Machine Translation Digest for Jul 19 2025
Here is today's selection of cs.CL papers exploring diverse aspects of language model advancements and their applications. The common themes include enhancing model performance through data augmentation, improving document understanding, and ethical considerations in cross-lingual model evaluations. These studies highlight the evolving landscape of language models in tackling specific language and document-related challenges.
Backtranslation and paraphrasing in the LLM era? Comparing data augmentation methods for emotion classification
Numerous domain-specific machine learning tasks struggle with data scarcity and class imbalance. This paper systematically explores data augmentation methods for NLP, particularly through large language models like GPT. The purpose of this paper is to examine and evaluate whether traditional methods such as paraphrasing and backtranslation can leverage a new generation of models to achieve comparable performance to purely generative methods. Methods aimed at solving the problem of data scarcity and utilizing ChatGPT were chosen, as well as an exemplary dataset. We conducted a series of experiments comparing four different approaches to data augmentation in multiple experimental setups. We then evaluated the results both in terms of the quality of generated data and its impact on classification performance. The key findings indicate that backtranslation and paraphrasing can yield comparable or even better results than zero and a few-shot generation of examples.
Docopilot: Improving Multimodal Models for Document-Level Understanding
Despite significant progress in multimodal large language models (MLLMs), their performance on complex, multi-page document comprehension remains inadequate, largely due to the lack of high-quality, document-level datasets. While current retrieval-augmented generation (RAG) methods offer partial solutions, they suffer from issues, such as fragmented retrieval contexts, multi-stage error accumulation, and extra time costs of retrieval. In this work, we present a high-quality document-level dataset, Doc-750K, designed to support in-depth understanding of multimodal documents. This dataset includes diverse document structures, extensive cross-page dependencies, and real question-answer pairs derived from the original documents. Building on the dataset, we develop a native multimodal model, Docopilot, which can accurately handle document-level dependencies without relying on RAG. Experiments demonstrate that Docopilot achieves superior coherence, accuracy, and efficiency in document understanding tasks and multi-turn interactions, setting a new baseline for document-level multimodal understanding. Data, code, and models are released at https://github.com/OpenGVLab/Docopilot
Optimizing Legal Document Retrieval in Vietnamese with Semi-Hard Negative Mining
Large Language Models (LLMs) face significant challenges in specialized domains like law, where precision and domain-specific knowledge are critical. This paper presents a streamlined two-stage framework consisting of Retrieval and Re-ranking to enhance legal document retrieval efficiency and accuracy. Our approach employs a fine-tuned Bi-Encoder for rapid candidate retrieval, followed by a Cross-Encoder for precise re-ranking, both optimized through strategic negative example mining. Key innovations include the introduction of the Exist@m metric to evaluate retrieval effectiveness and the use of semi-hard negatives to mitigate training bias, which significantly improved re-ranking performance. Evaluated on the SoICT Hackathon 2024 for Legal Document Retrieval, our team, 4Huiter, achieved a top-three position. While top-performing teams employed ensemble models and iterative self-training on large bge-m3 architectures, our lightweight, single-pass approach offered a competitive alternative with far fewer parameters. The framework demonstrates that optimized data processing, tailored loss functions, and balanced negative sampling are pivotal for building robust retrieval-augmented systems in legal contexts.
Mangosteen: An Open Thai Corpus for Language Model Pretraining
Pre-training data shapes a language model's quality, but raw web text is noisy and demands careful cleaning. Existing large-scale corpora rely on English-centric or language-agnostic pipelines whose heuristics do not capture Thai script or cultural nuances, leaving risky material such as gambling content untreated. Prior Thai-specific efforts customize pipelines or build new ones, yet seldom release their data or document design choices, hindering reproducibility and raising the question of how to construct a transparent, high-quality Thai corpus. We introduce Mangosteen: a 47 billion-token Thai corpus built through a Thai-adapted Dolma pipeline that includes custom rule-based language ID, revised C4/Gopher quality filters, and Thai-trained content filters, plus curated non-web sources such as Wikipedia, Royal Gazette texts, OCR-extracted books, and CC-licensed YouTube subtitles. Systematic ablations using GPT-2 show the pipeline trims CommonCrawl from 202M to 25M documents while raising SEA-HELM NLG from 3 to 11; an 8B-parameter SEA-LION model continually pre-trained on Mangosteen then surpasses SEA-LION-v3 and Llama-3.1 by about four points on Thai benchmarks. We release the full pipeline code, cleaning manifests, corpus snapshot, and all checkpoints, providing a fully reproducible foundation for future Thai and regional LLM research.
Rethinking Suicidal Ideation Detection: A Trustworthy Annotation Framework and Cross-Lingual Model Evaluation
Suicidal ideation detection is critical for real-time suicide prevention, yet its progress faces two under-explored challenges: limited language coverage and unreliable annotation practices. Most available datasets are in English, but even among these, high-quality, human-annotated data remains scarce. As a result, many studies rely on available pre-labeled datasets without examining their annotation process or label reliability. The lack of datasets in other languages further limits the global realization of suicide prevention via artificial intelligence (AI). In this study, we address one of these gaps by constructing a novel Turkish suicidal ideation corpus derived from social media posts and introducing a resource-efficient annotation framework involving three human annotators and two large language models (LLMs). We then address the remaining gaps by performing a bidirectional evaluation of label reliability and model consistency across this dataset and three popular English suicidal ideation detection datasets, using transfer learning through eight pre-trained sentiment and emotion classifiers. These transformers help assess annotation consistency and benchmark model performance against manually labeled data. Our findings underscore the need for more rigorous, language-inclusive approaches to annotation and evaluation in mental health natural language processing (NLP) while demonstrating the questionable performance of popular models with zero-shot transfer learning. We advocate for transparency in model training and dataset construction in mental health NLP, prioritizing data and model reliability.
| Curated by yukajii.com |