Results of the summarizing process for the arXiv paper: 2310.14855v1

In this study, we investigate the use of Large Language Models (LLM's) for Neural Machine Translation (NMT) and explore parameter-efficient fine-tuning techniques. While LLM's have shown success in various Natural Language Processing tasks, they have not yet achieved state-of-the-art performance in NMT. Our initial experiments found that fine-tuning for translation purposes led to performance degradation. To address this issue, we propose adapting LLM's as Automatic Post-Editors (APE) rather than direct translators. By leveraging their exceptional ability to process and generate lengthy sequences, we extend our approach to document-level translation. We also introduce Low-Rank-Adapter fine-tuning for APEs and observe significant improvements across both sentence and document-level metrics, even generalizing to out-of-domain data. Notably, our method achieves a state-of-the-art accuracy rate of 89% on the ContraPro test set which specifically assesses pronoun ambiguities when translating from English to German. Furthermore, we investigate a practical scenario involving manual post-editing for document-level translation with reference context provided. Our results demonstrate that incorporating human corrections can significantly reduce the number of edits required for subsequent translations. We explore various approaches such as Chunk-Based and Batched Sliding Window to enhance the translation process. Overall, our study highlights the potential of using LLM's as APEs for improving translation quality at both sentence and document levels. The findings suggest that integrating manual feedback can further enhance the efficiency and accuracy of machine translation systems.

• - Investigating use of Large Language Models (LLM's) for Neural Machine Translation (NMT)
• - Initial experiments showed performance degradation when fine-tuning LLM's for translation
• - Proposing adapting LLM's as Automatic Post-Editors (APE) instead of direct translators
• - Introducing Low-Rank-Adapter fine-tuning for APEs, leading to significant improvements in metrics and out-of-domain data generalization
• - Achieving state-of-the-art accuracy rate of 89% on ContraPro test set assessing pronoun ambiguities in English to German translation
• - Demonstrating the effectiveness of manual post-editing for document-level translation with reference context provided
• - Exploring Chunk-Based and Batched Sliding Window approaches to enhance translation process
• - Highlighting potential of using LLM's as APEs to improve translation quality at both sentence and document levels

Summary1. Scientists are studying big language models for helping computers translate languages better. 2. When they tried to make the models better at translating, the results got worse at first. 3. They suggest using the models to fix mistakes in translations instead of doing direct translations. 4. By making some adjustments, they made the fixing process much better and more accurate. 5. They did really well on a test that checks if pronouns are translated correctly from English to German. Definitions- Large Language Models (LLM's): Advanced computer programs that help with understanding and generating human languages. - Neural Machine Translation (NMT): Using artificial intelligence to automatically translate text from one language to another. - Automatic Post-Editors (APE): Tools that correct mistakes in translated text automatically. - Low-Rank-Adapter fine-tuning: A method of adjusting language models to improve their performance in specific tasks. - State-of-the-art accuracy rate: The highest level of correctness achieved compared to other methods or tools available. - ContraPro test set: A specific evaluation tool used for testing translation quality by focusing on pronoun ambiguities. - Manual post-editing: Correcting errors in translated text by hand after it has been generated by a machine translation system.

Large Language Models (LLM's) have been gaining attention in the field of Natural Language Processing (NLP) due to their exceptional ability to process and generate lengthy sequences. These models have shown success in various NLP tasks such as language generation, question-answering, and text summarization. However, their performance in Neural Machine Translation (NMT) has not yet reached state-of-the-art levels. In this study, we investigate the use of LLM's for NMT and explore parameter-efficient fine-tuning techniques. Our initial experiments found that directly fine-tuning LLM's for translation purposes led to a decrease in performance. To address this issue, we propose adapting LLM's as Automatic Post-Editors (APEs) instead of direct translators. The idea behind using LLM's as APEs is based on their ability to process and generate long sequences of text. This makes them well-suited for post-editing tasks where they can correct errors or improve translations generated by other systems. We extend our approach to document-level translation, where the input consists of multiple sentences rather than just one sentence. To further improve the performance of APEs, we introduce Low-Rank-Adapter fine-tuning. This technique involves adding low-rank adapters between layers of an existing pre-trained model during fine-tuning. Our experiments show significant improvements across both sentence and document-level metrics with this approach, even when tested on out-of-domain data. One notable result from our study is achieving a state-of-the-art accuracy rate of 89% on the ContraPro test set which specifically assesses pronoun ambiguities when translating from English to German. This demonstrates the effectiveness of using LLM's as APEs for improving translation quality at both sentence and document levels. Furthermore, we also investigate a practical scenario involving manual post-editing for document-level translation with reference context provided. In this scenario, human corrections are incorporated into the translation process, which can significantly reduce the number of edits required for subsequent translations. We explore various approaches such as Chunk-Based and Batched Sliding Window to enhance this process. Our findings highlight the potential of using LLM's as APEs for improving translation quality at both sentence and document levels. The results also suggest that integrating manual feedback can further enhance the efficiency and accuracy of machine translation systems. In conclusion, our study showcases the benefits of leveraging LLM's as APEs in NMT tasks. By utilizing their exceptional ability to process and generate lengthy sequences, we were able to achieve state-of-the-art performance on specific test sets and improve overall translation quality at both sentence and document levels. Our findings also demonstrate the potential of incorporating human corrections into machine translation systems to further enhance their efficiency and accuracy.