its efficacy for dialectical MT, particularly in comparison to fine-tuning, remains underexplored, especially for regional dialects that lack parallel training and evaluation data. This study presents a new parallel dataset between Modern Standard Arabic and the Irbid dialect, the largest city in northern Jordan, specifically within the travel domain. This dataset, an extension of the MADAR multi-dialect corpus , comprises 12,000 entries translated by native speakers of the Irbid dialect. We also describe the guidelines and evaluation process employed to collect this dataset and present several analyses within this article. Additionally, we investigate the effectiveness of prompting LLMs, particularly GPT-4o-mini, in performing MT under zero-shot and few-shot learning settings. We compare these methods to fine-tuning approaches. This includes the use of dialect-tolerant prompts and constraints. We compare these methods to fine-tuning approaches. Results indicate that prompting, particularly few-shot learning with an optimal number of exemplars, consistently outperforms fine-tuning in our tests. Utilizing several versions of T5 and mBART50 for fine-tuning, we compared their performance with that of GPT-4o-mini, which was employed for prompting. The comparative analysis reveals a notable improvement margin, with Bilingual Evaluation Understudy (BLEU), Crosslingual Optimized Metric for Evaluation of Translation (COMET), and Recall-Oriented Understudy for Gisting Evaluation–Longest Common Subsequence (ROUGE-L) scores surpassing those of the best fine-tuned model by margins of 11.89, 0.2476, and 1.18, respectively. These findings underscore the potential of Few-shot Prompting (FSP) in effectively addressing dialectical MT challenges.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it