使用 DeepSeek V3 和 R1 模型。 Prompt 是
A literature review on the impact of LLM on the translation industry, including interpretation, translation, video translation, etc
然后 Coordinator 把任务交给 Planner：
INFO [src.workflow] Starting workflow with user input: A literature review on the impact of LLM on the translation industry, including interpretation, translation, video translation, etc
INFO [src.graph.nodes] Coordinator talking.
DEBUG [src.graph.nodes] Current state messages: [HumanMessage(content='A literature review on the impact of LLM on the translation industry, including interpretation, translation, video translation, etc', additional_kwargs={}, response_metadata={}, id='597a842a-9618-4815-9f53-7a7b61cc7a7c')]
DEBUG [src.graph.nodes] Coordinator response: handoff_to_planner()
INFO [src.graph.nodes] Planner generating full plan
DEBUG [src.tools.decorators] Tool TavilySearchResults._run called with parameters: query=A literature review on the impact of LLM on the translation industry, including interpretation, translation, video translation, etc
接下来 Planner 调用 Tavily 在互联网上初步爬取信息：
DEBUG [src.tools.decorators] Tool TavilySearchResults._run called with parameters: query=A literature review on the impact of LLM on the translation industry, including interpretation, translation, video translation, etc
DEBUG [src.tools.decorators] Tool TavilySearchResults returned: ([{'title': 'The impact of LLMs on translation: a new impetus for evolving ...', 'url': 'https://www.rws.com/blog/the-evolution-of-translation/', 'content': 'A new type of natural language processing engine, called the Large Language Model, was being introduced to a diverse, global community of users.', 'score': 0.55796635}, {'title': 'Large Language Models No Match for Humans in Literary ... - Slator', 'url': 'https://slator.com/large-language-models-no-match-for-humans-in-literary-translation-study-finds/', 'content': 'They found a “substantial gap” between the quality of human and machine-generated literary translations, as “LLMs tend to produce more literal and less diverse', 'score': 0.41087836800000005}, {'title': 'The Role of Large Language Models in Machine Translation - Medium', 'url': 'https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d', 'content': 'Large language models (LLMs) have played a pivotal role in advancing machine translation technology, improving translation quality, speed, and versatility.', 'score': 0.33501504000000004}, {'title': 'The Path to LLM-based Machine Translation - Imminent', 'url': 'https://imminent.translated.com/llm-based-machine-translation', 'content': 'This article provides an in-depth analysis of the opportunities and current challenges this technology faces and a prediction of its emerging use in the near', 'score': 0.261809499}, {'title': 'If we train a LLM on a large amount of textual data in ... - Quora', 'url': 'https://www.quora.com/If-we-train-a-LLM-on-a-large-amount-of-textual-data-in-different-languages-but-the-data-contains-no-pieces-of-information-regarding-how-to-translate-one-into-the-other-or-examples-of-translated-text-will-the-LLM-be', 'content': "The demand for translators (and interpreters) will amost certainly diminish over time as AI's ability to translate and interpret improves,", 'score': 0.20677224}], {'query': 'A literature review on the impact of LLM on the translation industry, including interpretation, translation, video translation, etc', 'follow_up_questions': None, 'answer': None, 'images': [], 'results': [{'url': 'https://www.rws.com/blog/the-evolution-of-translation/', 'title': 'The impact of LLMs on translation: a new impetus for evolving ...', 'content': 'A new type of natural language processing engine, called the Large Language Model, was being introduced to a diverse, global community of users.', 'score': 0.55796635, 'raw_content': None}, {'url': 'https://slator.com/large-language-models-no-match-for-humans-in-literary-translation-study-finds/', 'title': 'Large Language Models No Match for Humans in Literary ... - Slator', 'content': 'They found a “substantial gap” between the quality of human and machine-generated literary translations, as “LLMs tend to produce more literal and less diverse', 'score': 0.41087836800000005, 'raw_content': None}, {'url': 'https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d', 'title': 'The Role of Large Language Models in Machine Translation - Medium', 'content': 'Large language models (LLMs) have played a pivotal role in advancing machine translation technology, improving translation quality, speed, and versatility.', 'score': 0.33501504000000004, 'raw_content': None}, {'url': 'https://imminent.translated.com/llm-based-machine-translation', 'title': 'The Path to LLM-based Machine Translation - Imminent', 'content': 'This article provides an in-depth analysis of the opportunities and current challenges this technology faces and a prediction of its emerging use in the near', 'score': 0.261809499, 'raw_content': None}, {'url': 'https://www.quora.com/If-we-train-a-LLM-on-a-large-amount-of-textual-data-in-different-languages-but-the-data-contains-no-pieces-of-information-regarding-how-to-translate-one-into-the-other-or-examples-of-translated-text-will-the-LLM-be', 'title': 'If we train a LLM on a large amount of textual data in ... - Quora', 'content': "The demand for translators (and interpreters) will amost certainly diminish over time as AI's ability to translate and interpret improves,", 'score': 0.20677224, 'raw_content': None}], 'response_time': 1.83})
DEBUG [src.graph.nodes] Current state messages: [HumanMessage(content='A literature review on the impact of LLM on the translation industry, including interpretation, translation, video translation, etc', additional_kwargs={}, response_metadata={}, id='597a842a-9618-4815-9f53-7a7b61cc7a7c')]
Planner 分析了初步爬取的信息并回应：I need to conduct a comprehensive literature review on the impact of Large Language Models (LLMs) on the translation industry, covering interpretation, text translation, video translation, and related areas. This requires gathering academic and industry insights, analyzing technological advancements, and evaluating challenges like translation quality and human-machine collaboration.
接下来，Planner 拆解了这个任务并生成了 Researcher、Browser、Reporter 几个 Agent 用于完成拆解的任务。
DEBUG [src.graph.nodes] Planner response: {
"thought": "I need to conduct a comprehensive literature review on the impact of Large Language Models (LLMs) on the translation industry, covering interpretation, text translation, video translation, and related areas. This requires gathering academic and industry insights, analyzing technological advancements, and evaluating challenges like translation quality and human-machine collaboration.",
"title": "Literature Review on LLM Impact in Translation Industry",
"steps": [
{
"agent_name": "researcher",
"title": "Gather foundational literature",
"description": "Search academic databases (e.g., IEEE, arXiv) and industry reports to collect peer-reviewed articles, case studies, and white papers on LLM applications in translation, interpretation, and video translation. Focus on sub-topics: quality comparison (human vs. LLM), efficiency gains, and niche challenges (e.g., literary translation)."
},
{
"agent_name": "browser",
"title": "Extract real-world examples",
"description": "Interact with platforms like GitHub (to find LLM-based translation tools), LinkedIn (industry expert opinions), and video platforms (e.g., YouTube auto-translate features) to gather practical use cases and community discussions.",
"note": "Prioritize open-source projects and professional forums for credible insights."
},
{
"agent_name": "researcher",
"title": "Analyze employment and ethical impacts",
"description": "Investigate studies or surveys on how LLMs affect translator jobs, freelance markets, and ethical concerns (e.g., bias in translations). Cross-reference findings from the initial search results about 'diminishing demand for translators.'"
},
{
"agent_name": "reporter",
"title": "Compile final report",
"description": "Synthesize all data into a structured report, highlighting key trends (e.g., LLMs enabling faster video localization), limitations (e.g., literal translations in literature), and future directions (e.g., hybrid human-LLM workflows)."
}
]
}
Supervisor 评估后续工作，把任务托管给 Researcher agent。
INFO [src.graph.nodes] Supervisor evaluating next action
DEBUG [src.graph.nodes] Current state messages: [HumanMessage(content='A literature review on the impact of LLM on the translation industry, including interpretation, translation, video translation, etc', additional_kwargs={}, response_metadata={}, id='597a842a-9618-4815-9f53-7a7b61cc7a7c'), HumanMessage(content='{"thought": "I need to conduct a comprehensive literature review on the impact of Large Language Models (LLMs) on the translation industry, covering interpretation, text translation, video translation, and related areas. This requires gathering academic and industry insights, analyzing technological advancements, and evaluating challenges like translation quality and human-machine collaboration.", "title": "Literature Review on LLM Impact in Translation Industry", "steps": [{"agent_name": "researcher", "title": "Gather foundational literature", "description": "Search academic databases (e.g., IEEE, arXiv) and industry reports to collect peer-reviewed articles, case studies, and white papers on LLM applications in translation, interpretation, and video translation. Focus on sub-topics: quality comparison (human vs. LLM), efficiency gains, and niche challenges (e.g., literary translation)."}, {"agent_name": "browser", "title": "Extract real-world examples", "description": "Interact with platforms like GitHub (to find LLM-based translation tools), LinkedIn (industry expert opinions), and video platforms (e.g., YouTube auto-translate features) to gather practical use cases and community discussions.", "note": "Prioritize open-source projects and professional forums for credible insights."}, {"agent_name": "researcher", "title": "Analyze employment and ethical impacts", "description": "Investigate studies or surveys on how LLMs affect translator jobs, freelance markets, and ethical concerns (e.g., bias in translations). Cross-reference findings from the initial search results about 'diminishing demand for translators.'"}, {"agent_name": "reporter", "title": "Compile final report", "description": "Synthesize all data into a structured report, highlighting key trends (e.g., LLMs enabling faster video localization), limitations (e.g., literal translations in literature), and future directions (e.g., hybrid human-LLM workflows)."}]}', additional_kwargs={}, response_metadata={}, name='planner', id='8236b05e-e869-47ae-9332-98bd71d9660a')]
DEBUG [src.graph.nodes] Supervisor response: {'next': 'researcher'}
INFO [src.graph.nodes] Supervisor delegating to: researcher
Researcher agent 开始做事，爬取相关信息：
INFO [src.graph.nodes] Research agent starting task
DEBUG [src.tools.decorators] Tool TavilySearchResults._run called with parameters: query=impact of Large Language Models on translation industry
DEBUG [src.tools.decorators] Tool TavilySearchResults._run called with parameters: query=LLM applications in interpretation and video translation
DEBUG [src.tools.decorators] Tool TavilySearchResults._run called with parameters: query=LLM impact on translator jobs and ethical concerns
DEBUG [src.tools.decorators] Tool TavilySearchResults returned: ([{'title': 'The Role of Large Language Models in Machine Translation - Medium', 'url': 'https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d', 'content': "However, as LLMs become more powerful, it's crucial to address ethical concerns such as bias, fairness, and transparency.", 'score': 0.8312667}, {'title': '[PDF] A Study on Ethical Issues and Mitigation Strategies in Translation ...', 'url': 'https://ijemh.com/issue_dcp/A%20Study%20on%20Ethical%20Issues%20and%20Mitigation%20Strategies%20in%20Translation%20with%20Large%20Language%20Models.pdf', 'content': 'By examining the technical principles of LLMs and translation ethics theory, this study provides a detailed analysis of the ethical issues and', 'score': 0.6583124930232558}, {'title': 'The Ethical Considerations of Large Language Models (LLMs)', 'url': 'https://www.linkedin.com/pulse/ethical-considerations-large-language-models-llms-navigating-jatan-kmslc', 'content': 'Ethical Concerns: · Bias and Discrimination : LLMs can perpetuate existing biases, reinforcing harmful stereotypes and prejudices.', 'score': 0.619895876744186}, {'title': 'The Ethical Implications of Large Language Models in AI', 'url': 'https://www.computer.org/publications/tech-news/trends/ethics-of-large-language-models-in-ai/', 'content': 'Large Language Models (LLMs) offer transformative potential but raise concerns about privacy, bias, misinformation, and societal impact.', 'score': 0.5400343604651162}, {'title': 'LLMs for Translation: Benefits, Challenges, and Use Cases', 'url': 'https://botpenguin.com/blogs/llms-for-translation-benefits-challenges-and-use-cases', 'content': 'Quality Concerns for Less Common Languages\u200b\u200b An LLM model for translation can produce less accurate outputs for less common languages. Also,', 'score': 0.510113634883721}], {'query': 'LLM impact on translator jobs and ethical concerns', 'follow_up_questions': None, 'answer': None, 'images': [], 'results': [{'url': 'https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d', 'title': 'The Role of Large Language Models in Machine Translation - Medium', 'content': "However, as LLMs become more powerful, it's crucial to address ethical concerns such as bias, fairness, and transparency.", 'score': 0.8312667, 'raw_content': None}, {'url': 'https://ijemh.com/issue_dcp/A%20Study%20on%20Ethical%20Issues%20and%20Mitigation%20Strategies%20in%20Translation%20with%20Large%20Language%20Models.pdf', 'title': '[PDF] A Study on Ethical Issues and Mitigation Strategies in Translation ...', 'content': 'By examining the technical principles of LLMs and translation ethics theory, this study provides a detailed analysis of the ethical issues and', 'score': 0.6583124930232558, 'raw_content': None}, {'url': 'https://www.linkedin.com/pulse/ethical-considerations-large-language-models-llms-navigating-jatan-kmslc', 'title': 'The Ethical Considerations of Large Language Models (LLMs)', 'content': 'Ethical Concerns: · Bias and Discrimination : LLMs can perpetuate existing biases, reinforcing harmful stereotypes and prejudices.', 'score': 0.619895876744186, 'raw_content': None}, {'url': 'https://www.computer.org/publications/tech-news/trends/ethics-of-large-language-models-in-ai/', 'title': 'The Ethical Implications of Large Language Models in AI', 'content': 'Large Language Models (LLMs) offer transformative potential but raise concerns about privacy, bias, misinformation, and societal impact.', 'score': 0.5400343604651162, 'raw_content': None}, {'url': 'https://botpenguin.com/blogs/llms-for-translation-benefits-challenges-and-use-cases', 'title': 'LLMs for Translation: Benefits, Challenges, and Use Cases', 'content': 'Quality Concerns for Less Common Languages\u200b\u200b An LLM model for translation can produce less accurate outputs for less common languages. Also,', 'score': 0.510113634883721, 'raw_content': None}], 'response_time': 1.5})
DEBUG [src.tools.decorators] Tool TavilySearchResults returned: ([{'title': 'The Role of Large Language Models in Machine Translation - Medium', 'url': 'https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d', 'content': 'LLMs can translate massive amounts of text quickly and consistently, making them ideal for real-time applications like chatbots or live video', 'score': 0.69845337}, {'title': 'Top 10 Real-Life Applications of Large Language Models - PixelPlex', 'url': 'https://pixelplex.io/blog/llm-applications/', 'content': 'Large language models (LLMs) are advanced artificial intelligence systems designed to understand, interpret, and generate human-like text. They are a type of deep learning model, to be exact, that have been trained on a wide variety of internet texts. LLM applications can perform numerous tasks including writing essays, creating poetry, coding, and even engaging in general conversation. [...] LLM applications are increasingly being used in education to personalize learning and provide tutoring.\nLLMs can adapt to individual student’s learning styles and pace, offering customized explanations and feedback. For instance, a model can generate interactive reading materials that adjust complexity based on the student’s comprehension level or provide real-time language translation to aid foreign students. [...] LLM applications are especially good at content generation. They can be used to automatically create texts for various purposes, including articles, blog posts, marketing copy, video scripts, and social media updates. Moreover, LLM-backed generative AI apps can adapt to different writing styles and tones, making them versatile for generating content that resonates with specific target audiences.', 'score': 0.6194264975490197}, {'title': 'How to use LLM such as Gemini and ChatGPT for video translation?', 'url': 'https://ghostcut.medium.com/how-to-use-llm-such-as-gemini-and-chatgpt-for-video-translation-e22ff076e885', 'content': 'This article will delve into the key steps of video translation, including audio recognition, text translation, text-to-speech (TTS), and video synthesis.', 'score': 0.5668633363398693}, {'title': 'Exploring Human-Like Translation Strategy with Large Language ...', 'url': 'https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00642/119992/Exploring-Human-Like-Translation-Strategy-with', 'content': 'used to generate summaries. Further innovations emerge in multimedia contexts. Zhu et al. (2023a) and Chen et al. (2023) empower LLMs to pose inquiries regarding provided images and videos to enrich the caption. Remarkably, MAPS extends the domain of self-prompting into machine translation for the first time. [...] 2.1\u2003Knowledge Mining\nAkin to the initial understanding and interpretation phase that human translators take (Gile, 2009), the knowledge mining step requires the LLM first to analyze the source text and elicit three aspects of knowledge generally beneficial to translation: [...] Large language models (LLMs) have demonstrated impressive capabilities in general scenarios, exhibiting a level of aptitude that approaches, in some aspects even surpasses, human-level intelligence. Among their numerous skills, the translation abilities of LLMs have received considerable attention. Compared to typical machine translation that focuses solely on source-to-target mapping, LLM-based translation can potentially mimic the human translation process, which might take preparatory steps', 'score': 0.45874644204793036}, {'title': 'Adapting translations with specialized LLMs - YouTube', 'url': 'https://www.youtube.com/watch?v=SbitjczIh-w', 'content': '... translation solutions help you translate documents, websites, apps, audio files, videos, and more at scale with best-in-class quality and', 'score': 0.45276585267973857}], {'query': 'LLM applications in interpretation and video translation', 'follow_up_questions': None, 'answer': None, 'images': [], 'results': [{'url': 'https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d', 'title': 'The Role of Large Language Models in Machine Translation - Medium', 'content': 'LLMs can translate massive amounts of text quickly and consistently, making them ideal for real-time applications like chatbots or live video', 'score': 0.69845337, 'raw_content': None}, {'url': 'https://pixelplex.io/blog/llm-applications/', 'title': 'Top 10 Real-Life Applications of Large Language Models - PixelPlex', 'content': 'Large language models (LLMs) are advanced artificial intelligence systems designed to understand, interpret, and generate human-like text. They are a type of deep learning model, to be exact, that have been trained on a wide variety of internet texts. LLM applications can perform numerous tasks including writing essays, creating poetry, coding, and even engaging in general conversation. [...] LLM applications are increasingly being used in education to personalize learning and provide tutoring.\nLLMs can adapt to individual student’s learning styles and pace, offering customized explanations and feedback. For instance, a model can generate interactive reading materials that adjust complexity based on the student’s comprehension level or provide real-time language translation to aid foreign students. [...] LLM applications are especially good at content generation. They can be used to automatically create texts for various purposes, including articles, blog posts, marketing copy, video scripts, and social media updates. Moreover, LLM-backed generative AI apps can adapt to different writing styles and tones, making them versatile for generating content that resonates with specific target audiences.', 'score': 0.6194264975490197, 'raw_content': None}, {'url': 'https://ghostcut.medium.com/how-to-use-llm-such-as-gemini-and-chatgpt-for-video-translation-e22ff076e885', 'title': 'How to use LLM such as Gemini and ChatGPT for video translation?', 'content': 'This article will delve into the key steps of video translation, including audio recognition, text translation, text-to-speech (TTS), and video synthesis.', 'score': 0.5668633363398693, 'raw_content': None}, {'url': 'https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00642/119992/Exploring-Human-Like-Translation-Strategy-with', 'title': 'Exploring Human-Like Translation Strategy with Large Language ...', 'content': 'used to generate summaries. Further innovations emerge in multimedia contexts. Zhu et al. (2023a) and Chen et al. (2023) empower LLMs to pose inquiries regarding provided images and videos to enrich the caption. Remarkably, MAPS extends the domain of self-prompting into machine translation for the first time. [...] 2.1\u2003Knowledge Mining\nAkin to the initial understanding and interpretation phase that human translators take (Gile, 2009), the knowledge mining step requires the LLM first to analyze the source text and elicit three aspects of knowledge generally beneficial to translation: [...] Large language models (LLMs) have demonstrated impressive capabilities in general scenarios, exhibiting a level of aptitude that approaches, in some aspects even surpasses, human-level intelligence. Among their numerous skills, the translation abilities of LLMs have received considerable attention. Compared to typical machine translation that focuses solely on source-to-target mapping, LLM-based translation can potentially mimic the human translation process, which might take preparatory steps', 'score': 0.45874644204793036, 'raw_content': None}, {'url': 'https://www.youtube.com/watch?v=SbitjczIh-w', 'title': 'Adapting translations with specialized LLMs - YouTube', 'content': '... translation solutions help you translate documents, websites, apps, audio files, videos, and more at scale with best-in-class quality and', 'score': 0.45276585267973857, 'raw_content': None}], 'response_time': 1.41})
DEBUG [src.tools.decorators] Tool TavilySearchResults returned: ([{'title': 'The impact of LLMs on translation: a new impetus for evolving ...', 'url': 'https://www.rws.com/blog/the-evolution-of-translation/', 'content': 'In the translation industry, initial experiments with LLMs cover a wide spectrum of use cases, from generating domain-specific content, through', 'score': 0.7526570693181819}, {'title': 'Use AI and large language models for translation - Globalization', 'url': 'https://learn.microsoft.com/en-us/globalization/localization/ai/ai-and-llms-for-translation', 'content': "Nontheless, advances in large language models (LLMs) are enabling new paradigms for natural language processing tasks, which include translation. LLMs have the potential to outperform NMT, while enabling natural language processing features in multilingual applications.\nUsing large language models in your translation workflow [...] These technologies have transformed the translation and localization industries. They’ve delivered increased productivity, reduced costs, improved consistency and scalability, and the ability to easily handle domain-specific terminology. [...] Unlike NMT, large language models (LLMs) weren't designed for translation. However, as LLMs are designed to excel at language processing tasks, they often perform well at translation, especially between high-resource language pairs.\nThere are similarities between NMT and LLM:", 'score': 0.7001014446969697}, {'title': 'The Role of Large Language Models in Machine Translation - Medium', 'url': 'https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d', 'content': 'In this article, we will explore the impact of LLMs on machine translation, their capabilities, challenges, and the future of this field.', 'score': 0.6995003232954546}, {'title': 'The Accessibility Impact of Large Learning Models in Translation ...', 'url': 'https://www.truelanguage.com/the-accessibility-impact-of-large-learning-models-in-translation-services/', 'content': 'One of the strengths of both Large Learning Models and Large Language Models lies in their ability to handle multiple languages. This comprehensive multilingual', 'score': 0.6892552704545455}, {'title': 'How Are LLMs Shaping the Future of Translation Technology? -', 'url': 'https://vistatec.com/how-are-llms-shaping-the-future-of-translation-technology/', 'content': 'Large Language Models (LLMs) have the potential to enhance translation and localization technology significantly. Learn how.', 'score': 0.6741185488636364}], {'query': 'impact of Large Language Models on translation industry', 'follow_up_questions': None, 'answer': None, 'images': [], 'results': [{'url': 'https://www.rws.com/blog/the-evolution-of-translation/', 'title': 'The impact of LLMs on translation: a new impetus for evolving ...', 'content': 'In the translation industry, initial experiments with LLMs cover a wide spectrum of use cases, from generating domain-specific content, through', 'score': 0.7526570693181819, 'raw_content': None}, {'url': 'https://learn.microsoft.com/en-us/globalization/localization/ai/ai-and-llms-for-translation', 'title': 'Use AI and large language models for translation - Globalization', 'content': "Nontheless, advances in large language models (LLMs) are enabling new paradigms for natural language processing tasks, which include translation. LLMs have the potential to outperform NMT, while enabling natural language processing features in multilingual applications.\nUsing large language models in your translation workflow [...] These technologies have transformed the translation and localization industries. They’ve delivered increased productivity, reduced costs, improved consistency and scalability, and the ability to easily handle domain-specific terminology. [...] Unlike NMT, large language models (LLMs) weren't designed for translation. However, as LLMs are designed to excel at language processing tasks, they often perform well at translation, especially between high-resource language pairs.\nThere are similarities between NMT and LLM:", 'score': 0.7001014446969697, 'raw_content': None}, {'url': 'https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d', 'title': 'The Role of Large Language Models in Machine Translation - Medium', 'content': 'In this article, we will explore the impact of LLMs on machine translation, their capabilities, challenges, and the future of this field.', 'score': 0.6995003232954546, 'raw_content': None}, {'url': 'https://www.truelanguage.com/the-accessibility-impact-of-large-learning-models-in-translation-services/', 'title': 'The Accessibility Impact of Large Learning Models in Translation ...', 'content': 'One of the strengths of both Large Learning Models and Large Language Models lies in their ability to handle multiple languages. This comprehensive multilingual', 'score': 0.6892552704545455, 'raw_content': None}, {'url': 'https://vistatec.com/how-are-llms-shaping-the-future-of-translation-technology/', 'title': 'How Are LLMs Shaping the Future of Translation Technology? -', 'content': 'Large Language Models (LLMs) have the potential to enhance translation and localization technology significantly. Learn how.', 'score': 0.6741185488636364, 'raw_content': None}], 'response_time': 2.44})
DEBUG [src.tools.decorators] Tool crawl_tool called with parameters: url=https://www.rws.com/blog/the-evolution-of-translation/
DEBUG [src.tools.decorators] Tool crawl_tool called with parameters: url=https://learn.microsoft.com/en-us/globalization/localization/ai/ai-and-llms-for-translation
DEBUG [src.tools.decorators] Tool crawl_tool called with parameters: url=https://medium.com/@davidfagb/the-role-of-large-language-models-in-machine-translation-5e1f6eeeb44d
DEBUG [src.tools.decorators] Tool crawl_tool called with parameters: url=https://www.truelanguage.com/the-accessibility-impact-of-large-learning-models-in-translation-services/
DEBUG [src.tools.decorators] Tool crawl_tool called with parameters: url=https://vistatec.com/how-are-llms-shaping-the-future-of-translation-technology/
WARNING [src.crawler.jina_client] Jina API key is not set. Provide your own key to access a higher rate limit. See https://jina.ai/reader for more information.
WARNING [src.crawler.jina_client] Jina API key is not set. Provide your own key to access a higher rate limit. See https://jina.ai/reader for more information.
WARNING [src.crawler.jina_client] Jina API key is not set. Provide your own key to access a higher rate limit. See https://jina.ai/reader for more information.
WARNING [src.crawler.jina_client] Jina API key is not set. Provide your own key to access a higher rate limit. See https://jina.ai/reader for more information.
WARNING [src.crawler.jina_client] Jina API key is not set. Provide your own key to access a higher rate limit. See https://jina.ai/reader for more information.
DEBUG [src.tools.decorators] Tool crawl_tool returned: {'role': 'user', 'content': [{'type': 'text', 'text': '# The impact of LLMs on translation: a new impetus for evolving machine translation\n\nThe approach we have taken with Evolve has a few interesting advantages:\n\n* All translation work is conducted using the dedicated enterprise-grade NMT models which are optimized for high-quality and high-scale applications, while offering reasonable compute requirements and low total costs of ownership. This technology has been successfully used by large user communities and is deployed across hundreds of commercial and public sector clients.\n* The quality estimation models have been calibrated using human-labeled examples, using our in-house expert linguistic teams. This allows us to tune the performance of the model and extend coverage to new languages as needed.\n* The automated post-editing service utilizes a dedicated, smaller LLM hosted by RWS. This allows us to tune the LLM performance, provide the highest levels of data security, and operate within a predictable cost structure. It is also invulnerable to any 3rd-party API instabilities.\n* Building the solution from three separate modules - translation, quality estimation, and post-editing, allows us to tweak not only the individual components but also how they work together. For example, Language Weaver can now iterate the evaluate/edit task loop several times until a desired outcome is achieved. When an edit task is completed, the translation is sent back for quality estimation – if the result is still found inadequate, the sentence is propagated to the post-editing module again. This time, however, the system captures additional context from the source document and uses it to generate a better translation. (So far, our tests have shown that allowing up to three iterations provides the best compromise between quality, speed, and cost for most types of content).\n* Evolve can be used in all use cases where traditional MT is used because it does not change the ways in which translations are consumed by external systems and workflows. Crucially, in the localization use case, where some degree of human intervention may still be required (or mandated, as it’s the case for a lot of regulatory content), Evolve can seamlessly integrate into current workflows to alleviate the post-editing burden presently shouldered by human linguists.\n* Finally, since Language Weaver keeps track of all the automated edits and estimation results, the by-product of the translate/evaluate/edit sequence is a fantastic source of feedback for the translation engine. The auto-adaptable language pairs monitor the incoming edits and automatically update their models to reflect the observed improvements.\n\n**You too can evolve with us**\n------------------------------\n\nOptimizing the task of post-editing is a major opportunity for all involved in the translation process, from enterprise customers, through language service providers, to individual linguists. Using a combination of auto-adaptive MT and LLMs to minimize manual post-editing effort allows limited resources to be prioritized for high value-added activities. It also increases the usefulness of automated translation in use cases with minimal room for human intervention – or where time-to-market or time-to-insight is the primary driver - for example the high-volume use cases related to legal eDiscovery, regulatory compliance, or digital forensics. For localization processes, the solution helps improve the ROI through significant productivity gains. And for organizations that want to benefit from adaptable MT models but cannot because they don’t have enough previously translated material – Language Weaver Evolve is a great option to jump-start their translation process and initiate a virtuous improvement cycle.'}, {'type': 'image_url', 'image_url': {'url': 'https://www.rws.com/media/images/Bart_tcm228-213657.jpg?v=20250313130328'}}, {'type': 'text', 'text': 'Author\n\nBart Maczynski\n--------------\n\nVP of Machine Learning, Solutions Consulting\n\nBart is VP of Machine Learning at RWS.\n\n[All from Bart Maczynski](/)'}]}
ERROR [src.tools.crawl] Failed to crawl. Error: TypeError('Incoming markup is of an invalid type: None. Markup must be a string, a bytestring, or an open filehandle.')
DEBUG [src.tools.decorators] Tool crawl_tool returned: Failed to crawl. Error: TypeError('Incoming markup is of an invalid type: None. Markup must be a string, a bytestring, or an open filehandle.')
DEBUG [src.tools.decorators] Tool crawl_tool returned: {'role': 'user', 'content': [{'type': 'text', 'text': "# Use AI and large language models for translation - Globalization\n\n[Skip to main content](#main)\n\nThis browser is no longer supported.\n\nUpgrade to Microsoft Edge to take advantage of the latest features, security updates, and technical support.\n\n\n\n\n[Learn](/en-us/)\n\n* + [Documentation](/en-us/docs/)\n\n In-depth articles on Microsoft developer tools and technologies\n + [Training](/en-us/training/)\n\n Personalized learning paths and courses\n + [Credentials](/en-us/credentials/)\n\n Globally recognized, industry-endorsed credentials\n + [Q&A](/en-us/answers/)\n\n Technical questions and answers moderated by Microsoft\n + [Code Samples](/en-us/samples/)\n\n Code sample library for Microsoft developer tools and technologies\n + [Assessments](/en-us/assessments/)\n\n Interactive, curated guidance and recommendations\n + [Shows](/en-us/shows/)\n\n Thousands of hours of original programming from Microsoft experts\n\n Microsoft Learn for Organizations\n\n [Boost your team's technical skills](/en-us/training/organizations/)\n\n Access curated resources to upskill your team and close skills gaps.\n\n\n[Sign in](#)\n\nUsing artificial intelligence and large language models for translation\n-----------------------------------------------------------------------\n\n* Article\n* 08/16/2024\n\nIn this article\n---------------\n\n1. [Large language models in software and globalization](#large-language-models-in-software-and-globalization)\n2. [Artificial intelligence and translation technology](#artificial-intelligence-and-translation-technology)\n3. [Using LLMs for localization tasks other than translation](#using-llms-for-localization-tasks-other-than-translation)\n\nWith recent advances in large language models (LLMs), there's much discussion around whether to use AI instead of existing machine translation (MT) systems or even as a replacement for human translation (HT). The latest LLMs are performing well, getting close to HT-level quality, especially for certain language pairs. However, LLM-based solutions have some drawbacks, which include:\n\n* LLMs might not perform as well as existing technologies, such as neural machine translation (NMT), especially for fields with specialized terminology such as healthcare.\n* LLMs take longer and are more expensive to train than NMT\n* LLMs are slower and require more processing power than NMT\n\nLLMs are evolving rapidly, costs are decreasing, and speed is increasing year over year, so many of the current concerns might be less relevant in the future.\n\nGenerative AI is a type of artificial intelligence focused on the ability of computers to use models to create content like text, synthetic data, and images. Generative AI applications are built on top of generative AI models such as large language models (LLMs).\n\nLLMs are deep learning models that consume and train on massive datasets, allowing them to excel in language processing tasks such as translation. After these models have completed their learning processes, they generate statistically probable outputs when prompted. The models create new combinations of text that mimic natural language based on their training data.\n\nThe development of LLMs has been a gradual process. The first LLMs were relatively small and could only perform simple language tasks. However, with the advances in deep neural networks, larger and more powerful LLMs were created. The 2020 release of the Generative Pre-trained Transformer 3 (GPT-3) model marked a significant milestone in the development of LLMs. GPT-3 demonstrated the ability to generate coherent and convincing text that was difficult to distinguish from text written by humans.\n\nGPT-3, and subsequent models, have been trained on datasets in multiple languages; therefore, these models are able to generate output in multiple languages. However, the quality of the output in each language is related to the amount of training data in that language. Languages where the LLMs were trained with a large set of data are considered *high-resource* languages. Languages that were trained with smaller sets of data are considered *low-resource* languages.\n\nAI and LLMs have the potential to be transformative technologies for globalization. While LLMs weren’t trained specifically for translation, their broad applicability to natural language tasks means that they perform well for translation, especially for high-resource languages. In addition, LLM features in a product often perform well for languages other than the original product language.\n\nMachine translation (MT) systems are applications or online services that use technology to translate text between any of their supported languages. Although the concepts behind machine translation technology and the interfaces to use it are relatively simple, the science behind it is complex and it brings together several leading-edge technologies. There has been an evolution in approaches to machine translation, including:\n\n* Rules based machine translation: machine translation based on dictionaries and grammar rules of each language\n* Statistical machine translation: machine translation based on statistical analysis of bilingual text corpora\n* Neural machine translation (NMT): NMT also uses statistical analysis to predict the likelihood of word sequences. It relies on neural networks to model entire sentences.\n\nThese technologies have transformed the translation and localization industries. They’ve delivered increased productivity, reduced costs, improved consistency and scalability, and the ability to easily handle domain-specific terminology.\n\nNontheless, advances in large language models (LLMs) are enabling new paradigms for natural language processing tasks, which include translation. LLMs have the potential to outperform NMT, while enabling [natural language processing features](localizing-ai-based-features) in multilingual applications.\n\nAdopting AI for translation is a forward-thinking approach that aligns with the latest advancements in technology. It’s essential to transition to this new process thoughtfully and incrementally, ensuring that it meets established benchmarks for each language before full implementation.\n\nWhen evaluating the case for shifting to AI-based translation, it’s crucial to consider various factors such as:\n\n* risk management\n* ensuring high-quality outputs\n* the total cost of ownership\n* the system’s performance\n* the impact on people and processes\n\nThe transition to AI should be a step-by-step process, tailored to the specifics of each product, content type, market, language, and customer expectations. This approach allows for a balanced and justified move towards AI, especially in cases where the return on investment might be minimal.\n\nIn terms of risk, AI-based translation carries a new set of challenges that require thorough human evaluation. Ensuring [responsible AI](about:blank/ai-and-localization#what-is-responsible-ai) usage is paramount, particularly for sensitive applications, to maintain the integrity of the brand and manage potential reputational risks. Special attention should be paid to new or updated terminology, and frequent spot-check validation of the LLM updates, as newer versions of the models might introduce degradation for some languages.\n\nQuality control is variable across different languages. While AI-based translation has exceeded or matched the quality of traditional methods in some languages, it still poses significant challenges in others. The focus of the quality reviews should include two factors: linguistic quality, and adequacy. Ensure that the text is appropriately written following the required linguistic quality required by your products and is an adequate translation for the source. The latter is specially important since, as opposed to MT, LLMs can introduce *fabrications* or *hallucinations*. Fabrications are words or phrases that aren't present in the source text but are generated by the model. The fabricated text might be factually correct, but it can also be incorrect or misleading, even when the text seems plausible.\n\nCost-wise, some of the latest AI models are slightly more cost-effective than their predecessors. However, the total cost of ownership, which includes both the operational and personnel costs, must be taken into account.\n\nMany of the current state of the art-translation applications, such as [Microsoft Translator](https://www.microsoft.com/translator/business/), are based on neural machine translation (NMT). NMT is an improvement on previous statistical machine translation (SMT)-based approaches as it uses far more *dimensions* to represent the tokens (such as words, morphemes, and punctuation) of the source and target text.\n\nUnlike NMT, large language models (LLMs) weren't designed for translation. However, as LLMs are designed to excel at language processing tasks, they often perform well at translation, especially between high-resource language pairs.\n\nThere are similarities between NMT and LLM:\n\n* Both are pretrained using bilingual (or multi-lingual) corpora\n* Both can be trained, or [fine-tuned](/en-us/ai/playbook/technology-guidance/generative-ai/working-with-llms/fine-tuning), to perform better for specific tasks\n\nHowever, there are also differences that means that NMT or LLMs might be the most appropriate technology, depending on the task:\n\n* It’s easier and cheaper to fine-tune NMT for specific domains, such as healthcare.\n* LLMs, in general, produce more natural-sounding text, while NMT produces more accurate text.\n* NMT typically processes segment by segment, while LLMs can work on entire documents at once. So, LLMs perform better with explicit context.\n* It can be easier to integrate existing glossaries and term bases with NMT than LLMs.\n* NMT performs faster than LLMs; however, newer LLMs perform better than previous LLMs. Speed might be a significant concern for processing large volumes of text.\n* Processing translations using LLMs is more expensive than NMT. This is especially true for low-resource languages.\n* NMT can be optimized for language variants. LLMs might have trouble differentiating between and producing text for language variants such as Portuguese for Portugal and Brazilian Portuguese.\n* NMT is optimized specifically for translation while LLMs can be used for various language processing tasks. For example, an LLM could be used to create a business email in Japanese.\n\nDue to their wide applicability for language processing tasks, consider using LLMs for other tasks in your localization workflow. For example,\n\n* LLMs might be suitable for linguistic review of human-translated or machine-translated text.\n* LLMs can be used to generate test data in multiple languages.\n* LLMs might produce better output than other machine translation methods for responses to technical support requests if your team can't support a language natively.\n\nAs you consider using LLMs for localization tasks, remember that LLMs need to be used [responsibly](about:blank/ai-and-localization#what-is-responsible-ai). Ensure that you're validating the output of the LLMs and that you're using diverse and representative datasets to train the LLMs."}]}
DEBUG [src.tools.decorators] Tool crawl_tool returned: {'role': 'user', 'content': [{'type': 'text', 'text': '# How Are LLMs Shaping the Future of Translation Technology?\n\nLarge language models (LLMs) are trending in the translation industry, and it’s easy to see why. Models such as OpenAI’s GPT-4 and Google’s BERT have the potential to generate human-like translations quickly, accurately, and with minimal intervention.\xa0\xa0\n\nCompared to machine translation (MT), [LLMs reduce reliance on bilingual data](https://slator.com/how-large-language-models-fare-against-classic-machine-translation-challenges/) during pre-training for major languages and produce better long-sentence translations. They excel at translating sentences with fewer than 80 words and consistently perform well on documents with about 500 words—a significant hurdle for MT.\xa0\n\nLLMs are still in their infancy and are prone to mistakes such as terminology mismatches, style discrepancies, and hallucinations. As they improve, their use in translation will no doubt increase. The question is: which tasks will they handle, and which ones will be left to humans?\n\nUse Cases in Localization\n-------------------------\n\nTo predict how LLMs will impact translation technology, we must consider their potential uses. Here are four ways the translation industry can integrate these models.\xa0\xa0\n\n### Multilingual customer support\n\nLLM-powered chatbots already respond to routine customer questions in multiple languages. In addition to providing instant answers, they gather information and escalate complex issues to human support.\xa0\n\nChatbots may even deliver more empathetic responses than humans when provided with the appropriate prompts. For example, [ChatGPT scored an unprecedented 100% on standardized empathy tests](https://www.thinkglobalforum.org/tgf-blog/llms-and-the-future-of-translation-technology-key-learnings), outperforming the average human score of 70%.\n\nHowever, linguists continue to play a crucial role in their development as they assist with:\xa0\n\n* **Intent recognition:** Linguists help identify the intent and purpose behind user inputs. They create a taxonomy of intents the chatbot should recognize, allowing the system to provide relevant responses based on user queries.\n\n* **Language generation:** Linguists craft linguistically accurate responses that are contextually appropriate and aligned with the chatbot’s brand or purpose, ensuring the chatbot’s language is natural and reflects the desired tone.\n* **User experience optimization:** Linguists optimize the overall user experience by refining conversation flow. This ensures that chatbot interactions are coherent, engaging, and meet user expectations.\n\n### Content personalization\n\nLLMs can also assist with AI-powered personalization by generating content such as emails, in-app messages, and recommendations based on user interactions, preferences, and historical data. This process allows companies to develop highly targeted, personalized campaigns that increase customer engagement and conversion rates. For example, an online retailer could send special offers based on past purchases and browsing behavior.\xa0\n\nLike chatbots, linguists are involved in key areas of the process, including:\n\n* **Content curation.** Linguists curate and categorize content. Using their linguistic and cultural expertise, they assess the relevance, quality, and appropriateness of different pieces of content.\n\n* **Fine-tuning algorithms.** Linguists work with data scientists and machine learning engineers to fine-tune algorithms. They provide insights into linguistic patterns, sentiment analysis, and other language-related features that can improve content recommendation accuracy and relevance.\n\n* **User feedback analysis.** Linguists analyze user feedback to understand how well personalized content aligns with user expectations. This feedback loop is valuable for continuous improvement in personalization algorithms.\n\n### Terminology consistency and glossary management\n\nLLMs can even help maintain consistent terminology across localized content. Localization often involves industry-specific or brand-specific terms. LLMs can analyze and suggest translations for such terms, helping localization teams adhere to established glossaries.\n\nLinguists help train and improve the model’s understanding of domain-specific terminology through\xa0\n\n* **Glossary creation and maintenance.** Linguists create and maintain glossaries containing industry-specific and brand-specific terms and other specialized vocabulary.\n* **Contextual understanding.** Linguists improve the contextual understanding of terminology usage, which helps models distinguish between multiple meanings and choose the best translation depending on the context.\n* **Fine-tuning for industry-specific terms.** Linguists fine-tune LLMs to understand better and generate content related to specific domains. This process involves training the models with additional data, which ensures more accurate and contextually appropriate translations.\n\n### Human-in-the-loop (HITL) systems\n\nHITL systems are another approach. This setup allows LLMs to provide suggestions or initial translations, which human translators can refine. Combining a human translator with a natural language processing system can overcome LLM limitations and biases and ensure higher quality and more accurate translations.\n\nThe Translation Challenges of LLMs\n----------------------------------\n\nYet, despite the potential of LLM translation technology, its use is largely experimental. These models remain limited due to a variety of challenges. Here are five issues LLMs must address before they become the norm in translation.\xa0\xa0\xa0\n\n### Translation accuracy\n\nWhile LLMs can produce translations, they weren’t designed to do it. English language corpora dominate [training data](https://vistatec.com/the-upstream-nature-of-language-data-for-llms/), which means translations into another language may not be as accurate. This is especially true for [languages with fewer resources](https://vistatec.com/how-to-overcome-the-need-for-data-for-low-resource-languages/), such as languages of lesser diffusion.\xa0\n\nOne possible exception is [CroissantLLM](https://arxiv.org/abs/2402.00786), which claims to be a truly bilingual French-English language model. Building multilingual models and training English-language models on more data could improve accuracy.\n\n### Pre-trained data sets\n\nThe next challenge lies in training. LLMs are trained using fixed data representing knowledge up to a given point in time. And problems can arise from inaccurate or outdated information.\xa0\n\nAlthough pre-trained data sets can provide valuable insights into language nuances, they may not consider a domain’s specific needs. As a result, they’re less relevant for dynamic fields such as technology, finance, and medicine.\xa0\n\nHumans are still better at translating specialized content. They’re better at capturing the nuances of the language and have a deeper understanding of these domains. LLMs need additional training in specific fields to achieve the same accuracy and quality.\xa0\n\n### Bias in training data\n\nHowever, training LLMs on additional data can also cause problems. These models tend to assimilate bias from their training data, resulting in sexist, racist, and ableist tendencies. As they become more powerful, bias increases. For example, a [model with 280 billion parameters showed 29% more toxicity](https://research.aimultiple.com/future-of-large-language-models/) than one with 117 million parameters.\xa0\n\nTo combat this problem, linguists must continually monitor training data to identify and eliminate bias. That includes employing regular data reviews and processes to detect training data bias.\n\n### Capacity limitations\n\nBesides data challenges, LLMs have capacity limitations that prevent them from translating larger, more complex texts.\xa0\n\nModels can process a maximum number of input tokens, which prevents LLMs from comprehending and producing outputs that surpass the token threshold. ChatGPT-3.5, for example, has a 2048-token limit, or approximately 1,500 words, while GPT-4 extends its capacity to 25,000 words. While this is a major improvement, LLMs must continue expanding their capacity before they can be widely used in the translation industry.\xa0\n\n### Prohibitive Costs\n\nLastly, LLMs can be prohibitively expensive, which poses a major challenge when developing software for a broader audience.\n\nModels use large amounts of computing power, making their size challenging to scale. Integrating them into existing translation systems may require significant software architecture changes.\xa0\n\nTheir size can also slow translation. On average, MT models require 0.3 seconds, while LLMs require 30 seconds. This delay raises concerns about real-time deployments. So, until LLMs can match the speed of MTs, they may not be a viable alternative for certain projects.\xa0\n\nWhat’s Next for Large Language Models?\n--------------------------------------\n\nFortunately, researchers are already exploring ways to solve common LLM issues. Here are a few ways they’re tackling inaccuracies, limited data sets, and inefficiency.\xa0\n\n### Fact-Checking\n\nWe noted that pre-trained data sets limit an LLM’s ability to provide accurate, up-to-date information. To overcome this challenge, models need access to external sources for reference. Google’s [REALM](https://blog.research.google/2020/08/realm-integrating-retrieval-into.html) and Facebook’s [RAG](https://ai.meta.com/blog/retrieval-augmented-generation-streamlining-the-creation-of-intelligent-natural-language-processing-models/) are two examples that use citations and references, similar to human researchers. \n \nOpenAI’s [WebGPT](https://openai.com/research/webgpt) is another promising option. A fine-tuned version of its GPT model, WebGPT uses Microsoft Bing to generate precise answers. Research shows that WebGPT outperforms GPT-3 in accuracy and informativeness.\n\n### Synthetic training data\n\nNext up is training data. Research is underway to develop large-scale language models to solve the problem of limited training data. In fact, a [Google study](https://arxiv.org/pdf/2210.11610.pdf) created a model capable of\n\n* Creating questions,\n* Generating comprehensive answers,\n* Filtering responses,\n* Fine-tuning itself.\n\nThe result? Superior performance across multiple language tasks. By optimizing themselves, models can reduce biases and toxicity of their outputs and fine-tune their performance with desired data sets.\xa0\n\n### Sparse expertise models\n\nLastly, sparse expertise models offer an efficient alternative to dense models that slow performance.\xa0\n\nSeveral language models use densely activated transformers, including OpenAI’s GPT-3, NVIDIA/Microsoft’s Megatron-Turing, and Google’s BERT. Dense models use all their parameters, making them less effective and user-friendly.\n\nIn sparse expertise models, such as Google GLaM. Despite its seven-fold larger size, [GLaM consumes two-thirds less energy for training and interference](https://www.forbes.com/sites/robtoews/2023/02/07/the-next-generation-of-large-language-models/), and it outperforms GPT-3 on natural language tasks. Future language models can benefit from this approach because it is more efficient and environmentally friendly.\n\nLLMs and the Future of Translation\n----------------------------------\n\nIn short, LLMs have the potential to significantly enhance translation and localization technology. However, they’ll likely require ongoing refinement to improve accuracy and efficiency. Linguists, translators, and other localization experts will continue to contribute to translation. LLMs are valuable tools, but they won’t replace human expertise.\xa0\n\nAt Vistatec, we combine cutting-edge technology with human expertise to provide the highest quality translations. [Contact us](https://vistatec.com/contact/) today to learn more.'}]}
DEBUG [src.tools.decorators] Tool crawl_tool returned: {'role': 'user', 'content': [{'type': 'text', 'text': '# The Role of Large Language Models in Machine Translation\n\n['}, {'type': 'image_url', 'image_url': {'url': 'https://miro.medium.com/v2/resize:fill:44:44/1*tQTjIfY643Gzl6ovvctb4A.jpeg'}}, {'type': 'text', 'text': '](/@davidfagb?source=post_page---byline--5e1f6eeeb44d---------------------------------------)\n\n7 min read\n\nApr 5, 2024\n\nIn today’s interconnected world, language barriers pose challenges for [communication](https://www.technologyreview.com/2021/05/20/1025135/ai-large-language-models-bigscience-project/), business, and global cooperation. Machine translation, the automatic conversion of text or speech from one language to another, has made great strides in breaking down these barriers. Over the years, large language models (LLMs) have played a pivotal role in advancing machine translation technology, improving translation quality, speed, and versatility.\n\nIn this article, we will explore the impact of [LLMs on machine translation](https://shafquatarefeen.com/gpt/), their capabilities, challenges, and the future of this field.'}, {'type': 'image_url', 'image_url': {'url': 'https://miro.medium.com/v2/resize:fit:600/1*FCua1Nf_N8uYTohzz7rVuw.png'}}, {'type': 'text', 'text': 'Overview of Machine Translation\n-------------------------------\n\n[Machine translation](https://www.nytimes.com/2023/05/30/science/ai-chatbots-language-learning-models.html) (MT) is the field of artificial intelligence concerned with automatically translating text or speech from one language to another. It aims to eliminate the need for human translators by using software and algorithms to bridge the linguistic gap.\n\n**Key characteristics of MT:**\n\nAutomatic: No human intervention is involved in the core translation process. \nText or Speech: Both written and spoken language can be translated. \nMultiple languages: Can translate between various language pairs, not just one-to-one. \nAdaptive: continuously learns and improves with new data and user feedback.\n\nLarge Language Models (LLMs)\n----------------------------\n\nLLMs are [revolutionizing](https://www.weforum.org/agenda/2023/07/harnessing-the-power-of-large-language-models-for-untapped-ai-apps-in-manufacturing/) the field of artificial intelligence with their ability to understand and generate human-like language. Let’s delve into their fascinating world:\n\n**Definition and characteristics of LLMs**: \nLLMs are a type of artificial intelligence (AI) that uses deep learning techniques and massive amounts of text data to understand, summarize, generate, and predict language. Think of them as super-powered language models trained on a library of books and the entire internet!\n\n**Key characteristics**:\n\n* **Large-scale**: trained on a massive corpus of text and code, often exceeding hundreds of billions of words.\n* **Deep learning**: Utilize complex neural networks to learn patterns and relationships within language.\n* **Generative**: can produce creative text formats like poems, code, scripts, emails, and even translate languages.\n* **Adaptive**: Continuously learn and improve with new data and interactions.\n\n**Notable examples of LLMs:**\n\n* GPT-3 (Generative Pre-trained Transformer 3): Developed by OpenAI, it’s known for its impressive text generation capabilities and ability to follow instructions and answer questions.\n* BERT (Bidirectional Encoder Representations from Transformers): Created by Google AI, it excels at understanding the context of words and phrases, making it valuable for tasks like sentiment analysis and question answering.\n* Jurassic-1 Jumbo (Jurassic-1): This behemoth from Google AI boasts a staggering 178B parameters, enabling it to perform complex reasoning and answer challenging questions in an informative way.\n\n**Training and Architecture:**\n\n* Training: LLMs are trained on massive datasets of text and code, often scraped from books, articles, websites, and code repositories. These datasets are then fed into complex neural networks that learn to identify patterns and relationships within the language.\n* Architecture: LLMs typically employ transformer-based architectures, which allow the model to analyze the relationships between words across long distances in a sentence, leading to more contextual understanding.\n\n**Capabilities and Applications of LLMs:**\n\n**Capabilities**Below are some of the capabilities of large language models:\n\n* Text generation: create poems, code, scripts, musical pieces, emails, letters, etc.\n* Translation: Translate languages accurately and fluently.\n* Question answering: Answer your questions in an informative way, even if they are open-ended, challenging, or strange.\n* Summarization: Condense large amounts of information into concise summaries.\n* Sentiment analysis: understand the emotional tone of the text.\n\n**Applications:** \nBelow are some of the applications of large language models:\n\n* Creative writing: assisting writers with ideas, generating different writing styles, and overcoming writer’s block.\n* Customer service: providing automated chatbots that understand and respond to customer inquiries.\n* Education: personalizing learning experiences and creating interactive educational content.\n* Code development: writing code, suggesting code completions, and debugging code.\n* Drug discovery: analyzing scientific literature to identify potential drug candidates.\n\nThe Role of LLMs in Machine Translation\n---------------------------------------\n\n**Overview of Traditional** [**Machine Translation**](https://arxiv.org/abs/2306.05969) **Approaches**\n\nMachine translation (MT) has come a long way, but traditional approaches still face limitations. This is where LLMs step in, offering a powerful new paradigm for accurate, nuanced, and even creative translation.\n\n**Overview of Traditional Machine Translation Approaches** \n**Rule-based MT**: Early systems relied on hand-crafted rules to map words and phrases from one language to another. This approach was inflexible and often produced unnatural translations. \n**Statistical MT**: Modern systems leverage statistical models, trained on large parallel corpora (aligned text in both languages), to predict the most likely target word or phrase for a given source text. While this improved accuracy, it still struggled with context, idioms, and complex sentence structures.\n\n**Enhanced Translation Quality with LLMs** \nLLMs bring several exciting [advancements](https://arxiv.org/abs/2107.00676) to the MT game:\n\n**1. Deeper Context Understanding**:\n\nLLMs analyze vast amounts of text, including non-parallel data, to grasp the nuances of language. This allows them to capture context, cultural references, and even humor, leading to more accurate and natural translations. \n**2. Handling Complexities:**\n\nLLMs excel at dealing with ambiguities, idioms, and intricate sentence structures that traditional MT often stumbles upon. They can choose the appropriate synonym or phrasing based on the specific context, producing more faithful and nuanced translations. \n**3. Creativity and Fluency:**\n\nLLMs aren’t just literal word-swappers. They can adjust sentence structure, rephrase for better flow, and even generate creative text formats like poems or scripts in the target language. This opens up new possibilities for artistic and marketing translations. \n**4. Personalized Translations:**\n\nLLMs can be fine-tuned to specific domains or individual preferences, tailoring translations to the intended audience and purpose. This is particularly beneficial for technical documents, marketing copy, or legal contracts. \n**5. Improved Efficiency and Scalability:**\n\nLLMs can translate massive amounts of text quickly and consistently, making them ideal for real-time applications like chatbots or live video translations. This efficiency can significantly lower translation costs and expand accessibility.\n\n**Verified References:**The [Role of Large Language Models](https://www.linkedin.com/pulse/role-llms-multilingual-communication-translation-neurocareai-ggj8f) in Multilingual Communication and Translation \nLarge Language Models and [Machine Translation](https://www.languagewire.com/en/blog/llm-hyper-personalization): The Era of Hyperpersonalization\n\nChallenges and Concerns\n-----------------------\n\nAs large language models (LLMs) continue to evolve, it’s crucial to address the ethical considerations that accompany their development and deployment. These concerns encompass potential biases, fairness, transparency, accountability, and societal impacts.\n\n**1. Bias and Fairness:**\n\n* Training Data Bias: LLMs are trained on massive datasets of text and code, which may reflect existing societal biases. These biases can perpetuate discrimination and marginalize certain groups if not carefully addressed.\n* Algorithmic Bias: The algorithms used to train LLMs may introduce their own biases, leading to unfair or discriminatory outcomes. This highlights the need for rigorous bias testing and mitigation techniques.\n\n**2. Transparency and Accountability:**\n\n* Black Box Nature: LLMs can be complex and opaque, making it difficult to understand how they make decisions. This lack of transparency raises concerns about accountability and potential misuse.\n* Attribution and Responsibility: When LLMs generate text or make decisions, determining who is responsible for the outcome can be challenging. Clear guidelines for attribution and accountability are needed.\n\n**3. Societal Impacts:**\n\n* Job Displacement: Automation through LLMs could lead to job displacement in certain sectors, particularly those involving routine tasks or data processing.\n* Misinformation and disinformation: LLMs can be used to create and spread misinformation or disinformation, posing a threat to informed decision-making and social harmony.\n\n**4. Human Control and Autonomy:**\n\n* Over-reliance on LLMs: Over-reliance on LLMs for decision-making could erode human expertise and judgment, potentially leading to suboptimal outcomes.\n* Preserving Human Autonomy: LLMs should be designed to augment human capabilities, not replace human judgment and decision-making.\n\nFuture Directions\n-----------------\n\nThe field of LLMs for [machine translation (MT)](https://arxiv.org/abs/2306.05969) is rapidly evolving, with researchers and developers exploring innovative techniques to enhance translation accuracy, fluency, and personalization. \nHere are some promising directions for future advancement:\n\n**1. Domain-Specific and Personalized LLMs:** Developing LLMs tailored to specific domains, such as legal, medical, or technical fields, can significantly improve translation accuracy and relevance. Additionally, fine-tuning LLMs based on individual preferences or user styles can personalize translations to better suit the target audience.\n\n**2. Multilingual LLMs:** Creating LLMs that can seamlessly translate between multiple languages, eliminating the need for intermediate language hops, will streamline communication across diverse global communities.\n\n**3.** **Multimodal MT with LLMs:** Incorporating multimedia data, such as images, videos, and audio, into LLMs can provide additional context and enhance translation accuracy for tasks like captioning, video localization, and multilingual product descriptions.\n\n**4. Continual Learning and Adaptation:** Developing LLMs that can continuously learn from new data and adapt to evolving language patterns will ensure their effectiveness in real-time translation applications and dynamic communication scenarios.\n\n**5. Explainable AI for LLMs:** Enhancing explainability in LLMs is crucial for building trust and understanding their decision-making processes. This can be achieved through techniques that provide insights into how LLMs translate different types of text and make choices between alternative translations.\n\nAddressing Ethical and Regulatory Concerns\n------------------------------------------\n\nAs LLMs become increasingly powerful and integrated into society, addressing ethical and regulatory concerns is paramount.\n\n**1. Bias Detection and Mitigation:** Implementing robust bias detection and mitigation techniques is essential to prevent LLMs from perpetuating or amplifying existing societal biases. This includes identifying and addressing biases in training data, algorithms, and evaluation metrics.\n\n**2. Transparency and Accountability:** Promoting transparency by providing clear documentation and explanations of LLM capabilities and limitations is crucial. Establishing accountability frameworks will ensure responsible AI development and deployment.\n\n**3. Public Engagement and Education:** Engaging the public in discussions about LLMs and educating them about potential risks and benefits can foster trust and informed decision-making.\n\n**4. Regulatory Oversight and Guidelines:** Developing clear regulatory frameworks and guidelines for LLM development and deployment will help ensure their alignment with ethical principles and societal values.\n\nConclusion\n----------\n\nLLMs are bringing a new level of sophistication to machine translation, offering more accurate, fluent, and personalized translations. They can handle complex language nuances, translate creatively, and adapt to specific domains or user preferences. However, as LLMs become more powerful, it’s crucial to address ethical concerns such as bias, fairness, and transparency. By working together, humans and AI can harness the power of LLMs to break down language barriers and promote global communication.'}]}
DEBUG [src.tools.decorators] Tool crawl_tool called with parameters: url=https://www.truelanguage.com/the-accessibility-impact-of-large-learning-models-in-translation-services/
WARNING [src.crawler.jina_client] Jina API key is not set. Provide your own key to access a higher rate limit. See https://jina.ai/reader for more information.
DEBUG [src.tools.decorators] Tool crawl_tool called with parameters: url=https://vistatec.com/how-are-llms-shaping-the-future-of-translation-technology/
WARNING [src.crawler.jina_client] Jina API key is not set. Provide your own key to access a higher rate limit. See https://jina.ai/reader for more information.
DEBUG [src.tools.decorators] Tool crawl_tool returned: {'role': 'user', 'content': [{'type': 'text', 'text': '# How Are LLMs Shaping the Future of Translation Technology?\n\nLarge language models (LLMs) are trending in the translation industry, and it’s easy to see why. Models such as OpenAI’s GPT-4 and Google’s BERT have the potential to generate human-like translations quickly, accurately, and with minimal intervention.\xa0\xa0\n\nCompared to machine translation (MT), [LLMs reduce reliance on bilingual data](https://slator.com/how-large-language-models-fare-against-classic-machine-translation-challenges/) during pre-training for major languages and produce better long-sentence translations. They excel at translating sentences with fewer than 80 words and consistently perform well on documents with about 500 words—a significant hurdle for MT.\xa0\n\nLLMs are still in their infancy and are prone to mistakes such as terminology mismatches, style discrepancies, and hallucinations. As they improve, their use in translation will no doubt increase. The question is: which tasks will they handle, and which ones will be left to humans?\n\nUse Cases in Localization\n-------------------------\n\nTo predict how LLMs will impact translation technology, we must consider their potential uses. Here are four ways the translation industry can integrate these models.\xa0\xa0\n\n### Multilingual customer support\n\nLLM-powered chatbots already respond to routine customer questions in multiple languages. In addition to providing instant answers, they gather information and escalate complex issues to human support.\xa0\n\nChatbots may even deliver more empathetic responses than humans when provided with the appropriate prompts. For example, [ChatGPT scored an unprecedented 100% on standardized empathy tests](https://www.thinkglobalforum.org/tgf-blog/llms-and-the-future-of-translation-technology-key-learnings), outperforming the average human score of 70%.\n\nHowever, linguists continue to play a crucial role in their development as they assist with:\xa0\n\n* **Intent recognition:** Linguists help identify the intent and purpose behind user inputs. They create a taxonomy of intents the chatbot should recognize, allowing the system to provide relevant responses based on user queries.\n\n* **Language generation:** Linguists craft linguistically accurate responses that are contextually appropriate and aligned with the chatbot’s brand or purpose, ensuring the chatbot’s language is natural and reflects the desired tone.\n* **User experience optimization:** Linguists optimize the overall user experience by refining conversation flow. This ensures that chatbot interactions are coherent, engaging, and meet user expectations.\n\n### Content personalization\n\nLLMs can also assist with AI-powered personalization by generating content such as emails, in-app messages, and recommendations based on user interactions, preferences, and historical data. This process allows companies to develop highly targeted, personalized campaigns that increase customer engagement and conversion rates. For example, an online retailer could send special offers based on past purchases and browsing behavior.\xa0\n\nLike chatbots, linguists are involved in key areas of the process, including:\n\n* **Content curation.** Linguists curate and categorize content. Using their linguistic and cultural expertise, they assess the relevance, quality, and appropriateness of different pieces of content.\n\n* **Fine-tuning algorithms.** Linguists work with data scientists and machine learning engineers to fine-tune algorithms. They provide insights into linguistic patterns, sentiment analysis, and other language-related features that can improve content recommendation accuracy and relevance.\n\n* **User feedback analysis.** Linguists analyze user feedback to understand how well personalized content aligns with user expectations. This feedback loop is valuable for continuous improvement in personalization algorithms.\n\n### Terminology consistency and glossary management\n\nLLMs can even help maintain consistent terminology across localized content. Localization often involves industry-specific or brand-specific terms. LLMs can analyze and suggest translations for such terms, helping localization teams adhere to established glossaries.\n\nLinguists help train and improve the model’s understanding of domain-specific terminology through\xa0\n\n* **Glossary creation and maintenance.** Linguists create and maintain glossaries containing industry-specific and brand-specific terms and other specialized vocabulary.\n* **Contextual understanding.** Linguists improve the contextual understanding of terminology usage, which helps models distinguish between multiple meanings and choose the best translation depending on the context.\n* **Fine-tuning for industry-specific terms.** Linguists fine-tune LLMs to understand better and generate content related to specific domains. This process involves training the models with additional data, which ensures more accurate and contextually appropriate translations.\n\n### Human-in-the-loop (HITL) systems\n\nHITL systems are another approach. This setup allows LLMs to provide suggestions or initial translations, which human translators can refine. Combining a human translator with a natural language processing system can overcome LLM limitations and biases and ensure higher quality and more accurate translations.\n\nThe Translation Challenges of LLMs\n----------------------------------\n\nYet, despite the potential of LLM translation technology, its use is largely experimental. These models remain limited due to a variety of challenges. Here are five issues LLMs must address before they become the norm in translation.\xa0\xa0\xa0\n\n### Translation accuracy\n\nWhile LLMs can produce translations, they weren’t designed to do it. English language corpora dominate [training data](https://vistatec.com/the-upstream-nature-of-language-data-for-llms/), which means translations into another language may not be as accurate. This is especially true for [languages with fewer resources](https://vistatec.com/how-to-overcome-the-need-for-data-for-low-resource-languages/), such as languages of lesser diffusion.\xa0\n\nOne possible exception is [CroissantLLM](https://arxiv.org/abs/2402.00786), which claims to be a truly bilingual French-English language model. Building multilingual models and training English-language models on more data could improve accuracy.\n\n### Pre-trained data sets\n\nThe next challenge lies in training. LLMs are trained using fixed data representing knowledge up to a given point in time. And problems can arise from inaccurate or outdated information.\xa0\n\nAlthough pre-trained data sets can provide valuable insights into language nuances, they may not consider a domain’s specific needs. As a result, they’re less relevant for dynamic fields such as technology, finance, and medicine.\xa0\n\nHumans are still better at translating specialized content. They’re better at capturing the nuances of the language and have a deeper understanding of these domains. LLMs need additional training in specific fields to achieve the same accuracy and quality.\xa0\n\n### Bias in training data\n\nHowever, training LLMs on additional data can also cause problems. These models tend to assimilate bias from their training data, resulting in sexist, racist, and ableist tendencies. As they become more powerful, bias increases. For example, a [model with 280 billion parameters showed 29% more toxicity](https://research.aimultiple.com/future-of-large-language-models/) than one with 117 million parameters.\xa0\n\nTo combat this problem, linguists must continually monitor training data to identify and eliminate bias. That includes employing regular data reviews and processes to detect training data bias.\n\n### Capacity limitations\n\nBesides data challenges, LLMs have capacity limitations that prevent them from translating larger, more complex texts.\xa0\n\nModels can process a maximum number of input tokens, which prevents LLMs from comprehending and producing outputs that surpass the token threshold. ChatGPT-3.5, for example, has a 2048-token limit, or approximately 1,500 words, while GPT-4 extends its capacity to 25,000 words. While this is a major improvement, LLMs must continue expanding their capacity before they can be widely used in the translation industry.\xa0\n\n### Prohibitive Costs\n\nLastly, LLMs can be prohibitively expensive, which poses a major challenge when developing software for a broader audience.\n\nModels use large amounts of computing power, making their size challenging to scale. Integrating them into existing translation systems may require significant software architecture changes.\xa0\n\nTheir size can also slow translation. On average, MT models require 0.3 seconds, while LLMs require 30 seconds. This delay raises concerns about real-time deployments. So, until LLMs can match the speed of MTs, they may not be a viable alternative for certain projects.\xa0\n\nWhat’s Next for Large Language Models?\n--------------------------------------\n\nFortunately, researchers are already exploring ways to solve common LLM issues. Here are a few ways they’re tackling inaccuracies, limited data sets, and inefficiency.\xa0\n\n### Fact-Checking\n\nWe noted that pre-trained data sets limit an LLM’s ability to provide accurate, up-to-date information. To overcome this challenge, models need access to external sources for reference. Google’s [REALM](https://blog.research.google/2020/08/realm-integrating-retrieval-into.html) and Facebook’s [RAG](https://ai.meta.com/blog/retrieval-augmented-generation-streamlining-the-creation-of-intelligent-natural-language-processing-models/) are two examples that use citations and references, similar to human researchers. \n \nOpenAI’s [WebGPT](https://openai.com/research/webgpt) is another promising option. A fine-tuned version of its GPT model, WebGPT uses Microsoft Bing to generate precise answers. Research shows that WebGPT outperforms GPT-3 in accuracy and informativeness.\n\n### Synthetic training data\n\nNext up is training data. Research is underway to develop large-scale language models to solve the problem of limited training data. In fact, a [Google study](https://arxiv.org/pdf/2210.11610.pdf) created a model capable of\n\n* Creating questions,\n* Generating comprehensive answers,\n* Filtering responses,\n* Fine-tuning itself.\n\nThe result? Superior performance across multiple language tasks. By optimizing themselves, models can reduce biases and toxicity of their outputs and fine-tune their performance with desired data sets.\xa0\n\n### Sparse expertise models\n\nLastly, sparse expertise models offer an efficient alternative to dense models that slow performance.\xa0\n\nSeveral language models use densely activated transformers, including OpenAI’s GPT-3, NVIDIA/Microsoft’s Megatron-Turing, and Google’s BERT. Dense models use all their parameters, making them less effective and user-friendly.\n\nIn sparse expertise models, such as Google GLaM. Despite its seven-fold larger size, [GLaM consumes two-thirds less energy for training and interference](https://www.forbes.com/sites/robtoews/2023/02/07/the-next-generation-of-large-language-models/), and it outperforms GPT-3 on natural language tasks. Future language models can benefit from this approach because it is more efficient and environmentally friendly.\n\nLLMs and the Future of Translation\n----------------------------------\n\nIn short, LLMs have the potential to significantly enhance translation and localization technology. However, they’ll likely require ongoing refinement to improve accuracy and efficiency. Linguists, translators, and other localization experts will continue to contribute to translation. LLMs are valuable tools, but they won’t replace human expertise.\xa0\n\nAt Vistatec, we combine cutting-edge technology with human expertise to provide the highest quality translations. [Contact us](https://vistatec.com/contact/) today to learn more.'}]}
到这里，由于上下文 Token 限制（DeepSeek 的上下文长度限制是 65536 Tokens）挂掉了：
openai.BadRequestError: Error code: 400 - {'error': {'message': "This model's maximum context length is 65536 tokens. However, you requested 89100 tokens (89100 in the messages, 0 in the completion). Please reduce the length of the messages or completion.", 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}
During task with name 'agent' and id 'e5ff4b4c-34a1-9d6c-2d3e-8030baa55579'
During task with name 'researcher' and id 'de03fdf6-2a31-c69e-8555-1f8f27b64865'
回头一看 Token 用量。干没了 5 万多。
