Zero-shot Translation Prompt

[10zinten]

Zero-shot Translation Prompt

This is our baseline for comparing other experiments.
LLM: claude sonet 3.5
Experiment Name: 01_zero_shot_translation

Hypothesis

None

Prompt

Translate the following Buddhist Tibetan passage into English: {tibetan_text} English:
## Core Instructions
   - Enclose final English translation in <t> tags

Evaluation Results

Corpus BLEU Score: 18.36
Corpus chrF++ Score: 44.11
Corpus TER Score: 100.55

Output:

use the experiment name to find the translation output from results.json

[peterolin]

Would it make sense in this and other prompts to add a request for Claude to highlight any ambiguities, for instance by adding something like the following to all prompts?

"I would like you to list any points where the interpretation is ambiguous and an unfounded choice of rendering has been made, list these as items within and tag. Summarize with a number of such issues between tags. Additionaly rank this translation from 1-10 based on uncertainties in interpretation and potential incorrenctness in rendering."