Investigating Multilingual Instruction-Tuning: Do Polyglot Models Demand for Multilingual Instructions?
This work addresses the problem of adapting multilingual LLMs for effective use across languages, providing incremental insights into instruction-tuning strategies.
The study investigated whether multilingual instruction-tuning on parallel datasets improves cross-lingual instruction-following in multilingual LLMs, finding benefits of up to 9.9% and showing that the Superficial Alignment Hypothesis does not hold for a 7B parameter model requiring large-scale datasets.
The adaption of multilingual pre-trained LLMs into eloquent and helpful assistants is essential to facilitate their use across different language regions. In that spirit, we are the first to conduct an extensive study of the performance of multilingual models instruction-tuned on different language compositions on parallel instruction-tuning benchmarks across a selection of the most spoken Indo-European languages. We systematically examine the effects of language and instruction dataset size on a mid-sized and a large, multilingual LLMs by instruction-tuning them on parallel instruction-tuning datasets. Our results demonstrate that instruction-tuning on parallel instead of monolingual corpora benefits cross-lingual instruction following capabilities by up to 9.9%. Furthermore, we show that the Superficial Alignment Hypothesis does not hold in general, as the investigated multilingual 7B parameter model presents a counter-example requiring large-scale instruction-tuning datasets. Finally, we conduct a human annotation study to understand the alignment between human-based and GPT-4-based evaluation within multilingual chat scenarios.