IC-1238Llama-2-13b produces off-target non-translation outputs in zero-shot English-to-foreign-language translation

Haoran Xu, Young Jin Kim, Amr Sharaf, Hany Hassan Awadalla

SourceA Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models

In zero-shot en-to-xx translation, Llama-2-13b frequently generates text unrelated to the translation task, such as 'I'm voting to close this question as off-topic because it is not about programming' when asked to translate English to Russian. This causes BLEU scores near zero for en-to-cs (0.87) and en-to-ru (0.59). The issue is substantially alleviated by using prompts in the target language (BLEU rises to 20.80 for en-to-cs) or by few-shot in-context learning, but both strategies still lag behind the authors' fine-tuning method by over 10 BLEU and COMET points.

Evidence
observational
Key metric
Llama-2-13b zero-shot en-to-cs BLEU 0.87, en-to-ru BLEU 0.59; with target-language prompt en-to-cs BLEU 20.80, en-to-ru BLEU 22.66
Caveat
The paper attributes the issue to the pre-training phase but does not provide a mechanistic explanation; the off-target behavior is observed in the zero-shot setting and is not quantified across all test sentences.
Model
Llama 2 / Llama 2 base Llama 2 13B
Concepts
Failure mode
Methods
BLEU / BLEU@4 [eval], COMET [eval]
Related findings
IC-1236, IC-1237, IC-1239
Extraction
automatic-extraction